Medical & Biological Engineering & Computing (2022) 60:2445–2462

https://doi.org/10.1007/s11517-022-02614-z

REVIEW ARTICLE

Analysis of red blood cells from peripheral blood smear images

for anemia detection: a methodological review
Navya K.T.1 · Keerthana Prasad2 · Brij Mohan Kumar Singh3

Received: 14 November 2020 / Accepted: 22 April 2022 / Published online: 15 July 2022
© The Author(s) 2022

Abstract
Anemia is a blood disorder which is caused due to inadequate red blood cells and hemoglobin concentration. It occurs in all
phases of life cycle but is more dominant in pregnant women and infants. According to the survey conducted by the World
Health Organization (WHO) (McLean et al., Public Health Nutr 12(4):444–454, 2009), anemia affects 1.62 billion people
constituting 24.8% of the population and is considered the world’s second leading cause of illness. The Peripheral Blood
Smear (PBS) examination plays an important role in evaluating hematological disorders. Anemia is diagnosed using PBS.
Being the most powerful analytical tool, manual analysis approach is still in use even though it is tedious, prone to errors,
time-consuming and requires qualified laboratorians. It is evident that there is a need for an inexpensive, automatic and
robust technique to detect RBC disorders from PBS. Automation of PBS analysis is very active field of research that moti-
vated many research groups to develop methods using image processing. In this paper, we present a review of the methods
used to analyze the characteristics of RBC from PBS images using image processing techniques. We have categorized these
methods into three groups based on approaches such as RBC segmentation, RBC classification and detection of anemia, and
classification of anemia. The outcome of this review has been presented as a list of observations.

Keywords Peripheral blood smear · Red blood cells · Image processing · Computer-aided system · Anemia diagnosis

1 Introduction morphology is a key tool for hematologists to recommend

appropriate clinical and laboratory follow-up and to select the
Anemia is a condition described by insufficient red blood cells best tests for definitive diagnosis. Anemia analysis can be done
or based on hemoglobin content in the blood below a specific based on RBC morphology and clinical parameters. Morpho-
range estimated for specific sex and age of a person. Anemia logical analysis using blood smear is performed by spreading
is diagnosed using PBS where microscopic examination of a drop of blood thinly onto a glass slide and stained with col-
blood smear provides useful information about alteration of oring agents such as Giemsa, Leishman, and Wright-Giemsa
RBC shape and size or presence of any inclusion bodies. RBC and examined under a microscope by a qualified lab technician
[93]. The blood smear contains different types of cells, namely
White Blood Cells (WBCs), RBCs and platelets. An image of
* Keerthana Prasad
keerthana.prasad@manipal.edu PBS indicating different blood cells is shown in Fig. 1.
It can be observed that RBCs are more in number in com-
Navya K.T.
navya.kt@manipal.edu parison with WBCs and platelets. During this examination
of the smear, the pathologists assess the size, shape, and
Brij Mohan Kumar Singh
brij.singh@manipal.edu color of the RBCs and WBCs. Also, they estimate the num-
ber of platelets present. The quality of RBC is character-
1
Manipal Institute of Technology, Manipal Academy ized by red cell indices and any deviation in size, volume,
of Higher Education, Manipal, India 576104 or shape of red cells represents an abnormal red blood cell
2
Manipal School of Information Sciences, Manipal Academy [78]. Anemia is classified based on the morphology of red
of Higher Education, Manipal, India cells, red cell indices and hemoglobin content as in Fig. 2.
3
Department of Pathology, Kasturba Medical College, Anemia is classified into hypochromic microcytic, nor-
Manipal Academy of Higher Education, Manipal, mochromic normocytic and macrocytic anemia. Further, it
India 576104

13
Vol.:(0123456789)
2446 Medical & Biological Engineering & Computing (2022) 60:2445–2462

Fig. 1  Microscopic view of blood smear image [95]

is categorized into Iron Deficiency Anemia (IDA), Sickle

Cell Anemia (SCA), thalassemia, Hereditary Spherocytosis
(HS), Hereditary Elliptocytosis (HE), aplastic anemia and
Hemolytic Anemia (HA) based on the RBC morphology.
Based on the morphology, types of normal and abnormal
red blood cells are shown in Fig. 3.
Anemia classification can also be performed based on the
clinical parameters such as RBC count, RBC indices, namely
Mean Corpuscular or Cell Volume (MCV in femtoliter),
Mean Cellular Hemoglobin Concentration (MCHC),
Mean Cell Hemoglobin Content (MCH in picograms), Fig. 3  Normal and abnormal RBCs [95]
hematocrit (HCT) or Packed Cell Volume (PCV) and Red
Cell Distribution Width (RDW). These parameters play an
important role in the detection and classification of anemia. counting, RBC classification and detection of anemia, and
Hematologists usually examine PBS if RBC indices are anemia classification.
abnormal [58]. The morphological classification of anemia
based on the clinical diagnosis is as shown in Table 1.
This paper presents a comprehensive review of automa- 2.1 RBC segmentation and counting
tion of PBS images for detection of anemia. Many research
groups have attempted this automation based on clinical or In this section, we provide information about RBC seg-
morphological analysis. mentation and counting using image processing methods
based on color and size variations. The segmentation tech-
niques are categorized into different sub-sections based on
2 Approaches the approaches.

This survey summarizes the various research works

involved in the automation of analysis of the PBS images.
The approaches are categorized as RBC segmentation and

Fig. 2  Classification of anemia

13
Medical & Biological Engineering & Computing (2022) 60:2445–2462 2447

Table 1  Classification of anemia based on the clinical parameters on RBC segmentation method using masking and watershed
Anemia Type MCV fL MCH pg MCHc%
algorithm for 20 images with 40X magnification. However,
this method needs improvement in segmentation for large
Macrocytic anemia > 100 > 32 32–35 overlap. Biswas et al. [43] proposed blood cell segmenta-
Normocytic anemia 80–100 27–32 32–35 tion method using Watershed Transform (WT) [118, 144]
Microcytic anemia < 80 < 27 < 32 and Sobel filter in the spatial frequency domain [60, 126]
and obtained 93% accuracy for 30 images measured using
a structure similarity index matrix. Habibzadeh et al. [86]
2.1.1 Thresholding and transform‑based segmentation proposed a method for WBC and RBC segmentation using
methods YIQ color space and WT and 90% accuracy was achieved for
RBC segmentation using 10 images. However, they reported
Thresholding is the simplest way of segmenting an image addressing large variations of blood cells and low quality
into foreground and background based on the different images in their future work. Cruz et al. [57] proposed RBC
intensities or colors. Transform methods are used to iden- counting method using blob analysis based on HSV com-
tify the features in the other domains. Prasad et al. [124] ponent and WT and obtained an average accuracy of 95.6%
developed a decision support system to detect malaria for 10 blood samples taken with 40X and 100X magnifica-
parasites in thin PBS images using color image analysis. tions. Segmentation of RBCs from PBS images using Hough
Morphological operations were used to detect RBCs and Transform (HT) was reported by many research groups [14,
color image processing techniques to extract the region of 72, 83, 110, 113, 149, 155]. They reported the accuracy in
interest. This method could detect around 96% of the para- the range of 94–96%. Mahamood et al. [104, 105] proposed
sites for 200 Giemsa stained images of 100X magnification color based blood cell segmentation in CIELAB color space
under uniform stain and illumination conditions [58] [90, and used CHT for cell extraction. The experiment was per-
125]. Bhavnani et al. [40] proposed a method to segment formed on ALL-IDB dataset with 108 Wright stain images
and count RBCs and WBCs using Otsu thresholding and of magnification ranging from 300X to 500X and obtained
morphological operations. WBC counting was performed by the average accuracy of 81% for WBCs and 64% for RBCs.
counting number of connected components and obtained an Sarrafzadeh et al. [139] presented a circlet transform-based
average accuracy of 94.25%. RBC counting was performed method to count RBCs and obtained a low error rate for
using Watershed segmentation and Circular Hough Trans- 100 images with 100X magnification. However, the authors
form (CHT) and accuracies of 92.67% and 91.07% were suggested to improve initial RBC mask for accurate seg-
obtained respectively. The principal objective of water- mentation. Yeldhos et al. [161] implemented FPGA based
shed segmentation [87] is to find the watershed lines which embedded system for counting RBC using CHT. YCbCr
forms continuous path giving rise to continuous bounda- color conversion and WT segmentation method were used.
ries between the regions. It extracts nearly uniform objects An accuracy of 90.98% was achieved for 108 blood smear
from the background. CHT is an image transform [53] that images of ALL-IDB dataset. Frejlichowski [80] proposed
extracts circular objects from an image. The transform can a method to detect RBCs based on pixel relationship and
measure radius and the centroid of each circular object in obtained 83% accuracy for 700 RBCs from May-Grunwald-
an image by searching a 3D Hough space. Maji et al. [107] Giemsa (MGG) stained images. Alomari et al. [28] proposed
presented RBC counting method using Otsu thresholding an iterative structured circle method to detect WBCs and
and mathematical morphology and classified into circular, RBCs and obtained average accuracy of 95.3% for RBCs and
non-circular, overlapped cells or artifacts. The average accu- 98.4% for WBCs from 100 images of different magnifica-
racy obtained was 96.9% for circular and 97.1% for non- tions ranging from 300 to 500X.
circular cells from 146 images. Ruberto et al. [65] proposed
a method for malarial parasite-infected blood cells using 2.1.2 Edge based segmentation methods
HSV component based on color similarity and Watershed
algorithm for 12 Giemsa stained images acquired at differ- Das et al. [60] proposed a method to identify RBCs and
ent magnifications with some variations in stain and lighting different types of WBCs using edge detection algorithms,
conditions. Ruberto et al. [66] proposed a method based on namely Canny, Laplacian of Gaussian (LOG), Sobel and
region proposal using edge boxes for detecting and quantify- obtained 85% accuracy for 20 images. Poomcokrak et al.
ing RBCs and obtained accuracy in the range of 96–98% for [123] proposed Canny edge algorithm based RBC counting
180 ALL-IDB images. The research groups [67] presented method. The method obtained 74% accuracy for 59 RBCs
the same method for another malarial parasite MP-IDB data- and 59 non-RBCs using Multilayer Perceptron (MLP). MLP
base with 100 images and achieved accuracy in the range of is a simple feed forward neural network that uses back prop-
89–99%. Sharif et al. [144] presented a preliminary study agation algorithm to train neurons [89, 92]. It consists of an

