cancers

Systematic Review
Oncologic Applications of Artificial Intelligence and Deep
Learning Methods in CT Spine Imaging—A Systematic Review
Wilson Ong 1, *,† , Aric Lee 1,† , Wei Chuan Tan 1 , Kuan Ting Dominic Fong 1 , Daoyong David Lai 1 , Yi Liang Tan 1 ,
Xi Zhen Low 1,2 , Shuliang Ge 1,2 , Andrew Makmur 1,2 , Shao Jin Ong 1,2 , Yong Han Ting 1,2 , Jiong Hao Tan 3 ,
Naresh Kumar 3 and James Thomas Patrick Decourcy Hallinan 1,2

1 Department of Diagnostic Imaging, National University Hospital, 5 Lower Kent Ridge Rd,
Singapore 119074, Singapore; aric.lee@mohh.com.sg (A.L.); weichuan.tan@mohh.com.sg (W.C.T.);
dominic.fong@mohh.com.sg (K.T.D.F.); david.lai@mohh.com.sg (D.D.L.); yiliang.tan@mohh.com.sg (Y.L.T.);
xi_zhen_low@nuhs.edu.sg (X.Z.L.); shuliang_ge@nuhs.edu.sg (S.G.); andrew_makmur@nuhs.edu.sg (A.M.);
shao_jin_ong@nuhs.edu.sg (S.J.O.); yonghan_ting@nuhs.edu.sg (Y.H.T.);
james_hallinan@nuhs.edu.sg (J.T.P.D.H.)
2 Department of Diagnostic Radiology, Yong Loo Lin School of Medicine, National University of Singapore,
10 Medical Drive, Singapore 117597, Singapore
3 National University Spine Institute, Department of Orthopaedic Surgery, National University Health System,
1E, Lower Kent Ridge Road, Singapore 119228, Singapore; jonathan_jh_tan@nuhs.edu.sg (J.H.T.);
dosksn@nus.edu.sg (N.K.)
* Correspondence: wilson.ong@mohh.com.sg; Tel.: +65-67725207
† These authors contributed equally to this work.

Simple Summary: In recent years, advances in deep learning have transformed the analysis of
medical imaging, especially in spine oncology. Computed Tomography (CT) imaging is crucial for
diagnosing, planning treatment, and monitoring spinal tumors. This review aims to comprehensively
explore the current uses of deep learning tools in CT-based spinal oncology. Additionally, potential
clinical applications of AI designed to address common challenges in this field will also be addressed.

Abstract: In spinal oncology, integrating deep learning with computed tomography (CT) imaging
Citation: Ong, W.; Lee, A.; Tan, W.C.;
has shown promise in enhancing diagnostic accuracy, treatment planning, and patient outcomes.
Fong, K.T.D.; Lai, D.D.; Tan, Y.L.; Low,
This systematic review synthesizes evidence on artificial intelligence (AI) applications in CT imaging
X.Z.; Ge, S.; Makmur, A.; Ong,
S.J.; et al. Oncologic Applications of
for spinal tumors. A PRISMA-guided search identified 33 studies: 12 (36.4%) focused on detecting
Artificial Intelligence and Deep spinal malignancies, 11 (33.3%) on classification, 6 (18.2%) on prognostication, 3 (9.1%) on treatment
Learning Methods in CT Spine planning, and 1 (3.0%) on both detection and classification. Of the classification studies, 7 (21.2%) used
Imaging—A Systematic Review. machine learning to distinguish between benign and malignant lesions, 3 (9.1%) evaluated tumor
Cancers 2024, 16, 2988. https:// stage or grade, and 2 (6.1%) employed radiomics for biomarker classification. Prognostic studies
doi.org/10.3390/cancers16172988 included three (9.1%) that predicted complications such as pathological fractures and three (9.1%)
Academic Editor: Dania Cioni
that predicted treatment outcomes. AI’s potential for improving workflow efficiency, aiding decision-
making, and reducing complications is discussed, along with its limitations in generalizability,
Received: 10 July 2024 interpretability, and clinical integration. Future directions for AI in spinal oncology are also explored.
Revised: 14 August 2024
In conclusion, while AI technologies in CT imaging are promising, further research is necessary to
Accepted: 26 August 2024
validate their clinical effectiveness and optimize their integration into routine practice.
Published: 28 August 2024

Keywords: artificial intelligence; deep learning; machine learning; spinal oncology; computed
tomography imaging; applications
Copyright: © 2024 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and 1. Introduction
conditions of the Creative Commons Spinal malignancy encompasses both primary spine tumors and secondary spine
Attribution (CC BY) license (https:// metastases, with the latter being more prevalent [1,2]. The repercussions of spinal malig-
creativecommons.org/licenses/by/
nancy on quality of life are profound, stemming from complications like pain resulting
4.0/).

Cancers 2024, 16, 2988. https://doi.org/10.3390/cancers16172988 https://www.mdpi.com/journal/cancers

Cancers 2024, 16, 2988 2 of 31

from fractures (mechanical), spinal cord compression, and neurological deficits [3]. More-
over, it can lead to diminished mobility, bone marrow aplasia, and hypercalcemia, which
manifest as symptoms such as constipation, polyuria, polydipsia, fatigue, and potentially
life-threatening conditions like cardiac arrhythmias and acute renal failure [4,5]. Therefore,
early detection, accurate diagnosis, and effective treatment of spinal malignancies, particu-
larly metastases, are imperative to minimize complications and enhance patients’ overall
well-being [6].
Artificial intelligence (AI) has made significant strides in the field of spinal oncology,
offering radiologists valuable assistive tools to enhance their diagnostic capabilities and
streamline workflows. In several applications, the findings in spinal oncology are either
promising or have already surpassed previous benchmarks. For instance, the deep learning
algorithm developed by Matohashi et al. [7] showed comparable sensitivity in detecting
spinal metastases compared to orthopedic and radiology experts, and improved the sensi-
tivity of in-training clinicians. AI has the potential to have a substantial impact on each
step of the imaging value chain [8]. At this early stage in the integration of AI into clinical
radiology, several studies have demonstrated its potential and utility. In recent years, there
has been a growing interest in various AI applications, particularly in automating the
detection and classification of spinal lesions [9]. CT, being a reliable imaging modality for
assessing osseous involvement and the degree of destruction in spine abnormalities [10,11],
has been widely studied for these reasons.
The purpose of this article is to introduce artificial intelligence techniques through
a review of recent research, emphasizing their use at various phases of spinal oncology
CT imaging, including image production and utilization. This narrative review aims to
evaluate the current achievements of artificial intelligence (AI) and its possible applications
to spinal oncology for both experts in the field and non-specialist readers. We hope to
provide an overview of the most clinically pertinent applications of machine learning,
including radiomics and convolutional neural networks (CNNs), in spinal oncology CT
imaging, and to discuss potential future innovations in its use.

2. Technical Aspects
2.1. AI, ML, and DL
While “AI”, “machine learning” (ML), and “deep learning” (DL) are often used in-
terchangeably, they actually represent distinct concepts. “AI” encompasses methods that
enable computers to mimic intelligent human behavior [12,13]. “ML” is a specialized field
within AI, utilizing statistical, mathematical, and computer science tools to enhance ma-
chine performance through experience [14–16]. “DL” further narrows down ML (Figure 1),
focusing on the utilization of deep neural networks to analyze large datasets [17]. This
particular domain is even more specialized compared to conventional machine learning,
and leverages artificial neural networks, which involve several layers, to solve complex
medical imaging challenges [18,19].
The complex layered structure allows the deep learning model or algorithm to learn in-
formation directly from imaging datasets and predict fundamental diagnostic patterns and
features that surpass human capabilities [20,21]. These advanced deep learning techniques
can then be used in accurate image classification (detecting the presence or absence of dis-
ease, as well as classifying disease severity), segmentation (based on individual pixels), and
detection. Unlike traditional machine learning approaches that rely on manually engineered
feature extraction from input images, deep learning methods autonomously learn image
features by directly analyzing input images using multi-layer neural networks [22,23],
which include convolutional neural networks (CNNs). This process transforms input im-
ages into valuable outputs. Deep learning systems not only excel at mapping image features
to desired outputs but also possess the capability to discern intrinsic image details that
often surpass human perception [24]. This advancement has significantly influenced the
emergence of radiomics, a specialized field within medical imaging. Radiomics leverages
various machine learning techniques, including deep learning, to extract and analyze a vast
 Cancers 2024, 16, 2988 3 of 31

array of quantitative features from medical images. By utilizing these extracted features,
radiomics enhances the identification, differentiation, and prognosis of lesions, enabling
more precise differentiation between pathological and histological tumor subtypes. This 3 of 31
Cancers 2024, 16, x FOR PEER REVIEW
capability underscores the pivotal role of deep learning in advancing radiomics, providing a
deeper understanding of complex medical images and improving diagnostic accuracy [25].

Figure1.1.Figure
Figure Figuredepicting
depicting
thethe hierarchical
hierarchical arrangement
arrangement of artificial
of artificial intelligence
intelligence (AI). Machine
(AI). Machine learning learn-
ing (ML),
(ML), a subset
a subset of AI,of AI,toaims
aims to empower
empower computers
computers to learn independently
to learn independently without
without explicit explicit pro-
program-
gramming.
ming. Deep learning
Deep learning (DL), a specialized
(DL), a specialized field
field within withinlearning,
machine machineinvolves
learning,
theinvolves the computa-
computation of
tion of neural networks comprising multiple
neural networks comprising multiple layers. layers.

2.2. Radiomics
The complex layered structure allows the deep learning model or algorithm to learn
Radiomics,
information an emerging
directly field within
from imaging AI and
datasets and machine learning, involves
predict fundamental convert-
diagnostic patterns
ing radiological images containing important tumor-related information
and features that surpass human capabilities [20,21]. These advanced deep learning tech- into quantifiable
data
niques [25].
can These
thendata assistinclinicians
be used accurateinimage
evaluating tumors beyond
classification subjective
(detecting visual inter-
the presence or absence
pretation, offering insights into tumor behavior and pathophysiology, including subtyping
of disease, as well as classifying disease severity), segmentation (based on individual pix-
and grading [26,27]. When combined with clinical and qualitative imaging data, radiomics
els), and detection. Unlike traditional machine learning approaches that rely on manually
enhances medical decision-making processes, aiding disease prediction, prognosis assess-
engineered
ment, feature response
and treatment extraction from input
monitoring images, deep learning methods autonomously
[28,29].
learnThe image features
workflow by directly
for building analyzing
a radiomics modelinput images using
[30] comprises multi-layer
several steps: imageneural
ac- net-
quisition, image processing (including segmentation and region of interest (ROI) selection), trans-
works [22,23], which include convolutional neural networks (CNNs). This process
forms feature
image input images intowithin
extraction valuable outputs.
the ROI, Deep learning
exploratory analysis systems notselection,
and feature only excelandat map-
finally
ping imagemodelfeatures
buildingto and classification
desired outputs (Figure 2). Factors
but also possesssuch as the imaging
the capability modality,
to discern intrinsic
acquisition
image details techniques,
that often software,
surpass segmentation methods, [24].
human perception and algorithm structure impact
This advancement has signifi-
the process and vary accordingly [31,32]. Machine learning
cantly influenced the emergence of radiomics, a specialized field within techniques like random
medicalde-
imaging.
cision forests can then be employed to validate and enhance the classification accuracy
Radiomics leverages various machine learning techniques, including deep learning, to ex-
of predictors [33,34]. Ultimately, these tools can then be applied in the clinical setting to
tract and analyze a vast array of quantitative features from medical images. By utilizing
improve diagnostic accuracy and prognosis prediction [28,35–38].
theseRadiomic
extracted features, consist
techniques radiomics enhances
of two the identification,
main approaches: differentiation,
handcrafted-feature and prog-
and deep
nosis of lesions, enabling more precise differentiation between pathological
learning-based analysis [39]. Handcrafted-feature radiomics [40,41] involves extracting and histolog-
numerical features from segmented regions of interest, categorized into shape, first-order in ad-
ical tumor subtypes. This capability underscores the pivotal role of deep learning
vancing radiomics,
statistics, providing
textural features, a deeper understanding
and higher-order statistical features.of Machine
complexlearning
medicalmodels,
images and
improving
such as a Coxdiagnostic
proportional accuracy
hazards[25].
model, single vector machine (SVM), decision trees, and
regression, are then developed to make clinical predictions based on these features [42,43]
and validated for efficiency and sensitivity. In contrast, deep learning techniques use
2.2. Radiomics
neural network architectures like CNNs to automatically extract important features from
Radiomics, an emerging field within AI and machine learning, involves converting
radiological images without prior descriptions [44–46]. These features undergo further
radiological
processing forimages
analysiscontaining
or are used to important tumor-related
generate machine learninginformation
models, which into
arequantifiable
then
data [25]. These data assist clinicians in evaluating tumors beyond
validated on larger datasets before they are applied in a clinical setting [47,48]. subjective visual inter-
pretation, offering insights into tumor behavior and pathophysiology, including subtyp-
ing and grading [26,27]. When combined with clinical and qualitative imaging data, radi-
omics enhances medical decision-making processes, aiding disease prediction, prognosis
assessment, and treatment response monitoring [28,29].
The workflow for building a radiomics model [30] comprises several steps: image
PEER REVIEW 4 of 31
Cancers 2024, 16, 2988 4 of 31

random decision forestsAlthough

can then beinemployed
still to of
the early stage validate and enhance
development, the adventthe classification
of radiomics has the
potential to revolutionize medical imaging by offering rapid and comprehensive
accuracy of predictors [33,34]. Ultimately, these tools can then be applied in the clinical tumor
characterization [49,50]. Radiomics holds the promise of enhancing traditional clinical
setting to improve diagnostic accuracy and
testing and prognostication, prognosis
making prediction
it a highly anticipated[28,35–38].
development for the future [51].

Figure 2. Diagram illustrating the basic

Figure 2. Diagram framework
illustrating the basicand essential
framework stages of
and essential radiomics,
stages including
of radiomics, im-
including image
age acquisition, imageacquisition,
processing image(including segmentation),
processing (including feature
segmentation), extraction
feature extraction within specified
within specified re-of
regions
gions of interest (ROIs), feature
interest selection,
(ROIs), exploratory
feature selection, analysis,
exploratory and
analysis, andsubsequent modeling.
subsequent modeling.

3. Materials and Methods

Radiomic techniques consist
3.1. Literature of Strategy
Search two main approaches: handcrafted-feature and deep
learning-based analysisWe [39]. Handcrafted-feature
systematically radiomics
searched major electronic [40,41]
databases involves
(PubMed, extracting
MEDLINE, Web of
Science, clinicaltrials.gov) following PRISMA guidelines [52].
numerical features from segmented regions of interest, categorized into shape, first-order The search strategy involved
using a combination of keywords and Medical Subject Headings (MeSH) to ensure com-
statistics, textural features, and higher-order statistical features. Machine learning models,
prehensive coverage. Specifically, we used the following query: (“Artificial intelligence”
such as a Cox proportional
OR “AI” OR hazards model,OR
“deep learning” single vector
“machine machine
learning” (SVM), decision
OR “convolutional trees,
neural network”
and regression, are OR then developed
“neural network”to ORmake clinical
“radiomics”) ANDpredictions
(“spine” ORbased
“spinal”onORthese features
“vertebral”) AND
(“CT” OR “CT imaging”) AND (“malignancy” OR “metastasis” OR “cancer” OR “tumor”
[42,43] and validated for efficiency and sensitivity. In contrast, deep learning techniques
OR “oncology”). This query was applied to the title and abstract fields of the articles to
use neural network ensure
architectures
the retrievallike CNNsmost
of studies to automatically
relevant to our reviewextract
topic.important
Two authorsfeatures
(W.O. and
from radiological images without prior descriptions [44–46]. These features undergo
A.L.) independently reviewed the collected references and selected studies for fur-
detailed
full-text screening. The literature search was conducted
ther processing for analysis or are used to generate machine learning models, which are up to 31 March 2024.

then validated on larger datasets

3.2. Study before
Screening they are
and Selection applied in a clinical setting [47,48].
Criteria
Although still in the In early stagethis
conducting of systematic
development,review,thewe advent
adopted ofan radiomics hasstrategy
extensive search the po-to
tential to revolutionize medical imaging by offering rapid and comprehensive tumor char-no
ensure a comprehensive evaluation of the literature. Our search process involved
acterization [49,50]. specific
Radiomics restrictions, aiming to encompass a wide range of relevant studies. We focused on
holds the promise of enhancing traditional clinical testing
identifying scientific research that utilized radiomic techniques, artificial intelligence (AI),
and prognostication,ormaking
deep learningit a highly
methods anticipated development
within the context of CT imagingfor forthe future
spinal [51].
oncology.
To be included in the review, studies had to meet several criteria. First, the research
3. Materials and Methodsto involve CT imaging analysis, specifically applying radiomics, AI, or deep learn-
needed

3.1. Literature Search Strategy

We systematically searched major electronic databases (PubMed, MEDLINE, Web of
Science, clinicaltrials.gov) following PRISMA guidelines [52]. The search strategy in-
Cancers 2024, 16, 2988 5 of 31

ing techniques to the study of spinal oncology. Second, the studies had to assess the
predictive and diagnostic capabilities of these technologies, with a particular emphasis on
their integration into clinical practice and their potential to enhance diagnostic accuracy or
provide prognostic insights. Third, only studies involving human subjects were considered
to ensure the applicability of findings to real-world clinical settings. Finally, we limited
our review to publications in English to maintain consistency and facilitate a thorough
evaluation of the literature.
We excluded studies that did not align with these criteria. Specifically, case reports,
editorial correspondence (such as letters, commentaries, and opinion pieces), and review
articles were not included, as these do not present original research findings. Additionally,
we excluded publications that focused on non-imaging radiomic techniques or did not
utilize AI for the clinical analysis of CT images.
To ensure a thorough and exhaustive search, we concluded our literature review
by examining the bibliographies of the selected publications. This additional step was
undertaken to identify any relevant studies that might have been missed during the
initial search, thereby enhancing the comprehensiveness of our review on the oncologic
applications of AI and deep learning methods in CT spine imaging.

