Hindawi

Scientific Programming
Volume 2021, Article ID 9933481, 14 pages
https://doi.org/10.1155/2021/9933481

Review Article
Machine Learning in Detection and Classification of Leukemia
Using Smear Blood Images: A Systematic Review

Mustafa Ghaderzadeh ,1 Farkhondeh Asadi ,1 Azamossadat Hosseini ,1

Davood Bashash ,2 Hassan Abolghasemi ,3 and Arash Roshanpour4
1
Department of Health Information Technology and Management, School of Allied Medical Sciences,
Shahid Beheshti University of Medical Sciences, Tehran, Iran
2
Department of Hematology and Blood Banking, School of Allied Medical Sciences,
Shahid Beheshti University of Medical Sciences, Tehran, Iran
3
Pediatric Congenital Hematologic Disorders Research Center, Shahid Beheshti University of Medical Sciences, Tehran, Iran
4
Department of Computer Science, Sama Technical and Vocational Training College, Tehran Branch (Tehran),
Islamic Azad University (IAU), Tehran, Iran

Correspondence should be addressed to Farkhondeh Asadi; asadifar@sbmu.ac.ir and Azamossadat Hosseini; souhosseini@
sbmu.ac.ir

Received 8 March 2021; Revised 13 May 2021; Accepted 27 May 2021; Published 25 June 2021

Academic Editor: Pengwei Wang

Copyright © 2021 Mustafa Ghaderzadeh et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
Introduction. The early detection and diagnosis of leukemia, i.e., the precise differentiation of malignant leukocytes with minimum
costs in the early stages of the disease, is a major problem in the domain of disease diagnosis. Despite the high prevalence of
leukemia, there is a shortage of flow cytometry equipment, and the methods available at laboratory diagnostic centers are time-
consuming. Motivated by the capabilities of machine learning (machine learning (ML)) in disease diagnosis, the present sys-
tematic review was conducted to review the studies aiming to discover and classify leukemia by using machine learning. Methods.
A systematic search in four databases (PubMed, Scopus, Web of Science, and ScienceDirect) and Google Scholar was performed
via a search strategy using Machine Learning (ML), leukemia, peripheral blood smear (PBS) image, detection, diagnosis, and
classification as the keywords. Initially, 116 articles were retrieved. After applying the inclusion and exclusion criteria, 16 articles
remained as the population of the study. Results. This review study presents a comprehensive and systematic view of the status of
all published ML-based leukemia detection and classification models that process PBS images. The average accuracy of the ML
methods applied in PBS image analysis to detect leukemia was >97%, indicating that the use of ML could lead to extraordinary
outcomes in leukemia detection from PBS images. Among all ML techniques, deep learning (DL) achieved higher precision and
sensitivity in detecting different cases of leukemia, compared to its precedents. ML has many applications in analyzing different
types of leukemia images, but the use of ML algorithms to detect acute lymphoblastic leukemia (ALL) has attracted the greatest
attention in the fields of hematology and artificial intelligence. Conclusion. Using the ML method to process leukemia smear
images can improve accuracy, reduce diagnosis time, and provide faster, cheaper, and safer diagnostic services. In addition to the
current diagnostic methods, clinical and laboratory experts can also adopt ML methods in laboratory applications and tools.

1. Introduction cells, which can damage red blood cells, bone marrow, and
the defense system. In the United States, more than 3.5% of
Among all types of blood cancers, leukemia is the most new cancer cases are leukemia, and in 2018 alone, this
common form of malignancy in different age groups, es- country reported more than 60,000 new cases of this cancer.
pecially in children. This abnormal phenomenon is caused Malignant white blood cells, or lymphoblast, in the blood
by excessive proliferation and immature growth of blood reach other organs, such as the spleen, brain, liver, and
2 Scientific Programming

kidneys, and then metastasize to important tissues of the (2) Which ML algorithm has achieved high efficiency in
body [1–3]. There are different types of leukemia that he- PBS analysis?
matologists in cell transplant laboratories can differentiate/ (3) For the diagnosis and classification of what types of
diagnose based on microscopic images. If the slide is cor- leukemia, has ML achieved better results?
rectly stained, some types of leukemia can be more easily
(4) How can healthcare systems benefit from using ML
identified and distinguished than others, but more equip-
methods for leukemia detection and diagnosis?
ment is needed to determine underlying leukemia. Figure 1
shows the stained slides of the most common different types By surveying electronic databases that provide scientific
of leukemia. articles on two domains of medicine and computer sciences,
An early diagnosis of leukemia has always been a the researcher concluded that PubMed, Web of Science,
challenge to researchers, doctors, and hematologists. En- Scopus, and ScienceDirect contain the highest number of
largement of lymph nodes, pallor, fever, and weight loss are articles relevant to the title and objectives of this study. The
the symptoms of leukemia, but they can also be associated search was performed by using leukemia, leukemia diag-
with other diseases. Leukemia diagnosis is difficult in its nosis, and detection and ML keywords, based on the in-
early stages due to the mild nature of the symptoms. The clusion and exclusion criteria, from 2015 to 10 November
most common leukemia diagnosis method is the micro- 2020, and relevant articles were extracted from the said
scopic evaluation of PBS, but the golden standard for leu- databases. EMBASE and IEEE databases were removed from
kemia diagnosis only involves taking and analyzing bone the domain of search due to the similarity in publications.
marrow samples [3–6]. Table 1 lists the inclusion and exclusion criteria.
In the last two decades, various studies have adopted
machine learning (ML) and computer-aided diagnostic
methods for laboratory image analysis, hoping to overcome 2.2. Data Extraction. By examining the previous articles,
the limitations of a late leukemia diagnosis and determine its details of their methods and results were extracted and
subgroups. These studies have analyzed blood smears images recorded in specially designed forms [15]. Two re-
for diagnosing, differentiating, and counting the cells in searchers extracted the data, and the disagreements were
various types of leukemia [7, 8]. resolved upon discussions. The extracted data elements
ML is a well-known branch of artificial intelligence, included the title of the article, country, year of publi-
comprising algorithms and mathematical relations, which cation, the studied population, ML technique, evaluation
was quickly introduced to the domain of clinical research. method, and results.
ML enables computers to be programmed without explicit
experience and learns from that experience. The outcome of
using these methods in medical data processing has been 2.3. Quality Assessment. The quality of the eligible studies
extraordinary, and they have made remarkable success in was assessed by the criteria proposed by Qiao [15]. The
disease diagnosis [9–11]. Research indicates that, in medical assessment was performed based on five categories: unmet
image processing, ML methods greatly aid complex medical need (limits in current non-ML approach), reproducibility
decision-making processes by extracting and then analyzing (feature engineering methods, platforms/packages, hyper-
the features of these images [12–14]. As the number of parameters), robustness (valid methods to overcome overfit,
medical diagnosis tools increased and a large volume of the stability of results), generalizability (external data vali-
high-quality data was produced, there was an urgent need dation), and clinical significance (predictors’ explanation
for more advanced data analysis methods. Traditional and suggested clinical use). A quality assessment table was
methods could not analyze such a large volume of data or provided by listing “yes” or “no” for the corresponding items
find data patterns. in each category.