13
2448 Medical & Biological Engineering & Computing (2022) 60:2445–2462

input layer, an arbitrary number of hidden layers and an out- Dice coefficient score of 0.96 was obtained for WBC seg-
put layer. The hidden layer processes the input information mentation from 155 images. Ritter et al. [133] proposed a
and transmits to the output layer. MLPs are often applied to blood cell segmentation method using a graph algorithm and
supervised learning problems. Backpropagation is used to obtained a success rate of 90% for 98 images. However, due
adjust weights and biases to minimize the error. Hafiz et al. to the diffuse area, this method failed to detect all platelets.
[19] proposed RBC segmentation algorithm using boundary- Cai et al. [47] presented an RBC segmentation method based
based thresholding and Canny edge detection methods and on an active appearance model incorporating shape and tex-
obtained average accuracy of 87.9% for five images from ture information of the cell.
the Broad Bioimage Benchmark Collection (BBBC) dataset.
2.1.5 Machine learning‑based segmentation methods
2.1.3 Clustering based segmentation methods
Sadafi et al. [135] presented a fully convolutional neural
Abbas et al. [11] presented a method to segment blood cells network-based RBC segmentation method and obtained
using the YCbCr color space and K-means clustering method 90% accuracy for 5772 raw images of different stains. In
[45] for 90 Giemsa stained blood smear images. Blood cells this work, the authors suggested to use post-processing
were easily identified by this method using a unique color methods for touching cell split up to improve the accuracy.
of every component. Wei et al. [157] proposed a method Kimbahune et al. [98] and Jun et al. [108] proposed blood
to detect and count overlapped RBCs in microscopic blood cell image segmentation and counting method using Pulse-
smear images. The H and S components were used to dif- Coupled Neural Network and found that the method is time
ferentiate between WBCs and segmented RBCs. H and S efficient. A machine learning approach based on the YOLO
components are closely related to the way humans feel color. algorithm was presented by Alam et al. [20] to identify and
H is the color sensed due to the wavelength. S indicates the count blood cells and obtained an accuracy of 96.09% for
purity of the color [126]. Watershed and K-means clustering RBCs and 86.89% for WBCs from 364 100x magnified
algorithms were applied for segmentation. An accuracy of annotated images of Blood Cell Count Dataset (BCCD).
92.9% was obtained for 100 Wright-Giemsa stained images. Adagale et al. [16] proposed an overlapped RBC count-
However, the authors of this paper suggested to fine tune the ing algorithm using Pulse Coupled Neural Network with
segmentation method for robustness. Acharya et al. [15] pre- a template matching technique and obtained 90% aver-
sented a method to separate RBCs from other components age accuracy for 40 images. Chari et al. [51] presented
of blood using K-medoids and obtained 98% accuracy for a pilot study on the analysis of MGG stained normal
1000 Wright stained images. Savkare et al. [140] proposed images using S ­ honitTM artificial intelligence system. The
a method to segment blood cells using K-means cluster- extracted cells were classified using three different deep
ing algorithm and WT and obtained 95.5% accuracy for 78 neural network models based on images annotated by three
Giemsa stained microscopic images. However, they reported experts. The precision of 93.9% was achieved for all WBC
that if cells are not well-stained and have low contrast, this classes from 6000 WBCs. RBCs and platelets were iden-
method does not work well. Ruberto et al. [68] presented tified based on the estimation of indices for 100 images
a fuzzy set based optimal threshold selection approach for from every 100 samples and obtained estimation within
blood cell segmentation. The local threshold was set using 10% reported value of Sysmex XN 3­ 000TM hematology
a histogram and average accuracy of 98% was obtained with analyzer. Loddo et al. [101] proposed a blood cell count-
a computation time less than a second. ing method using nearest neighbor and SVM techniques.
This method used ALL-IDB dataset with 368 images and
2.1.4 Contour and matching based segmentation methods obtained an average accuracy of 99.2% for WBCs and 98%
for RBCs. Tran et al. [150, 151] presented deep learning
Bronkorsta et al. [46] proposed a parametric deformable semantic segmentation method for RBC and WBC seg-
template-based online detection method to detect RBC mentation and counting. It is pixel level segmentation of
shapes of 900 cells in 100X magnification and obtained the image. An experiment was conducted on 42 ALL-IDB
accuracy of 95.7% for 10 images. This technique is based database images with 380 training images post augmenta-
on the prior knowledge about the shape and appearance of tion. SegNet architecture was utilized to segment blood
the object. A template prototype and according energy func- cells by labeling each pixel. It is a deep architecture for
tion is defined for template description. However, a good the segmentation of multi-class based on assigning each
initial guess for the shape, size, and location of the object is pixel of an image into a corresponding class [38]. The seg-
needed to find global minimum in this method. Bergen et al. mentation accuracy for WBCs, RBCs and the background
[39] proposed a method for WBC and RBC segmentation reached 94.93%, 91.11% and 87.32%, respectively. For cell
using template matching and level set algorithm [141]. A counting, Euclidean distance transform and binary dilation

13
Medical & Biological Engineering & Computing (2022) 60:2445–2462 2449

were used and accuracy of 93.3% for RBCs and 97.29%

for WBCs was obtained. Shahzad et al. [143] designed
semantic segmentation using a convolutional encoder-
decoder framework along with VGG16 network and the
model was trained and tested on the ALL-IDB dataset with
108 images. The proposed system achieved accuracies of
97.45%, 93.34%, and 85.11% for RBCs, WBCs, and plate-
lets respectively. Amin et al. [33] presented a comparison
of different classification algorithms using WEKA tool for
hematological data. The experiment was conducted on two
datasets from a total of 900 samples with CBC parameters.
Three data mining classifiers were tried, namely J48 deci-
sion tree, MLP, and Naive Bayes, using which accuracies
of 97.2%, 86.6% and 70% were achieved respectively.

2.1.6 Miscellaneous category

Gupta et al. [84] identified RBCs using blob detection

method and obtained 75% accuracy for 88 RBCs. However,
a few RBCs were left undetected due to the lighting condi-
tions. Hidalgo et al. [81] proposed a novel method to count
the number of circular and elongated RBCs using circum-
ference and ellipse adjustment algorithms for 66 Giemsa
stained images from erythrocytesIDB database. They used
Fig. 4  Distribution of blood cell segmentation methods
k-curvature for separating clustered RBCs and obtained 98%
accuracy without pre-processing steps. Hegde et al. [90] pre-
sented a review on WBC, RBC, platelet analysis techniques
2.2.1 Shape feature and region based RBC classification
and highlighted the importance of illumination and color
methods
shade variation correction to develop a robust system for
PBS analysis.
Wheeless et al. [158] presented a method to classify RBCs
A lot of work has been carried out to segment blood cells
into normal, sickle or other abnormal cell using recursive
from PBS images. The distribution of segmentation tech-
partitioning and form factor. The recursive partitioning tech-
niques used in the literature is depicted in Fig. 4.
nique is based on the concept of finding the cutpoints for the
It is observed from the distribution that most of the litera-
features that best isolate different disease cases. The data are
ture used transform method, color thresholding and machine
then divided according to these cutpoints [44]. Form factor
learning techniques for the segmentation. The summary of
provides a measure of circularity as given in equation. If
the literature described in Section 2.1 highlighting results of
more is the departure from the perfect circle value 1, lower
each method which is listed in Table 2.
is the form factor value.

FormFactor = (4𝜋Area∕Perimeter)2 (1)

2.2 RBC classification and detection of anemia
An accuracy of 85% for normal RBCs, 83% for abnormal
In this section, we provide the details of image processing cells, and 81% for sickle cells from 3878 cell images was
techniques used to classify RBCs based on shape, size and obtained. Safca et al. [136] proposed a method to classify
texture variations in order to detect anemia. RBCs into sickle cells, echinocytes and elliptocytes
Classification of RBCs into normal and abnormal was using morphological operations and shape features such
presented using image processing techniques [18, 23, 24, 54, as diameter, area and perimeter [13]. Morphological
69, 99, 112, 121, 127, 147, 148, 154]. Various methods such imaging operations done on a binary image to remove
as Otsu thresholding, CHT, statistical and moment invari- small objects, fill the cell holes and clearing the border to
ants, and geometric texture features were used. The accura- avoid edge touching cells [147]. An average accuracy of
cies in the range of 83–94% were obtained using ANN, SVM 96% was achieved for 34 images. Deb et al. [64] proposed
and BPNN classifiers. an algorithm to classify RBCs using aspect ratio and

13
2450 Medical & Biological Engineering & Computing (2022) 60:2445–2462

Table 2  RBC segmentation and counting methods

Methods No. of images(Stain) Accuracy(%) Remarks Ref.