3.3. Data Extraction and Reporting

All chosen research articles were collected and organized into a spreadsheet using
Microsoft Excel (Microsoft Corporation, Redmond, WA, USA). Information gathered from
each research article included:
• Details of the research article: authors, publication date, journal name;
• Primary clinical application: detection, classification, segmentation, treatment, or
prognosis prediction;
• Study specifics: study type, patient characteristics or imaging modalities, body parts
scanned, and specific bone areas segmented for analysis (e.g., internal or
external datasets);
• Machine learning methodologies utilized: radiomics, artificial neural networks, con-
volutional neural networks, etc.

4. Results
4.1. Search Results
The initial search across major electronic medical databases (Figure 3) identified a
total of 226 relevant articles, which were screened based on the specified criteria. Articles
were excluded if they were published more than 15 years ago, not written in English, were
non-journal articles, or were conference abstracts/articles. Based on these exclusion criteria,
18 publications were initially excluded, leaving 208 articles for further full-text analysis
to determine inclusion. For the remaining 208 articles, full-text reviews were performed,
and a total of 177 publications were excluded as they focused on cancer sites other than
the spine or did not utilize imaging data. An additional two articles were included after
manually reviewing the bibliographies of selected articles. This process resulted in a total
of 33 articles (Figure 1) selected for thorough analysis. Key findings from these studies were
compiled (Table 1) and summarized in this review, with the main tasks of the AI models
categorized into four broad categories: (1) Detection, (2) Classification, (3) Prognosis, and
(4) Treatment Planning [53]. Most studies did not provide sufficient data to construct
2 × 2 contingency tables, precluding a formal meta-analysis.
Our search revealed that 12/33 studies (36.4%) focused on using AI techniques for
detecting spinal malignancy, while 11/33 studies (33.3%) concentrated on classification.
Additionally, 6/33 studies (18.2%) utilized radiomics for prognostication, 3/33 studies
(9.1%) focused on treatment planning, and 1/33 study (3.0%) addressed both detection
and classification. Among the studies focusing on classifying spinal malignancy using
AI, 7/33 studies (21.2%) employed machine learning methods to distinguish between
benign and malignant lesions, while 3/33 studies (9.1%) evaluated the stage or grade
EER REVIEW 6 of 31

Cancers 2024, 16, 2988 6 of 31

total of 33 articles (Figure 1) selected for thorough analysis. Key findings from these stud-
ies were compiled (Table 1) and summarized in this review, with the main tasks of the AI
models categorized of spinal
into fourmalignancy, and 2/33 studies
broad categories: (6.1%) classified
(1) Detection, the presence of certain
(2) Classification, tumor
(3) Prog-
biomarkers using radiomic applications. Of the papers focusing on predicting prognosis,
nosis, and (4) Treatment Planning [53]. Most studies did not provide sufficient data to
3/33 studies (9.1%) dealt with predicting complications such as adverse outcomes or
construct 2 × 2 contingency tables,
pathological precluding
fractures, while 3/33astudies
formal meta-analysis.
(9.1%) aimed to predict treatment outcomes.

Figure 3. PRISMA flowchart

Figure 3.(adapted from PRISMA
PRISMA flowchart group,
(adapted from 2020)
PRISMA outlining
group, the process
2020) outlining of selecting
the process of selecting
pertinent articles for analysis.
pertinent articles for analysis.

Our search revealed that 12/33 studies (36.4%) focused on using AI techniques for
detecting spinal malignancy, while 11/33 studies (33.3%) concentrated on classification.
Additionally, 6/33 studies (18.2%) utilized radiomics for prognostication, 3/33 studies
(9.1%) focused on treatment planning, and 1/33 study (3.0%) addressed both detection and
classification. Among the studies focusing on classifying spinal malignancy using AI, 7/33
Cancers 2024, 16, 2988 7 of 31

4.2. Performance Assessment

It is important to be aware of the various methodologies used to assess the performance
of AI systems [54]. The most common key metrics used to assess the performance of AI
systems include sensitivity, specificity, accuracy, precision, Area Under the Curve (AUC),
Figure of Merit (FOM), F1-score [55,56], and kappa. Each of these metrics provides different
insights into the performance of AI algorithms and their clinical relevance, with values
closer to 1.0 indicating good performance:
• Sensitivity (Recall) and Specificity: sensitivity (recall) measures the proportion of
true positives (spinal tumors correctly identified by the AI system) out of all actual
positives (all spinal tumors present in the CT images) [57]. Specificity, on the other
hand, quantifies the ability of the AI system to correctly identify true negatives (normal
spinal conditions) out of all actual negatives (all non-pathological conditions). These
metrics are essential to assessing how well AI algorithms detect both positive and
negative cases in spinal oncology;
• Accuracy and Precision: accuracy indicates the overall correctness of the AI system’s
predictions, measuring the ratio of correctly predicted cases (both true positives and
true negatives) to the total number of cases evaluated. Precision, meanwhile, focuses
on the AI system’s ability to accurately identify positive cases among all predicted
positive instances, minimizing false positives. These metrics provide a comprehensive
view of the AI algorithm’s reliability and correctness in clinical diagnosis;
• Area Under the Curve (AUC): AUC evaluates the performance of AI models in binary
classification tasks [58,59], such as distinguishing between diseased and healthy spinal
conditions based on CT imaging features. A high AUC value indicates that the AI
model effectively ranks diseased cases higher than healthy ones, demonstrating its
discriminatory power in spinal oncology diagnostics;
• Figure of Merit (FOM) and F1-score [60]: Figure of Merit encompasses a range of
metrics including sensitivity, specificity, accuracy, and precision, tailored to the specific
diagnostic challenges presented by spinal tumors. F1-score, a harmonic mean of
precision and recall, balances the trade-off between these metrics and is particularly
useful in scenarios where there is an imbalance between positive and negative cases
in the dataset [61]. A model with a high F1 score indicates both good precision and
recall, reflecting a robust model;
• The kappa statistic [62], often denoted as κ, is a measure used to assess the level of
agreement between two or more raters or classifiers beyond what would be expected
by chance alone. It is particularly useful in the context of classification tasks, where
it evaluates how well the AI system’s predictions align with the true classifications
compared to random chance [63]. Kappa values range from −1 to 1. A kappa value
of 1 indicates perfect agreement between the raters, 0 indicates agreement no better
than chance, and negative values suggest worse-than-chance agreement. High kappa
values indicate that the AI system’s predictions are consistently aligned with expert
assessments or ground truth, which is vital for the system’s clinical applicability
and trustworthiness.
These metrics provide a framework for assessing the clinical relevance and perfor-
mance of AI algorithms in interpreting CT scans. By quantifying these metrics, clinicians
and researchers can effectively evaluate the diagnostic accuracy, reliability, and potential
impact of these AI-driven approaches in enhancing spinal tumor detection and management.
Cancers 2024, 16, 2988 8 of 31

Table 1. Key characteristics of the selected articles.

Sample Size Performance of

Authors AI Method Publication Year Main Objectives Journal Main Task
(No. of CTs/Patients) AI Model
Using supervised
learning methods to
SPIE Medical
Tatjana W. et al. [64] SVM classifier 2012 detect sclerotic bone Detection 22 Sensitivity: 71.2–87.0%
Imaging
metastases in
CT imaging.
Automated detection of
osteolytic and Sensitivity 83.0%
Hammon M. et al. [65] CADe (RF classifier) 2013 European Radiology Detection 134
osteoblastic spine (lytic), 88.0% (blastic)
metastases on CT.
Automated detection of
CADe (SVM sclerotic metastases in
Burns J. et al. [66] 2013 Radiology Detection 59 Sensitivity 79.0–90.0%
classifier) the thoracolumbar spine
on CT
Using Deep CNN Recent Advances in
methods to detect Computational
Sensitivity: 79.0%;
Roth. H. et al. [67] CNN (DropConnect) 2014 sclerotic spinal Methods and Clinical Detection 59
AUC 0.834
metastases on Applications for
CT imaging. Spine Imaging
Detection and
ResNet-50 with Detect and classify bone
Journal of Classification Sensitivity: 81.0%;
Masoudi S et al. [68] DC-GAN 2020 lesions in CT images of 56
Clinical Oncology (Benign accuracy 89.0%
augmentation prostate cancer patients.
vs. malignant)
Using Deep Learning
Methods to Identify
Fan X et al. [69] AlexNet 2021 Spinal Metas in Lung Hindawi Detection 36 Sensitivity: 66.0–81.4%
Cancer using
CT Images.
Sensitivity: 82.4–89.6%;
Deep learning to
improved radiologist’s
improve radiologists’
2D U-Net, 3D U-Net, sensitivity by 15.3%;
Noguchi S. et al. [70] 2022 performance in spinal European Radiology Detection 732
and ResNet reduced mean
metastases detection
interpretation time by
on CT
83 s (p < 0.05).
Cancers 2024, 16, 2988 9 of 31

Table 1. Cont.

Sample Size Performance of

Authors AI Method Publication Year Main Objectives Journal Main Task
(No. of CTs/Patients) AI Model
FOM from 0.848 to
0.876 (radiologist) and
Detecting vertebral European Journal
Hoshiai S. et al. [71] DL Model 2022 Detection 130 from 0.752 to 0.799
bone metastases on CT of Radiology
(resident), (p < 0.05
for both)
Detecting prostate
Accuracy 85.0%;
cancer bone/spinal Current
Musa A. et al. [72] PyRadiomics 2022 Detection 53 sensitivity 78.0–91.0%;
metastases invisible Medical Imaging
specificity: 88.0–93.0%
in CT
Deep learning to
improve Radiologist’s Sensitivity: 75.0%;
Gilberg L. et al. [73] DLA (U-Net) 2023 detection of spinal Applied Sciences Detection 32 improved radiologist’s
malignancies in sensitivity by 20.8%
CT imaging.
Deep learning to
Sensitivity: 89.4%;
improve radiologists’
Frontiers improved radiologist’s
Huo T. et al. [74] 3D U-Net (DCNN) 2023 performance in lung Detection 126
in Oncology sensitivity by 22.2%
cancer spinal metastases
and accuracy of 26.2%
detection on CT
International Journal Accuracy 87.2%;
AI-aided lytic spinal
YOLOv5m of Computer precision 94.8%; recall:
Koike Y. et al. [75] 2023 bone metastasis Detection 2125
and InceptionV3 Assisted Radiology 74.1%; F1-score 74.1%;
detection on CT scans
and Surgery AUC 0.940
Using Deep Learning
Sensitivity: 75.0–78.0%;
Algorithm to Detect
Motohasi M. et al. [7] U-Net (DeepLabv3+) 2024 Spine Detection 435 precision: 36.0–68.0%;
Spinal Metastases on
F1 score: 0.48–0.72
CT Images.
Segmentation and
Classification AUC 0.780–0.800;
classification of Medical
Chmelik J. et al. [76] CNN 2018 (Benign 31 sensitivity: 71.0–74.0%;
metastatic spinal lesions Image Analysis
vs. malignant) specificity: 82.0–88.0%
in 3D CT data
Cancers 2024, 16, 2988 10 of 31

Table 1. Cont.

Sample Size Performance of

Authors AI Method Publication Year Main Objectives Journal Main Task
(No. of CTs/Patients) AI Model
Differentiate benign and
Classification Sensitivity: 95.0%;
malignant vertebral
Li Y. et al. [77] ResNet50 2021 European Radiology (Benign 433 specificity: 80.0%;
fracture on CT using
vs. malignant) accuracy: 88.0%
deep learning
2D ResNet-50,
Differentiate benign Classification
ResNeXt-50, 3D Accuracy: 79.4–92.2%;
Masoudi S. et al. [78] 2021 versus malignant spinal IEEE Access (Benign 114
ResNet-18, 3D F1-Score: 0.755–0.923
lesions on CT. vs. malignant)
ResNet-50
Deep learning
algorithm for grading kappas (κ: 0.873–0.911);
cord compression Classification AUC: 0.953–0.971;
Hallinan J. et al. [79] R-CNN (ResNet50) 2022 Cancers 444
secondary to spinal (Stage/Grade) sensitivity: 92.6–98.0%;
metastasis/epidural specificity: 94.8–99.8%
disease on CT
Radiomics-based
machine learning Classification
Naseri H. et al. [80] PyRadiomics 2022 models to distinguish Scientific Report (Benign 170 AUC 0.640–0.950
between metastatic and vs. malignant)
healthy bone.
Automated
segmentation of the
Classification Dice similarity
fractured vertebrae on
Park, T. et al. [81] U-Net (CNN) 2022 Nature (Benign 158 coefficient: 0.930–0.940;
CT and using radiomics
vs. malignant) AUC 0.800–0.930
to predict benign versus
malignant.
Using Machine learning AUC: 0.658–0.880
techniques to predict Classification (Pre- sensitivity: 65.7–97.9%;
Wang, Q. et al. [82] Research Portal V1.1 2022 Cancers 107
RANKL expression of dict biomarkers) specificity: 23.3–71.9%;
Spinal GCTB. accuracy: 64.6–80.2%
Cancers 2024, 16, 2988 11 of 31

Table 1. Cont.

Sample Size Performance of

Authors AI Method Publication Year Main Objectives Journal Main Task
(No. of CTs/Patients) AI Model
Clinical and CT-Based
Radiomics based
techniques to predict Frontiers Classification (Pre-
Wang, Q. et al. [83] PyRadiomics 2022 80 AUC 0.790–0.880
p53 and VEGF in Oncology dict biomarkers)
expression in
Spinal GCBT.
Deep learning method
kappa (κ = 0.879);
to diagnose epidural
European Classification sensitivity: 91.8%;
Hallinan J. et al. [84] R-CNN (ResNet50) 2023 spinal cord compression 223
Spine Journal (Stage/Grade) specificity: 92.0%;
using
AUC: 0.919
thoracolumbar CT.
Assess for metastatic
spinal cord compression Almost-perfect
(mainly epidural Frontiers Classification inter-rater agreement
Hallinan J. et al. [85] R-CNN (ResNeXt50) 2023 420
extension) on CT in Oncology (Stage/Grade) (κ = 0.813);
imaging with sensitivity: 94.0%
external validation
Differentiating benign
Classification
and malignant vertebral European Journal AUC: 0.890–0.990;
Duan S. et al. [86] Inception_V3 2023 (Benign 280
compression fracture on of Radiology accuracy: 88.0–99.0%
vs. malignant)
spinal CT imaging.
Radiomic modeling to
Sensitivity: 84.4%;
predict risk of vertebral
Journal Prognosis (Predict- specificity 80.0%, AUC
Gui C. et al. [87] PyRadiomics 2021 compression fracture 74
of Neurosurgery ing complications) 0.844–0.878 of 0.844,
after SBRT for
specificity of 0.800
spinal metastases
Using Radiomics-based
technique to predict
Accuracy: 89.0%;
recurrence in spinal Journal of Prognosis (Predict
Wang, Q. et al. [88] PyRadiomics 2021 62 AUC 0.780
GCBT from Bone Oncology treatment outcome)
(predicting recurrence)
pre-operative
CT imaging.
Cancers 2024, 16, 2988 12 of 31

Table 1. Cont.

Sample Size Performance of

Authors AI Method Publication Year Main Objectives Journal Main Task
(No. of CTs/Patients) AI Model
Using machine learning Body composition
methods to derive body analysis using machine
CNN (Densenet Journal of Prognosis (Predict
Massaad E. et al. [89] 2022 composition analysis to 484 learning help predict
and U-Net) Neurosurgery Spine treatment outcome)
predict complications in risk for
spine tumor surgery. inferior outcomes.
Radiomics-based
prediction of vertebral Accuracy: 78.8–82.9%;
European Prognosis (Predict-
Seou Y. et al. [90] PyRadiomics 2023 compression fracture 85 precision: 60.0–62.5%;
Spine Journal ing complications)
prior to spinal SBRT F1 score: 0.573–0.650
from planning CT.
Automated body
composition analysis
using L3 as reference on Frontiers in Prognosis (Treat- DICE similarity
Delrieu L. et al. [91] U-Net 2024 352
CT scans to predict Nuclear Medicine ment Outcome) coefficient: 0.850–0.940
treatment outcome in
cancer patients.
Detection of sarcopenic
DL detected sarcopenia
obesity and association
patients increased odds
with adverse outcomes Journal of Prognosis (Predict-
Khalid S. et al. [92] DL Model 2024 62 of non-home discharge,
in patients undergoing Neurosurgery Spine ing complications)
readmission, and post-
surgical treatment for
operative mortality.
spinal metastases
Clinical Utility of CNN Physics and Imaging DSC 96.7%
Sebastiaan et al. [93] DeepMedic 2022 for treatment planning in Treatment Planning 782 HD: 3.6 mm.
in spinal metastases. Radiation Oncology Acceptable: 77.0%
Dice-similarity
coefficient: 0.850
Automating Treatment
International Journal (cervical), 0.903
Planning for Spinal
Netherton T. et al. [94] CNN (X-Net) 2022 of Oncology, Biology Treatment Planning 220 (thoracic), 93.7
Radiation Therapy
and Physics (lumbar); AUC: 0.820;
using CT Imaging
end-to-end treatment
planning time <8 min
Cancers 2024, 16, 2988 13 of 31

Table 1. Cont.