3. Results
2. Methodology
A total of 116 articles were extracted from the four credible
The present systematic review aimed to identify the studies
databases based on the search strategy. After reading the
on leukemia detection and diagnosis by using ML tech-
articles’ abstract and full text, applying the inclusion and
niques for peripheral blood smear (PBS) image analysis. The
exclusion criteria, and selecting articles relevant to the title of
systematic search strategy was developed based on previous
the present study, 17 full-text articles were finally deemed
studies and the criteria selected by the authors.
eligible and were and selected. This process was performed
based on the PRISMA flowchart (Figure 2). As ML methods
and their applications in blood smear image analysis have
2.1. Search Criteria. This study mainly aimed to answer the
newly emerged, this systematic search was conducted over
following questions:
the past five years. A review of the articles showed that, over
(1) To what extent has ML been efficient in leukemia time, the use of ML methods in PBS image analysis has
diagnosis and classification by using PBS images? expanded; seven articles in 2020, five articles in 2019, and
Scientific Programming 3

(a) (b)

(c) (d)

Figure 1: (a) AML (M1), (b) AML (M2), (c) B-ALL (pre-B), and (d) B-ALL (pro-B).

Table 1: Inclusion and exclusion criteria for review.

Inclusion criteria Exclusion criteria
(i) Machine learning (i) DNA
(ii) Deep learning (ii) RNA
(iii) Leukemia (iii) Microarray
(iv) Blood smear image (iv) Other diseases
(v) Microscopic WBC image (v) Non-English reference abstracts were excluded
(vi) Full-text article available (vi) Histology and molecular images
(vii) Written in English (vii) Models not based on ML algorithms
(viii) Published after 2015 (viii) Did not cover an original analysis (e.g., review, narrative)
(ix) Detection (ix) Did not cover full details on methods
(x) Classification

four articles in 2018 have focused on the diagnosis and have employed a combination of these datasets as a cross-
classification of leukemia PBS images. dataset. Sharif has employed three datasets to achieve a
system with high precision and efficiency in diagnosing
various leukocytes [22]. Some researchers have also used
3.1. Leukemia Image Datasets. Diagnosis of leukemia in local datasets in their studies. Among all types of leukemia
peripheral blood images is dependent on stained slide diagnosed and classified by using ML, the most frequent type
quality. Hence, a large number of quality standard datasets was ALL [23–26]. Figure 3 displays the diagnostic goals of
are not available. The majority of studies have employed various types of leukemia based on PBS image processing. In
published public datasets. To design and develop ML al- some articles, image analysis has been performed to count
gorithms, hematologists have made some of these datasets the leukocytes [19].
(that include PBS images) available to researchers. ALL-IDB,
one of the most well-known datasets published in two
versions, has been utilized in many articles, most of which 3.2. Overview of Machine Vision Techniques in PBS Image
have diagnosed and classified acute lymphoblastic leukemia Analysis. Examining the methods adopted by the reviewed
(ALL) via different ML techniques [16–21]. There is another studies indicated that two categories of machine vision
published leukemia dataset called Benchmark for the de- techniques have been used in PBS image analysis; machine
velopment of ML algorithms, used by some studies. Most learning and its important subclass, deep learning, are two
researchers have tested their proposed model only on ho- categories of learning algorithms. The first strategy relies on
mogeneous databases or private databases. However, a selective image feature extraction. These methods are
major challenge in a robust detection and classification common in the extraction of a volume of image features via
model is the ability to diagnose the disease in databases with mathematical and ML algorithms. In this view, the goal of
distinct characteristics [22]. Hence, to present a robust feature extraction is to obtain a set of image descriptors. By
model and achieve reliable and valid results, some studies finding the relationship between these descriptors, the
4 Scientific Programming

PubMed result Web of science result ScienceDirect result Scopus result

Identification
(n = 11) (n = 10) (n = 45) (n = 50)

Collection result
(n = 116)
Screening

Records after duplicates removed

(n = 108) Article excluded with
causes:
Were not about diagnosis
and detection the
leukemia.
Article assessed for eligibility
Were about other methods
(n = 15)
and sampling to
identify ALL
Eligibility

Were about WBC

component separation

Full-text articles assessed Extra research after

for eligibility search again
(n = 15) (n = 1)
Included

Studies included in
qualitative synthesis
(n = 16)

Figure 2: PRISMA flow diagram of the review process and exclusion of papers.

patterns determining the images can be discovered [17, 19]. [28, 29]. Then, the features are classified by feeding the
Several classes of features have been considered by re- features to a layer containing one or more classifiers. These
searchers and analyzed via ML algorithms to select the most methods extract important features and neglect less im-
valuable and most effective classification performance. The portant ones. A review of the studies revealed that, to extract
features extracted from the cytomorphological structure can and process the features of PBS images for leukemia de-
include cell form, nucleus structure, chromatin, etc. Many tection, many studies have employed the CNN algorithm
articles consider other features as well. Table 2 summarizes and its state-of-the-art models [30–33]. The features of
the most common features in the field of blast analysis. leukocytes by Vogado [22] simultaneously achieved using
Al-jaboriy et al. used the nuclear-to-cytoplasmic ratio, CaffeNet, AlexNet, and Vgg-f architectures, which, at that
nucleus compactness, nucleus form factors, nucleus ec- time, were among the most efficient CNN [22]. Figure 4
centricity, nucleus elongation, and nucleus rigidity illustrates the frequency of use of both methods, ML and DL.
[17, 23, 24, 27]. Among seven studies, which used traditional The frequency of using ML for medical data analysis is daily
ML algorithm, four used the SVM method alone and with increasing.
other algorithms [18–20, 24] and three utilized ANN and
other algorithms [17]. Note that these algorithms are among
the most popular algorithms in medical image processing. 3.3. Segmentation in PBS Images. Segmentation is a common
The second view comprises methods in which feature task in natural and medical image analysis. The researchers
extraction is performed automatically, and the researcher to achieve better classification rates use different types of
plays no role in feature selection. In these methods, building segmentation. Segmentation is a method for image pre-
blocks of convolution neural network, including convolu- processing applied for feature extraction and selection and
tion and pooling layers, process the values corresponding to could be considered as the first stage of feature extraction.
the pixels; in this way, features are extracted automatically Segmentation with the goal of extracting a cell from context
Scientific Programming 5

AML

12.5%
62%

25%

ALL
Other leukemia
Figure 3: The aim of studies in processing different types of leukemia PBS using ML.