K-means clustering, WT 60 (Giemsa )100 (Wright– 93–98.9 Robustness is not explained [11, 140, 157]
Giemsa)
Iterative structured circle 100 95.3 Incorrect hole filling leads [28, 143]
detection, circlet transform to errors To improve initial
RBCs mask for accurate
segmentation
Graph algorithm 98 99 Considered only non-over- [133]
lapped cells
Parametric template matching, 900 cells 90–95.7 Require prior knowledge [16, 46, 98]
PCNN about the appearance of the
cell
YOLO algorithm 364 96.1 Satisfactory performance [20]
HSV conversion, morphologi- 200 (Giemsa) 96 Used uniform staining and [125]
cal operations illumination
Pixel relationship 10 (MGG) 83 Occluded objects are rejected [80]
before the later stages
Canny, LOG, Sobel 20–30 85–93 Normal RBCs Less samples [43, 60]
K-curvature, circumference 66 98 Images are not preprocessed [81]
and ellipse adjustments to reduce execution time
Blob analysis, WT 10 blood samples 90–96 Need optimization to get [57, 86]
accurate results
CNN AlexNet 5772 90 Average execution time was [135]
227 ms
Canny edge, MLP 59 RBCs and 59 non RBCs 74–88 Increase training images [123] [19]
K-medoids, distance transform 1000 (Wright) 98 Processing of central pallor of [15]
RBCs consume more time
HT 500 subjects 91–94.9 Many tunable parameters [72, 110, 148, 155]
Deep neural network models 100 (MGG) Indices lie within the 10% Considered only normal blood [51]
of Sysmex reported smear images
value
CHT, NN, SVM 368 98 Achieved low false negative [101]
rate
LAB, YCbCr color space, 108(Wright) 81–91 Computational time is more [105] [161]
CHT
Region proposal 180 (Wright) 96-98 Tested on ALL-IDB and MP- [65]
IDB datasets
Semantic segmentation 108 (Wright) 91–97 More labeled images are [143, 150, 151]
required

Fourier descriptors and obtained average accuracy of 92% A recognition rate of 93% was obtained for 55 MGG
for 33 images. The authors also presented a method to stained images. Arnau et al. [82] presented a method for
count NRBCs and WBCs based on the roundness factor. RBC classification using an active contour segmentation.
Rezatofighi et al. [132] proposed RBC detection method This method classified RBCs into normal, sickle cells and
using polar transformation and run-length matrix and other cell deformations and obtained 95% accuracy for
a true positive rate of 97.73% was obtained for 22 blood 45 images. Aruna et al. [34] proposed a method to detect
smears. However, objects with large size variations failed sickle cells using Canny Edge, LOG, Prewitt, Robert and
to be detected by this method. Soltanzadeh et al. [146] Sobel operators and found that the Canny edge detection
presented a method to classify three types of RBCs using method was preferable. Rakshit et al. [128] presented sickle
morphological methods. The method obtained 98.63% for cell detection method using Sobel edge detector and region
elliptocyte, 96.7% for discocyte and 95.36% accuracy for properties and obtained an overall accuracy of 95.8%.
echinocyte recognition for 200 images based on Euclidean Ahmadzadeh et al. [17] presented a method to group RBCs
distance. Frejlichowski [79] proposed RBC classification into three clusters (biconcave, stomatocyte, and sphero-
method using template matching and Fourier transform. echinocyte) using K-medoids, and K-means clustering and

13
Medical & Biological Engineering & Computing (2022) 60:2445–2462 2451

obtained 95% accuracy for Digital Holographic Microscopic cells. An accuracy of 91% was achieved by this method for
(DHM) images. Chandrasiri et al. [48, 49] presented an a dataset consisting of 200 normal and 200 abnormal single
algorithm to identify four common types of anemia, namely cells. Rodrigues et al. [134] proposed a method to classify
elliptocytes, microcytes, macrocytes and spherocytes, using RBCs into normal, sickle cells, and erythrocytes with other
HT and morphological operations. The researchers could deformations using morphological properties and obtained
obtain an accuracy in the range of 91–97% based on cell 94.6% accuracy using SVM classifier for 626 images from
features for 40 images. the erythrocytesIDB dataset. They used ANOVA for fea-
ture selection and suggested to study unsupervised methods
2.2.2 Machine learning‑based RBC classification methods to identify the patterns in cells. Hirimutugoda et al. [91]
presented a method to detect malarial parasites and thalas-
Bhowmick et al. [41] presented classification of RBCs in semia using 3-layered ANN for 200 Giemsa images of
scanning electron microscopic images using a Marker-con- each case and obtained 86.54% correct recognition rate by
trolled watershed segmentation method. It is combinational defining ROIs. Aliyu et al. [26] presented RBC classifica-
approach of edge-based segmentation and morphological tion method using SVM and obtained 100% accuracy for
operation methods that uses markers on some set of norms. normal, acanthocyte, teardrop cells and 73% for elliptocyte
A marker is a connected component that can easily segment and 90% for sickle cells using SVM and 33% using deep
boundaries from an image. With this approach, the regional learning for 250 images. They reported that the SVM clas-
minimal values occur only at marked locations [117]. The sifier outperformed DL due to limited datasets. A research
authors projected both structural and textural feature clas- group [24] also proposed a method to detect sickle cell
sification in this work and obtained an accuracy of 88.99% using Otsu thresholding and shape features and obtained
for 132 anemic blood samples using Bayesian classifier. 88% accuracy for 30 Giemsa stained images. Dalvi et al.
Bayesian approach classifies the new instance by assigning [59] proposed a method to classify RBCs into four abnormal
the most possible target value, given the attribute values types, namely elliptocyte, echinocyte, teardrop and macro-
that represent the instance on the principle of Bayes’ Theo- cyte, using thirteen geometric features and achieved better
rem [122]. Das et al. [61] proposed a method for RBC char- accuracy using ANN than the decision tree. The accuracy
acterization in anemia using Marker-controlled watershed obtained was 96.04% for RBC counting and 90.54% for RBC
segmentation and morphological features. The algorithm classification. Razzak et al. [130] presented contour aware
classified five different types of RBCs such as elliptocyte, segmentation method based on CNN and extreme learn-
echinocyte, acanthocyte, sickle cell and teardrop cell in ane- ing. The experiment was conducted on 64,000 blood cells
mia and obtained an accuracy of 86.87% for 715 abnormal from ALL-IDB database. RBCs and WBCs were segregated
and 290 normal RBCs using logistic regression classifier. based on the color intensity features and were cropped to
A method to recognize abnormal RBC shapes such as tear- extract features using CNN and given for ELM for subtypes
drop, echinocyte and elliptocyte using Hu’s moments for 300 classification. The segmentation accuracy of 98.12% and
anemia and 100 Leishman stained images of thalassemia 98.16% and classification accuracy of 94.71% and 98.68%
cases was proposed by [62]. Elsalamony [74, 75] proposed was achieved for RBCs and WBCs respectively. Mundhra
a geometrical shape signature method to detect sickle and ­ honitTM system to
et al. [116] proposed deep learning-based S
elliptocytosis using CHT and watershed segmentation and localize and classify blood cells. U-net deep learning archi-
100% accuracy was achieved for 30 images. Elsalamony tecture was used to localize WBCs and platelets from 300
[76] proposed benign and distorted cell detection methods MGG and Leishman stained training images. Otsu thresh-
using HT and WT and obtained 96.9% accuracy using NN olding in the green channel was used to identify RBCs and
and 92.9% using Classification and Regression (C&R) tree clumped cells were rejected. CNN architecture was used to
for 180 cells from 45 images and reported that NN was classify WBCs and RBCs based on size and shape. The sen-
preferred over C&R tree to detect sickle cells. In another sitivity for WBC extraction was 99.5%. The sensitivity and
paper, Elsalamony [77] used Self-Organising Map (SOM) specificity of identification for the common cell types were
along with the above mentioned methods and reported that above 91% and 98% respectively. Alom et al. [27] presented
SOM does not require any target variables but gets slower deep learning-based inception recurrent residual convolu-
in training the neurons. A neural network-based algorithm tional neural network for WBC and RBC classification. The
was proposed by Kim et al. [97] to distinguish abnormali- recognition accuracies of 100% for 352 WBC images and
ties in RBCs and WBCs using Principal Component Analy- 99.94% for 3737 RBC images were achieved. It is mentioned
sis (PCA) and obtained 91% average recognition rate for that the model requires a large number of network param-
RBCs in classifying 12 classes from 680 RBCs. Lee et al. eters. Durant et al. [71] proposed a method for RBC clas-
[100] proposed RBC classification method using a hybrid sification based on morphology using CNN for 10 classes.
neural network and identified sickle, horn and elliptocyte Around 3737, 100X magnified labeled cells were used and

13
2452 Medical & Biological Engineering & Computing (2022) 60:2445–2462

correct classification frequency of 90.60% was achieved. The Figure 5 shows the distribution of RBC classification
researchers reported that the distribution of labels for cell methods used in the related works to detect anemia. We
types was not homogeneous. observe from the distribution that, majority of the research-
ers used shape features and machine learning to classify
2.2.3 Clinical parameters based RBC classification methods RBCs. A brief overview of the RBC classification and ane-
mia detection methods are listed in Table 3.
Zahir et al. [162] presented an ANN-based method to detect
RBC disorders anemia and polycythemia using Hb value, 2.3 Classification of anemia
MCH and RBC count and obtained significant results for more
than 90% of the 1000 blood samples with training time less This section provides details of image processing methods
than 15 minutes. Bacus et al. [36, 37] presented RBC classi- used to classify anemia based on the morphology and hema-
fication method by extracting the features and obtained cor- tological parameters of RBCs.
relation coefficient of 0.965 for 100 cells from 4 specimens.
Red cell indices along with the red cell differential counts 2.3.1 Morphology based anemia classification methods
were considered in this work. Maity et al. [106] presented a
method to generate an anemia diagnosis report based on the Chen et al. [52] presented a method to classify hemolytic
CBC report and RBC morphology using red cell indices and anemia based on differential value and variation of chain
shape features. A precision of 98.2% was achieved in clas- codes in eight directions and irregularity of erythrocytes.
sifying microcytic, macrocytic, sickle, teardrop, elliptocyte, Accuracy in the range of 95–97% was achieved for 24 micro-
and normal cells from 1500 Leishman blood smear images. scopic images using Bayes classifier, logistic model trees and

Table 3  RBC classification and anemia detection methods

Methods No. of images (Stain) Performance metric Remarks Ref.