Sample Size Performance of

Authors AI Method Publication Year Main Objectives Journal Main Task
(No. of CTs/Patients) AI Model
Dice similarity
Automating Treatment
coefficient: 0.650–0.980;
Planning for Paediatric Paediatric
Hernandez S. et al. [95] nn-UNet 2023 Treatment Planning 143 end-to-end treatment
Craniospinal Radiation Blood Cancer
planning time:
Therapy using CT.
3.5 ± 0.4 min
Single Vector Machine (SVM), Convolutional Neural Network (CNN), Deep Learning Algorithm (DLA), Random Forest (RF), Computer-Aided Detection (CADe), Deep Learning (DL),
Dice Similarity Coefficient (DSC), Hausdorff Distance (HD), Figure of Merit (FOM), Area Under Curve (AUC), Receiver Operating Characteristic (ROC), Giant Cell Bone Tumour (GCBT).
Cancers 2024, 16, 2988 14 of 31

4.3. Applications
4.3.1. Detection of Spinal Lesions
Early identification and diagnosis of spinal lesions is crucial in clinical practice [96,97].
This plays a pivotal role in determining the disease stage in patients with malignancy,
which significantly influences the treatment strategies and prognosis [98]. Spinal lesions,
including metastases, are associated with increased morbidity, with over half of patients
needing radiotherapy or invasive procedures due to complications such as spinal cord or
nerve root compression [99–101]. Therefore, early diagnosis and targeted treatment before
permanent neurological and functional deficits develop, is critical to achieve favorable
outcomes [102–104].
However, detecting spinal lesions manually via various imaging techniques is often
time-consuming, laborious, and challenging [105] due to factors such as overlapping
imaging features with other pathologies [106] and variations in imaging quality, resulting
in artifacts which can obscure lesions and hinder their detection [10,107]. Automated
lesion detection is widely recognized for its potential to enhance radiologists’ sensitivity
in identifying osseous metastases. Computer-aided detection (CADe) software and AI
models have been shown to be as effective, if not more effective, than manual detection
by radiologists.
The first CADe study using CT for spinal metastasis detection was conducted by
O’Connor et al. in 2007 [108], specifically targeting osteolytic spinal metastases. This
pioneering work laid the foundation for subsequent CADe research, which has been
expanded to include other types of spinal metastases, such as osteoblastic and mixed
lesions. Recently, advancements in artificial intelligence, particularly in DL and CNNs,
have significantly enhanced the accuracy of CADe in detecting spinal metastases. These
improvements have led to notable reductions in both false positive and false negative
rates, thereby increasing the reliability of automated detection across different imaging
techniques [67,69,109,110].
Several efforts employing diverse methods have been undertaken to detect spinal
lesions in CT scans using AI techniques. For instance, Burns et al. [66] used a combination
of watershed segmentation and a support vector machine classifier, achieving sensitivity
of 79.0–90.0% in detecting osteoblastic spinal metastases. Hammon et al. [65] employed
sequences of random forest classifiers that analyze local image features to evaluate regions
for spinal metastases, attaining sensitivity of 83.0% for detecting osteolytic lesions and
88.0% in detecting osteoblastic lesions. Matohashi et al. [7] utilized deep learning methods
using the “Deep Lab V3+” segmentation model, attaining sensitivity of 75.0–78.0% in
detecting osteolytic lesions in thoracolumbar spine CT, which is comparable to that of
orthopedic or radiology experts, and superior to that of in-training clinicians.
Several other techniques have been utilized to further improve sensitivity for detecting
spinal metastases from CT. For instance, Hoshiai et al. [71] investigated the application
of temporal subtraction CT for bone segmentation, using a multi-atlas-based method
combined with spatial registration of two images via advanced computational techniques
and deep learning methods showing improved overall field of merit (FOM) compared to
both a board-certified radiologist and resident groups (p < 0.05). Both Noguchi et al. [9]
and Huo et al. [59] developed deep convolutional neural network (DCNN) models that
showed statistically significant improvements (p < 0.05) in the figure of merit (FOM) from
0.746 to 0.899 for radiologists, and in the detection accuracy and sensitivity of in-training
radiologists, while also reducing the average interpretation time per case.
AI has already demonstrated its potential in oncology for improving the detection of
various cancers. For example, in breast cancer screening, AI systems can accurately identify
suspicious lesions on mammograms [111–113], reducing the workload for radiologists
while maintaining high sensitivity and lowering recall rates [114,115]. In lung cancer,
AI algorithms analyze CT scans to detect early-stage nodules with precision, facilitating
timely intervention [116–118]. Additionally, AI tools in prostate cancer aid the detection
of malignant tissues on MRI, improving diagnostic accuracy [119] and aiding treatment
Cancers 2024, 16, 2988 15 of 31

planning [119–121]. Although AI applications for detecting spinal lesions on CT scans

are still in their infancy, their potential integration into clinical practice holds promise for
helping alleviate radiologists’ workload [122,123] and reducing error rates [16]. By serving
as a secondary reviewer, AI can enhance diagnostic accuracy and efficiency, providing a
dependable safety net. Additionally, with improved detection accuracy and sensitivity, AI
has the potential to minimize the necessity for further imaging modalities such as MRI or
bone scans [124,125], potentially curbing healthcare costs.

4.3.2. Classification of Spinal Lesions

AI has shown promise in differentiating benign versus malignant lesions across various
cancers, leveraging advanced algorithms to improve diagnostic accuracy [126–129]. In
the context of spinal lesions, machine learning has been applied to differentiate between
various spinal pathologies, such as metastases, by identifying key radiomic features in
vertebral lesions and incorporating these into diverse machine learning models [130].
Spinal lesion classification has been predominantly studied on MRI due to its superior
soft tissue contrast and various sequences allowing for more precise characterization
of lesion morphology and signal characteristics [131–134]. For example, Liu et al. [135]
and Xiong et al. [136] utilized MRI-based radiomics to differentiate between multiple
myeloma and spinal metastases. Yin et al. [137] came out with a radiomics model for the
differentiation of primary Chordoma, Giant Cell Tumor, and metastatic sacral tumors.
Distinguishing benign from malignant spinal lesions on CT remains challenging. The
complexity of spinal anatomy, the variability of lesion appearance, and the overlapping
features between benign and malignant lesions pose significant hurdles [138,139]. AI mod-
els can potentially overcome these challenges to provide reliable differentiation through
extensive training on diverse datasets and continuous refinement to improve accuracy
and reduce false positives. For example, the deep learning method developed by Ma-
soudi et al. [78], using lesion-based average 2D ResNet-50 and 3D ResNet-18 architectures
with texture, volumetric, and morphologic information, achieved an accuracy of 92.2% for
the classification of benign versus malignant sclerotic bony lesions in patients with prostate
cancer. Chmelik et al. [76] developed a deep CNN-based segmentation and classification
of difficult-to-define spinal lesions using 3D CT data, achieving an AUC of 0.780–0.800
in distinguishing metastatic (both osteoblastic and osteolytic) bone lesions versus benign
bone lesions.
Differentiating vertebral fractures due to benign causes, such as osteoporosis, from
those caused by malignancy on CT imaging poses significant challenges. Both types of
fracture can present with similar radiologic features, including vertebral body collapse and
cortical disruption, complicating accurate diagnosis [140]. Often, CT imaging alone fails
to reliably distinguish these conditions, necessitating further work-up such as contrast-
enhanced MRI or isotope bone scans [141,142]. AI, leveraging radiomics and deep learning
techniques, offers a promising solution by analyzing complex patterns in CT images that
are not easily discernible to the human eye [143]. Radiomics and deep learning models can
be trained to identify subtle differences in image characteristics, such as bone density and
lesion shape, and by integrating these advanced analytics AI can improve the differentiation
between benign osteoporotic fractures and pathological vertebral fractures.
For example, Li Y. et al. [77] developed a deep learning model using a ResNet 50
network with ten-fold cross-validation to achieve 85.0–88.0% accuracy in distinguishing
benign from malignant compression fractures. Duan S. et al. [86] came up with radiomics
and deep learning methods using Inception_V3 to differentiate these two entities through
CT features and clinical characteristics, achieving an AUC of up to 0.990. Both studies
also concluded that visual features such as the presence of a soft tissue mass and bone
destruction were highly suggestive of malignancy, while the presence of a transverse fracture
line is highly suggestive of a benign fracture—consistent with the literature [140,144,145].
AI applications go beyond mere tumor detection and differentiation, and they can
also automatically generate significant parameters, such as grading spinal lesions and
 their complications. For example, Hallinan et al. [79] employed a deep learning mo
automate the classification of metastatic epidural disease and/or spinal cord compre
on CT scans based on the Bilsky classification [101,103]. The model showed near p
agreement when compared to trained radiologists, with kappas of 0.873–0.911 (p < 0
Cancers 2024, 16, 2988 16 of 31
Subsequent studies by the same author [85] showed that their deep learning algorith
metastatic spinal cord compression on CT showed superior performance to the CT r
with almost-perfect
their complications.inter-rater
For example, agreement
Hallinan et (κ
al. = 0.813)
[79] employedandahigh sensitivity
deep learning model (94.0%)
to as
automate the classification of metastatic epidural disease and/or spinal cord compression
pared to CT reports issued by experienced radiologists which had only slight inter
on CT scans based on the Bilsky classification [101,103]. The model showed near perfect
agreement (κ = when
agreement 0.027) and low
compared sensitivity
to trained (44.0%)
radiologists, with(pkappas
< 0.001). Precise and
of 0.873–0.911 consistent
(p < 0.001).
fication of metastatic
Subsequent epidural
studies by the samespinal cord
author [85] compression
showed that their deepwilllearning
help clinicians
algorithm dete
whether initial treatment should involve radiotherapy or surgicaltointervention
for metastatic spinal cord compression on CT showed superior performance the CT
report with almost-perfect inter-rater agreement (κ = 0.813) and high sensitivity (94.0%)
These as
studies
compared demonstrate the potential
to CT reports issued of deep
by experienced learning
radiologists whichtohad
assist
only clinicians
slight inter- in the
diagnosis
raterand grading
agreement (κ = of metastatic
0.027) spinal disease,
and low sensitivity (44.0%) (pensuring that appropriate
< 0.001). Precise and consistent treatm
classification
delivered promptly. of metastatic epidural spinal cord compression will help clinicians deter-
mine whether initial treatment should involve radiotherapy or surgical intervention [146].
Radiogenomics, which combines “Radiomics” and “Genomics”, entails utilizin
These studies demonstrate the potential of deep learning to assist clinicians in the early
aging traits or and
diagnosis substitutes
grading of to identify
metastatic genomic
spinal disease, patterns
ensuring thatand advanced
appropriate bio-markers
treatment is w
tumorsdelivered
[147,148]. These markers subsequently inform clinical decisions, encompa
promptly.
Radiogenomics, which combines “Radiomics” and “Genomics”, entails utilizing imag-
the prognosis, diagnosis, and predictive accuracy of tumor subtypes [149]. The t
ing traits or substitutes to identify genomic patterns and advanced bio-markers within
workflow
tumorsof[147,148].
a radiogenomics
These markers study (Figureinform
subsequently 4) encompasses
clinical decisions,five key stages:
encompassing theimage
sition and pre-processing,
prognosis, diagnosis, andfeature
predictive extraction
accuracy ofand tumorselection
subtypes from both
[149]. The medical
typical work-imagin
flow of a radiogenomics study (Figure 4) encompasses five key stages:
genotype data, association of radiomic and genomic features, data analysis utilizin image acquisition
and pre-processing, feature extraction and selection from both medical imaging and geno-
chine learning models, of
type data, association and the establishment
radiomic and genomic features,of a data
finalanalysis
radiogenomics outcome m
utilizing machine
[150]. learning models, and the establishment of a final radiogenomics outcome model [150].

Figure Figure
4. Diagram
4. Diagramillustrating
illustrating aatypical
typical radiogenomics
radiogenomics process, encompassing
process, encompassing image acquisitionimage
and acqu
and pre-processing, feature
pre-processing, feature extraction
extraction and selection
and selection from
from medical medical
imaging imaging
and genotype and
data, genotype dat
correlation
between radiomic and genomic features, data analysis using machine
relation between radiomic and genomic features, data analysis using machine learninglearning models, and the mode
determination of the final radiogenomics outcome.
the determination of the final radiogenomics outcome.
In the realm of oncologic spinal CT imaging, the focus of radiogenomics research has
Inpredominantly
the realm ofbeen oncologic spinal
on giant cell tumorCT imaging,
of the the focus
bone (GCTB). Wang etofal.
radiogenomics
[82,83] pioneered researc
the application of machine learning and radiomic techniques to forecast the presence
predominantly been on giant cell tumor of the bone (GCTB). Wang et al. [82,83] pion
of key biomarkers—RANKL (receptor activator of the nuclear factor kappa B ligand),
the application
p53, and VEGFof machine learninggrowth
(vascular endothelial and radiomic techniques
factor)—in spinal to forecast
GCTB, achieving the prese
an AUC
key biomarkers—RANKL
of 0.658–0.880. Elevated (receptor activator
levels of RANKL, p53, of
andthe nuclear
VEGF factor
expression kappa
[151] B ligand)
in GCTB
have been correlated with more aggressive tumor behavior and
and VEGF (vascular endothelial growth factor)—in spinal GCTB, achieving an A increased recurrence
risk [152–154]. These biomarkers also form the basis of targeted molecular therapies, such
0.658–0.880. Elevated
as Denosumab, levels ofantibody
a monoclonal RANKL, p53, and
targeting RANKL,VEGFusedexpression [151] in GCTB
to treat challenging-to-
been correlated with
operate spinal GCTBmore
casesaggressive tumor behavior
[155,156]. Identification and increased
of these biomarkers aids in recurrence
prognosis risk
154]. These biomarkers also form the basis of targeted molecular therapies, su
Cancers 2024, 16, 2988 17 of 31

prediction and facilitates the selection of optimal disease management strategies. However,
their assessment typically necessitates invasive tissue biopsies. The ability to predict
these biomarkers non-invasively offers a quantitative and convenient approach to support
predictive decision-making, ultimately enhancing patient outcomes [157].

4.3.3. Prognosis and Predicting Complications

The integration of artificial intelligence (AI), deep learning, and radiomic models
into CT spinal imaging holds promise for predicting outcomes and anticipating treatment
complications in oncologic scenarios [158,159]. In addition to tumor characteristics, the
assessment of sarcopenia and body composition plays a crucial role in the prognosis of
patients with spinal metastases [160–162]. However, manual evaluation of these parameters
can be tedious and exhausting, often requiring significant time and expertise [163]. AI and
deep learning algorithms offer a transformative solution by automating the analysis of
body composition from CT scans, providing rapid and accurate assessments [164]. These
technologies can quantify muscle mass, adipose tissue distribution, and other relevant
parameters with high precision, facilitating risk stratification and treatment planning.
For instance, Delrieu L et al. [91] developed a deep learning algorithm to automati-
cally detect the L3 vertebra and segment body tissues to evaluate body composition and
sarcopenia, achieving a median DICE similarity coefficient of up to 0.940 in relation to
the sarcopenia metrics of the patient pool. Khalid et al. [92] came out with a machine
learning-based technique for detection of sarcopenic obesity (SO) using CT in patients
undergoing surgery for spinal metastases. Their algorithm effectively identified patients
with SO, who exhibited increased odds of non-home discharge, re-admission, and post-
operative mortality. These advancements enable clinicians to potentially recognize patients
requiring nutritional interventions prior to invasive procedures, thereby enhancing overall
treatment outcomes [165].
In addition to assessing body composition and sarcopenia, artificial intelligence (AI)
holds promise for predicting the risk of recurrence, a crucial aspect in determining treatment
outcomes for patients with spinal metastases. By analyzing radiomic features extracted
from CT spinal imaging data, AI algorithms can identify subtle patterns and biomarkers
associated with tumor aggressiveness and likelihood of recurrence [166,167]. For example,
Wang Q et al. [88] developed a deep learning model to predict the risk of recurrence of
spinal GCTB using images from pre-operative CT, achieving an accuracy of 89.0% and
an AUC of 0.780. These predictive models leverage deep learning techniques to process
vast amounts of imaging data and generate personalized risk assessments for individual
patients [168]. Early identification of patients at high risk of recurrence enables clinicians to
tailor treatment strategies, such as implementing adjuvant therapies or closer surveillance
protocols, to mitigate the risk of disease progression and improve long-term outcomes [169].
Furthermore, AI-driven predictive models provide valuable insights into the under-
lying biological mechanisms driving tumor recurrence, facilitating the development of
targeted therapeutic interventions aimed at preventing disease relapse and enhancing
patient survival rates [170]. For instance, Gong et al. [171] came out with a radiomics
model to predict pre-operative expression of PD-1 and PD-L1 in hepatocellular carcinoma
(HCC), which was associated with more aggressive tumor behavior in the form of increased
recurrence and distal metastatic risk [172]. The ability to use such an imaging biomarker
may potentially help identify patients who will benefit from immune checkpoint inhibitor
(ICI)-based treatment before surgery. Similarly, Bove et al. [173] developed a predictive
model using radiomic features extracted from pre-treatment CT scans to forecast the like-
lihood of recurrence in patients with non-small cell lung cancer (NSCLC). Their findings
demonstrated that specific radiomic signatures from the peri-tumor region correlated sig-
nificantly with higher recurrence rates. By harnessing the power of AI-driven predictive
models in spinal oncology, there is potential for clinicians to better stratify patients based
on their risk of recurrence, tailor treatment strategies accordingly, and ultimately improve
long-term survival rates.
Cancers 2024, 16, 2988 18 of 31

AI and deep learning methodologies offer promising avenues to enhance the identifi-
cation of patients at high risk of radiation-induced vertebral compression fracture (VCF)
in the context of stereotactic body radiation therapy (SBRT) for spinal metastases. By
extracting radiomic features from pre-treatment CT imaging data, these advanced tech-
niques can provide more nuanced insights into the underlying tissue characteristics and
microenvironment, enabling more accurate prediction of VCF risk [174]. Gui et al. [87]
developed a machine learning model based on clinical characteristics and radiomic features
from pre-treatment CT imaging to predict the risk of vertebral fracture following radiation
therapy. Their model achieved sensitivity of 84.4%, specificity of 80.0%, and an area under
the receiver operating characteristic (ROC) curve (AUC) of 0.878. The model developed
by Seol et al., using similar methods, achieved an accuracy of 81.8% with an AUC of
0.870 in predicting VCF prior to spinal SBRT using pre-treatment planning CT. Integrating
these AI-driven predictive models into clinical practice could potentially assist clinicians in
implementing timely prophylactic measures to mitigate the occurrence of VCF and mini-
mize associated morbidities before treatment [175]. This personalized approach to patient
care will help optimize treatment strategies and improve overall treatment outcomes in
individuals undergoing SBRT for spinal metastases.