Table 2: The most common features used in blast analysis using subtypes. Several studies have diagnosed leukemia by seg-
ML. menting and then extracting the features of some of these
Class of features Parameters segments; other studies, however, have diagnosed leukemia
(i) Mean color intensity for red component
without segmentation and merely by extracting the features
(ii) Mean color intensity for green from the entire image [12, 34–37].
component
(iii) Mean color intensity for blue
component 3.3.1. Overview of Segmentation Techniques. Several studies
Color features (iv) Mean color intensity for hue trying to detect and differentiate leukocytes used ML
component techniques to segment and extract this cell and its nuclei
(v) Mean color intensity for saturation from other blood cells. The main types of segmentation
component techniques include thresholding methods, boundary-based
(vi) Mean color intensity for value segmentation, region-based segmentation, and hybrid
component technology combining boundary and region standards, and
(i) Homogeneity most of the techniques combine boundary and region cri-
(ii) Contrast teria [38–40]. Two techniques of blood smear image seg-
Texture features
(iii) Correlation mentation are more prominent and have received more
(iv) Energy attention from researchers. In the first view, which is based
(i) Mean on the concept of thresholding and change color channels in
Statistical features (ii) Variance the scope of cell sets, only the extraction of blasts without
(iii) Skewness
considering blasts feature is considered in research, and then
(i) Mean of HARR A coefficient the model is trained on these blasts [41]. In this method, the
(ii) Mean of HARR H coefficient rest of the blood components like RBC are removed from the
(iii) Mean of HARR V coefficient
Wavelet features context of the images and, therefore, from the machine
(iv) Variance of HARR A coefficient
(v) Variance of HARR H coefficient learning input. Al-jaboriy et al. using this type of segmen-
(vi) Variance of HARR V coefficient tation removed all other blood components such as RBC
(i) Area, perimeter, roundness, elongation, cells and other erythrocyte lines and extracted only WBC
Morphological cells, which include lymphocytes and lymphoblast. Figure 5
form factor, length-to-diameter ratio,
features shows a view of this type of segmentation.
compactness, discrete Fourier transform
Another class of segmentation is object detection, in
which segmentation is not performed from the edge of the
Deep learning cell, and the crop is done around the ROI surrounding the
37% cell frame, accommodating other cellular components. In
this type of segmentation, this entire box is fed to the
model to learn its usage. This segmentation model has
44% been used in many studies due to its high similarity be-
19%
tween blood cells and their sensitivity to differentiation.
This segmentation has been referred to as localization in
some studies. In this type of segmentation, the noise
Machine learning Hybrid algorithm components in the learning process are minimized. Fig-
ure 6 shows this type of segmentation. Other ML methods
Figure 4: Different machine learning views in PBS image analysis.
of segmentation are clustering [42], Gram–Schmidt or-
thogonalization method [43], edge detection, region
or its nuclei from the cytoplasm provides an accurate view of growing [44], and optimization-based method [45]. In
the structure and features of the blast, and the ML technique blood cell segmentation, more traditional ML algorithms
by analysis the origin of the blast can identify leukemia or its have been used.
6 Scientific Programming

Figure 5: Localization, preprocessing, and thresholding segmentation [23].

100 pixel

Figure 6: Blast segmentation-based object detection.

3.3.2. Overview of the ML Algorithm in Blood Cell leukemia segmentation [46]. The other popular ML al-
Segmentation. Machine learning plays an important role gorithm is watershed algorithm, which separating com-
in blood image segmentation, and segmentation is one of ponent-based morphological or other features presented
the first steps in identifying leukemia in blood smear in Table 2 treats pixels values as a local topography. The
images. Different machine learning algorithms have been application of watershed segmentation to a distance map
used in most segmentation techniques. The purpose of cell increases efficiency. Watershed segmentation is based on
segmentation is to identify the boundary between the the idea of a catchment basin of a contour map. In other
nucleus and the cytoplasm for further characterization, words, the water droplets follow the image gradient flow
such as the characterization of the nuclear properties, the along the path to reach a local minimum. Many studies
properties of the cytoplasm, and the nuclear-to-cyto- have used the watershed algorithm for segmentation.
plasmic ratio, which is useful for explosive identification Using this algorithm has been easier and more acceptable
[39, 46, 47]. Many segmentation algorithms have been than other algorithms [49–51]. Other ML algorithms such
presented in the literature and the traditional ML algo- as SVM, ANN, and decision tree have been used fre-
rithms based on selected features were the main and quently to segment blast in blood smear images. Table 3
popular algorithms. Machine learning algorithms are used lists the studies that have performed segmentation using
in the computational core of two categories of segmen- the ML algorithm to extract blasts or their features for
tation types. They are pixel-based image segmentation and specific purposes, not just for leukemia detection or
region-based segmentation. Some other studies used classification. Several of this research uses segmentation
shaped-based segmentation (threshold-based, edge- to extract nuclei of blast or other WBC cells.
based, and region-based techniques) instead of region- Segmentation for leukemia detection or diagnosis is
based segmentation. Among the different types of ma- particularly much crucial. The accurate feature extrac-
chine algorithms, clustering class algorithms had the most tion and leukemia classification are proportionately
acceptance and efficiency. Kim et al. used clustering al- dependent on the correct segmentation of the maximized
gorithms in the threshold, edge detection, pixel clustering, and cropped lymphocytes. Table 4 presents the studies
and region growing segmentation [48]. Kekre et al. used which have analyzed PBS images to diagnose (detect) or
k-mean and fuzzy c-mean algorithm vector quantization classify different kinds of leukemia based on the indi-
on the color pixel to segment the blood cells [44], and also cators considered in the present study. Some studies have
Viswanathan used morphological contours (edge detec- merely diagnosed leukemia and aimed for its primary
tion, erosion, and dilation) as features in the fuzzy c-mean detection based on cell morphological changes
algorithm to achieve a high-performance model in [19, 21, 22, 25, 26, 75, 76].
Scientific Programming 7

Table 3: Studies that used ML algorithms for blast segmentation.