CHT, Heywood circularity 150–1000 samples 80–99% accuracy for normal & Lacks robustness [18, 69,
factor, ANN, moment invari- abnormal RBCs 147,
ants, inclusion-tree structure, 149,
BPNN, PCA, SVM 154]
Morphological properties, 626 94.6–96% accuracy for normal Consider unsupervised classi- [128, 134]
Naive Bayes, K-NN, SVM, and sickle cells fiers for more RBC patterns
Sobel edge
CHT, WT, NN, decision tree, 30–45 (Giemsa) 97–100% accuracy for sickle Geometrical shape signature is [74–76]
SOM, SVM and elliptocytosis used for detection process
Recursive partitioning, form 3878 cells 85% for discocytes, 83% for Form factor invariant to cell [121, 158]
factor abnormal cells and 81% for size and provides useful
sickle cells information on cell shape
Hybrid neural network 200 normal and 200 abnormal 91% accuracy for sickle, horn Considered only convexity [100]
cells and elliptocytes index feature
DL, SVM 105 normal and 250 abnormal Normal—100%, achanto- SVM classifier outperformed [23, 26]
cyte—100%, sickle cell— DL
90%, teardrop—100% and
elliptocyte—73% accuracy
using SVM
Rolling ball background, shape 1500 (Leishman) 98.2% precision for microcytic, Decision from CBC test [106]
features, Naive Bayes, Bayes- macrocytic, sickle, teardrop, measures is semi-automatic
ian classifier elliptocyte operation
ANN 1000 blood samples Less computational time Used RBG values—from Hb, [162]
MCH and RBC count
CNN , ELM 64,000 blood cells 94.71% accuracy Images from multiple sources [130]
are used
U-Net 300 (MGG) and (Leishman) 91% sensitivity and 98% Results are shown for a variety [116]
specificity of smear and stain
Inception recurrent residual 352 WBCs and 3737 RBCs 100% for WBC and 99.94% Model require larger number of [27]
CNN accuracy for RBC network parameters
CNN 3737 labeled Cells 90.6% accuracy for 10 RBC Label distribution was not [71]
classes homogeneous

13
Medical & Biological Engineering & Computing (2022) 60:2445–2462 2453

using CHT and WT for 233 cells and proved that CHT is
better than WT. Elsalamony [73] proposed a method to diag-
nose SCA using HT and obtained segmentation accuracy of
99.98%. This method also achieved a classification accuracy
of 96.9% and 92.9% using NN and C&R tree respectively.
Lotfi et al. [102] presented a technique for the automatic
detection of IDA by identifying three types of abnormal
red cells using region and Fourier descriptors. This method
obtained accuracies of 99%, 97% and 100% for dacrocytes,
elliptocytes and schistocyte cells respectively for 100 cells
of each case using NN, SVM and KNN classifier. Tyagi et al.
[152] presented a method to detect poikilocyte cells in IDA
using GLCM features and moment invariants and obtained
accuracy in the range of 75–81% for 100 images using ANN.
Several methods have been proposed to detect the sickle-
shaped RBC in SCD patients by [21, 22, 29, 55, 56, 88, 114,
115, 120, 131, 159, 160] using methods such as random
walk, Sobel edge, geometric features, Fuzzy C means clus-
tering, LOG, WT, HT and morphological filters. The average
accuracy reported was in the range of 85–95%. Alzubaidi
et al. [30] developed deep learning models for SCA diag-
nosis and achieved an accuracy of 99.54%. The research-
ers proposed three CNN models with different layers and
filters and used data from erythrocytesIDB, ALL-IDB and
other internet sources. The extracted features were used for
training multi-class SVM and accuracy around 98–99% was
obtained. Zhang et al. [163] proposed a method to segment
subtypes of RBCs in sickle cell disease using deformable
U-net on 266 raw microscopy images and obtained 99.12%
Fig. 5  Distribution of RBC classification methods
accuracy. This method could segment blurred, clustered and
heterogeneous shaped RBCs and performed better than base-
rule based classifiers. Nithyaa et al. [119] proposed a method line U-net. Aliyu et al. [25] proposed the Alexnet deep learn-
to detect various blood disorders such as malaria, elephantia- ing model for the classification of RBCs in SCA. Around
sis, trypanosomiasis, SCA and polycythemia using statisti- 750 single RBCs from Giemsa stained blood smears were
cal features and Euclidean distance for 40 images. Azam acquired for the experiment and classification accuracy of
et al. [35] presented a method to detect seven RBC types of 95.92% was obtained. They reported low specificity due to
anemic diseases using shape descriptors and obtained 92% less normal cells. Haan et al. [85] presented a deep learning
accuracy for 100 instances using MLP and random forest framework based screening of sickle cells using a smart-
classifier. Tyas et al. [153] presented a semi-automated phone microscope. U-net architecture for image normaliza-
algorithm to classify four types of abnormal RBCs such tion and enhancement network and semantic segmentation
as teardrop, acanthocytes, sickle cell and target cell using for sickle cells were used and approximately 98% accuracy
GLCM features in the minor thalassemia cases. The ROI was was achieved from 96 unique patient samples. Das et al. [63]
selected manually in this method and accuracies of 93.22% presented an overview of enhancement, segmentation and
and 92.55% were obtained using BPNN and CNN respec- classification techniques used for SCA detection. The review
tively for 256 images. Various methods [109, 129] have been also highlights clinical uses, hardware implementation and
used to detect thalassemia using K-means clustering, active future scope for the analysis of SCD.
contour, neural network, decision tree and an average accu-
racy in the range 82–95% was reported. Sharma et al. [145] 2.3.2 Hematological parameters based anemia
proposed a method to detect SCA and thalassemia using classification
Marker-controlled watershed segmentation and geometric
features and accuracy of 80.6% was obtained for 100 images Birndorf et al. [42] presented ANN-based hybrid system to
with KNN classifier. Fadhel et al. [13] proposed an algo- evaluate microcytic anemia such as IDA, hemoglobinopathy
rithm to count normal and abnormal RBCs in the SCA slide and anemia of chronic disease using HCT, MCV, RDW and

13
2454 Medical & Biological Engineering & Computing (2022) 60:2445–2462

obtained 96.5% accuracy for 473 cases of microcytic anemia study in IDA detection. Also, there is a need for other ane-
and anemia of chronic disease. Dogan et al. [70] proposed mia subtypes detection and diagnosis in peripheral blood
IDA detection method based on hematology parameters, smear analysis.
namely Serum iron and total iron-binding capacity, using
decision trees for 96 patients and the results got perfectly
matched with the physician’s decisions. Lund et al. [103]
presented an algorithm to classify microcytic and macro- 3 Database
cytic anemia using image analysis techniques based on MCV
and RBC size and obtained 95% accuracy for 4000 cells. Many researchers have used proprietary datasets with
Sanap et al. [137] proposed an anemia classification method blood smear images of different stains. A few publicly
based on CBC reports using C4.5 decision tree algorithm available databases are used for the performance analysis
and SVM using WEKA tool with 514 instances and obtained of the developed algorithm. An overview of these data-
accuracy of 99.42% and 88.13% respectively. Abdullah et al. bases are given in Table 5.
[12] presented anemia types prediction method based on BCCD is a small-scale publicly available dataset [4]
CBC reports using data mining techniques using WEKA that has 364 annotated images for blood cell detection
tool for 41 patients. The experiment showed that the J48 taken originally from cosmicad and akshaylamba open
decision tree performed better with 97% precision among sources. The erythrocytesIDB [6] contains 196 full field
Naive Bayes, MLP and SVM algorithms. Jaiswal et al. [94] and 629 individual Giemsa stained peripheral blood
presented anemia prediction method based on CBC reports smear images taken from SCD patients. ASH image bank
using supervised machine learning algorithms. This method [2] is a web-based image library that has a collection of
used eighteen attributes from 200 samples and obtained hematologic images consists of normal and abnormal
maximum accuracy of 96.09%. In this work, it was reported blood cells. However, studies have not explored available
that Naive Bayes outperformed C4.5 and random forest. images for anemia detection. The atlas of hematology
Khalaf et al. [96] presented machine learning approaches [3] provides normal and abnormal Leishman stained
for the classification of SCD dosage levels using 13 attrib- blood smear images for the morphological study of
utes from 1168 sample points. They concluded that the cells. Medical Image and Signal Processing (MISP)
random forest classifier performed overall better than RNN Research Center and Department of Pathology at Isfahan
and feedforward neural networks. Amendolia et al. [31, 32] University of Medical Sciences [9] contributed for the
presented ANN-based method to detect α and β thalassemia dataset consists of 148 microscopic blood smear images.
using hemochromic parameters. A specialized ANN was Public Health Image Library (PHIL) [8] contains a few
used in the method and accuracy of 94% was achieved for blood smear images created by the Centers for Disease
304 cases. Setsirichok et al. [142] proposed a method for Control and Prevention (CDC) for reference. The Broad
classifying thalassemia using Hb and MCV parameters and Bioimage Benchmark Collection (BBBC) [5] consists of
obtained an average accuracy in the range of 93–99% for publicly available image sets such as annotated biological
8054 clinical trial samples using C4.5 decision tree, Naive images for the analysis of algorithms. Telepathology
Bayes classifier and MLP. However, they mentioned that Hb 2012 [10] consists of webmicroscope to acquire malarial
parameter is redundant for the study. parasite data along with annotation tool. Leukocyte
Table 4 summarizes the anemia classification methods Images for Segmentation and Classification (LISC) [7]
used in literature. for identification of different WBCs with ground truth
It is evident from Fig. 6 that, most of the research groups for only 250 images. Acute Lymphoblastic Leukemia
used the traditional machine learning approach for ane- Image Database (ALL-IDB) [1] is a free, publicly
mia classification. It can also be observed that, there is an available dataset for the evaluation of segmentation and
increasing tendency towards the usage of deep learning clas- classification methods. There are two datasets specifically
sifier models. for lymphoblasts detection. ALL-IDB1 consists of 108
Occurrence of anemia worldwide according to WHO [50, blood smear images with labeled lymphocytes taken
111] and papers on classification of subtypes of anemia are with 300 to 500 microscope magnifications. ALL-IDB2
shown in Fig. 7. is a collection of 260 cropped normal and blast cells that
Out of many anemia subtypes, the frequency of SCA belongs to ALL-IDB1 dataset.
and IDA detection was high by the majority of the Even though a lot of work has been carried out on PBS
researchers which is depicted in Fig. 7. It can be observed images, annotations are not available for the publicly
from figure that occurrence of IDA globally is much higher available datasets. A comparison of work would not be
than SCA. However, number of studies seen in the litera- fair because ground truth depends on annotation done by
ture is not in the same proportion. There is scope for more individuals.

13
Medical & Biological Engineering & Computing (2022) 60:2445–2462 2455

Table 4  Anemia classification methods

Methods No. of images Accuracy (%) Remarks Ref.