4.3.4. Treatment Planning

Accurate delineation of spinal lesions is paramount for the effective planning of radio-
therapy for spinal neoplasms and metastases [176–178]. However, manual outlining often
proves time-consuming and susceptible to variability between observers. Deep learning
methods present a transformative approach by automating the identification and segmen-
tation of spinal metastases and surrounding anatomical structures [179]. This automation
not only facilitates precise radiation dose calculation and treatment plan optimization but
also significantly streamlines the workflow [180,181]. By reducing manual intervention
and mitigating inter-observer variability, deep learning expedites the treatment planning
process, offering potential benefits in clinical settings [182,183].
For instance, Hernandez et al. [95] developed a comprehensive automated contour
and planning tool for 3D-conformal craniospinal irradiation therapy using CT images for
pediatric patients with medulloblastoma. Their model achieved remarkable DICE similarity
coefficients (DSC) ranging from 0.650 to 0.980, with an average end-to-end treatment
planning time of 3.5 ± 0.4 min. Similarly, Sebastiaan et al. [93] devised a CNN-based model
for automating segmentation and delineation of vertebral bodies in radiotherapy treatment
planning for spinal metastases. Their model exhibited an average computational time of
less than 5 min compared to an average of 20 min for manual contouring, with DSC scores
ranging from 0.944 to 0.967, and 77.0% of cases deemed clinically acceptable. Although
still in its infancy and not yet fully validated in clinical settings, these studies underscore
the potential of AI and deep learning methods for enhancing the efficiency of radiotherapy
workflows without compromising accuracy, ultimately improving patient care.
In current clinical practice, deep learning tools such as RayStation by RaySearch Labs
are actively used for tumor segmentation in gamma radiotherapy planning. For instance,
Riguad et al. [184] utilized RayStation for optimizing dose planning in cervical cancer
treatment, while Almberg et al. [185] applied it in breast cancer treatment, demonstrating
significant improvements in dose optimization. These examples highlight the real-world
effectiveness of advanced segmentation technologies. Additionally, other commercial solu-
tions, such as Varian’s Eclipse [186] and Elekta’s Monaco [187], also employ sophisticated
deep learning algorithms for radiotherapy planning.

5. Discussion
5.1. Interpretation and Implications of Findings
In our systematic review of oncologic applications of artificial intelligence (AI) and
deep learning methods in CT spine imaging, we observed a broad spectrum of applications
including detection, classification, prognosis, and treatment planning. Our review indicated
Cancers 2024, 16, 2988 19 of 31

that AI models show considerable promise in these areas, demonstrating relatively reliable
performance and considerable improvement in clinical practice. However, several critical
insights and interpretations emerged from our findings.
AI models have proven effective in enhancing the detection and classification of
spinal lesions. Despite these advancements, the performance metrics reported across
studies exhibit notable variability, reflecting a significant limitation. This variability is
partly due to the fact that many studies were constrained by small sample sizes and were
conducted at single centers. Such limitations raise concerns about the generalizability of the
results [188]. Specifically, while AI models may perform well within specific study settings,
their effectiveness and reliability across diverse clinical environments remain uncertain
without broader validation [189]. In the realm of prognosis and treatment planning, AI
models have demonstrated substantial potential by improving the prediction of patient
outcomes and aiding in the formulation of treatment strategies. Nonetheless, translating
these capabilities into routine clinical practice requires further validation. The current
evidence base, constrained by the limited scope of existing studies, necessitates larger,
multi-center trials to confirm the models’ effectiveness and ensure their applicability to a
wider range of patient populations.
Overall, while our review underscores the potential of AI and deep learning in ad-
vancing CT spine imaging, addressing the identified limitations through comprehensive
validation and integration efforts is crucial for realizing their full clinical potential. These
steps will ensure that AI technologies can provide reliable, actionable insights, and support
the management of spinal malignancies.

5.2. Integration into Clinical Practice

The integration of AI models into clinical workflows presents several significant
challenges that must be carefully addressed. Current research often suffers from a lack
of external validation and does not fully tackle the practicalities of implementing these
models in routine clinical settings [190,191]. To address these barriers, future research
should prioritize the standardization of imaging protocols to reduce variability [192],
expand datasets through collaborative and multi-center studies, and develop AI systems
that seamlessly integrate into existing clinical practices [193]. Additionally, obtaining FDA
and CE approvals for these tools is a critical step, ensuring they meet rigorous safety
and efficacy standards before widespread adoption. To date, only a limited number of
AI applications in CT spinal oncology have received such approvals, with most focusing
on treatment planning [185] in general, and not specific to spinal oncology. However,
we anticipate that more AI tools will achieve regulatory clearance in the future as the
technology matures and more evidence of its clinical utility emerges.
Furthermore, the integration of AI and machine learning (ML) into healthcare intro-
duces a range of ethical concerns. These include issues related to privacy and data security,
the risk of biases in AI algorithms, and the challenges of ensuring transparency and explain-
ability of AI systems [194]. Additional concerns involve patient consent, autonomy, and the
equitable access to these technologies [195]. The complexity of these issues is compounded
by problems such as limited data availability, data drift, and the need for ongoing retraining
and regulatory updates [196]. Ensuring the safety and clinical validation of AI tools is
essential to prevent misdiagnoses, avoid inappropriate treatments, and mitigate potential
health disparities. Therefore, a balanced approach is crucial to leverage the benefits of AI
while addressing these significant ethical and practical challenges effectively.

5.3. Other Potential Applications

5.3.1. Improving Image Quality
Oncologic spinal imaging poses significant challenges, particularly in complex cases
such as post-surgical patients with metallic implants, which induce artifacts and degrade
image quality, thereby impeding accurate evaluation of spinal lesions. These artifacts can
obscure critical details (e.g., mass effect on the spinal cord or collections), complicating
Cancers 2024, 16, 2988 20 of 31

diagnosis, and treatment planning. However, deep learning reconstruction methods have
emerged as valuable tools for enhancing CT imaging quality in oncology. Numerous
studies have demonstrated their efficacy in mitigating artifacts and improving image
clarity. For instance, Arabi et al. [197] developed a deep learning-based reconstruction
algorithm that effectively reduced metal artifacts in PET/CT scans of post-surgical spinal
patients, enhancing visualization of adjacent structures and facilitating more accurate
lesion assessment. Similarly, Rui et al. [198] employed a deep learning-based metal artifact
correction (MAC) algorithm, achieving significantly higher subjective scores as compared
to conventional MAC and virtual monochromatic imaging (VMI) techniques.
Building upon these successes, there is potential for Generative Adversarial Networks
(GANs) to further advance the field. GANs offer the ability to generate realistic and artifact-
corrected images by learning from large datasets, potentially providing a solution to the
challenges posed by various artifacts in CT spinal imaging. For instance, Lu et al. [199]
utilized GAN to correct motion artefacts on CT Coronary Angiogram images. Their
GAN-generated images showed statistically significant improvement in motion artifact
alleviation score (4/5 vs. 1/5, p < 0.001) and overall image quality score (4/5 vs. 1/5,
p < 0.001), with high accuracy in identifying stenosis (81.0% vs. 66.0%) in the mid-right
coronary artery as compared to motion-affected images. Goli et al. [200] utilized GAN to
improve CT images of the head and neck affected by metallic artefacts from dental implants,
achieving 16.8% improvement in assessing the oral cavity region, which is important for
treatment planning. While traditional non-AI-based metal artifact correction methods,
such as i-MAR used in Siemens CT scanners [201,202], are commonly employed in clinical
settings, deep learning reconstruction methods offer several significant advantages. These
methods enhance image quality by more effectively reducing metal artifacts and improving
the visualization of adjacent structures [203]. They provide greater adaptability to new
data and imaging modalities, reduce the need for manual adjustments, and often result in
faster processing times and better quantitative accuracy [204]. Consequently, deep learning
techniques offer potentially more robust solutions for addressing metallic artifacts in CT
imaging, leading to more reliable and accurate diagnostic outcomes [205]. By synthesizing
artifact-free images, GANs hold promise for improving diagnostic accuracy and enhancing
oncologic evaluations in patients with spinal lesions.

5.3.2. Predicting Primary Malignancy from Spinal Metastases

Spinal metastasis originating from an unidentified primary tumor presents a common
clinical challenge, affecting up to 30% of patients on initial presentation [206,207]. While
conventional CT scans can accurately detect vertebral metastases, distinguishing between
cancers of various origins can be challenging as they often appear similar. Consequently,
these patients often have to undergo additional PET/CT imaging for primary cancer diag-
nosis and comprehensive whole-body staging scans prior to treatment initiation [208]. In
rare cases, even with further imaging, the primary tumor remains unidentified, necessitat-
ing an invasive biopsy to determine the probable primary tumor site and to explore more
targeted treatment options.
Several studies have demonstrated the effectiveness of deep learning models in pre-
dicting primary tumor sites in patients with spinal metastases using MRI-derived imaging
features. For example, Liu et al. [209] explored the feasibility of a ResNet-50 convolutional
neural network model for this purpose, achieving an AUC–ROC of 0.770 and 53.0% accu-
racy in classifying spinal metastases originating from the lung, kidney, prostate, breast, and
thyroid. Lang et al. [210] utilized radiomics and deep learning techniques to distinguish
spinal metastases from lung cancer and other origins using dynamic contrast-enhanced
(DCE) sequences from a spinal MRI database. Their findings indicated that the DCE kinetic
measurement of the washout slope from a hotspot within the spinal metastatic lesion was
the most reliable parameter for diagnosing primary lung cancer versus other tumor types,
achieving accuracies of up to 81.0%.
Cancers 2024, 16, 2988 21 of 31

Cao et al. [211] investigated radiomics for distinguishing primary tumors from brain
metastases using CT images. Shang et al. [184] developed a similar model for identifying
primary tumor types from lung metastases in thoracic CT scans. Despite bone metastases
being common, no studies have specifically addressed primary tumor differentiation in
spinal metastases using CT imaging. This gap presents a significant opportunity for future
research in spinal oncology. With promising results seen in MRI studies, applying deep
learning techniques to CT spinal imaging shows potential. CT scans offer widespread
availability and detailed anatomical information, making them valuable for assessing
primary malignancies. By training deep learning models on large datasets of CT spinal
images, robust algorithms could potentially differentiate and identify primary malignancies
solely from CT scans. Leveraging radiomic approaches from other metastatic contexts
could assist with developing predictive models specific to spinal metastases. Advanced
imaging analytics applied to CT scans could reveal distinct patterns and biomarkers for
different primary tumors, enhancing diagnostic accuracy and treatment planning. These
advancements may provide clinicians with a non-invasive method to efficiently diagnose
and plan treatments for spinal metastases.

5.3.3. Quantifying Tumor Burden to Predict Treatment Response

Artificial intelligence (AI) applications hold promise for enhancing the measurement
of tumor burden for assessing treatment response and monitoring tumor progression. AI-
assisted segmentation allows for precise lesion and/or tumor volumetry, facilitating more
accurate evaluations. For instance, Goehler et al. [212] implemented a deep learning method
to estimate overall tumor burden for neuroendocrine neoplasia on MRI, achieving a con-
cordance of 91% with manual clinician assessment and DSC of up to 0.81. Belal et al. [213]
developed a fully automated CNN-based model to calculate skeletal tumor burden in pa-
tients with prostate cancer on PET/CT, achieving a moderately strong PET index correlation
with that estimated by the physician (mean r = 0.69).
Assessing tumor burden and volume in spinal metastases presents challenges due to
variations in vertebral shape and involvement across multiple levels [185] Manual volu-
metric assessment is not only time-consuming but also prone to variability among different
observers, and even within the same observer. In clinical practice, evaluating radiological
images by oncology specialists and radiologists is hindered by the labor-intensive nature
of manual analysis, the absence of standardized quantification, and challenges related
to reproducibility in both evaluation and measurement. Despite that, assessing tumor
burden in spinal metastases is pivotal for predicting prognosis and treatment efficacy [214].
While there are studies which have looked into using deep learning to assess bone tumor
burden in PET/CT [213,215] and bone scintigraphy [216], the current literature lacks stud-
ies exploring the use of CT imaging to comprehensively evaluate spinal tumor burden.
This gap is significant, especially given the frequent use of whole-body CT scans in oncol-
ogy for monitoring treatment response. Spinal metastases, including those affecting the
spine, necessitate a holistic assessment of the entire skeleton due to the systemic nature of
metastatic disease.
Recent advancements in deep learning applied to CT imaging have automated the
segmentation of bone metastases, including those affecting the spine. Studies such as those
by Motohashi et al. [7] and Saeed et al. [217] have shown deep learning algorithms’ effec-
tiveness in identifying and quantifying metastatic lesions across skeletal regions. Despite
these capabilities in segmenting bone tumors, there remains a lack of research specifically
focused on assessing tumor burden. These AI-driven approaches could streamline the volu-
metric evaluation of spinal metastases and pave the way for improved clinical management
by providing a comprehensive view of tumor burden throughout the body. Integrating
AI into the assessment of bone metastases using CT holds the potential to significantly
enhance the timely and accurate monitoring of treatment response. By automating tumor
burden quantification from whole-body CT scans, AI can potentially guide oncologists to
Cancers 2024, 16, 2988 22 of 31

more effectively evaluate overall disease burden, which is crucial for assessing treatment
efficacy and making informed decisions in patient care [218,219].

5.4. Study Limitations

Due to the heterogeneity among the studies reviewed and insufficient data, a for-
mal meta-analysis could not be conducted. Consequently, our review is presented as a
descriptive analysis rather than as a quantitative synthesis. This limitation affects our
ability to perform statistical aggregation of results and derive more generalized conclusions.
Additionally, the quality of data in the individual studies was not analyzed in detail, which
could influence the robustness of the findings. However, this does not diminish the value
of our review. Despite these limitations, our descriptive review provides valuable insights
into the current state of oncologic applications of artificial intelligence and deep learning
methods in CT spine imaging. It highlights key trends, identifies research gaps, and offers
a comprehensive overview of technological advancements and their clinical implications.
By synthesizing and summarizing findings from a diverse range of studies, our review
contributes significantly to the understanding of these emerging technologies and provides
a foundation for future research directions and clinical applications in the field. The qual-
itative synthesis presented serves as a useful resource for researchers and practitioners,
guiding further investigation and application of AI and deep learning in this domain.
Furthermore, our review acknowledges a notable gap in addressing perspectives from
patients and clinicians regarding the use of AI in spinal oncology. Incorporating feedback
from these key stakeholders is essential for understanding the real-world implications of
AI applications. Patients’ experiences [220] and clinicians’ insights [221,222] are critical
for evaluating the practical benefits and limitations of AI technologies, as well as their
acceptance and integration into routine clinical practice. Additionally, our review did
not extensively cover the long-term impact and cost-effectiveness of AI applications [223].
Evaluating these factors is important for assessing the overall value of AI to spinal oncology,
as it can influence decision-making processes, resource allocation, and the sustainability
of these technologies in clinical settings. Future research should address these aspects to
provide a more comprehensive evaluation of AI’s role in spinal oncology, ensuring that
the benefits of AI applications are balanced with considerations of their economic and
long-term impacts.

6. Conclusions
In conclusion, the integration of deep learning techniques with computed tomography
(CT) imaging in spinal oncology presents a promising avenue for enhancing diagnostic accu-
racy, treatment planning, and improving patient outcomes. This review has demonstrated
the diverse applications of artificial intelligence (AI) in CT imaging of spinal metastases,
including detection, classification, grading, and treatment planning. AI technologies have
demonstrated notable performance across these domains, offering the potential to support
clinicians by improving workflow efficiency and minimizing complications. However,
despite promising findings, additional research is warranted to validate the clinical effec-
tiveness of these AI tools and streamline their integration into everyday clinical workflows.
Ultimately, the continued exploration and refinement of AI applications in CT spinal imag-
ing holds immense promise for advancing the field of spinal oncology and improving
patient care.

