[7] E. Ergin, D. Karaarslan, S. Şahan, and Ü.
Bingöl, “Can artificial intelligence and robotic nurses replace
operating room nurses? The quasi-experimental research,” Journal of Robotic Surgery, vol. 17, no. 4.
Springer Nature, pp. 1847–1855, Aug. 01, 2023. doi: 10.1007/s11701-023-01592-0.
[8] A. A. Gumbs et al., “Artificial intelligence surgery: How do we get to autonomous actions in surgery?,”
Sensors, vol. 21, no. 16, Aug. 2021, doi: 10.3390/s21165526.
[9] A. A. Gumbs et al., “The Advances in Computer Vision That Are Enabling More Autonomous Actions in
Surgery: A Systematic Review of the Literature,” Sensors, vol. 22, no. 13. MDPI, Jul. 01, 2022. doi:
10.3390/s22134918.
[10] H. Saeidi et al., “Autonomous robotic laparoscopic surgery for intestinal anastomosis,” 2022. [Online].
Available: https://www.science.org
[11] A. Attanasio, B. Scaglioni, E. De Momi, P. Fiorini, and P. Valdastri, “Annual Review of Control,
Robotics, and Autonomous Systems Autonomy in Surgical Robotics,” Annu. Rev. Control Robot. Auton.
Syst, vol. 4, pp. 651–679, 2021, doi: 10.1146/annurev-control-062420.
3.4 AI in imaging
Since ancient times, mankind strived to unlock the hidden secrets of nature and especially the intricacies
of their own complexion. Hippocrates, although perhaps wasn’t the first to do so, was the oldest reported
clinician who tried to peer through the skin by the means of palpitation and auscultation. For thousands of
years to follow doctors were limited to those rudimentary tools to visualize the internal workings of the
living body in their imagination. Practically unable to connect the lifeless anatomical structures with their
proper physiological function it was extremely difficult to understand the causes of ailments, let alone
propose a valid treatment. Eventually, thought, Roentgen brought about a new era of understanding,
diagnosing and management when he invented the X-ray machine in the 19 th century, the mainstay of
imaging. The study of the atom during the 20th century lead to yet another triumph, the MRI and the
isotopes based imaging of nuclear medicine. The ultrasound machine was also made common place
throughout the world and despite its challenging aspect for amateur operators it is undoubtedly the corner
store of most departments that treasure speed and safety. Having harness most of the possible imaging
methods from the entirety of Physics’ spectrum it is not easy to conclude what could revolutionize the
future, unless we turn our sights on the field of informatics. A new power is in the rise, a far superior
computing power than ever before which holds many promises for Radiology; if not medicine as a whole.
Radiology stands out as a prime candidate for the implementation and integration of AI due to a wide
range of pivotal factors. First and foremost, AI demonstrates exceptional proficiency in the analysis and
categorization of images. Unlike other medical specialties relying on imaging, radiology boasts a well-
established digital workflow and adheres to universal standards for image storage. This inherent digital
aspect makes the integration of AI more straightforward and effortless due to the already existing
infrastructure.
3.4.1 Current breakthroughs
No need to worry about the experience of the new radiologist anymore; an AI’s data bank will hold
thousands upon thousands of exams of the same findings reaching an unprecedented accuracy. No need
either for the new radiologist to take all the risk on its own with such a trusted and foolproof tool. It is
inevitable for the brightest minds in each and every imaging technique t condense their knowledge and
76
experience into such machine. The successful integration of AI in radiology relies on a seamless
collaboration between clinicians and AI systems. In practice, incorporating AI algorithms to aid
radiologists facilitates the development of a collaborative workflow, emphasizing the synergistic strengths
of both human expertise and AI capabilities. Studies indicate that the assistance of AI in interpreting
medical images proves more advantageous for specific clinicians, especially the least experienced. In
particular, it is worth mentioning that the almost cryptic nature of the processing “logic” of AI allows
them to follow on paths of pattern recognition and formation that human cognition could hardly conceive
let alone exploit in their pursuit of accuracy and diagnostical significance. AI plays a crucial role in
radiology by analyzing images across various techniques, including radiography, CT, ultrasonography,
and MRI. Radiologic AI algorithms are tailored to perform specific narrow image-analysis functions,
aiding radiologists in tasks like quantification, workflow triage, and image enhancement. Quantification
algorithms engage in segmentation and measurement of anatomical structures or abnormalities, such as
measuring breast density, identifying brain structures, quantifying cardiac flow, and assessing local lung-
tissue density. Workflow triage includes the identification and communication of suspected positive
findings, such as intracranial hemorrhage, large-vessel occlusion, pneumothorax, and pulmonary
embolism. Additionally, AI contributes to the detection, localization, and classification of conditions like
pulmonary nodules and breast abnormalities. Moreover, AI algorithms enhance processes, encompassing
image reconstruction, image acquisition, and noise mitigation. Notably, the use of CNNs (convolutional
neural network) in the segmentation of CNS tumors, as well as in temporal bone imaging has yielded
increasingly promising results. A non-exhaustive list of different software for 3D image segmentation
includes the following: Materialize Mimics, AH-Net, ResNet, YOLACT, W‐Net, and 3D cGANs. U-Net
is currently the most popular CNN for precise pixel-level segmentation. The name stems from the shape
of the network’s architecture resembling the letter "U" when visualized graphically. It consists of an
encoder path and a corresponding decoder path, with skip connections in between. Its ability to produce
optimal results while handling limited training data has earned its place as the method of choice in
multiple clinical studies. Moreover, AI can assist on crucial clinical radiology tasks in oncology:
abnormality detection, characterization, and monitoring of change.[1] These tasks require a blend of
medical expertise for diagnosis and technical skills for image processing, presenting opportunities for AI
to enhance outcomes by identifying phenotypic characteristics.