GLCM, CNN 256 BPNN—93.2, CNN—92.6 for Sample size is less [153]
minor thalassemia case
SVM, KNN, MLP 304 records MLP—92 , SVM—83 sensitivity Using RBC, Hb, HCT, MCV [31, 32]
for thalassemia parameters
ANN 473 cases 96.5 for IDA, HA, ACD Using HCT, MCV, RDW [42]
Active contour, NN, DT 15 groups 82–93 for thalassaemia False-positive and false negative [?, 109]
errors are less than 1% and 2%
C4.5 DT, Naive Bayes classifier 8054 samples 99.4 for 18 classes of thalassaemia Using six Hb attributes and MCV [142]
and MLP
Marker-controlled Watershed 100 80.6 for SCA and thalassaemia Developed combined method [145]
segmentation, KNN
Fuzzy C means clustering, geo- 80 KNN—73.3, SVM—83.3, Fuzzy C means overcomes the [55, 56]
metrical and statistical features ELM—87.7 for SCD disadvantages of threshold
segmentation
HSI color space, K-means cluster- 60 94.6 for thalassemia Detected α, β thalassemia, [129]
ing β-thalassemia trait
ANN, GLCM features 100 75–81 for IDA Classified 4 types of poikilocytes [152]
CHT, marker-controlled WT, 8–20 91.1 for SCD CHT performed better, need [13] [21, 22,
LOG, Fuzzy thresholding improvement in de-noising 88, 114,
method 115]
CLAHE, MLP and random forest 100 instances 92 for IDA and HA Persistent results for any luminos- [35]
ity conditions
Deformable U-Net 266 raw 99.12 for SCD RBC Method could segment blurred, [163]
clustered, heterogeneous shaped
RBCs
Chain codes, Bayes classifier, 24 96.6 for HA HA is classified based on differen- [52]
logistic model trees and rules tial value of chain codes
classifier
Naive Bayes, C4.5 and random 200 samples 96.1 for anemia detection Used 18 attributes from CBC [94]
forest classifier reports
DL, multi-class SVM 100–250 99.5 for SCD Proposed three CNN models with [30]
different layers and filters
DL-Alexnet 750 single RBCs 95.9 for SCA Specificity was low due to less [25]
normal cells
U-net architecture, semantic seg- 96 unique samples 98 for SCD Developed smartphone microscope [85]
mentation

4 Discussion and future scope to accept the results. It is also observed that the number of
images in dataset is in the range from 100 to 1000. As deep
Diagnosis of anemia is challenging, particularly in inad- learning is becoming popular and many of classification
equate resource settings. Various state-of-the-art methods algorithms are supervised learning approaches, it is desir-
used in the literature for PBS analysis are mentioned in Sec- able to have large PBS datasets with annotations. It can be
tion 2 and some of them are listed in Tables 2, 3 and 4. In observed from Fig. 7 that the occurrence of IDA is much
this paper, we have provided a detailed report of the use higher than SCA and studies related to IDA is lesser. Hence,
of image processing methodologies to automate peripheral there is a need for more study of IDA and other subtypes of
blood smear analysis for diagnosing morphology-based RBC anemia. Figure 8 depicts the ways of classification of ane-
disorders. Also, we can notice from Fig. 6 that detection and mia. We can observe that around 88% of researchers used
classification of anemia using deep learning is increasing morphological parameters and only a few research groups
over traditional machine learning approaches. From Table 5, used hematology parameters for the anemia diagnosis.
it is evident that most of the research groups used a proprie- To detect anemia cases, mostly the shape and size of
tary dataset for the analysis of the algorithms. As researchers the abnormal RBCs are considered. The methods used by
used a publicly available datasets with different images and the research groups did not focus on other abnormalities of
annotations, a comparison of the algorithms is not possible blood cells.

13
2456 Medical & Biological Engineering & Computing (2022) 60:2445–2462

refers to the papers which are categorized into three divi-

sions depending on the purpose and approaches mentioned
in Section 2. All these research groups focused on develop-
ing computer-aided automated systems to reduce the task of
hematologists in analyzing the peripheral blood smears. A
major impediment of automated microscopic evaluation is
that they are affected due to imaging and staining variations.
An accurate identification of normal and abnormal cases is
essential to assist pathologists for further diagnosis. Images
are acquired from blood smear stained using a specific stain
with different microscopic settings and lab arrangements.
However, this poses many challenges for the automation of
blood smear images. For example, as mentioned in Section 2
and Table 2, the methods presented by various research-
ers for RBC segmentation and counting were implemented
on specific stained blood smear images under a controlled
environment. However, it is observed from the past studies
described in the listed papers that results vary due to lack of
Fig. 6  Application of traditional machine learning and deep learning
for anemia classification robustness in the methods. Although an automated decision
support system is developed to reduce the burden on hema-
tologists by eliminating manual inspection of blood smears,
It can be noticed that multiple methods have been imple- there is no integrated approach that has been developed to
mented to detect a few anemia cases like SCA, IDA and handle both standard and inconsistent microscopic blood
thalassemia that mainly depend on the input image taken smear images acquired from both the manual and automated
from different lab setups and conditions. The summary table workflow.

Fig. 7  Occurrence and classifi-

cation of anemia subtypes

Table 5  Outline of the publicly available databases

Database No. of images Annotations Studies

BCCD [4] 364 smear images Available for RBCs, WBCs and platelets [20]
erythrocytesIDB [6] 196 smear images and 629 Giemsa stained Available for sickle cells of 80 smear images [29, 81, 134]
single RBCs
ASH image bank [3] 2100 hematologic Leishman stained images Not available -
Isfahan MISP [9] 148 Not available [139]
PHIL [8] 100 Not available [14]
BBBC [5] 18 biological image sets Available for RBC’s only [19]
Telepathology 2012 [10] Malarial parasite images Tool available [156]
LISC [7] 400 Wright-Giemsa stained images Available for WBC’s from 250 images only [138]
ALL-IDB [1] 108 smear images and 260 cropped normal Available for WBC’s only [29, 65, 101, 104, 105,
and blast single cell images 130, 143, 150, 151,
161]

13
Medical & Biological Engineering & Computing (2022) 60:2445–2462 2457

Funding Open access funding provided by Manipal Academy of

Higher Education, Manipal

Declarations

Conflict of interest The authors declare no competing interests.

Open Access This article is licensed under a Creative Commons Attri-

bution 4.0 International License, which permits use, sharing, adapta-
tion, distribution and reproduction in any medium or format, as long
as you give appropriate credit to the original author(s) and the source,
provide a link to the Creative Commons licence, and indicate if changes
were made. The images or other third party material in this article are
included in the article's Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in
the article's Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a
copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.
Fig. 8  Ways of anemia classification

5 Conclusion References

This review provides a summary of the current develop- 1. ALL-IDB dataset. https://​homes.​di.​unimi.​it/​scotti/​all/
2. ASH image bank. http://​image​bank.​hemat​ology.​org/
ments in computerized PBS analysis using microscopic
3. Atlas of hematology. http://​hemat​ology​atlas.​com/​princ​ipalp​
blood smear images. The paper comprises of introduction age.​htm
to PBS analysis and anemia, an RBC disorder. Different 4. Blood cell count and detection. https://​github.​com/​Sheng​gan/​
sections have been summarized in this paper on existing BCCDD​ataset
5. Broad bioimage benchmark collection. http://​www.​broad​insti​
automated image processing methods for identification, seg-
tute.​org/​bbbc/
mentation, feature extraction and classification of RBCs for 6. ErythrocytesIDB dataset. http://​eryth​rocyt​esidb.​uib.​es/
further diagnosis of anemia. The literature on approaches of 7. LISC database. http://​users.​cecs.​anu.​edu.​au/​hreza​tofig​hi/​Data/​
the classification of anemia also has been included in this Leuko​cyte2​0Data.​htm
8. Public health image library (PHIL). http://​phil.​cdc.​gov/​phil/​
review. Although, manual microscopic evaluation is a gold
home.​asp
standard for PBS analysis, for quick and accurate diagnosis, 9. Red blood cells. https://​hrabb​ani.​site1​23.​me/​avail​able-​datas​ets/​
an automated decision system is essential to overcome the red-​blood-​cells
limitations of microscopic analysis. Hence, the observations 10. Telepathology 2012-webmicroscope. http://​fimm.​webmi​crosc​
ope.​net/​Resea​rch/​Momic/​tp2012
made during the process of the review are listed below.
11. Abbas N, Mohamad D, et al. (2013) Microscopic RGB color
images enhancement for blood cells segmentation in YCbCr
– To analyze RBC disorders rigorously, a large publicly color space for k-means clustering. J Theor Appl Inf Technol
available dataset with annotations of various types of 55(1):117–125
12. Abdullah M, Al-Asmari S (2017) Anemia types prediction
blood cells is needed.
based on data mining classification algorithms, communica-
– A robust system that can handle staining and imaging tion, management and information technology–Sampaio de
variations is desirable in PBS analysis for anemia detec- Alencar (ed.)
tion. 13. AbdulraheemFadhel M, Humaidi AJ, RazzaqOleiwi S (2017)
Image processing-based diagnosis of sickle cell anemia in eryth-
– A hybrid method which considers both morphological
rocytes. In: 2017 Annual conference on new trends in informa-
and clinical features would play an important role to tion & communications technology applications (NTICT), pp
improve the efficiency of classification of anemia sub- 203–207. IEEE
types. 14. Acharjee S, Chakrabartty S, Alam MI, Dey N, Santhi V, Ash-
our AS (2016) A semiautomated approach using GUI for the
detection of red blood cells. In: 2016 International conference on
An automated system plays a very important role in the PBS electrical, electronics, and optimization techniques (ICEEOT),
analysis. However, it is observed from the review that, there pp 525–529. IEEE
is a need for a large PBS publicly available database with 15. Acharya V, Kumar P (2018) Identification and red blood cell
automated counting from blood smear images using computer-
appropriate annotations and robust system to assist clinicians
aided system. Med Biol Eng Comput 56(3):483–489
for further diagnosis of disease.