Author Contributions: Conceptualization, methodology, supervision, and writing: A.L., W.O., W.C.T.,
X.Z.L., D.D.L., S.G. and J.T.P.D.H. Investigation and project administration: W.O., A.L., K.T.D.F.,
Y.L.T., A.M., Y.H.T. and J.T.P.D.H. Resources and software: W.O., A.L., J.H.T., N.K. and J.T.P.D.H.
Formal analysis and validation: W.O., A.L., W.C.T., X.Z.L., S.G., K.T.D.F., Y.L.T., S.J.O. and J.T.P.D.H.
All authors have read and agreed to the published version of the manuscript.
Funding: This research was directly funded by MOH/NMRC, Singapore. Specifically, this study
received support from the Singapore Ministry of Health National Medical Research Council under the
Cancers 2024, 16, 2988 23 of 31

NMRC Clinician Innovator Award (CIA). The grant was awarded for the project titled “Deep learning
pipeline for augmented reporting of MRI whole spine” (Grant ID: CIAINV23jan-0001, MOH-001405,
J.T.P.D.H).
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Mundy, G.R. Metastasis to bone: Causes, consequences and therapeutic opportunities. Nat. Rev. Cancer 2002, 2, 584–593.
[CrossRef] [PubMed]
2. Witham, T.F.; Khavkin, Y.A.; Gallia, G.L.; Wolinsky, J.P.; Gokaslan, Z.L. Surgery insight: Current management of epidural spinal
cord compression from metastatic spine disease. Nat. Clin. Pract. Neurol. 2006, 2, 87–94. [CrossRef] [PubMed]
3. Coleman, R.E. Metastatic bone disease: Clinical features, pathophysiology and treatment strategies. Cancer Treat. Rev. 2001, 27,
165–176. [CrossRef]
4. Cecchini, M.G.; Wetterwald, A.; van der Pluijm, G.; Thalmann, G.N. Molecular and Biological Mechanisms of Bone Metastasis.
EAU Update Ser. 2005, 3, 214–226. [CrossRef]
5. Cuccurullo, V.; Cascini, G.L.; Tamburrini, O.; Rotondo, A.; Mansi, L. Bone metastases radiopharmaceuticals: An overview. Curr.
Radiopharm. 2013, 6, 41–47. [CrossRef] [PubMed]
6. Yu, H.H.; Tsai, Y.Y.; Hoffe, S.E. Overview of diagnosis and management of metastatic disease to bone. Cancer Control 2012,
19, 84–91. [CrossRef] [PubMed]
7. Motohashi, M.; Funauchi, Y.; Adachi, T.; Fujioka, T.; Otaka, N.; Kamiko, Y.; Okada, T.; Tateishi, U.; Okawa, A.; Yoshii, T.; et al.
A New Deep Learning Algorithm for Detecting Spinal Metastases on Computed Tomography Images. Spine 2024, 49, 390–397.
[CrossRef]
8. Cui, Y.; Zhu, J.; Duan, Z.; Liao, Z.; Wang, S.; Liu, W. Artificial Intelligence in Spinal Imaging: Current Status and Future Directions.
Int. J. Environ. Res. Public Health 2022, 19, 11708. [CrossRef]
9. Martín-Noguerol, T.; Oñate Miranda, M.; Amrhein, T.J.; Paulano-Godino, F.; Xiberta, P.; Vilanova, J.C.; Luna, A. The role of
Artificial intelligence in the assessment of the spine and spinal cord. Eur. J. Radiol. 2023, 161, 110726. [CrossRef]
10. Kim, G.U.; Chang, M.C.; Kim, T.U.; Lee, G.W. Diagnostic Modality in Spine Disease: A Review. Asian Spine J. 2020, 14, 910–920.
[CrossRef]
11. Ariyaratne, S.; Jenko, N.; Iyengar, K.P.; James, S.; Mehta, J.; Botchu, R. Primary Osseous Malignancies of the Spine. Diagnostics
2023, 13, 1801. [CrossRef] [PubMed]
12. Thrall, J.H.; Li, X.; Li, Q.; Cruz, C.; Do, S.; Dreyer, K.; Brink, J. Artificial Intelligence and Machine Learning in Radiology:
Opportunities, Challenges, Pitfalls, and Criteria for Success. J. Am. Coll. Radiol. 2018, 15, 504–508. [CrossRef] [PubMed]
13. Oren, O.; Gersh, B.J.; Bhatt, D.L. Artificial intelligence in medical imaging: Switching from radiographic pathological data to
clinically meaningful endpoints. Lancet Digit. Health 2020, 2, e486–e488. [CrossRef]
14. Erickson, B.J.; Korfiatis, P.; Akkus, Z.; Kline, T.L. Machine Learning for Medical Imaging. Radiographics 2017, 37, 505–515.
[CrossRef] [PubMed]
15. Wang, S.; Summers, R.M. Machine learning and radiology. Med. Image Anal. 2012, 16, 933–951. [CrossRef]
16. Najjar, R. Redefining Radiology: A Review of Artificial Intelligence Integration in Medical Imaging. Diagnostics 2023, 13, 2760.
[CrossRef]
17. Cheng, P.M.; Montagnon, E.; Yamashita, R.; Pan, I.; Cadrin-Chênevert, A.; Romero, F.P.; Chartrand, G.; Kadoury, S.; Tang, A. Deep
Learning: An Update for Radiologists. RadioGraphics 2021, 41, 1427–1445. [CrossRef]
18. Zaharchuk, G.; Gong, E.; Wintermark, M.; Rubin, D.; Langlotz, C.P. Deep Learning in Neuroradiology. Am. J. Neuroradiol. 2018,
39, 1776–1784. [CrossRef]
19. Yamashita, R.; Nishio, M.; Do, R.K.G.; Togashi, K. Convolutional neural networks: An overview and application in radiology.
Insights Imaging 2018, 9, 611–629. [CrossRef]
20. Jussupow, E.; Spohrer, K.; Heinzl, A. Radiologists’ Usage of Diagnostic AI Systems. Bus. Inf. Syst. Eng. 2022, 64, 293–309.
[CrossRef]
21. Kaka, H.; Zhang, E.; Khan, N. Artificial intelligence and deep learning in neuroradiology: Exploring the new frontier. Can. Assoc.
Radiol. J. 2021, 72, 35–44. [CrossRef] [PubMed]
22. Daugaard Jørgensen, M.; Antulov, R.; Hess, S.; Lysdahlgaard, S. Convolutional neural network performance compared to
radiologists in detecting intracranial hemorrhage from brain computed tomography: A systematic review and meta-analysis. Eur.
J. Radiol. 2022, 146, 110073. [CrossRef]
23. Mutasa, S.; Sun, S.; Ha, R. Understanding artificial intelligence based radiology studies: CNN architecture. Clin. Imaging 2021,
80, 72–76. [CrossRef]
24. Manakitsa, N.; Maraslidis, G.S.; Moysis, L.; Fragulis, G.F. A Review of Machine Learning and Deep Learning for Object Detection,
Semantic Segmentation, and Human Action Recognition in Machine and Robotic Vision. Technologies 2024, 12, 15. [CrossRef]
25. van Timmeren, J.E.; Cester, D.; Tanadini-Lang, S.; Alkadhi, H.; Baessler, B. Radiomics in medical imaging—”How-to” guide and
critical reflection. Insights Imaging 2020, 11, 91. [CrossRef] [PubMed]
Cancers 2024, 16, 2988 24 of 31

26. Shur, J.D.; Doran, S.J.; Kumar, S.; ap Dafydd, D.; Downey, K.; O’Connor, J.P.B.; Papanikolaou, N.; Messiou, C.; Koh, D.-M.; Orton,
M.R. Radiomics in Oncology: A Practical Guide. RadioGraphics 2021, 41, 1717–1732. [CrossRef] [PubMed]
27. Aerts, H.J. The Potential of Radiomic-Based Phenotyping in Precision Medicine: A Review. JAMA Oncol. 2016, 2, 1636–1642.
[CrossRef]
28. Rogers, W.; Thulasi Seetha, S.; Refaee, T.A.G.; Lieverse, R.I.Y.; Granzier, R.W.Y.; Ibrahim, A.; Keek, S.A.; Sanduleanu, S.; Primakov,
S.P.; Beuque, M.P.L.; et al. Radiomics: From qualitative to quantitative imaging. Br. J. Radiol. 2020, 93, 20190948. [CrossRef]
29. Cobo, M.; Menéndez Fernández-Miranda, P.; Bastarrika, G.; Lloret Iglesias, L. Enhancing radiomics and Deep Learning systems
through the standardization of medical imaging workflows. Sci. Data 2023, 10, 732. [CrossRef]
30. Avery, E.; Sanelli, P.C.; Aboian, M.; Payabvash, S. Radiomics: A Primer on Processing Workflow and Analysis. Semin. Ultrasound
CT MR 2022, 43, 142–146. [CrossRef]
31. Sharma, N.; Aggarwal, L.M. Automated medical image segmentation techniques. J. Med. Phys. 2010, 35, 3–14. [CrossRef]
32. Rana, M.; Bhushan, M. Machine learning and deep learning approach for medical image analysis: Diagnosis to detection.
Multimed. Tools Appl. 2023, 82, 26731–26769. [CrossRef] [PubMed]
33. Valladares, A.; Beyer, T.; Rausch, I. Physical imaging phantoms for simulation of tumor heterogeneity in PET, CT, and MRI: An
overview of existing designs. Med. Phys. 2020, 47, 2023–2037. [CrossRef]
34. Breiman, L. Random Forests. Mach. Learn. 2001, 45, 5–32. [CrossRef]
35. Rizzo, S.; Botta, F.; Raimondi, S.; Origgi, D.; Fanciullo, C.; Morganti, A.G.; Bellomi, M. Radiomics: The facts and the challenges of
image analysis. Eur. Radiol. Exp. 2018, 2, 36. [CrossRef] [PubMed]
36. Lambin, P.; Leijenaar, R.T.H.; Deist, T.M.; Peerlings, J.; de Jong, E.E.C.; van Timmeren, J.; Sanduleanu, S.; Larue, R.; Even, A.J.G.;
Jochems, A.; et al. Radiomics: The bridge between medical imaging and personalized medicine. Nat. Rev. Clin. Oncol. 2017, 14,
749–762. [CrossRef] [PubMed]
37. Larsen, M.; Olstad, C.F.; Lee, C.I.; Hovda, T.; Hoff, S.R.; Martiniussen, M.A.; Mikalsen, K.; Lund-Hanssen, H.; Solli, H.S.;
Silberhorn, M.; et al. Performance of an Artificial Intelligence System for Breast Cancer Detection on Screening Mammograms
from BreastScreen Norway. Radiol. Artif. Intell. 2024, 6, e230375. [CrossRef]
38. Khawaja, A.; Bartholmai, B.J.; Rajagopalan, S.; Karwoski, R.A.; Varghese, C.; Maldonado, F.; Peikert, T. Do we need to see
to believe?—Radiomics for lung nodule classification and lung cancer risk stratification. J. Thorac. Dis. 2020, 12, 3303–3316.
[CrossRef] [PubMed]
39. Astaraki, M.; Yang, G.; Zakko, Y.; Toma-Dasu, I.; Smedby, Ö.; Wang, C. A Comparative Study of Radiomics and Deep-Learning
Based Methods for Pulmonary Nodule Malignancy Prediction in Low Dose CT Images. Front. Oncol. 2021, 11, 737368. [CrossRef]
40. Liu, Z.; Jiang, Z.; Meng, L.; Yang, J.; Liu, Y.; Zhang, Y.; Peng, H.; Li, J.; Xiao, G.; Zhang, Z.; et al. Handcrafted and Deep
Learning-Based Radiomic Models Can Distinguish GBM from Brain Metastasis. J. Oncol. 2021, 2021, 5518717. [CrossRef]
41. Salmanpour, M.R.; Rezaeijo, S.M.; Hosseinzadeh, M.; Rahmim, A. Deep versus Handcrafted Tensor Radiomics Features: Prediction
of Survival in Head and Neck Cancer Using Machine Learning and Fusion Techniques. Diagnostics 2023, 13, 1696. [CrossRef]
42. Huang, Y.; Li, J.; Li, M.; Aparasu, R.R. Application of machine learning in predicting survival outcomes involving real-world
data: A scoping review. BMC Med. Res. Methodol. 2023, 23, 268. [CrossRef] [PubMed]
43. Xiao, J.; Mo, M.; Wang, Z.; Zhou, C.; Shen, J.; Yuan, J.; He, Y.; Zheng, Y. The Application and Comparison of Machine Learning
Models for the Prediction of Breast Cancer Prognosis: Retrospective Cohort Study. JMIR Med. Inform. 2022, 10, e33440. [CrossRef]
[PubMed]
44. Zhang, Z.; Sejdić, E. Radiological images and machine learning: Trends, perspectives, and prospects. Comput. Biol. Med. 2019, 108,
354–370. [CrossRef]
45. Alzubaidi, L.; Zhang, J.; Humaidi, A.J.; Al-Dujaili, A.; Duan, Y.; Al-Shamma, O.; Santamaría, J.; Fadhel, M.A.; Al-Amidie, M.;
Farhan, L. Review of deep learning: Concepts, CNN architectures, challenges, applications, future directions. J. Big Data 2021,
8, 53. [CrossRef] [PubMed]
46. Puttagunta, M.; Ravi, S. Medical image analysis based on deep learning approach. Multimed. Tools Appl. 2021, 80, 24365–24398.
[CrossRef]
47. Arbet, J.; Brokamp, C.; Meinzen-Derr, J.; Trinkley, K.E.; Spratt, H.M. Lessons and tips for designing a machine learning study
using EHR data. J. Clin. Transl. Sci. 2020, 5, e21. [CrossRef]
48. Zhang, A.; Xing, L.; Zou, J.; Wu, J.C. Shifting machine learning for healthcare from development to deployment and from models
to data. Nat. Biomed. Eng. 2022, 6, 1330–1345. [CrossRef]
49. Liu, Z.; Wang, S.; Dong, D.; Wei, J.; Fang, C.; Zhou, X.; Sun, K.; Li, L.; Li, B.; Wang, M.; et al. The Applications of Radiomics in
Precision Diagnosis and Treatment of Oncology: Opportunities and Challenges. Theranostics 2019, 9, 1303–1322. [CrossRef]
50. Limkin, E.J.; Sun, R.; Dercle, L.; Zacharaki, E.I.; Robert, C.; Reuzé, S.; Schernberg, A.; Paragios, N.; Deutsch, E.; Ferté, C. Promises
and challenges for the implementation of computational medical imaging (radiomics) in oncology. Ann. Oncol. 2017, 28,
1191–1206. [CrossRef]
51. Li, S.; Zhou, B. A review of radiomics and genomics applications in cancers: The way towards precision medicine. Radiat. Oncol.
2022, 17, 217. [CrossRef]
52. Page, M.J.; McKenzie, J.E.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.;
Brennan, S.E.; et al. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ 2021, 372, n71.
[CrossRef]
Cancers 2024, 16, 2988 25 of 31