3.4.2 Abnormality Detection:
Radiologists manually spot abnormalities, but Computer-Aided Detection (CADe) systems, though
proposed, often lack generalizability. Recent efforts in deep learning-based CADe, especially using
CNNs, show promise. For instance, they excel in detecting pulmonary nodules in CT and prostate cancer
in multiparametric MRI. Deep learning, particularly convolutional neural networks (CNNs), outperforms
traditional CADe systems in mammograms, indicating its potential for robust and high-performance
detection.[2]
3.4.3 Characterization:
Characterization involves disease segmentation, diagnosis, and staging based on radiological features. AI
can consider numerous quantitative features for reproducible analysis. Automated segmentation, using
probabilistic atlases and deep learning architectures like U-net, proves effective across various modalities.
Examples include brain MRI segmentation for locating glioma, prostate MRI for volume estimation, and
head and neck CT for radiotherapy planning.[3]
3.4.4 Monitoring of Change:
77
Monitoring involves tracking changes in radiological features over time. AI's role in this aspect is crucial
for assessing treatment response and disease progression. Deep learning methods, such as convolutional
neural networks, have shown promise in consistently analyzing changes in various imaging modalities.
3.4.5 Additional tasks
In addition to the primary clinical tasks in oncology, AI is poised to impact various image-based tasks in
the radiology workflow. Notably, advancements are expected in image reconstruction, with deep learning
addressing gaps in current CT reconstruction algorithms. AI methods are also being employed to enhance
registration processes, offering more efficient and accurate solutions, especially in handling complex
tissue deformations. The generation of radiology reports, often a time-consuming task, could benefit from
AI-driven automatic report-generation tools. These tools may bring standardization to terminology and
replace traditional text-based approaches with interactive, quantitative formats. This shift could improve
collaboration between medical professionals and facilitate more effective communication in monitoring
lesions over time. Beyond individual tasks, AI-based integrated diagnostics hold the potential to
assimilate data from multiple sources, offering a comprehensive analysis of a patient's health. This
includes not only radiology reports but also data from pathology, genomics testing, wearables, social
media, and lifestyle sources.[3], [4] This comprehensive approach, facilitated by AI biomarkers, aims to
contribute to improved clinical decision-making, as evidenced by advancements in lung cancer diagnosis
and care. Such extended information integration in a single program will take a truly long time but in the
end we will have created the best possible interpreter The truly mindboggling fact though is that a devise
like that could available to every institution in every longitude and latitude of the world. It won’t matter if
it’s John Hopkins or a small clinic in central Africa. All it would take is some internet access and the
accumulated experience after millions of images will instantly be at their disposal. The pattern reading of
the neuronic networks –the AI’s backbone – makes them absolutely perfect for this job, With each new
image they improve, much like a human brain, but one that can never forget nor get tired. It would de
difficult to say for how long the AI will assist radiologist before surpassing them in speed, efficiency and
financial viability ; although when that last milestone is achieved assisting will transform in a gradual take
over globally. It is also worth noted that apart from judging the results from an imaging machine the AI
could as easily control the machine itself. It is self-apparent once again the way this set up would
eventually surpass human capabilities , both due to programming developments as well as data base
enrichments. Although humans will have to begin this process manually it is rather certain that eventually
the AI-powered machine will operate on this one, choosing optimal imaging aspects , setting the needed
parameters for the most accurate possible image which will be interpreted by the machine itself. Fair to
say, similar techniques could be applied to other viewing based specialties like pathology or even
dermatology . In truth , most of medicine is a pattern based trial and error science and if current
technology continues to improve exponentially , it may be an one way street for medicine as we know it .
As sad as this may sound , eventually it might be just another noble sacrifice for maximizing patient
welfare , the same way a teacher is honestly successful only when his students has surpassed him.
78