13
2458 Medical & Biological Engineering & Computing (2022) 60:2445–2462

16. Adagale S, Pawar S (2013) Image segmentation using PCNN and 34. Aruna N, Hariharan S (2014) Edge detection of sickle cells in red
template matching for blood cell counting. In: 2013 IEEE Inter- blood cells. Int J Comput Sci Inform Technol 5(3):4140–4144
national conference on computational intelligence and computing 35. Azam B, Rahman S, Ullah S, Hanan F (2017) Detection of the
research, pp 1–5. IEEE top anemic diseases in blood smear images using image quantiza-
17. Ahmadzadeh E, Jaferzadeh K, Lee J, Moon I (2017) Automated tion followed by ensemble of classifiers. In: Proceedings of the
three-dimensional morphology-based clustering of human eryth- 2017 4th international conference on biomedical and bioinfor-
rocytes with regular shapes: stomatocytes, discocytes, and echi- matics engineering, pp 115–120
nocytes. J Biomed Opt 22(7):076015 36. Bacus J, Belanger M, Aggarwal R, Trobaugh Jr F (1976) Image
18. Akrimi JA, Suliman A, George LE, Ahmad AR (2014) Classifica- processing for automated erythrocyte classification. J Histochem
tion red blood cells using support vector machine. In: Proceed- Cytochem 24(1):195–201
ings of the 6th international conference on information technol- 37. Bacus J, Weens J (1977) An automated method of differential
ogy and multimedia, pp 265–269. IEEE red blood cell classification with application to the diagnosis of
19. Al-Hafiz F, Al-Megren S, Kurdi H (2018) Red blood cell seg- anemia. J Histochem Cytochem 25(7):614–632
mentation by thresholding and Canny detector. Procedia Comput 38. Badrinarayanan V, Kendall A, Cipolla R (2017) Segnet: a deep
Sci 141:327–334 convolutional encoder-decoder architecture for image segmenta-
20. Alam MM, Islam MT (2019) Machine learning approach of tion. IEEE Trans Pattern Anal Mach Intell 39(12):2481–2495
automatic identification and counting of blood cells. Healthcare 39. Bergen T, Steckhan D, Wittenberg T, Zerfass T (2008) Segmenta-
Technol Lett 6(4):103–108 tion of leukocytes and erythrocytes in blood smear images. In:
21. Albayrak B, Darici MB, Kiraci F, Ougrenci AS, Ozmen A, Ertez 2008 30th Annual international conference of the IEEE engineer-
K (2018) Sickle cell anemia detection. IEEE Institute of Electri- ing in medicine and biology society, pp 3075–3078. IEEE
cal and Electronics Engineers Inc 40. Bhavnani LA, Jaliya UK, Joshi MJ (2016) Segmentation and
22. Algailani H, Hamad MES (2018) Detection of sickle cell disease counting of WBCs and RBCs from microscopic blood sample
based on an improved Watershed segmentation. In: 2018 Interna- images. Int J Image Graph Signal Process 8(11):2016
tional conference on computer, control, electrical, and electronics 41. Bhowmick S, Das D, Maiti A, Chakraborty C (2013) Structural
engineering (ICCCEEE), pp 1–4. IEEE and textural classification of erythrocytes in anaemic cases: a
23. Aliyu HA, Razak MAA, Sudirman R (2019) Normal and abnor- scanning electron microscopic study. Micron 44:384–394
mal red blood cell recognition using image processing. Indone- 42. Birndorf NI, Pentecost JO, Coakley JR, Spackman KA (1996)
sian J Electr Eng Comput Sci 14(1):100–104 An expert system to diagnose anemia and report results directly
24. Aliyu HA, Razak MAA, Sudirman R (2019) Segmentation and on hematology forms. Comput Biomed Res 29(1):16–26
detection of sickle cell red blood image. In: AIP Conference pro- 43. Biswas S, Ghoshal D (2016) Blood cell detection using thresh-
ceedings, vol 2173, p 020004. AIP Publishing LLC olding estimation based watershed transformation with sobel
25. Aliyu HA, Razak MAA, Sudirman R, Ramli N (2020) A deep filter in frequency domain. Proced Comput Sci 89:651–657
learning AlexNet model for classification of red blood cells in 44. Breiman L, Friedman JH, Olshen RA, Stone CJ (1984) Classifi-
sickle cell anemia. Int J Artif Intell 9(2):221–228 cation and regression trees. Belmont, CA: Wadsworth. Int Group
26. Aliyu HA, Sudirman R, Razak MAA, Abd Wahab MA (2018) 432:151–166
Red blood cell classification: deep learning architecture versus 45. Brewka G (1996) Artificial intelligence—a modern approach by
support vector machine. In: 2018 2nd International conference Stuart Russell and Peter Norvig, Prentice Hall. Series in artificial
on biosignal analysis, processing and systems (ICBAPS), pp intelligence, Englewood Cliffs, NJ. Knowl Eng Rev 11(1):78–79
142–147. IEEE 46. Bronkorsta P, Reinders MJ, Hendriks EA, Grimbergen J,
27. Alom MZ, Yakopcic C, Taha TM, Asari VK (2018) Microscopic Heethaar RM, Brakenhoff G (2000) On-line detection of red
blood cell classification using inception recurrent residual con- blood cell shape using deformable templates. Pattern Recogn
volutional neural networks. In: NAECON 2018-IEEE national Lett 21(5):413–424
aerospace and electronics conference, pp 222–227. IEEE 47. Cai R, Wu Q, Zhang R, Fan L, Ruan C (2012) Red blood cell
28. Alomari YM, Abdullah S, Huda SN, Zaharatul Azma R, Omar segmentation using active appearance model. In: 2012 IEEE
K (2014) Automatic detection and quantification of WBCs and 11th International conference on signal processing, vol 3, pp
RBCs using iterative structured circle detection algorithm. Com- 1641–1644. IEEE
put Math Methods Med, 2014 48. Chandrasiri S, Samarasinghe P (2014) Automatic anemia identi-
29. Alzubaidi L, Fadhel MA, Al-Shamma O, Zhang J (2018) Robust fication through morphological image processing. In: 7th Inter-
and efficient approach to diagnose sickle cell anemia in blood. national conference on information and automation for sustain-
In: International conference on intelligent systems design and ability, pp 1–5. IEEE
applications, pp 560–570. Springer 49. Chandrasiri S, Samarasinghe P (2014) Morphology based auto-
30. Alzubaidi L, Fadhel MA, Al-Shamma O, Zhang J, Duan Y (2020) matic disease analysis through evaluation of red blood cells. In:
Deep learning models for classification of red blood cells in 2014 5th International conference on intelligent systems, model-
microscopy images to aid in sickle cell anemia diagnosis. Elec- ling and simulation, pp 318–323. IEEE
tronics 9(3):427 50. Chaparro CM, Suchdev PS (2019) Anemia epidemiology, patho-
31. Amendolia SR, Brunetti A, Carta P, Cossu G, Ganadu M, Golo- physiology, and etiology in low-and middle-income countries.
sio B, Mura GM, Pirastru MG (2002) A real-time classification Ann N Y Acad Sci 1450(1):15
system of thalassemic pathologies based on artificial neural net- 51. Chari PS, Prasad S (2018) Pilot study on the performance of
works. Med Decis Making 22(1):18–26 a new system for image based analysis of peripheral blood
32. Amendolia SR, Cossu G, Ganadu M, Golosio B, Masala GL, smears on normal samples. Indian J Hemat Blood Transfus
Mura GM (2003) A comparative study of k-nearest neighbour, 34(1):125–131
support vector machine and multi-layer perceptron for thalas- 52. Chen HM, Tsao YT, Tsai SN (2014) Automatic image segmen-
semia screening. Chemom Intell Lab Syst 69(1-2):13–20 tation and classification based on direction texton technique
33. Amin MN, Habib MA (2015) Comparison of different classifica- for hemolytic anemia in thin blood smears. Mach Vis Appl
tion techniques using WEKA for hematological data. Amer J Eng 25(2):501–510
Res 4(3):55–61