53. Pinto-Coelho, L. How Artificial Intelligence Is Shaping Medical Imaging Technology: A Survey of Innovations and Applications.
Bioengineering 2023, 10, 1435. [CrossRef] [PubMed]
54. Tanguay, W.; Acar, P.; Fine, B.; Abdolell, M.; Gong, B.; Cadrin-Chênevert, A.; Chartrand-Lefebvre, C.; Chalaoui, J.; Gorgos, A.;
Chin, A.S.-L.; et al. Assessment of Radiology Artificial Intelligence Software: A Validation and Evaluation Framework. Can.
Assoc. Radiol. J. 2023, 74, 326–333. [CrossRef] [PubMed]
55. Handelman, G.S.; Kok, H.K.; Chandra, R.V.; Razavi, A.H.; Huang, S.; Brooks, M.; Lee, M.J.; Asadi, H. Peering into the Black Box
of Artificial Intelligence: Evaluation Metrics of Machine Learning Methods. Am. J. Roentgenol. 2019, 212, 38–43. [CrossRef]
56. McKinney, S.M.; Sieniek, M.; Godbole, V.; Godwin, J.; Antropova, N.; Ashrafian, H.; Back, T.; Chesus, M.; Corrado, G.S.; Darzi,
A.; et al. International evaluation of an AI system for breast cancer screening. Nature 2020, 577, 89–94. [CrossRef] [PubMed]
57. Hajian-Tilaki, K. Receiver Operating Characteristic (ROC) Curve Analysis for Medical Diagnostic Test Evaluation. Casp. J. Intern.
Med. 2013, 4, 627–635.
58. Hanley, J.A.; McNeil, B.J. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982,
143, 29–36. [CrossRef]
59. Obuchowski, N.A. Receiver operating characteristic curves and their use in radiology. Radiology 2003, 229, 3–8. [CrossRef]
60. Powers, D. Evaluation: From Precision, Recall and F-Factor to ROC, Informedness, Markedness & Correlation. Int. J. Mach. Learn.
Technol. 2011, 2, 37–63. [CrossRef]
61. Sokolova, M.; Lapalme, G. A systematic analysis of performance measures for classification tasks. Inf. Process. Manag. 2009, 45,
427–437. [CrossRef]
62. Czodrowski, P. Count on kappa. J. Comput.-Aided Mol. Des. 2014, 28, 1049–1055. [CrossRef]
63. Yang, F.; Zamzmi, G.; Angara, S.; Rajaraman, S.; Aquilina, A.; Xue, Z.; Jaeger, S.; Papagiannakis, E.; Antani, S.K. Assessing
Inter-Annotator Agreement for Medical Image Segmentation. IEEE Access 2023, 11, 21300–21312. [CrossRef]
64. Wiese, T.; Yao, J.; Burns, J.E.; Summers, R.M. Detection of sclerotic bone metastases in the spine using watershed algorithm and
graph cut. In Proceedings of the SPIE Medical Imaging 2012, San Diego, CA, USA, 4–9 February 2012; SPIE: Bellingham, WA,
USA, 2012; Volume 831512. [CrossRef]
65. Hammon, M.; Dankerl, P.; Tsymbal, A.; Wels, M.; Kelm, M.; May, M.; Suehling, M.; Uder, M.; Cavallaro, A. Automatic detection of
lytic and blastic thoracolumbar spine metastases on computed tomography. Eur. Radiol. 2013, 23, 1862–1870. [CrossRef] [PubMed]
66. Burns, J.E.; Yao, J.; Wiese, T.S.; Muñoz, H.E.; Jones, E.C.; Summers, R.M. Automated detection of sclerotic metastases in the
thoracolumbar spine at CT. Radiology 2013, 268, 69–78. [CrossRef] [PubMed]
67. Roth, H.R.; Yao, J.; Lu, L.; Stieger, J.; Burns, J.E.; Summers, R.M. Detection of Sclerotic Spine Metastases via Random Aggregation
of Deep Convolutional Neural Network Classifications. In Recent Advances in Computational Methods and Clinical Applications for
Spine Imaging; Yao, J., Glocker, B., Klinder, T., Li, S., Eds.; Springer International Publishing: Cham, Switzerland, 2015; pp. 3–12.
68. Masoudi, S.; Mehralivand, S.; Harmon, S.; Walker, S.; Pinto, P.A.; Wood, B.J.; Citrin, D.E.; Karzai, F.; Gulley, J.L.; Madan, R.A.; et al.
Artificial intelligence assisted bone lesion detection and classification in computed tomography scans of prostate cancer patients.
J. Clin. Oncol. 2020, 38, e17567. [CrossRef]
69. Fan, X.; Zhang, X.; Zhang, Z.; Jiang, Y. Deep Learning-Based Identification of Spinal Metastasis in Lung Cancer Using Spectral CT
Images. Sci. Program. 2021, 2021, 2779390. [CrossRef]
70. Noguchi, S.; Nishio, M.; Sakamoto, R.; Yakami, M.; Fujimoto, K.; Emoto, Y.; Kubo, T.; Iizuka, Y.; Nakagomi, K.; Miyasa, K.; et al.
Deep learning-based algorithm improved radiologists’ performance in bone metastases detection on CT. Eur. Radiol. 2022, 32,
7976–7987. [CrossRef]
71. Hoshiai, S.; Hanaoka, S.; Masumoto, T.; Nomura, Y.; Mori, K.; Okamoto, Y.; Saida, T.; Ishiguro, T.; Sakai, M.; Nakajima, T.
Effectiveness of temporal subtraction computed tomography images using deep learning in detecting vertebral bone metastases.
Eur. J. Radiol. 2022, 154, 110445. [CrossRef]
72. Musa Aguiar, P.; Zarantonello, P.; Aparisi Gómez, M.P. Differentiation between Osteoporotic and Neoplastic Vertebral Fractures:
State of the Art and Future Perspectives. Curr. Med. Imaging 2022, 18, 187–207. [CrossRef]
73. Gilberg, L.; Teodorescu, B.; Maerkisch, L.; Baumgart, A.; Ramaesh, R.; Ataide, E.; Koç, A. Deep Learning Enhances Radiologists’
Detection of Potential Spinal Malignancies in CT Scans. Appl. Sci. 2023, 13, 8140. [CrossRef]
74. Huo, T.; Xie, Y.; Fang, Y.; Wang, Z.; Liu, P.; Duan, Y.; Zhang, J.; Wang, H.; Xue, M.; Liu, S.; et al. Deep learning-based algorithm
improves radiologists’ performance in lung cancer bone metastases detection on computed tomography. Front. Oncol. 2023, 13,
1125637. [CrossRef] [PubMed]
75. Koike, Y.; Yui, M.; Nakamura, S.; Yoshida, A.; Takegawa, H.; Anetai, Y.; Hirota, K.; Tanigawa, N. Artificial intelligence-aided lytic
spinal bone metastasis classification on CT scans. Int. J. Comput. Assist. Radiol. Surg. 2023, 18, 1867–1874. [CrossRef] [PubMed]
76. Chmelik, J.; Jakubicek, R.; Walek, P.; Jan, J.; Ourednicek, P.; Lambert, L.; Amadori, E.; Gavelli, G. Deep convolutional neural
network-based segmentation and classification of difficult to define metastatic spinal lesions in 3D CT data. Med. Image Anal.
2018, 49, 76–88. [CrossRef] [PubMed]
77. Li, Y.; Zhang, Y.; Zhang, E.; Chen, Y.; Wang, Q.; Liu, K.; Yu, H.J.; Yuan, H.; Lang, N.; Su, M.Y. Differential diagnosis of benign and
malignant vertebral fracture on CT using deep learning. Eur. Radiol. 2021, 31, 9612–9619. [CrossRef]
78. Masoudi, S.; Mehralivand, S.; Harmon, S.A.; Lay, N.; Lindenberg, L.; Mena, E.; Pinto, P.A.; Citrin, D.E.; Gulley, J.L.; Wood,
B.J.; et al. Deep Learning Based Staging of Bone Lesions From Computed Tomography Scans. IEEE Access 2021, 9, 87531–87542.
[CrossRef]
Cancers 2024, 16, 2988 26 of 31

79. Hallinan, J.T.; Zhu, L.; Zhang, W.; Kuah, T.; Lim, D.S.; Low, X.Z.; Cheng, A.J.L.; Eide, S.E.; Ong, H.Y.; Muhamat Nor, F.E.; et al.
Deep Learning Model for Grading Metastatic Epidural Spinal Cord Compression on Staging CT. Cancers 2022, 14, 3219. [CrossRef]
80. Naseri, H.; Skamene, S.; Tolba, M.; Faye, M.D.; Ramia, P.; Khriguian, J.; Patrick, H.; Andrade Hernandez, A.X.; David, M.; Kildea,
J. Radiomics-based machine learning models to distinguish between metastatic and healthy bone using lesion-center-based
geometric regions of interest. Sci. Rep. 2022, 12, 9866. [CrossRef]
81. Park, T.; Yoon, M.A.; Cho, Y.C.; Ham, S.J.; Ko, Y.; Kim, S.; Jeong, H.; Lee, J. Automated segmentation of the fractured vertebrae on
CT and its applicability in a radiomics model to predict fracture malignancy. Sci. Rep. 2022, 12, 6735. [CrossRef]
82. Wang, Q.; Chen, Y.; Qin, S.; Liu, X.; Liu, K.; Xin, P.; Zhao, W.; Yuan, H.; Lang, N. Prognostic Value and Quantitative CT Analysis
in RANKL Expression of Spinal GCTB in the Denosumab Era: A Machine Learning Approach. Cancers 2022, 14, 5201. [CrossRef]
83. Wang, Q.; Zhang, Y.; Zhang, E.; Xing, X.; Chen, Y.; Nie, K.; Yuan, H.; Su, M.Y.; Lang, N. A Multiparametric Method Based on
Clinical and CT-Based Radiomics to Predict the Expression of p53 and VEGF in Patients with Spinal Giant Cell Tumor of Bone.
Front. Oncol. 2022, 12, 894696. [CrossRef]
84. Hallinan, J.; Zhu, L.; Tan, H.W.N.; Hui, S.J.; Lim, X.; Ong, B.W.L.; Ong, H.Y.; Eide, S.E.; Cheng, A.J.L.; Ge, S.; et al. A deep
learning-based technique for the diagnosis of epidural spinal cord compression on thoracolumbar CT. Eur. Spine J. 2023, 32,
3815–3824. [CrossRef]
85. Hallinan, J.T.P.D.; Zhu, L.; Zhang, W.; Ge, S.; Muhamat Nor, F.E.; Ong, H.Y.; Eide, S.E.; Cheng, A.J.L.; Kuah, T.; Lim, D.S.W.; et al.
Deep learning assessment compared to radiologist reporting for metastatic spinal cord compression on CT. Front. Oncol. 2023,
13, 1151073. [CrossRef]
86. Duan, S.; Hua, Y.; Cao, G.; Hu, J.; Cui, W.; Zhang, D.; Xu, S.; Rong, T.; Liu, B. Differential diagnosis of benign and malignant
vertebral compression fractures: Comparison and correlation of radiomics and deep learning frameworks based on spinal CT
and clinical characteristics. Eur. J. Radiol. 2023, 165, 110899. [CrossRef]
87. Gui, C.; Chen, X.; Sheikh, K.; Mathews, L.; Lo, S.L.; Lee, J.; Khan, M.A.; Sciubba, D.M.; Redmond, K.J. Radiomic modeling to
predict risk of vertebral compression fracture after stereotactic body radiation therapy for spinal metastases. J. Neurosurg. Spine
2021, 36, 294–302. [CrossRef] [PubMed]
88. Wang, Q.; Zhang, Y.; Zhang, E.; Xing, X.; Chen, Y.; Su, M.Y.; Lang, N. Prediction of the early recurrence in spinal giant cell tumor
of bone using radiomics of preoperative CT: Long-term outcome of 62 consecutive patients. J. Bone Oncol. 2021, 27, 100354.
[CrossRef] [PubMed]
89. Massaad, E.; Bridge, C.P.; Kiapour, A.; Fourman, M.S.; Duvall, J.B.; Connolly, I.D.; Hadzipasic, M.; Shankar, G.M.; Andriole,
K.P.; Rosenthal, M.; et al. Evaluating frailty, mortality, and complications associated with metastatic spine tumor surgery using
machine learning-derived body composition analysis. J. Neurosurg. Spine 2022, 37, 263–273. [CrossRef]
90. Seol, Y.; Song, J.H.; Choi, K.H.; Lee, Y.K.; Choi, B.O.; Kang, Y.N. Predicting vertebral compression fracture prior to spinal SBRT
using radiomics from planning CT. Eur. Spine J. 2023, 33, 3221–3229. [CrossRef]
91. Delrieu, L.; Blanc, D.; Bouhamama, A.; Reyal, F.; Pilleul, F.; Racine, V.; Hamy, A.S.; Crochet, H.; Marchal, T.; Heudel, P.E. Automatic
deep learning method for third lumbar selection and body composition evaluation on CT scans of cancer patients. Front. Nucl.
Med. 2024, 3, 1292676. [CrossRef]
92. Khalid, S.I.; Massaad, E.; Kiapour, A.; Bridge, C.P.; Rigney, G.; Burrows, A.; Shim, J.; de la Garza Ramos, R.; Tobert, D.G.;
Schoenfeld, A.J.; et al. Machine learning-based detection of sarcopenic obesity and association with adverse outcomes in patients
undergoing surgical treatment for spinal metastases. J. Neurosurg. Spine 2024, 40, 291–300. [CrossRef]
93. Arends, S.R.S.; Savenije, M.H.F.; Eppinga, W.S.C.; van der Velden, J.M.; van den Berg, C.A.T.; Verhoeff, J.J.C. Clinical utility of
convolutional neural networks for treatment planning in radiotherapy for spinal metastases. Phys. Imaging Radiat. Oncol. 2022, 21,
42–47. [CrossRef]
94. Netherton, T.J.; Nguyen, C.; Cardenas, C.E.; Chung, C.; Klopp, A.H.; Colbert, L.E.; Rhee, D.J.; Peterson, C.B.; Howell, R.; Balter,
P.; et al. An Automated Treatment Planning Framework for Spinal Radiation Therapy and Vertebral-Level Second Check. Int. J.
Radiat. Oncol. Biol. Phys. 2022, 114, 516–528. [CrossRef] [PubMed]
95. Hernandez, S.; Nguyen, C.; Parkes, J.; Burger, H.; Rhee, D.J.; Netherton, T.; Mumme, R.; Vega, J.G.; Duryea, J.; Leone, A.; et al.
Automating the treatment planning process for 3D-conformal pediatric craniospinal irradiation therapy. Pediatr. Blood Cancer
2023, 70, e30164. [CrossRef]
96. Mok, S.; Chu, E.C. The Importance of Early Detection of Spinal Tumors Through Magnetic Resonance Imaging in Chiropractic
Practices. Cureus 2024, 16, e51440. [CrossRef] [PubMed]
97. Costelloe, C.M.; Rohren, E.M.; Madewell, J.E.; Hamaoka, T.; Theriault, R.L.; Yu, T.-K.; Lewis, V.O.; Ma, J.; Stafford, R.J.; Tari,
A.M.; et al. Imaging bone metastases in breast cancer: Techniques and recommendations for diagnosis. Lancet Oncol. 2009, 10,
606–614. [CrossRef]
98. Curtin, M.; Piggott, R.P.; Murphy, E.P.; Munigangaiah, S.; Baker, J.F.; McCabe, J.P.; Devitt, A. Spinal Metastatic Disease: A Review
of the Role of the Multidisciplinary Team. Orthop. Surg. 2017, 9, 145–151. [CrossRef]
99. Tomita, K.; Kawahara, N.; Kobayashi, T.; Yoshida, A.; Murakami, H.; Akamaru, T. Surgical strategy for spinal metastases. Spine
2001, 26, 298–306. [CrossRef]
100. Delank, K.S.; Wendtner, C.; Eich, H.T.; Eysel, P. The treatment of spinal metastases. Dtsch. Arztebl. Int. 2011, 108, 71–79. [CrossRef]
101. Barzilai, O.; Fisher, C.G.; Bilsky, M.H. State of the Art Treatment of Spinal Metastatic Disease. Neurosurgery 2018, 82, 757–769.
[CrossRef]
Cancers 2024, 16, 2988 27 of 31

102. Clemons, M.; Gelmon, K.A.; Pritchard, K.I.; Paterson, A.H. Bone-targeted agents and skeletal-related events in breast cancer
patients with bone metastases: The state of the art. Curr. Oncol. 2012, 19, 259–268. [CrossRef] [PubMed]
103. Bilsky, M.H.; Lis, E.; Raizer, J.; Lee, H.; Boland, P. The diagnosis and treatment of metastatic spinal tumor. Oncologist 1999, 4,
459–469. [CrossRef]
104. Chen, B.; Cai, L.; Zhou, F. Management of acute spinal cord compression in multiple myeloma. Crit. Rev. Oncol. Hematol. 2021,
160, 103205. [CrossRef]
105. Cardobi, N.; Dal Palù, A.; Pedrini, F.; Beleù, A.; Nocini, R.; De Robertis, R.; Ruzzenente, A.; Salvia, R.; Montemezzi, S.; D’Onofrio,
M. An Overview of Artificial Intelligence Applications in Liver and Pancreatic Imaging. Cancers 2021, 13, 2162. [CrossRef]
106. Ong, W.; Zhu, L.; Zhang, W.; Kuah, T.; Lim, D.S.W.; Low, X.Z.; Thian, Y.L.; Teo, E.C.; Tan, J.H.; Kumar, N.; et al. Application of
Artificial Intelligence Methods for Imaging of Spinal Metastasis. Cancers 2022, 14, 4025. [CrossRef] [PubMed]
107. Hallinan, J.T.P.D.; Ge, S.; Zhu, L.; Zhang, W.; Lim, Y.T.; Thian, Y.L.; Jagmohan, P.; Kuah, T.; Lim, D.S.W.; Low, X.Z.; et al. Diagnostic
Accuracy of CT for Metastatic Epidural Spinal Cord Compression. Cancers 2022, 14, 4231. [CrossRef]
108. O’Connor, S.D.; Yao, J.; Summers, R.M. Lytic metastases in thoracolumbar spine: Computer-aided detection at CT—Preliminary
study. Radiology 2007, 242, 811–816. [CrossRef] [PubMed]
109. Wang, J.; Fang, Z.; Lang, N.; Yuan, H.; Su, M.Y.; Baldi, P. A multi-resolution approach for spinal metastasis detection using deep
Siamese neural networks. Comput. Biol. Med. 2017, 84, 137–146. [CrossRef] [PubMed]
110. Chang, C.Y.; Buckless, C.; Yeh, K.J.; Torriani, M. Automated detection and segmentation of sclerotic spinal lesions on body CTs
using a deep convolutional neural network. Skeletal Radiol. 2022, 51, 391–399. [CrossRef]
111. Yoen, H.; Jang, M.J.; Yi, A.; Moon, W.K.; Chang, J.M. Artificial Intelligence for Breast Cancer Detection on Mammography: Factors
Related to Cancer Detection. Acad. Radiol. 2024, 31, P2239–P2247. [CrossRef]
112. Dembrower, K.; Crippa, A.; Colón, E.; Eklund, M.; Strand, F. Artificial intelligence for breast cancer detection in screening
mammography in Sweden: A prospective, population-based, paired-reader, non-inferiority study. Lancet Digit. Health 2023, 5,
e703–e711. [CrossRef]
113. Díaz, O.; Rodríguez-Ruíz, A.; Sechopoulos, I. Artificial Intelligence for breast cancer detection: Technology, challenges, and
prospects. Eur. J. Radiol. 2024, 175, 111457. [CrossRef]
114. Kim, Y.S.; Jang, M.J.; Lee, S.H.; Kim, S.Y.; Ha, S.M.; Kwon, B.R.; Moon, W.K.; Chang, J.M. Use of Artificial Intelligence for Reducing
Unnecessary Recalls at Screening Mammography: A Simulation Study. Korean J. Radiol. 2022, 23, 1241–1250. [CrossRef] [PubMed]
115. Lee, S.E.; Kim, G.R.; Yoon, J.H.; Han, K.; Son, W.J.; Shin, H.J.; Moon, H.J. Artificial intelligence assistance for women who had
spot compression view: Reducing recall rates for digital mammography. Acta Radiol. 2023, 64, 1808–1815. [CrossRef] [PubMed]
116. Zhang, Y.; Jiang, B.; Zhang, L.; Greuter, M.J.W.; de Bock, G.H.; Zhang, H.; Xie, X. Lung Nodule Detectability of Artificial
Intelligence-assisted CT Image Reading in Lung Cancer Screening. Curr. Med. Imaging. 2022, 18, 327–334. [CrossRef] [PubMed]
117. Hendrix, W.; Hendrix, N.; Scholten, E.T.; Mourits, M.; Trap-de Jong, J.; Schalekamp, S.; Korst, M.; van Leuken, M.; van Ginneken,
B.; Prokop, M.; et al. Deep learning for the detection of benign and malignant pulmonary nodules in non-screening chest CT
scans. Commun. Med. 2023, 3, 156. [CrossRef] [PubMed]
118. Goncalves, S.; Fong, P.C.; Blokhina, M. Artificial intelligence for early diagnosis of lung cancer through incidental nodule detection
in low- and middle-income countries-acceleration during the COVID-19 pandemic but here to stay. Am. J. Cancer Res. 2022,
12, 1–16. [PubMed]
119. Twilt, J.J.; van Leeuwen, K.G.; Huisman, H.J.; Fütterer, J.J.; de Rooij, M. Artificial Intelligence Based Algorithms for Prostate
Cancer Classification and Detection on Magnetic Resonance Imaging: A Narrative Review. Diagnostics 2021, 11, 959. [CrossRef]
120. Bhattacharya, I.; Khandwala, Y.S.; Vesal, S.; Shao, W.; Yang, Q.; Soerensen, S.J.C.; Fan, R.E.; Ghanouni, P.; Kunder, C.A.; Brooks,
J.D.; et al. A review of artificial intelligence in prostate cancer detection on imaging. Ther. Adv. Urol. 2022, 14, 17562872221128791.
[CrossRef]
121. Li, H.; Lee, C.H.; Chia, D.; Lin, Z.; Huang, W.; Tan, C.H. Machine Learning in Prostate MRI for Prostate Cancer: Current Status
and Future Opportunities. Diagnostics 2022, 12, 289. [CrossRef]
122. Schalekamp, S.; van Leeuwen, K.; Calli, E.; Murphy, K.; Rutten, M.; Geurts, B.; Peters-Bax, L.; van Ginneken, B.; Prokop, M.
Performance of AI to exclude normal chest radiographs to reduce radiologists’ workload. Eur. Radiol. 2024, 1–9. [CrossRef]
123. Raya-Povedano, J.L.; Romero-Martín, S.; Elías-Cabot, E.; Gubern-Mérida, A.; Rodríguez-Ruiz, A.; Álvarez-Benito, M. AI-based
Strategies to Reduce Workload in Breast Cancer Screening with Mammography and Tomosynthesis: A Retrospective Evaluation.
Radiology 2021, 300, 57–65. [CrossRef] [PubMed]
124. Seah, J.; Brady, Z.; Ewert, K.; Law, M. Artificial intelligence in medical imaging: Implications for patient radiation safety. Br. J.
Radiol. 2021, 94, 20210406. [CrossRef]
125. van Leeuwen, K.G.; de Rooij, M.; Schalekamp, S.; van Ginneken, B.; Rutten, M. How does artificial intelligence in radiology
improve efficiency and health outcomes? Pediatr. Radiol. 2022, 52, 2087–2093. [CrossRef]
126. Hu, H.T.; Wang, W.; Chen, L.D.; Ruan, S.M.; Chen, S.L.; Li, X.; Lu, M.D.; Xie, X.Y.; Kuang, M. Artificial intelligence assists
identifying malignant versus benign liver lesions using contrast-enhanced ultrasound. J. Gastroenterol. Hepatol. 2021, 36, 2875–2883.
[CrossRef]
127. Anaya-Isaza, A.; Mera-Jiménez, L.; Verdugo-Alejo, L.; Sarasti, L. Optimizing MRI-based brain tumor classification and detection
using AI: A comparative analysis of neural networks, transfer learning, data augmentation, and the cross-transformer network.
Eur. J. Radiol. Open 2023, 10, 100484. [CrossRef]
Cancers 2024, 16, 2988 28 of 31