No. of data Performance
Author Type of feature extracted ML segmentation technique
and details (accuracy %)
Morphological operations
Begum and Razak Not
erosion, dilation, opening, and SVM 83%
[52] mentioned
closing of nuclei
Naıve Bayes, linear discriminant analysis, K-
Morphological, wavelet, color,
nearest neighbor, support vector machine,
Jothi et al. [53] texture, and statistical features, 300 (60%–100%)
decision tree, and ensemble random under
and other features
sampling boost
Gajul and Shelke
Not mentioned K-mean clustering and morphological operations 40 —
[54]
Vogado et al. [22] Not identified Automatic feature extraction Three datasets 99%
Color, texture, shape, and
Agaian et al. [55] Using k-means clustering 80 98%
Hausdorff dimension feature
Geometric, color, texture, and
Negm et al. [51] Using k-means clustering 75 99.5%
size feature of blast
k-means cluster and constructing a cell image by
Su et al. [39] Color and morphology features 61 96%
hidden Markov random field
Goutam and
LDP feature Using k-means clustering 90 98%
Sailaja [56]
Shankar et al. [57] Color, shape, and texture Threshold by using the Zack algorithm 33 96%
Morphological contour (edge
Viswanathan [46] Fuzzy c-means — 98%
detection, erosion, and dilation)
Patel and Mishra Geometric, color, texture, and
K-mean clustering and the Zack algorithm 7 93%
[58] size feature of cell
Morphological operation and
Zhao et al. [12] granularity feature are selected CNN and SVM 9 94%
automatically
Karthikeyan and
Geometrical, texture, and color Fuzzy c-means 19 90%
Poornima [59]
MoradiAmin et al. Geometrical and statistical
Fuzzy c-means 21 98%
[37] feature
Rawat et al. [60] Morphological operation Global thresholding and morphological opining 260 (79%–95.4%)
Mishra et al. [61] Texture and color Marker-controlled watershed segmentation 190 96%
Bhattacharjee and Morphological operations erosion, dilation,
Morphological operation 120 96%
Saini [36] opening, and closing
Khobragade et al. Not
Geometrical and statistical Otsus’s thresholding and Sobel operator 90%
[62] mentioned
Patil and Raskar Not Not
Color, shape, and texture Thresholding by using Otsu’s method
[41] mentioned mentioned
Rawat et al. [34] Shape features Global thresholding and morphological opining 420 96.75%
K-mean clustering followed by expectation Not
Neelam et al. Texture features 80%
maximization algorithm mentioned
ALL-IDB (no:
Singh et al. [63] Shape and texture features ANN 97.2%
108)
Singhal and Singh ALL-IDB (no:
Texture features SVM 93.8%
[64] 260)
Local (not Not
Zhang et al. [65] Shape features Fuzzy system
mentioned) mentioned
ALL-IDB (no:
Neoh et al. [66] Shape, texture, and color features Dempster–Shafer 96.7%
180)
Amin et al. [67] Shape and texture features SVM Local (no: 21) 97%
ALL-IDB (no:
Viswanathan [46] Shape, color, and texture features Fuzzy c-means classifier 98.0%
108)
8 Scientific Programming

Table 3: Continued.
No. of data Performance
Author Type of feature extracted ML segmentation technique
and details (accuracy %)
Bhattacharjee and ALL-IDB (no:
Shape features ANN 95.2%
Saini [36] 120)
ElDahshan et al. ALL-IDB (no: Not
Not mentioned Field
[68] 300) mentioned
ALL-IDB (no:
Rawat et al. [60] Shape and texture features SVM 89.8%
196)
ALL-IDB (no:
Putzu et al. [35] Shape, color, and texture features SVM 92.0%
267)
Mohapatra et al. Local dataset
Shape and texture features Ensemble classifier 94.7%
[70] (no: 104)
Local dataset
Nasir et al. [71] Shape and color features MLP_BR 95.7%
(no: 230)
Mohapatra et al. Local dataset Not
Shape and texture features ANN
[72] (no: 100) mentioned
Madhloom et al. Local dataset
Shape and texture features kNN clustering 92.5%
[73] (no: 260)
Multiple clinical and laboratorial Local dataset
Pedreira et al. [74] ANN 98.2%
features (no: 189)

Table 4: Characteristics of studies that used machine learning algorithms in the detection and classification of blood smears.
First author,
year of
Aim of the study Data ML method Validation results More information
publication,
and country
Al-jaboriy The proposed model
ALL Blood smear
et al., 2019, ANN Accuracy � 97% detected 625 cells out
segmentation images (ALL-IDB)
Malaysia [17] of 540 WBC
Quadratic SVM has
Al-Tahhan Blood smear KNN the best performance
Automatic Accuracy of testing � 100%
et al., 2020, images (ALL- SVM in detecting ALL
detection ALL F1-score � 100%
Egypt [19] IDB2) ANN among ALL-IDB2
dataset
Artificial neural
Bodzas et al.,
Automated Blood smear SVM Sensitivity � 100% network has the best
2020, Czech
detection of ALL images (local) ANN Specificity � 95.31% performance in
[24]
detecting ALL
Automatic Six groups of cell Classification
Boldú et al.,
recognition of Peripheral blood accuracy � 85% and for accuracy for the six
2019, Spain LDA
different types of images (local) some class, accuracy was groups of cell types
[23]
blast 97% was 85.8
Single-cell
Detection and Segmentation,
morphological Accuracy of detection of
classification of feature extraction,
Dasariraju dataset of immature � 92.99%
immature detection and
et al., 2020, leukocytes from RF Accuracy for classification
leukocytes for classification, and
USA [27] AML patients and of immature leukocytes for
diagnosis of calculation modules
nonmalignant types � 93.45%
AML were applied
(public)
Goal of this research
was to design a
Fathi et al., Classification of Accuracy � 96.2%
Blood smear SVM with a Gaussian radial framework for
2018, Egypt ALL from Sensitivity � 97.3%
dataset (ALL-IDB) basis kernel classification of
[16] normal cases Specificity � 95.3%
cancer based on
medical images
Scientific Programming 9

Table 4: Continued.
First author,
year of
Aim of the study Data ML method Validation results More information
publication,
and country
Dataset 1:
precision � 0.995%
Four commonly Proposed Leukocyte
Dataset 2:
Fan et al., used blood smear Mask architecture to
Localization and precision � 0.994%
2019, China dataset (BCISC, DNN gain best precision
segmentation Dataset 3:
[77] LISC, and 2 other result with all
precision � 0.989%
released datasets) datasets
Dataset 4:
precision � 0.984%
Proposed model
Khilji et al.,
compared with other
2020, ALL dataset CNN-based different models
Detection of ALL Accuracy � 77.934% state-of-the-art
Bangladesh (C_NMC) (encrypted)
model and gain
[75]
better accuracy
Different types of
Nagiub pretrained (CNN)
Abdelsalam Detection of all Leukemia models were applied
CNN Accuracy � 99.98%
et al., 2018, types of leukemia microscopic and Inception-v3
Egypt [26] model had the
highest accuracy
Even deep neural
Nagiub networks: AlexNet,
Abdelsalam AML microscopic Accuracy � 100% GoogLeNet, VGG16,
AML detection CNN (ResNet-101)
et al., 2019, images (local) Sensitivity � 100% VGG19, Inception-
Egypt [25] v3, ResNet50, and
ResNet10
Grey Wolf-based
Praveena and Segmentation Sparse-FCM and deep Accuracy � 93.5% Jaya optimization
Singh, 2020, and classification ALL-IDB2 convolutional neural Sensitivity � 95.28% algorithm was
India [78] of ALL network Specificity � 93.89% applied for training
CNN
Accuracy for ALL- Naı̈ve Bayes and
Sharif et al.,
Recognition of LISC, ALL-IDB1, IDB1 � 97.2% discriminant analysis
2020, Pakistan Localization using YOLOv2.
different types of and ALL-IDB2 Accuracy for ALL- and particle swarm
and Qatar Classification using PSO
leukocytes malignant (public) IDB2 � 100% optimization was
[76]
Accuracy for LISC > 99% used
Goal of this research
was to design a
framework for
ALL-IDB1 classification of
Vogado et al., CNN
ALL-IDB2 Accuracy � 100% cancer based on
2018, Brazil Diagnosis of ALL (AlexNet + CaffeNet + Vgg-
Leukocytes Precision � 100% medical images (3
[22] f ) and SVM
CellaVision architectures were
used in feature
extraction, SVM for
classification)
ALL detection After detection, ALL
Shafique and Detection and accuracy � 99.50%, ALL subtype was
ALL-IDB1 and
Tehsin, 2020 classification of CNN (AlexNet) subtype classified based FAB
ALL-IDB2
Pakistan [79] ALL classification � 96.06%, classification system.
dataset precision � 0.984% Datasets
Detection of After segmentation,
Hegde et al.,
nuclei and Accuracy of detection of the nucleus of WBC
2018, India Leishman SVM
classification of lymphocyte � 100% cells different kinds
[18]
WBC of them was classified
10 Scientific Programming