13
Medical & Biological Engineering & Computing (2022) 60:2445–2462 2459

53. Chen X, Lu L, Gao Y (2012) A new concentric circle detection 71. Durant TJ, Olson EM, Schulz WL, Torres R (2017) Very deep
method based on Hough transform. In: 2012 7th International convolutional neural networks for morphologic classification of
conference on computer science & education (ICCSE), pp 753– erythrocytes. Clin Chem 63(12):1847–1855
758. IEEE 72. Ejaz Z, Hassan A, Aslam H (2018) Automatic red blood cell
54. Chintawar I, Aishvarya M, Kuhikar C (2016) Detection of sickle detection and counting system using Hough transform. Indo
cells using image processing. Int J Sci Technol Eng 2(9):335–339 Amer J Pharmaceut Sci 5(7):7104–7110
55. Chy TS, Rahaman MA (2018) Automatic sickle cell anemia 73. Elsalamony HA (2014) Sickle anemia and distorted blood cells
detection using image processing technique. In: 2018 Interna- detection using hough transform based on neural network and
tional conference on advancement in electrical and electronic decision tree. In: Proceedings of the international conference
engineering (ICAEEE), pp 1–4. IEEE on image processing, computer vision, and pattern recognition
56. Chy TS, Rahaman MA (2019) A comparative analysis by KNN, (IPCV). The Steering Committee of The World Congress in
SVM & ELM classification to detect sickle cell anemia. In: 2019 Computer Science, Computer ..., p 1
International conference on robotics, electrical and signal pro- 74. Elsalamony HA (2015) Detecting distorted and benign blood
cessing techniques (ICREST), pp 455–459. IEEE cells using the Hough transform based on neural networks and
57. Cruz D, Jennifer C, Castor LC, Mendoza CMT, Jay BA, Jane decision trees. In: Emerging trends in image processing, com-
LSC, Brian PTB et al (2017) Determination of blood components puter vision and pattern recognition, pp 457–473. Elsevier
(WBCs, RBCs, and platelets) count in microscopic images using 75. Elsalamony HA (2016) Healthy and unhealthy red blood cell
image processing and analysis. In: 2017 IEEE 9th International detection in human blood smears using neural networks. Micron
conference on humanoid, nanotechnology, information technol- 83:32–41
ogy, communication and control, environment and management 76. Elsalamony HA (2017) Anaemia cells detection based on shape
(HNICEM), pp 1–7. IEEE signature using neural networks. Measurement 104:50–59
58. Dacie JV (2006) Dacie and Lewis practical haematology. Elsevier 77. Elsalamony HA (2018) Detection of anaemia disease in human
Health Sciences red blood cells using cell signature, neural networks and SVM.
59. Dalvi PT, Vernekar N (2016) Computer aided detection of abnor- Multimed Tools Appl 77(12):15047–15074
mal red blood cells. In: 2016 IEEE International conference on 78. Ford J (2013) Red blood cell morphology. Int J Lab Hematol
recent trends in electronics, information & communication tech- 35(3):351–357
nology (RTEICT), pp 1741–1746. IEEE 79. Frejlichowski D (2010) Pre-processing, extraction and rec-
60. Das BK, Jha KK, Dutta HS (2014) A new approach for segmenta- ognition of binary erythrocyte shapes for computer-assisted
tion and identification of disease affected blood cells. In: 2014 diagnosis based on MGG images. In: International conference
International conference on intelligent computing applications, on computer vision and graphics, pp 368–375. Springer
pp 208–212. IEEE 80. Frejlichowski D (2012) Detection of erythrocyte cells
61. Das D, Chakraborty C, Mitra B, Maiti A, Ray A (2013) Quantita- in microscopy images. Przeglad Elektrotechniczny
tive microscopy approach for shape-based erythrocytes charac- 88(10b):264–267
terization in anaemia. J Microsc 249(2):136–149 81. Gonzalez-Hidalgo M, Guerrero-Pena F, Herold-Garcia S, Jaume-
62. Das D, Ghosh M, Chakraborty C, Pal M, Maity AK (2010) i Capó A., Marrero-Fernández PD (2014) Red blood cell cluster
Invariant moment based feature analysis for abnormal erythro- separation from digital images for use in sickle cell disease. IEEE
cyte recognition. In: 2010 International conference on systems J Biomed Health Inform 19(4):1514–1525
in medicine and biology, pp 242–247. IEEE 82. Gual-Arnau X, Herold-Garcia S, Simo A (2015) Erythrocyte
63. Das PK, Meher S, Panda R, Abraham A (2019) A review of shape classification using integral-geometry-based methods.
automated methods for the detection of sickle cell disease. IEEE Med Biol Eng Comput 53(7):623–633
Rev Biomed Eng 13:309–324 83. Guan PP, Yan H (2011) Blood cell image segmentation based on
64. Deb N, Chakraborty S (2014) A noble technique for detecting the Hough transform and fuzzy curve tracing. In: 2011 Interna-
anemia through classification of red blood cells in blood smear. tional conference on machine learning and cybernetics, vol 4, pp
In: International conference on recent advances and innovations 1696–1701. IEEE
in engineering (ICRAIE-2014), pp 1–9. IEEE 84. Gupta M (2012) Cell identification by blob detection. UACEE
65. Di Ruberto C, Dempster A, Khan S, Jarra B (2002) Analysis of Int J Adv Electon Eng 2:56–59
infected blood cell images using morphological operators. Image 85. de Haan K, Koydemir HC, Rivenson Y, Tseng D, Van Dyne E,
Vis Comput 20(2):133–146 Bakic L, Karinca D, Liang K, Ilango M, Gumustekin E et al
66. Di Ruberto C, Loddo A, Putzu L (2019) A region proposal (2020) Automated screening of sickle cells using a smartphone-
approach for cells detection and counting from microscopic based microscope and deep learning. NPJ Digit Med 3(1):1–9
blood images. In: International conference on image analysis 86. Habibzadeh M, Krzyzak A, Fevens T (2011) Application of pat-
and processing, pp 47–58. Springer tern recognition techniques for the analysis of thin blood smear
67. Di Ruberto C, Loddo A, Putzu L (2020) Detection of red and images. Journal of Medical Informatics and Technologies 18
white blood cells from microscopic blood images using a region (2011)
proposal approach. Comput Biol Med 116:103530 87. Hari J, Prasad AS, Rao SK (2014) Separation and counting of
68. Di Ruberto C, Putzu L (2014) Accurate blood cells segmentation blood cells using geometrical features and distance transformed
through intuitionistic fuzzy set threshold. In: 2014 Tenth interna- watershed. In: 2014 2nd International conference on devices,
tional conference on signal-image technology and internet-based circuits and systems (ICDCS), pp 1–5. IEEE
systems, pp 57–64. IEEE 88. Hariharan S, Parvathy H, Aruna S (2016) An overview of sickle
69. Diaz G, Gonzalez FA, Romero E (2009) A semi-automatic cell anemia–a special emphasis on image processing on SEM
method for quantification and classification of erythrocytes images. Int J Appl Eng Res 11(1):201–8
infected with malaria parasites in microscopic images. J Biomed 89. Haykin S, Network N (2004) A comprehensive foundation. Neur
Inform 42(2):296–307 Netw 2(2004):41
70. Dogan S, Turkoglu I (2008) Iron-deficiency anemia detection 90. Hegde RB, Prasad K, Hebbar H, Sandhya I (2018) Peripheral
from hematology parameters by using decision trees. Int J Sci blood smear analysis using image processing approach for diag-
Technol 3(1):85–92 nostic purposes: a review. Biocybern Biomed Eng 38(3):467–480

13
2460 Medical & Biological Engineering & Computing (2022) 60:2445–2462

91. Hirimutugoda Y, Wijayarathna G (2010) Image analysis system on artificial intelligence, modelling and simulation, pp 320–324.
for detection of red cell disorders using artificial neural networks. IEEE
Sri Lanka J Bio-Med Inform 1(1):2010 111. McLean E, Cogswell M, Egli I, Wojdyla D, De Benoist B (2009)
92. Hornik K, Stinchcombe M, White H (1989) Multilayer feed- Worldwide prevalence of anaemia, WHO vitamin and mineral
forward networks are universal approximators. Neur Netw nutrition information system, 1993–2005. Public Health Nutr
2(5):359–366 12(4):444–454
93. Houwen B (2002) Blood film preparation and staining proce- 112. Md Tomari MR, Wan Zakaria WN, et al. (2015) An empirical
dures. Clin Lab Med 22(1):1–14 framework for automatic red blood cell morphology identifica-
94. Jaiswal M, Srivastava A, Siddiqui TJ (2019) Machine learn- tion and counting ARPN. Journal of Engineering and Applied
ing algorithms for anemia disease prediction. In: Recent trends Sciences 10(2015)
in communication, computing, and electronics, pp 463–469. 113. Mogra M, Bansel A, Srivastava V (2014) Comparative analysis
Springer of extraction and detection of RBCs and WBCs using Hough
95. Jones KW (2009) Evaluation of cell morphology and introduc- transform and k-means clustering algorithm. Int J Eng Res Gen
tion to platelet and white blood cell morphology. Clinical Hema- Sci 2(5):670–674
tology and Fundamentals of Hemostasis, 93–116 114. Mohamad AS, Halim NSA, Nordin MN, Hamzah R, Sathar J
96. Khalaf M, Hussain AJ, Keight R, Al-Jumeily D, Fergus P, (2018) Automated detection of human RBC in diagnosing sickle
Keenan R, Tso P (2017) Machine learning approaches to the cell anemia with laplacian of gaussian filter. In: 2018 IEEE Con-
application of disease modifying therapy for sickle cell using ference on systems, process and control (ICSPC), pp 214–217.
classification models. Neurocomputing 228:154–164 IEEE
97. Kim K, Kim P, Song J, Park Y (2000) Analyzing blood cell 115. Mohamad AS, Hamzah R, Mokhtar AS, Sathar J (2017) Sickle
image to distinguish its abnormalities. In: Proceedings of the cell disease verification via Sobel edge algorithms for image
eighth ACM international conference on multimedia, pp 395–397 processing. In: 2017 International conference on engineering
98. Kimbahune VV, Uke N (2011) Blood cell image segmentation technology and technopreneurship (ICE2T), pp 1–4. IEEE
and counting. Int J Eng Sci Technol (IJEST) 3(3):2448 116. Mundhra D, Cheluvaraju B, Rampure J, Dastidar TR (2017)
99. Kulasekaran S, Sheeba F, Mammen JJ, Saivigneshu B, Mohan- Analyzing microscopic images of peripheral blood smear using
kumar S (2015) Morphology based detection of abnormal red deep learning. In: Deep learning in medical image analysis and
blood cells in peripheral blood smear images. In: 7th WACBE multimodal learning for clinical decision support, pp 178–185.
world congress on bioengineering 2015, pp 57–60. Springer Springer
100. Lee H, Chen YPP (2014) Cell morphology based classifica- 117. Navya K, Pradeep G (2018) Lung nodule segmentation using
tion for red cells in blood smear images. Pattern Recogn Lett adaptive thresholding and watershed transform. In: 2018 3rd
49:155–161 IEEE International conference on recent trends in electronics,
101. Loddo A, Putzu L, Di Ruberto C, Fenu G (2016) A computer- information & communication technology (RTEICT), pp 630–
aided system for differential count from peripheral blood cell 633. IEEE
images. In: 2016 12th International conference on signal-image 118. Nee LH, Mashor MY, Hassan R (2012) White blood cell seg-
technology & internet-based systems (SITIS), pp 112–118. IEEE mentation for acute leukemia bone marrow images. J Med Imag
102. Lotfi M, Nazari B, Sadri S, Sichani NK (2015) The detection Health Inform 2(3):278–284
of dacrocyte, schistocyte and elliptocyte cells in iron deficiency 119. Nithyaa A, Premkumar R, Kanchana D, Krishnan NA (2013)
anemia. In: 2015 2nd International conference on pattern recog- Automated detection and classification of blood diseases. In:
nition and image analysis (IPRIA), pp 1–5. IEEE Recent advancements in system modelling applications, pp
103. Lund P, Barnes R (1972) Automated classification of anaemia 393–404. Springer
using image analysis. The Lancet 300(7775):463–464 120. Parvathy H, Hariharan S, Aruna S (2016) A real time system
104. Mahmood NH, Lim PC, Mazalan SM, Razak MAA (2013) Blood for the analysis of sickle cell anemia blood smear images using
cells extraction using color based segmentation technique. Int J image processing. Int J Innov Res Sci Eng Technol 5:6200–6207
Life Sci Biotechnol Pharma Res 2(2):2250–3137 121. Patil DN, Khot UP (2015) Image processing based abnormal
105. Mahmood NH, Mansor MA (2012) Red blood cells estimation blood cells detection. Int J Tech Res Applic 31:37–43
using Hough transform technique. Signal Image Process 3(2):53 122. Poole DL, Mackworth AK (2010) Artificial intelligence: founda-
106. Maity M, Sarkar P, Chakraborty C (2012) Computer-assisted tions of computational agents. Cambridge University Press
approach to anemic erythrocyte classification using blood patho- 123. Poomcokrak J, Neatpisarnvanit C (2008) Red blood cells extrac-
logical information. In: 2012 Third international conference on tion and counting. In: The 3rd international symposium on bio-
emerging applications of information technology, pp 116–121. medical engineering, pp 199–203
IEEE 124. Prasad K, Winter J, Bhat UM, Acharya RV, Prabhu GK (2012)
107. Maji P, Mandal A, Ganguly M, Saha S (2015) An automated Image analysis approach for development of a decision support
method for counting and characterizing red blood cells using system for detection of malaria parasites in thin blood smear
mathematical morphology. In: 2015 Eighth international confer- images. J Digit Imaging 25(4):542–549
ence on advances in pattern recognition (ICAPR), pp 1–6. IEEE 125. Prasad MN, Prasad K, Navya K (2018) Color transfer method
108. Mao-Jun S, Zhao-bin W, Hong-Juan Z, Yi-de M (2008) A new for efficient enhancement of color images and its application to
method for blood cell image segmentation and counting based peripheral blood smear analysis. In: International conference on
on pcnn and autowave. In: 2008 3rd International symposium on recent trends in image processing and pattern recognition, pp
communications, control and signal processing, pp 6–9. IEEE 134–142. Springer
109. Marzuki NIBC, bin Mahmood NH, bin Abdul Razak MA (2017) 126. Rafael C (1992) Gonzalez, and Richard E. Woods. Digital Image
Identification of thalassemia disorder using active contour. Indo- Processing, 793
nesian J Electr Eng Comput Sci 6(1):160–165 127. Rahmat R, Wulandari F, Faza S, Muchtar M, Siregar I (2018) The
110. Mazalan SM, Mahmood NH, Razak MAA (2013) Automated morphological classification of normal and abnormal red blood
red blood cells counting in peripheral blood smear image using cell using self organizing map. In: IOP Conf. Series: Mater. Sci.
circular Hough transform. In: 2013 1st International conference Eng, vol 308, p 012015