128. Wan, Y.L.; Wu, P.W.; Huang, P.C.; Tsay, P.K.; Pan, K.T.; Trang, N.N.; Chuang, W.Y.; Wu, C.Y.; Lo, S.B. The Use of Artificial
Intelligence in the Differentiation of Malignant and Benign Lung Nodules on Computed Tomograms Proven by Surgical
Pathology. Cancers 2020, 12, 2211. [CrossRef]
129. Ong, W.; Zhu, L.; Tan, Y.L.; Teo, E.C.; Tan, J.H.; Kumar, N.; Vellayappan, B.A.; Ooi, B.C.; Quek, S.T.; Makmur, A.; et al. Application
of Machine Learning for Differentiating Bone Malignancy on Imaging: A Systematic Review. Cancers 2023, 15, 1837. [CrossRef]
[PubMed]
130. Filograna, L.; Lenkowicz, J.; Cellini, F.; Dinapoli, N.; Manfrida, S.; Magarelli, N.; Leone, A.; Colosimo, C.; Valentini, V. Identification
of the most significant magnetic resonance imaging (MRI) radiomic features in oncological patients with vertebral bone marrow
metastatic disease: A feasibility study. Radiol. Med. 2019, 124, 50–57. [CrossRef] [PubMed]
131. Chung, W.J.; Chung, H.W.; Shin, M.J.; Lee, S.H.; Lee, M.H.; Lee, J.S.; Kim, M.J.; Lee, W.K. MRI to differentiate benign from
malignant soft-tissue tumours of the extremities: A simplified systematic imaging approach using depth, size and heterogeneity
of signal intensity. Br. J. Radiol. 2012, 85, e831–e836. [CrossRef]
132. van Rijswijk, C.S.; Kunz, P.; Hogendoorn, P.C.; Taminiau, A.H.; Doornbos, J.; Bloem, J.L. Diffusion-weighted MRI in the
characterization of soft-tissue tumors. J. Magn. Reson. Imaging 2002, 15, 302–307. [CrossRef]
133. Vande Berg, B.C.; Malghem, J.; Lecouvet, F.E.; Maldague, B. Classification and detection of bone marrow lesions with magnetic
resonance imaging. Skeletal Radiol. 1998, 27, 529–545. [CrossRef]
134. Davies, A.M.; Sundaram, M.; James, S.L. Imaging of Bone Tumors and Tumor-like Lesions: Techniques and Applications; Springer:
Berlin/Heidelberg, Germany, 2009.
135. Liu, J.; Guo, W.; Zeng, P.; Geng, Y.; Liu, Y.; Ouyang, H.; Lang, N.; Yuan, H. Vertebral MRI-based radiomics model to differentiate
multiple myeloma from metastases: Influence of features number on logistic regression model performance. Eur. Radiol. 2022, 32,
572–581. [CrossRef]
136. Xiong, X.; Wang, J.; Hu, S.; Dai, Y.; Zhang, Y.; Hu, C. Differentiating Between Multiple Myeloma and Metastasis Subtypes of
Lumbar Vertebra Lesions Using Machine Learning-Based Radiomics. Front. Oncol. 2021, 11, 601699. [CrossRef] [PubMed]
137. Yin, P.; Mao, N.; Zhao, C.; Wu, J.; Chen, L.; Hong, N. A Triple-Classification Radiomics Model for the Differentiation of Primary
Chordoma, Giant Cell Tumor, and Metastatic Tumor of Sacrum Based on T2-Weighted and Contrast-Enhanced T1-Weighted MRI.
J. Magn. Reson. Imaging 2019, 49, 752–759. [CrossRef]
138. Sciubba, D.M.; Petteys, R.J.; Dekutoski, M.B.; Fisher, C.G.; Fehlings, M.G.; Ondra, S.L.; Rhines, L.D.; Gokaslan, Z.L. Diagnosis and
management of metastatic spine disease. A review. J. Neurosurg. Spine 2010, 13, 94–108. [CrossRef] [PubMed]
139. Andreula, C.; Murrone, M. Metastatic disease of the spine. Eur. Radiol. 2005, 15, 627–632. [CrossRef]
140. Mauch, J.T.; Carr, C.M.; Cloft, H.; Diehn, F.E. Review of the Imaging Features of Benign Osteoporotic and Malignant Vertebral
Compression Fractures. AJNR Am. J. Neuroradiol. 2018, 39, 1584–1592. [CrossRef] [PubMed]
141. Abowarda, M.H.; Abdel-Rahman, H.M.; Taha, M.M. Differentiation of acute osteoporotic from malignant vertebral compression
fractures with conventional MRI and diffusion MR imaging. Egypt. J. Radiol. Nucl. Med. 2017, 48, 207–213. [CrossRef]
142. Uetani, M.; Hashmi, R.; Hayashi, K. Malignant and benign compression fractures: Differentiation and diagnostic pitfalls on MRI.
Clin. Radiol. 2004, 59, 124–131. [CrossRef]
143. Tang, X. The role of artificial intelligence in medical imaging research. BJR Open 2020, 2, 20190031. [CrossRef]
144. Laredo, J.D.; Lakhdari, K.; Bellaïche, L.; Hamze, B.; Janklewicz, P.; Tubiana, J.M. Acute vertebral collapse: CT findings in benign
and malignant nontraumatic cases. Radiology 1995, 194, 41–48. [CrossRef]
145. Kubota, T.; Yamada, K.; Ito, H.; Kizu, O.; Nishimura, T. High-resolution imaging of the spine using multidetector-row computed
tomography: Differentiation between benign and malignant vertebral compression fractures. J. Comput. Assist. Tomogr. 2005, 29,
712–719. [CrossRef]
146. Perry, J.; Chambers, A.; Laperriere, N. Systematic Review of the Diagnosis and Management of Malignant Extradural Spinal Cord
Compression: The Cancer Care Ontario Practice Guidelines Initiative‘s Neuro-Oncology Disease Site Group. J. Clin. Oncol. Off. J.
Am. Soc. Clin. Oncol. 2005, 23, 2028–2037. [CrossRef]
147. Liu, Z.; Duan, T.; Zhang, Y.; Weng, S.; Xu, H.; Ren, Y.; Zhang, Z.; Han, X. Radiogenomics: A key component of precision cancer
medicine. Br. J. Cancer 2023, 129, 741–753. [CrossRef] [PubMed]
148. Shui, L.; Ren, H.; Yang, X.; Li, J.; Chen, Z.; Yi, C.; Zhu, H.; Shui, P. The Era of Radiogenomics in Precision Medicine: An Emerging
Approach to Support Diagnosis, Treatment Decisions, and Prognostication in Oncology. Front. Oncol. 2020, 10, 570465. [CrossRef]
[PubMed]
149. Rabasco Meneghetti, A.; Zwanenburg, A.; Linge, A.; Lohaus, F.; Grosser, M.; Baretton, G.B.; Kalinauskaite, G.; Tinhofer, I.;
Guberina, M.; Stuschke, M.; et al. Integrated radiogenomics analyses allow for subtype classification and improved outcome
prognosis of patients with locally advanced HNSCC. Sci. Rep. 2022, 12, 16755. [CrossRef] [PubMed]
150. Singh, G.; Manjila, S.; Sakla, N.; True, A.; Wardeh, A.H.; Beig, N.; Vaysberg, A.; Matthews, J.; Prasanna, P.; Spektor, V. Radiomics
and radiogenomics in gliomas: A contemporary update. Br. J. Cancer 2021, 125, 641–657. [CrossRef]
151. Ferrara, N. Vascular endothelial growth factor: Basic science and clinical progress. Endocr. Rev. 2004, 25, 581–611. [CrossRef]
152. Quattrini, I.; Pollino, S.; Pazzaglia, L.; Conti, A.; Novello, C.; Ferrari, C.; Pignotti, E.; Picci, P.; Benassi, M.S. Prognostic role of
nuclear factor/IB and bone remodeling proteins in metastatic giant cell tumor of bone: A retrospective study. J. Orthop. Res. 2015,
33, 1205–1211. [CrossRef]
Cancers 2024, 16, 2988 29 of 31

153. Li, B.; Li, J.; Miao, W.; Zhao, Y.; Jiao, J.; Wu, Z.; Yang, X.; Wei, H.; Xiao, J. Prognostic Analysis of Clinical and Immunohistochemical
Factors for Patients with Spinal Schwannoma. World Neurosurg. 2018, 120, e617–e627. [CrossRef]
154. Ghani, S.A.; Wan Ismail, W.F.; Md Salleh, M.S.; Yahaya, S.; Syahrul Fitri, Z.M. The Values of Receptor Activator Nuclear Kappa-B
Ligand Expression in Stage III Giant Cell Tumor of the Bone. Indian J. Orthop. 2018, 52, 31–34. [CrossRef] [PubMed]
155. Tsukamoto, S.; Mavrogenis, A.F.; Kido, A.; Errani, C. Current Concepts in the Treatment of Giant Cell Tumors of Bone. Cancers
2021, 13, 3647. [CrossRef]
156. Bukata, S.V.; Blay, J.Y.; Rutkowski, P.; Skubitz, K.; Henshaw, R.; Seeger, L.; Dai, T.; Jandial, D.; Chawla, S. Denosumab Treatment
for Giant Cell Tumor of the Spine Including the Sacrum. Spine 2021, 46, 277–284. [CrossRef] [PubMed]
157. Darvish, L.; Bahreyni-Toossi, M.-T.; Roozbeh, N.; Azimian, H. The role of radiogenomics in the diagnosis of breast cancer: A
systematic review. Egypt. J. Med. Hum. Genet. 2022, 23, 99. [CrossRef]
158. Zhang, B.; Shi, H.; Wang, H. Machine Learning and AI in Cancer Prognosis, Prediction, and Treatment Selection: A Critical
Approach. J. Multidiscip. Healthc. 2023, 16, 1779–1791. [CrossRef]
159. Bera, K.; Braman, N.; Gupta, A.; Velcheti, V.; Madabhushi, A. Predicting cancer outcomes with radiomics and artificial intelligence
in radiology. Nat. Rev. Clin. Oncol. 2022, 19, 132–146. [CrossRef]
160. Tan, H.; Gao, X.; Li, X.; Huang, Y.; Cao, Q.; Wan, T. Sarcopenia in Patients With Spinal Metastasis: A Systematic Review and
Meta-Analysis of Retrospective Cohort Studies. Front. Oncol. 2022, 12, 864501. [CrossRef]
161. Nishigori, T.; Obama, K.; Sakai, Y. Assessment of body composition and impact of sarcopenia and sarcopenic obesity in patients
with gastric cancer. Transl. Gastroenterol. Hepatol. 2020, 5, 22. [CrossRef]
162. Bongers, M.E.R.; Groot, O.Q.; Buckless, C.G.; Kapoor, N.D.; Twining, P.K.; Schwab, J.H.; Torriani, M.; Bredella, M.A. Body
composition predictors of mortality on computed tomography in patients with spinal metastases undergoing surgical treatment.
Spine J. 2022, 22, 595–604. [CrossRef]
163. Xie, W.-Q.; Xiao, G.-L.; Hu, P.-W.; He, Y.-Q.; Lv, S.; Xiao, W.-F. Possible sarcopenia: Early screening and intervention-narrative
review. Ann. Palliat. Med. 2020, 9, 4283–4293. [CrossRef]
164. Pickhardt, P.J.; Summers, R.M.; Garrett, J.W. Automated CT-Based Body Composition Analysis: A Golden Opportunity. Korean J.
Radiol. 2021, 22, 1934–1937. [CrossRef] [PubMed]
165. Wall, J.; Paul, M.; Phillips, B.E. Nutritional interventions in prehabilitation for cancer surgery. Curr. Opin. Clin. Nutr. Metab. Care
2023, 26, 497–507. [CrossRef] [PubMed]
166. Cepeda, S.; Luppino, L.T.; Pérez-Núñez, A.; Solheim, O.; García-García, S.; Velasco-Casares, M.; Karlberg, A.; Eikenes, L.;
Sarabia, R.; Arrese, I.; et al. Predicting Regions of Local Recurrence in Glioblastomas Using Voxel-Based Radiomic Features of
Multiparametric Postoperative MRI. Cancers 2023, 15, 1894. [CrossRef]
167. Libling, W.A.; Korn, R.; Weiss, G.J. Review of the use of radiomics to assess the risk of recurrence in early-stage non-small cell
lung cancer. Transl. Lung Cancer Res. 2023, 12, 1575–1589. [CrossRef]
168. Dixon, D.; Sattar, H.; Moros, N.; Kesireddy, S.R.; Ahsan, H.; Lakkimsetti, M.; Fatima, M.; Doshi, D.; Sadhu, K.; Junaid Hassan,
M. Unveiling the Influence of AI Predictive Analytics on Patient Outcomes: A Comprehensive Narrative Review. Cureus 2024,
16, e59954. [CrossRef] [PubMed]
169. Liao, J.; Li, X.; Gan, Y.; Han, S.; Rong, P.; Wang, W.; Li, W.; Zhou, L. Artificial intelligence assists precision medicine in cancer
treatment. Front. Oncol. 2022, 12, 998222. [CrossRef]
170. Bhinder, B.; Gilvary, C.; Madhukar, N.S.; Elemento, O. Artificial Intelligence in Cancer Research and Precision Medicine. Cancer
Discov. 2021, 11, 900–915. [CrossRef]
171. Gong, X.-Q.; Liu, N.; Tao, Y.-Y.; Li, L.; Li, Z.-M.; Yang, L.; Zhang, X.-M. Radiomics models based on multisequence MRI for
predicting PD-1/PD-L1 expression in hepatocellular carcinoma. Sci. Rep. 2023, 13, 7710. [CrossRef]
172. Li, Q.; Han, J.; Yang, Y.; Chen, Y. PD-1/PD-L1 checkpoint inhibitors in advanced hepatocellular carcinoma immunotherapy. Front.
Immunol. 2022, 13, 1070961. [CrossRef]
173. Bove, S.; Fanizzi, A.; Fadda, F.; Comes, M.C.; Catino, A.; Cirillo, A.; Cristofaro, C.; Montrone, M.; Nardone, A.; Pizzutilo, P.; et al.
A CT-based transfer learning approach to predict NSCLC recurrence: The added-value of peritumoral region. PLoS ONE 2023, 18,
e0285188. [CrossRef]
174. Yoo, G.S.; Park, H.C.; Yu, J.I.; Lim, D.H.; Cho, W.K.; Lee, E.; Jung, S.H.; Han, Y.; Kim, E.S.; Lee, S.H.; et al. Stereotactic ablative
body radiotherapy for spinal metastasis from hepatocellular carcinoma: Its oncologic outcomes and risk of vertebral compression
fracture. Oncotarget 2017, 8, 72860–72871. [CrossRef] [PubMed]
175. Al Taha, K.; Lauper, N.; Bauer, D.E.; Tsoupras, A.; Tessitore, E.; Biver, E.; Dominguez, D.E. Multidisciplinary and Coordinated
Management of Osteoporotic Vertebral Compression Fractures: Current State of the Art. J. Clin. Med. 2024, 13, 930. [CrossRef]
176. Hammoudeh, L.; Abunimer, A.M.; Lee, H.Y.; Dee, E.C.; Brennan, S.V.; Yaguang, P.; Shin, K.Y.; Chen, Y.H.; Huynh, M.A.; Spektor,
A.; et al. Spinal Cord Delineation Based on Computed Tomography Myelogram Versus T2 Magnetic Resonance Imaging in Spinal
Stereotactic Body Radiation Therapy. Adv. Radiat. Oncol. 2023, 8, 101158. [CrossRef] [PubMed]
177. Sacino, A.N.; Chen, H.; Sahgal, A.; Bettegowda, C.; Rhines, L.D.; Maralani, P.; Redmond, K.J. Stereotactic body radiation therapy
for spinal metastases: A new standard of care. Neuro-Oncology 2024, 26, S76–S87. [CrossRef]
178. Kuah, T.; Vellayappan, B.; Makmur, A.; Nair, S.; Song, J.; Tan, J.H.; Kumar, N.; Quek, S.; Hallinan, J. State-of-the-Art Imaging
Techniques in Metastatic Spinal Cord Compression. Cancers 2022, 14, 3289. [CrossRef] [PubMed]
Cancers 2024, 16, 2988 30 of 31