Table 4: Continued.
First author,
year of
Aim of the study Data ML method Validation results More information
publication,
and country
Nucleus and
cytoplasm
segmentation using
Jha and Dutta, Proposed hybrid
Accuracy � 99% Chrono-SCA-ACNN
2019, India Detection of ALL ALL-IDB2 segmentation + Chrono-
Sensitivity � 100% (chronological sine
[21] SCA-ACNN
cosine algorithm-
based actor-critic
neural network)

4. Discussion as a definitive method for leukemia differential diagnosis

[1, 24, 82]. Thus, the selection of several important features
Microscopic evaluation of PBS images is the most common from among a large number of features is a completely
primary method of leukemia diagnosis in its early stages. algorithmic process, and promoting the efficiency of feature
Still, a manual examination of these smears can cause errors selection depends on the algorithm’s method. The studies
in determining the type of the disease and lead to non- demonstrated that methods extracting fewer cell features
standard reports. Moreover, the examination of these smears have attained a lower precision in leukemia diagnosis. It
is tiresome and time-consuming, thus influencing the di- seems that, to achieve better results in leukemia detection
agnostic precision. Accordingly, there is a need for an au- and diagnosis, one can adopt feature extraction methods
tomatic method to provide a precise diagnosis, without based on hybrid algorithms or swarm intelligence and pay
being affected by the technicians’ experience or the opera- attention to further coverage of the feature space. It is also
tor’s fatigue and job pressures [49, 80]. recommended that a set of various features, including
Upon a search in scientific databases, it was found that geometrical, statistical, and morphological ones, be used for
no comprehensive systematic review had been conducted on leukemia detection. ML methods require manual feature
PBS image analysis via ML methods. Therefore, the authors extraction and selection; if the number of images is ac-
conducted a review study on the applications of ML in the ceptable for DL, instead of ML, it is better to use the DL
diagnosis and classification of different types of leukemia method owing to its mechanism.
based on PBS images. By comparing the previous studies, the A major problem associated with leukemia diagnosis via
present research answered the questions posed by the re- ML algorithms in different studies is the lack of compre-
searcher at the outset. hensive datasets of leukemia smear images, an issue which
In terms of smear preparation, several factors (e.g., il- causes problems for the ML methods, e.g., overfitting. Based
lumination condition, staining time, blood film thickness, on the studies, and with respect to the data-driven nature of
and a defect in the film) lead to undesirable visual artifacts or these methods, one can show that diagnostic errors are
different color distributions in the laboratory images. These higher in the case of smaller datasets. This is why the results
issues complicate the precise detection and monitoring of of many studies cannot be confirmed because small/local
blood smears. As processing these smear images by ML is datasets have been used. Thus, to have a robust ML method
problematic, preprocessing is necessary [81]. As for leuke- for leukemia diagnosis/classification, a comprehensive
mia detection using ML algorithms, data preprocessing (e.g., dataset with sufficient data is required, yet the datasets
preparation, normalization, and segmentation) can promote existing in the reviewed studies did not satisfy this basic
the precision of leukemia detection. For precise leukemia need. Of course, there are techniques for increasing the data,
detection with minimum error via ML methods, it is sug- which, by processing the main images, create new images
gested that a set of preprocessing techniques be adopted for that maintain the features of the main images. To overcome
dataset preparation. this problem in DL, numerous studies have reported that
The selection of effective features is the bedrock of augmentation techniques can lead to better results in terms
preliminary processing of blood smears via ML methods. In of pattern recognition [47–49]. It seems that image aug-
cases where the researcher could control the selection and mentation can lead to better coverage of data space and
analysis of blood cell features, the main problem was markedly improve the results of leukemia detection by using
selecting these features to determine leukemia. Some studies these methods. Based on the review of previous studies and
have used color and shape, while others have utilized texture the results of smear processing, it can be concluded that ML
and different texture metrics as the features of blast cells. The methods and techniques have received more attention for
manual selection of the most important features is always the diagnosis and classification of acute leukemia, whether
associated with some degree of error, and this process is AML or ALL, compared to other types. No comprehensive
always viewed as a major challenge. Medical texts have not study has examined the performance of traditional and
mentioned any of these features selected by manual methods visual leukemia diagnosis by using smear images. However,
Scientific Programming 11