13
Medical & Biological Engineering & Computing (2022) 60:2445–2462 2461

128. Rakshit P, Bhowmik K (2013) Detection of abnormal findings 144. Sharif JM, Miswan M, Ngadi M, Salam MSH, bin Abdul Jamil
in human rbc in diagnosing sickle cell anaemia using image pro- MM (2012) Red blood cell segmentation using masking and
cessing. Proced Technol 10:28–36 Watershed algorithm: a preliminary study. In: 2012 International
129. Rashid NZN, Mashor MY, Hassan R (2015) Unsupervised color conference on biomedical engineering (ICoBE), pp 258–262.
image segmentation of red blood cell for thalassemia disease. In: IEEE
2015 2nd International conference on biomedical engineering 145. Sharma V, Rathore A, Vyas G (2016) Detection of sickle cell
(ICoBE), pp 1–6. IEEE anaemia and thalassaemia causing abnormalities in thin smear
130. Razzak MI, Naz S (2017) Microscopic blood smear segmenta- of human blood sample using image processing. In: 2016 Inter-
tion and classification using deep contour aware cnn and extreme national conference on inventive computation technologies
machine learning. In: 2017 IEEE Conference on computer vision (ICICT), vol 3, pp 1–5. IEEE
and pattern recognition workshops (CVPRW), pp 801–807. IEEE 146. Soltanzadeh R, Rabbani H (2010) Classification of three types
131. Revathi T, Jeevitha S (2016) Efficient Watershed based red blood of red blood cells in peripheral blood smear based on morphol-
cell segmentation from digital images in sickle cell disease. Int ogy. In: IEEE 10th International conference on signal processing
J Sci Eng Appl Sci 2:300–317 proceedings, pp 707–710. IEEE
132. Rezatofighi S, Roodaki A, Zoroofi R, Sharifian R, Soltanian- 147. Taherisadr M, Nasirzonouzi M, Baradaran B, Mehdizade A, Shi-
Zadeh H (2008) Automatic detection of red blood cells in hema- raz I (2013) New approach to red blood cell classification using
tological images using polar transformation and run-length morphological image processing. Shiraz E-Med J 14(1):44–53
matrix. In: 2008 9th International conference on signal process- 148. Tomari M, Lias J, Musa R, Wan Zakaria WN, et al. (2015)
ing, pp 806–809. IEEE Development of red blood cell analysis system using NI vision
133. Ritter N, Cooper J (2007) Segmentation and border identification builder AI ARPN. Journal of Engineering and Applied Sciences
of cells in images of peripheral blood smear slides. In: Proceed- 10(2015)
ings of the thirtieth Australasian conference on computer science, 149. Tomari R, Zakaria WNW, Ngadengon R, Wahab MHA (2015)
vol 62, pp 161–169. Australian Computer Society, Inc Red blood cell counting analysis by considering an overlapping
134. Rodrigues LF, Naldi MC, Mari JF (2016) Morphological analy- constraint Ⓒ 2006-2015 Asian Research Publishing Network
sis and classification of erythrocytes in microscopy images. In: (ARPN) 10(3)
Proceedings of the 2016 Workshop de Visao Computacional, 150. Tran T, Binh Minh L, Lee SH, Kwon KR (2019) Blood cell count
Campo Grande, Brazil, pp 9–11 using deep learning semantic segmentation. https://​doi.​org/​10.​
135. Sadafi A, Radolko M, Serafeimidis I, Hadlak S (2018) Red blood 20944/​prepr​ints2​01909.​0075.​v1
cells segmentation: a fully convolutional network approach. 151. Tran T, Kwon O, Kwon K, Lee S, Kang K (2018) Blood cell
In: 2018 IEEE Intl Conf on Parallel & Distributed Processing images segmentation using deep learning semantic segmenta-
with Applications, Ubiquitous Computing & Communications, tion. In: 2018 IEEE International conference on electronics and
Big Data & Cloud Computing, Social Computing & Network- communication engineering (ICECE), pp 13–16. IEEE
ing, Sustainable Computing & Communications (ISPA/IUCC/ 152. Tyagi M, Saini LM, Dahyia N (2016) Detection of poikilocyte
BDCloud/SocialCom/SustainCom), pp 911–914. IEEE cells in iron deficiency anaemia using artificial neural network.
136. Safca N, Popescu D, Ichim L, Elkhatib H, Chenaru O (2018) In: 2016 International conference on computation of power,
Image processing techniques to identify red blood cells. In: energy information and commuincation (ICCPEIC), pp 108–112.
2018 22nd International conference on system theory, control IEEE
and computing (ICSTCC), pp 93–98. IEEE 153. Tyas DA, Ratnaningsih T, Harjoko A, Hartati S (2017) The clas-
137. Sanap SA, Nagori M, Kshirsagar V (2011) Classification of ane- sification of abnormal red blood cell on the minor thalassemia
mia using data mining techniques. In: International conference case using artificial neural network and convolutional neural net-
on swarm, evolutionary, and memetic computing, pp 113–121. work. In: Proceedings of the international conference on video
Springer and image processing, pp 228–233
138. Sapna S, Renuka A (2017) Techniques for segmentation and clas- 154. Veluchamy M, Perumal K, Ponuchamy T (2012) Feature extrac-
sification of leukocytes in blood smear images-a review. In: 2017 tion and classification of blood cells using artificial neural net-
IEEE International conference on computational intelligence and work. Am J Appl Sci 9(5):615
computing research (ICCIC), pp 1–5. IEEE 155. Venkatalakshmi B, Thilagavathi K (2013) Automatic red blood
139. Sarrafzadeh O, Dehnavi AM, Rabbani H, Ghane N, Talebi A cell counting using Hough transform. In: 2013 IEEE Conference
(2015) Circlet based framework for red blood cells segmenta- on information & communication technologies, pp 267–271.
tion and counting. In: 2015 IEEE workshop on signal processing IEEE
systems (SiPS), pp 1–6. IEEE 156. Walliander M, Turkki R, Linder N, Lundin M, Konsti J, Ojansivu
140. Savkare S, Narote S (2015) Blood cell segmentation from micro- V, Meri T, Holmberg V, Lundin J (2013) Automated segmenta-
scopic blood images. In: 2015 International conference on infor- tion of blood cells in giemsa stained digitized thin blood films.
mation processing (ICIP), pp 502–505. IEEE In: Diagnostic pathology, vol 8, p. S37. Springer
141. Sethian JA (1996) A fast marching level set method for mono- 157. Wei X, Cao Y, Fu G, Wang Y (2015) A counting method for
tonically advancing fronts. Proc Natl Acad Sci 93(4):1591–1595 complex overlapping erythrocytes-based microscopic imaging.
142. Setsirichok D, Piroonratana T, Wongseree W, Usavanarong T, J Innov Opt Health Sci 8(06):1550033
Paulkhaolarn N, Kanjanakorn C, Sirikong M, Limwongse C, 158. Wheeless LL, Robinson RD, Lapets OP, Cox C, Rubio A, Wein-
Chaiyaratana N (2012) Classification of complete blood count traub M, Benjamin LJ (1994) Classification of red blood cells as
and haemoglobin typing data by a C4.5 decision tree, a naive normal, sickle, or other abnormal, using a single image analysis
Bayes classifier and a multilayer perceptron for thalassaemia feature. Cytometry: J Int Soc Anal Cytol 17(2):159–166
screening. Biomed Signal Process Control 7(2):202–212 159. Xu M, Papageorgiou DP, Abidi SZ, Dao M, Zhao H, Karniadakis
143. Shahzad M, Umar AI, Khan MA, Shirazi SH, Khan Z, Yousaf W GE (2017) A deep convolutional neural network for classifica-
(2020) Robust method for semantic segmentation of whole-slide tion of red blood cells in sickle cell anemia. PLoS Comput Biol
blood cell microscopic images. Comput Math Methods Med, 13(10):e1005746
2020 160. Xu M, Yang J, Zhao H (2017) A multivariate shape quantification
approach for sickle red blood cell in patient-specific microscopy

13
2462 Medical & Biological Engineering & Computing (2022) 60:2445–2462

image data. In: Ninth international conference on digital image IEEE International symposium on signal processing and informa-
processing (ICDIP 2017), vol 10420, p 104203W. International tion technology, pp 776–780. IEEE
Society for Optics and Photonics 163. Zhang M, Li X, Xu M, Li Q (2018) Rbc semantic segmentation
161. Yeldhos M (2018) Red blood cell counter using embedded image for sickle cell disease based on deformable U-Net. In: Interna-
processing techniques. Research Reports 2(2018) tional conference on medical image computing and computer-
162. Zahir S, Chowdhury R, Payne GW (2006) Automated assessment assisted intervention, pp 695–702. Springer
of erythrocyte disorders using artificial neural network. In: 2006
Publisher’s note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

13