179. McGrath, H.; Li, P.; Dorent, R.; Bradford, R.; Saeed, S.; Bisdas, S.; Ourselin, S.; Shapey, J.; Vercauteren, T. Manual segmentation
versus semi-automated segmentation for quantifying vestibular schwannoma volume on MRI. Int. J. Comput. Assist. Radiol. Surg.
2020, 15, 1445–1455. [CrossRef]
180. Matoska, T.; Patel, M.; Liu, H.; Beriwal, S. Review of Deep Learning Based Autosegmentation for Clinical Target Volume: Current
Status and Future Directions. Adv. Radiat. Oncol. 2024, 9, 101470. [CrossRef]
181. Lastrucci, A.; Wandael, Y.; Ricci, R.; Maccioni, G.; Giansanti, D. The Integration of Deep Learning in Radiotherapy: Exploring
Challenges, Opportunities, and Future Directions through an Umbrella Review. Diagnostics 2024, 14, 939. [CrossRef]
182. Samarasinghe, G.; Jameson, M.; Vinod, S.; Field, M.; Dowling, J.; Sowmya, A.; Holloway, L. Deep learning for segmentation in
radiation therapy planning: A review. J. Med. Imaging Radiat. Oncol. 2021, 65, 578–595. [CrossRef]
183. Wang, M.; Zhang, Q.; Lam, S.; Cai, J.; Yang, R. A Review on Application of Deep Learning Algorithms in External Beam
Radiotherapy Automated Treatment Planning. Front. Oncol. 2020, 10, 580919. [CrossRef]
184. Rigaud, B.; Anderson, B.M.; Yu, Z.H.; Gobeli, M.; Cazoulat, G.; Söderberg, J.; Samuelsson, E.; Lidberg, D.; Ward, C.; Taku, N.; et al.
Automatic Segmentation Using Deep Learning to Enable Online Dose Optimization During Adaptive Radiation Therapy of
Cervical Cancer. Int. J. Radiat. Oncol. Biol. Phys. 2021, 109, 1096–1110. [CrossRef]
185. Almberg, S.S.; Lervåg, C.; Frengen, J.; Eidem, M.; Abramova, T.M.; Nordstrand, C.S.; Alsaker, M.D.; Tøndel, H.; Raj, S.X.;
Wanderås, A.D. Training, validation, and clinical implementation of a deep-learning segmentation model for radiotherapy of
loco-regional breast cancer. Radiother. Oncol. 2022, 173, 62–68. [CrossRef]
186. Li, J.; Anne, R. Comparison of Eclipse Smart Segmentation and MIM Atlas Segment for liver delineation for yttrium-90 selective
internal radiation therapy. J. Appl. Clin. Med. Phys. 2022, 23, e13668. [CrossRef]
187. Buchanan, L.; Hamdan, S.; Zhang, Y.; Chen, X.; Li, X.A. Deep learning-based prediction of deliverable adaptive plans for
MR-guided adaptive radiotherapy: A feasibility study. Front. Oncol. 2023, 13, 939951. [CrossRef] [PubMed]
188. Eche, T.; Schwartz, L.H.; Mokrane, F.-Z.; Dercle, L. Toward Generalizability in the Deployment of Artificial Intelligence in
Radiology: Role of Computation Stress Testing to Overcome Underspecification. Radiol. Artif. Intell. 2021, 3, e210097. [CrossRef]
189. Khan, B.; Fatima, H.; Qureshi, A.; Kumar, S.; Hanan, A.; Hussain, J.; Abdullah, S. Drawbacks of Artificial Intelligence and Their
Potential Solutions in the Healthcare Sector. Biomed. Mater. Devices 2023, 1, 731–738. [CrossRef] [PubMed]
190. Petersson, L.; Larsson, I.; Nygren, J.M.; Nilsen, P.; Neher, M.; Reed, J.E.; Tyskbo, D.; Svedberg, P. Challenges to implementing
artificial intelligence in healthcare: A qualitative interview study with healthcare leaders in Sweden. BMC Health Serv. Res. 2022,
22, 850. [CrossRef]
191. Wubineh, B.Z.; Deriba, F.G.; Woldeyohannis, M.M. Exploring the opportunities and challenges of implementing artificial
intelligence in healthcare: A systematic literature review. Urol. Oncol. Semin. Orig. Investig. 2024, 42, 48–56. [CrossRef] [PubMed]
192. Hagiwara, A.; Fujita, S.; Ohno, Y.; Aoki, S. Variability and Standardization of Quantitative Imaging: Monoparametric to
Multiparametric Quantification, Radiomics, and Artificial Intelligence. Investig. Radiol. 2020, 55, 601–616. [CrossRef]
193. Kotter, E.; Ranschaert, E. Challenges and solutions for introducing artificial intelligence (AI) in daily clinical workflow. Eur. Radiol.
2021, 31, 5–7. [CrossRef]
194. Brady, A.P.; Neri, E. Artificial Intelligence in Radiology-Ethical Considerations. Diagnostics 2020, 10, 231. [CrossRef]
195. Acar, E.; Leblebici, A.; Ellidokuz, B.E.; Basbinar, Y.; Kaya, G.C. Machine learning for differentiating metastatic and completely
responded sclerotic bone lesion in prostate cancer: A retrospective radiomics study. Br. J. Radiol. 2019, 92, 20190286. [CrossRef]
[PubMed]
196. Joshi, G.; Jain, A.; Araveeti, S.R.; Adhikari, S.; Garg, H.; Bhandari, M. FDA-Approved Artificial Intelligence and Machine Learning
(AI/ML)-Enabled Medical Devices: An Updated Landscape. Electronics 2024, 13, 498. [CrossRef]
197. Arabi, H.; Zaidi, H. Deep learning-based metal artefact reduction in PET/CT imaging. Eur. Radiol. 2021, 31, 6384–6396. [CrossRef]
198. Guo, R.; Zou, Y.; Zhang, S.; An, J.; Zhang, G.; Du, X.; Gong, H.; Xiong, S.; Long, Y.; Ma, J. Preclinical validation of a novel
deep learning-based metal artifact correction algorithm for orthopedic CT imaging. J. Appl. Clin. Med. Phys. 2023, 24, e14166.
[CrossRef] [PubMed]
199. Zhang, L.; Jiang, B.; Chen, Q.; Wang, L.; Zhao, K.; Zhang, Y.; Vliegenthart, R.; Xie, X. Motion artifact removal in coronary CT
angiography based on generative adversarial networks. Eur. Radiol. 2023, 33, 43–53. [CrossRef]
200. Khaleghi, G.; Hosntalab, M.; Sadeghi, M.; Reiazi, R.; Mahdavi, S.R. Metal artifact reduction in computed tomography images
based on developed generative adversarial neural network. Inform. Med. Unlocked 2021, 24, 100573. [CrossRef]
201. Lim, S.-W.; Choi, W.-K.; Kim, S. Evaluation of the Effects of Differences in Metal Artifact Type and Location on Image Quality in
Computed Tomography Scans. J. Med. Phys. 2023, 48, 80–84. [PubMed]
202. Schmitt, N.; Weyland, C.S.; Wucherpfennig, L.; Herweh, C.; Bendszus, M.; Möhlenbruch, M.A.; Vollherbst, D.F. Iterative Metal
Artifact Reduction (iMAR) of the Non-adhesive Liquid Embolic Agent Onyx in Computed Tomography: An Experimental Study.
Clin. Neuroradiol. 2022, 32, 695–703. [CrossRef]
203. Koetzier, L.R.; Mastrodicasa, D.; Szczykutowicz, T.P.; van der Werf, N.R.; Wang, A.S.; Sandfort, V.; van der Molen, A.J.;
Fleischmann, D.; Willemink, M.J. Deep Learning Image Reconstruction for CT: Technical Principles and Clinical Prospects.
Radiology 2023, 306, e221257. [CrossRef]
204. Njiti, M.M.; Osman, N.D.; Mansor, M.S.; Rabaiee, N.A.; Abdul Aziz, M.Z. Potential of Metal Artifact Reduction (MAR) and Deep
Learning-based Reconstruction (DLR) algorithms integration in CT Metal Artifact Correction: A review. Radiat. Phys. Chem. 2024,
218, 111541. [CrossRef]
Cancers 2024, 16, 2988 31 of 31

205. Selles, M.; van Osch, J.A.C.; Maas, M.; Boomsma, M.F.; Wellenberg, R.H.H. Advances in metal artifact reduction in CT images: A
review of traditional and novel metal artifact reduction techniques. Eur. J. Radiol. 2024, 170, 111276. [CrossRef] [PubMed]
206. Paholpak, P.; Sirichativapee, W.; Wisanuyotin, T.; Kosuwon, W.; Jeeravipoolvarn, P. Prevalence of known and unknown primary
tumor sites in spinal metastasis patients. Open Orthop. J. 2012, 6, 440–444. [CrossRef]
207. Piccioli, A.; Maccauro, G.; Spinelli, M.S.; Biagini, R.; Rossi, B. Bone metastases of unknown origin: Epidemiology and principles
of management. J. Orthop. Traumatol. 2015, 16, 81–86. [CrossRef]
208. Lawrenz, J.M.; Gordon, J.; George, J.; Haben, C.; Rubin, B.P.; Ilaslan, H.; Mesko, N.W.; Nystrom, L.M. Does PET/CT Aid in
Detecting Primary Carcinoma in Patients with Skeletal Metastases of Unknown Primary? Clin. Orthop. Relat. Res. 2020, 478,
2451–2457. [CrossRef] [PubMed]
209. Liu, K.; Qin, S.; Ning, J.; Xin, P.; Wang, Q.; Chen, Y.; Zhao, W.; Zhang, E.; Lang, N. Prediction of Primary Tumor Sites in Spinal
Metastases Using a ResNet-50 Convolutional Neural Network Based on MRI. Cancers 2023, 15, 2974. [CrossRef]
210. Lang, N.; Zhang, Y.; Zhang, E.; Zhang, J.; Chow, D.; Chang, P.; Yu, H.J.; Yuan, H.; Su, M.Y. Differentiation of spinal metastases
originated from lung and other cancers using radiomics and deep learning based on DCE-MRI. Magn. Reson. Imaging 2019, 64,
4–12. [CrossRef]
211. Cao, G.; Zhang, J.; Lei, X.; Yu, B.; Ai, Y.; Zhang, Z.; Xie, C.; Jin, X. Differentiating Primary Tumors for Brain Metastasis with
Integrated Radiomics from Multiple Imaging Modalities. Dis. Markers 2022, 2022, 5147085. [CrossRef]
212. Goehler, A.; Harry Hsu, T.-M.; Lacson, R.; Gujrathi, I.; Hashemi, R.; Chlebus, G.; Szolovits, P.; Khorasani, R. Three-Dimensional
Neural Network to Automatically Assess Liver Tumor Burden Change on Consecutive Liver MRIs. J. Am. Coll. Radiol. 2020, 17,
1475–1484. [CrossRef] [PubMed]
213. Lindgren Belal, S.; Larsson, M.; Holm, J.; Buch-Olsen, K.M.; Sörensen, J.; Bjartell, A.; Edenbrandt, L.; Trägårdh, E. Automated
quantification of PET/CT skeletal tumor burden in prostate cancer using artificial intelligence: The PET index. Eur. J. Nucl. Med.
Mol. Imaging 2023, 50, 1510–1520. [CrossRef]
214. Oprea-Lager, D.E.; Cysouw, M.C.F.; Boellaard, R.; Deroose, C.M.; de Geus-Oei, L.F.; Lopci, E.; Bidaut, L.; Herrmann, K.; Fournier,
L.S.; Bäuerle, T.; et al. Bone Metastases Are Measurable: The Role of Whole-Body MRI and Positron Emission Tomography. Front.
Oncol. 2021, 11, 772530. [CrossRef]
215. Moreau, N.; Rousseau, C.; Fourcade, C.; Santini, G.; Ferrer, L.; Lacombe, M.; Guillerminet, C.; Campone, M.; Colombié, M.;
Rubeaux, M.; et al. Deep learning approaches for bone and bone lesion segmentation on 18FDG PET/CT imaging in the context
of metastatic breast cancer. In Proceedings of the 2020 42nd Annual International Conference of the IEEE Engineering in Medicine
& Biology Society (EMBC), Montreal, QC, Canada, 20–24 July 2020; pp. 1532–1535.
216. Liu, S.; Feng, M.; Qiao, T.; Cai, H.; Xu, K.; Yu, X.; Jiang, W.; Lv, Z.; Wang, Y.; Li, D. Deep Learning for the Automatic Diagnosis and
Analysis of Bone Metastasis on Bone Scintigrams. Cancer Manag. Res. 2022, 14, 51–65. [CrossRef] [PubMed]
217. Saeed, M.U.; Dikaios, N.; Dastgir, A.; Ali, G.; Hamid, M.; Hajjej, F. An Automated Deep Learning Approach for Spine Segmentation
and Vertebrae Recognition Using Computed Tomography Images. Diagnostics 2023, 13, 2658. [CrossRef]
218. Schutte, K.; Brulport, F.; Harguem-Zayani, S.; Schiratti, J.-B.; Ghermi, R.; Jehanno, P.; Jaeger, A.; Alamri, T.; Naccache, R.;
Haddag-Miliani, L.; et al. An artificial intelligence model predicts the survival of solid tumour patients from imaging and clinical
data. Eur. J. Cancer 2022, 174, 90–98. [CrossRef] [PubMed]
219. Gålne, A.; Enqvist, O.; Sundlöv, A.; Valind, K.; Minarik, D.; Trägårdh, E. AI-based quantification of whole-body tumour burden
on somatostatin receptor PET/CT. Eur. J. Hybrid Imaging 2023, 7, 14. [CrossRef]
220. Moy, S.; Irannejad, M.; Manning, S.J.; Farahani, M.; Ahmed, Y.; Gao, E.; Prabhune, R.; Lorenz, S.; Mirza, R.; Klinger, C. Patient
Perspectives on the Use of Artificial Intelligence in Health Care: A Scoping Review. J. Patient Cent. Res. Rev. 2024, 11, 51–62.
[CrossRef]
221. Alanazi, A. Clinicians’ Views on Using Artificial Intelligence in Healthcare: Opportunities, Challenges, and Beyond. Cureus 2023,
15, e45255. [CrossRef]
222. Daniyal, M.; Qureshi, M.; Marzo, R.R.; Aljuaid, M.; Shahid, D. Exploring clinical specialists’ perspectives on the future role of AI:
Evaluating replacement perceptions, benefits, and drawbacks. BMC Health Serv. Res. 2024, 24, 587. [CrossRef] [PubMed]
223. Khanna, N.N.; Maindarkar, M.A.; Viswanathan, V.; Fernandes, J.F.E.; Paul, S.; Bhagawati, M.; Ahluwalia, P.; Ruzsa, Z.; Sharma, A.;
Kolluri, R.; et al. Economics of Artificial Intelligence in Healthcare: Diagnosis vs. Treatment. Healthcare 2022, 10, 2493. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.