studies that have diagnosed leukemia via ML techniques Conflicts of Interest

have achieved extraordinary results, with a disease detection
mean accuracy of >96%. Although the applications of The authors declare that they have no conflicts of interest.
machine learning in disease diagnosis and blood cell imaging
are still evolving, the use of these algorithms in cell counting References
and blood cell type differentiation is expanding in the
healthcare industry. Nowadays, the use of cell counter de- [1] R. B. Hegde, K. Prasad, H. Hebbar, B. M. K. Singh, and
vices to determine and count blood components based on I. Sandhya, “Automated decision support system for detection
ML is becoming more common. It is thought that, in the of leukemia from peripheral blood smear images,” Journal of
Digital Imaging, vol. 33, pp. 361–374, 2019.
near future, bone marrow transplant laboratories could
[2] “Hematology TAS of facts-and-statistics,” 2018, http://www.
replace traditional devices with applications and software hematology.org/.
based on ML, especially DL, to offer a timely method and [3] S. M. Namayandeh, Z. Khazaei, M. Lari Najafi, E. Goodarzi,
assist a diagnosis with high certainty and low detection error and A. Moslem, “GLOBAL Leukemia in children 0–14 sta-
in the early stages. tistics 2018, incidence and mortality and human development
index (HDI): GLOBOCAN sources and methods,” Asian
5. Conclusion Pacific Journal of Cancer Prevention, vol. 21, no. 5,
pp. 1487–1494, 2020.
Blood smear image analysis is a vital role in the diagnosis of [4] O. Wolach and R. M. Stone, “Mixed-phenotype acute leu-
many blood-related diseases. The diagnosis of leukemia in its kemia,” Current Opinion in Hematology, vol. 24, no. 2,
pp. 139–145, 2017.
early stages and the first smears can lead to immediate
[5] C.-H. Pui, “Childhood leukemias,” New England Journal of
diagnosis and the quick initiation of the treatment. Blood Medicine, vol. 332, no. 24, pp. 1618–1630, 1995.
smear image analysis by ML methods can aid the diagnosis [6] “Childhood leukemias,” 2020, https://www.nejm.org/doi/full/
of early-onset leukemia and the determination of subtypes 10.1056/NEJM199506153322407.
with a minimum error at the shortest time, so that the [7] F. Xing and L. Yang, “Robust nucleus/cell detection and
process of treatment can be immediately started. A prom- segmentation in digital pathology and microscopy images: a
ising future direction for research can be the application of comprehensive review,” IEEE Reviews in Biomedical Engi-
novel ML algorithms, in particular, DL, in computer-aided neering, vol. 9, pp. 234–263, 2016.
detection (CAD) systems, whole-slide imaging (WSI), and [8] J. Wen, Y. Xu, Z. Li, Z. Ma, and Y. Xu, “Inter-class sparsity
even apps and software at hematology laboratories, to help based discriminative least square regression,” Neural Net-
the pathologists and oncologists in better detecting leuke- works, vol. 102, pp. 36–47, 2018.
[9] Z. Obermeyer and E. J. Emanuel, “Predicting the future—big
mia. In the 2018 meeting of the American Society of He-
data, machine learning, and clinical medicine,” New England
matology, Höllein et al. investigated 43 roles of AI in MFC Journal of Medicine, vol. 375, no. 13, pp. 1216–1219, 2016.
for B cell lymphoma and leukemia diagnosis. By using the [10] V. Ehrenstein, H. Nielsen, A. B. Pedersen, S. P. Johnsen, and
data of 38416 patients and control groups, a model was L. Pedersen, “Clinical epidemiology in the era of big data: new
developed by using neural networks. This system achieved opportunities, familiar challenges,” Clinical Epidemiology,
97% precision in determining normal and abnormal cells. vol. 9, pp. 245–250, 2017.
Still, the precision of B cell lymphoma and leukemia clas- [11] M. Ghaderzadeh, F. Rebecca, and A. Standring, “Comparing
sification was 74%. Thus, it is recommended that, in the near performance of different neural networks for early detection
future, the use of ML algorithms for the analysis of blood of cancer from benign hyperplasia of prostate,” Applied
smear images progresses from the phase of modeling to the Medical Informatics, vol. 33, pp. 45–54, 2013.
[12] J. Zhao, M. Zhang, Z. Zhou, J. Chu, and F. Cao, “Automatic
phase of implementation.
detection and classification of leukocytes using convolutional
neural networks,” Medical & Biological Engineering &
Data Availability Computing, vol. 55, no. 8, pp. 1287–1301, 2017.
[13] I. Oikonomidis, N. Kyriazis, and A. A. Argyros, “Efficient
No data were used to support the findings of this study. model-based 3D tracking of hand articulations using Kinect,”
in Proceedings of the 22nd British Machine Vision Conference,
Ethical Approval BMVC’2011, Dundee, UK, 2011.
[14] T. W. Reader, R. Flin, K. Mearns, and B. H. Cuthbertson,
This study was approved by Iran National Committee for “Developing a team performance framework for the intensive
Ethics in Biomedical Research with Approval ID care unit,” Critical Care Medicine, vol. 37, no. 5, pp. 1787–
1793, 2009.
IR.SBMU.RETECH.REC.1399.735.
[15] N. Qiao, “A systematic review on machine learning in sellar
region diseases: quality and reporting items,” Endocrine
Disclosure Connections, vol. 8, no. 7, pp. 952–960, 2019.
[16] E. Fathi, M. J. Rezaee, R. Tavakkoli-Moghaddam, A. Alizadeh,
This study was part of a PhD project conducted at Shahid and A. Montazer, “Design of an integrated model for diag-
Beheshti University of Medical Sciences, Tehran, Iran. nosis and classification of pediatric acute leukemia using
12 Scientific Programming

machine learning,” Proceedings of the Institution of Me- [30] S. Koitka and C. M. Friedrich, “Traditional feature engi-
chanical Engineers, Part H: Journal of Engineering in Medicine, neering and deep learning approaches at medical classification
vol. 234, no. 10, pp. 1051–1069, 2020. task of ImageCLEF 2016,” 2016, http://ceur-ws.org/Vol-1609/
[17] S. S. Al-jaboriy, N. N. A. Sjarif, S. Chuprat, W. M. Abduallah, 16090304.pdf.
and W. M. Abduallah, “Acute lymphoblastic leukemia seg- [31] L. Oakden-Rayner, G. Carneiro, T. Bessen, J. C. Nascimento,
mentation using local pixel information,” Pattern Recognition A. P. Bradley, and L. J. Palmer, “Precision radiology: pre-
Letters, vol. 125, pp. 85–90, 2019. dicting longevity using feature engineering and deep learning
[18] R. B. Hegde, K. Prasad, H. Hebbar, and B. M. K. Singh, methods in a radiomics framework,” Scientific Reports, vol. 7,
“Development of a robust algorithm for detection of nuclei pp. 1–13, 2017.
and classification of white blood cells in peripheral blood [32] G. E. Hinton and R. R. Salakhutdinov, “Reducing the di-
smear images,” Journal of Medical Systems, vol. 42, no. 6, mensionality of data with neural networks,” Science, vol. 313,
p. 110, 2018. pp. 504–507, 2006.
[19] F. E. Al-Tahhan, M. E. Fares, A. A. Sakr, and D. A. Aladle, [33] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature,
“Accurate automatic detection of acute lymphatic leukemia vol. 521, no. 7553, pp. 436–444, 2015.
using a refined simple classification,” Microscopy Research & [34] J. Rawat, A. Singh, H. S. Bhadauria, J. Virmani, and
Technique, vol. 83, no. 10, pp. 1178–1189, 2020. J. S. Devgun, “Computer assisted classification framework for
[20] E. M. F. E. Houby, “Framework of computer aided diagnosis prediction of acute lymphoblastic and acute myeloblastic
systems for cancer classification based on medical images,” leukemia,” Biocybernetics and Biomedical Engineering, vol. 37,
Journal of Medical Systems, vol. 42, p. 157, 2018. no. 4, pp. 637–654, 2017.
[21] K. K. Jha and H. S. Dutta, “Nucleus and cytoplasm-based [35] L. Putzu, G. Caocci, and C. Di Ruberto, “Leucocyte classifi-
segmentation and actor-critic neural network for acute cation for leukaemia detection using image processing
lymphocytic leukaemia detection in single cell blood smear techniques,” Artificial Intelligence in Medicine, vol. 62, no. 3,
images,” Medical & Biological Engineering & Computing, pp. 179–191, 2014.
vol. 58, pp. 171–186, 2019. [36] R. Bhattacharjee and L. M. Saini, “Robust technique for the
[22] L. H. S. Vogado, R. M. S. Veras, F. H. D. Araujo, R. R. V. Silva, detection of acute lymphoblastic leukemia,” in Proceedings of
and K. R. T. Aires, “Leukemia diagnosis in blood slides using the 2015 IEEE Power, Communication and Information
transfer learning in CNNs and SVM for classification,” En- Technology Conference (PCITC), Bhubaneswar, India, 2015.
[37] M. MoradiAmin, N. Samadzadehaghdam, S. Kermani, and
gineering Applications of Artificial Intelligence, vol. 72,
A. Talebi, “Enhanced recognition of acute lymphoblastic
pp. 415–422, 2018.
leukemia cells in microscopic images based on feature re-
[23] L. Boldú, A. Merino, S. Alférez, A. Molina, A. Acevedo, and
duction using principle component analysis,” Frontiers in
J. Rodellar, “Automatic recognition of different types of acute
Biomedical Technologies, vol. 2, pp. 128–136, 2015.
leukaemia in peripheral blood by image analysis,” vol. 72,
[38] K. Krishnaveni, S. Allwin, S. P. K. Kenny, and G. Mariappan,
no. 11, pp. 755–761, 2019.
“Analysis for textural features in nuclei of cervical cyto im-
[24] A. Bodzas, P. Kodytek, and J. Zidek, “Automated detection of
ages,” in Proceedings of the 2010 IEEE International Confer-
acute lymphoblastic leukemia from microscopic images based
ence on Computational Intelligence and Computing Research,
on human visual perception,” Frontiers in Bioengineering and
Coimbatore, India, 2010.
Biotechnology, vol. 8, pp. 1005–1013, 2020.
[39] J. Su, S. Liu, and J. Song, “A segmentation method based on
[25] E. M. Nagiub Abdelsalam, K. F. Hussain, N. M. Omar, and
HMRF for the aided diagnosis of acute myeloid leukemia,”
Q. T. Ali, “Acute myeloid leukemia diagnosis using deep Computer Methods and Programs in Biomedicine, vol. 152,
learning,” Clinical Lymphoma Myeloma and Leukemia, pp. 115–123, 2017.
vol. 19, p. S206, 2019. [40] F. Scotti, “Robust segmentation and measurements tech-
[26] E. M. Nagiub Abdelsalam, K. F. Hussain, N. M. Omar, niques of white cells in blood microscope images,” in Pro-
Q. T. Ali, T. Dao, and D. Scheinberg, “RES-069 improved ceedings of the 2006 IEEE Instrumentation and Measurement
sensitivity and specificity for mutated and unmutated class I Technology Conference, Sorrento, Italy, 2006.
HLA ligand identifications lead to better immunotherapy [41] T. G. Patil and V. B. Raskar, “Automated leukemia detection
target discovery RES-064 computer aided leukemia detection by using contour signature method,” International Journal of
using microscopic blood image based machine learning,” Advance Foundation and Research in Computer, vol. 2, 2015.
Clinical Lymphoma Myeloma and Leukemia, vol. 18, p. S297, [42] H. Sheikh, B. Zhu, and E. Micheli-Tzanakou, “Blood cell
2018. identification using neural networks,” in Proceedings of the
[27] S. Dasariraju, M. Huo, and S. Mccalla, “Detection and clas- 22nd Annual Northeast Bioengineering Conference, New
sification of immature leukocytes for diagnosis of acute Brunswick, NJ, USA, 1996.
myeloid leukemia using random forest algorithm,” Bioengi- [43] S. H. Rezatofighi and H. Soltanian-Zadeh, “Automatic rec-
neering (Basel), vol. 7, no. 4, p. 120, 2020. ognition of five types of white blood cells in peripheral blood,”
[28] M. Ghaderzadeh, F. Asadi, R. Jafari, D. Bashash, Computerized Medical Imaging and Graphics, vol. 35, no. 4,
H. Abolghasemi, and M. Aria, “Deep convolutional neural pp. 333–343, 2011.
network-based computer-aided detection system for COVID- [44] H. Kekre, B. Archana, and H. R. Galiyal, “Segmentation of
19 using multiple lung scans: design and implementation blast using vector quantization technique,” International
study,” Journal of Medical Internet Research, vol. 23, no. 4, Journal of Computer Applications, vol. 72, no. 15, 2013.
Article ID e27468, 2021. [45] E. Cuevas, M. Dı́az, M. Manzanares, D. Zaldivar, and
[29] M. Ghaderzadeh and F. Asadi, “Deep learning in the detection M. Perez-Cisneros, “An improved computer vision method
and diagnosis of COVID-19 using radiology modalities: a for white blood cells detection,” Computational and Mathe-
systematic review,” Journal of Healthcare Engineering, matical Methods in Medicine, vol. 2013, Article ID 137392,
vol. 2021, Article ID 6677314, 10 pages, 2021. 14 pages, 2013.
Scientific Programming 13

[46] P. Viswanathan, “Fuzzy C means detection of leukemia based Signal Processing and Information Technology (ISSPIT),
on morphological contour segmentation,” Procedia Computer Noida, India, 2014.
Science, vol. 58, pp. 84–90, 2015. [61] S. Mishra, B. Majhi, P. K. Sa, and L. Sharma, “Gray level co-
[47] Y. Li, R. Zhu, L. Mi, Y. Cao, and D. Yao, “Segmentation of occurrence matrix and random forest based acute lympho-
white blood cell from acute lymphoblastic leukemia images blastic leukemia detection,” Biomedical Signal Processing and
using dual-threshold method,” Computational and Mathe- Control, vol. 33, pp. 272–280, 2017.
matical Methods in Medicine, vol. 2016, Article ID 9514707, [62] S. Khobragade, D. D. Mor, and C. Y. Patil, “Detection of
12 pages, 2016. leukemia in microscopic white blood cell images,” in Pro-
[48] K. Kim, J. Jeon, W. Choi, P. Kim, and Y.-S. Ho, “Automatic ceedings of the 2015 International Conference on Information
cell classification in human’s peripheral blood images based Processing (ICIP), Pune, India, 2015.
on morphological image processing,” in Proceedings of the [63] G. Singh, G. Bathla, and S. Kaur, “Design of new architecture
14th Australian Joint Conference on Artificial Intelligence, to detect leukemia cancer from medical images,” International
Adelaide, Australia, 2001. Journal of Applied Engineering Research, vol. 11, pp. 7087–
[49] E. A. Mohammed, M. M. A. Mohamed, C. Naugler, and 7094, 2016.
B. H. Far, “Chronic lymphocytic leukemia cell segmentation [64] V. Singhal and P. Singh, “Texture features for the detection of
from microscopic blood images using watershed algorithm acute lymphoblastic leukemia,” in Proceedings of the 2016
and optimal thresholding,” in Proceedings of the 2013 26th International Conference on ICT for Sustainable Development,
IEEE Canadian Conference on Electrical and Computer En- Ahmedabad, India, 2016.
gineering (CCECE), Regina, Canada, 2013. [65] L. Zhang, Q. G. Wang, and J. P. Qi, “Processing technology in
[50] J. Hari, A. S. Prasad, and S. K. Rao, “Separation and counting microscopic images of cancer cells in pleural fluid based on
of blood cells using geometrical features and distance fuzzy edge detection method,” Journal of Physics: Conference
transformed watershed,” in Proceedings of the 2014 2nd In- Series, vol. 48, p. 61, 2006.
ternational Conference on Devices, Circuits and Systems [66] S. C. Neoh, W. Srisukkham, L. Zhang et al., “An intelligent
decision support system for leukaemia diagnosis using mi-
(ICDCS), Coimbatore, India, 2014.
[51] A. S. Negm, O. A. Hassan, and A. H. Kandil, “A decision croscopic blood images,” Scientific Reports, vol. 5, Article ID
14938, 2015.
support system for acute leukaemia classification based on
[67] M. M. Amin, S. Kermani, A. Talebi, and M. G. Oghli,
digital microscopic images,” Alexandria Engineering Journal,
“Recognition of acute lymphoblastic leukemia cells in mi-
vol. 57, no. 4, pp. 2319–2332, 2018.
croscopic images using k-means clustering and support vector
[52] A. R. Begum and T. A. Razak, “A proposed novel method for
machine classifier,” Journal of Medical Signals and Sensors,
detection and classification of leukemia using blood micro-
vol. 5, pp. 49–58, 2015.
scopic images,” International Journal of Advanced Research in
[68] K. ElDahshan, M. Youssef, E. Masameer, and M. A. Mustafa,
Computer Science, vol. 8, 2017.
“An efficient implementation of acute lymphoblastic leukemia
[53] G. Jothi, H. H. Inbarani, A. T. Azar, and K. R. Devi, “Rough set
images segmentation on the FPGA,” Advances in Image and
theory with Jaya optimization for acute lymphoblastic leu-
Video Processing, vol. 3, p. 8, 2015.
kemia classification,” Neural Computing and Applications, [69] J. Rawat, A. Singh, H. S.. Bhadauria, and J. Virmani,
vol. 31, no. 9, pp. 5175–5194, 2019. “Computer aided diagnostic system for detection of leukemia
[54] Y. A. Gajul and R. Shelke, “Computerized detection system for using microscopic images,” Procedia Computer Science,
acute myelogenous leukemia in blood microscopic images,” vol. 70, pp. 748–756, 2015.
International Journal of Innovative Research in Science, En- [70] S. Mohapatra, D. Patra, and S. Satpathy, “An ensemble
gineering and Technology, vol. 5, no. 6, pp. 10073–10079, 2016. classifier system for early diagnosis of acute lymphoblastic
[55] S. Agaian, M. Madhukar, and A. T. Chronopoulos, “Auto- leukemia in blood microscopic images,” Neural Computing
mated screening system for acute myelogenous leukemia and Applications, vol. 24, no. 7-8, pp. 1887–1904, 2014.
detection in blood microscopic images,” IEEE Systems Jour- [71] A. A. Nasir, M. Y. Mashor, and R. Hassan, “Classification of
nal, vol. 8, no. 3, pp. 995–1004, 2014. acute leukaemia cells using multilayer perceptron and sim-
[56] D. Goutam and S. Sailaja, “Classification of acute myeloge- plified fuzzy ARTMAP neural networks,” International Arab
nous leukemia in blood microscopic images using supervised Journal of Information Technology, vol. 10, pp. 1–9, 2013.
classifier,” in Proceedings of the 2015 IEEE International [72] S. Mohapatra, D. Patra, S. Kumar, and S. Satpathy, “Lym-
Conference on Engineering and Technology (ICETECH), phocyte image segmentation using functional link neural
Cimbatore, India, 2015. architecture for acute leukemia detection,” Biomedical Engi-
[57] V. Shankar, M. M. Deshpande, N. Chaitra, and S. Aditi, neering Letters, vol. 2, no. 2, pp. 100–110, 2012.
“Automatic detection of acute lymphoblasitc leukemia using [73] H. T. Madhloom, S. A. Kareem, and H. Ariffin, “A robust
image processing,” in Proceedings of the 2016 IEEE Interna- feature extraction and selection method for the recognition
tional Conference on Advances in Computer Applications of lymphocytes versus acute lymphoblastic leukemia,” in
(ICACA), Coimbatore, India, 2016. Proceedings of the 2012 International Conference on Advanced
[58] N. Patel and A. Mishra, “Automated leukaemia detection Computer Science Applications and Technologies (ACSAT),
using microscopic images,” Procedia Computer Science, Kuala Lumpur, Malaysia, 2012.
vol. 58, pp. 635–642, 2015. [74] C. E. Pedreira, L. Macrini, M. G. Land, and E. S. Costa, “New
[59] T. Karthikeyan and N. Poornima, “Microscopic image seg- decision support tool for treatment intensity choice in
mentation using fuzzy c means for leukemia diagnosis,” childhood acute lymphoblastic leukemia,” IEEE Transactions
Leukemia, vol. 4, pp. 3136–3142, 2017. on Information Technology in Biomedicine, vol. 13, pp. 284–
[60] J. Rawat, A. Singh, and H. S. Bhadauria, “An approach for 290, 2008.
leukocytes nuclei segmentation based on image fusion,” in [75] I. Q. Khilji, K. Saha, J. A. Shonon, and M. I. Hossain, “Ap-
Proceedings of the 2014 IEEE International Symposium on plication of homomorphic encryption on neural network in
14 Scientific Programming

prediction of acute lymphoid leukemia,” International Journal

of Advanced Computer Science and Applications (IJACSA),
vol. 11, no. 6, 2020.
[76] M. Sharif, J. Amin, A. Siddiqa et al., “Recognition of different
types of leukocytes using YOLOv2 and optimized bag-of-
features,” Emerging Deep Learning Theories and Methods for
Biomedical Engineering, vol. 8, pp. 167448–167459, 2020.
[77] H. Fan, F. Zhang, L. Xi, Z. Li, and G. Liu, “LeukocyteMask: an
automated localization and segmentation method for leu-
kocyte in blood smear images using deep neural networks,”
Journal of Biophotonics, vol. 12, no. 7, 2019.
[78] S. Praveena and S. P. Singh, “Sparse-FCM and deep con-
volutional neural network for the segmentation and classi-
fication of acute lymphoblastic leukaemia,” Biomedical
Engineering/Biomedizinische Technik, vol. 65, no. 6, pp. 759–
773, 2020.
[79] S. Shafique and S. Tehsin, “Acute lymphoblastic leukemia
detection and classification of its subtypes using pretrained
deep convolutional neural networks,” Technology in Cancer
Research & Treatment, vol. 17, 2018.
[80] H. C. W. De Vet, J. Koudstaal, W.-S. Kwee, D. Willebrand,
and J. W. Arends, “Efforts to improve interobserver agree-
ment in histopathological grading,” Journal of Clinical Epi-
demiology, vol. 48, no. 7, pp. 869–873, 1995.
[81] G. Dı́az and A. Manzanera, “Automatic analysis of micro-
scopic images in hematological cytology applications,” Clin-
ical Technologies: Concepts, Methodologies, Tools and
Applications, pp. 325–52, IGI Global, Philadelphia, PA, USA,
2011.
[82] S. Wan and M.-W. Mak, Machine Learning for Protein
Subcellular Localization Prediction, Walter de Gruyter, Berlin,
Germany, 2015.