sensors

Review
Critical Analysis of the Current Medical Image-Based Processing
Techniques for Automatic Disease Evaluation: Systematic
Literature Review
Baidaa Mutasher Rashed and Nirvana Popescu *

Computer Science Department, University Politehnica of Bucharest, 060042 Bucharest, Romania

* Correspondence: [email protected]

Abstract: Medical image processing and analysis techniques play a significant role in diagnosing
diseases. Thus, during the last decade, several noteworthy improvements in medical diagnostics
have been made based on medical image processing techniques. In this article, we reviewed articles
published in the most important journals and conferences that used or proposed medical image
analysis techniques to diagnose diseases. Starting from four scientific databases, we applied the
PRISMA technique to efficiently process and refine articles until we obtained forty research articles
published in the last five years (2017–2021) aimed at answering our research questions. The medical
image processing and analysis approaches were identified, examined, and discussed, including
preprocessing, segmentation, feature extraction, classification, evaluation metrics, and diagnosis
techniques. This article also sheds light on machine learning and deep learning approaches. We also
focused on the most important medical image processing techniques used in these articles to establish
the best methodologies for future approaches, discussing the most efficient ones and proposing in
this way a comprehensive reference source of methods of medical image processing and analysis that
can be very useful in future medical diagnosis systems.
Citation: Rashed, B.M.; Popescu, N.
Critical Analysis of the Current Keywords: medical image analysis; machine learning; deep learning; diagnosis system
Medical Image-Based Processing
Techniques for Automatic Disease
Evaluation: Systematic Literature
Review. Sensors 2022, 22, 7065. 1. Introduction
https://doi.org/10.3390/s22187065
Classification methods have increased in importance and now play a significant role
Academic Editors: Alessandro in image processing. Their importance stems from their applications in various fields,
Bevilacqua and Margherita Mottola particularly in medicine. Given the importance of classification in medicine, new and
Received: 20 August 2022
sophisticated classification tools and methods are needed to diagnose and classify medical
Accepted: 14 September 2022
images efficiently [1]. Several classification algorithms encompass hundreds of different
Published: 18 September 2022
classification issues, and no single classification method can successfully and efficiently
address all classification problems. As a result, answering the question concerning which
Publisher’s Note: MDPI stays neutral
classification approach is best for a particular study is challenging. The fast growth in
with regard to jurisdictional claims in
medical data and imagery in recent years has necessitated the employment of new method-
published maps and institutional affil-
ologies depending on big data technology, artificial intelligence, and machine learning in
iations.
health care, making it an important research area [2]. Given the importance of classification
in the medical field, new approaches for rapidly identifying and evaluating medical images
are required. As a result, this research aims to compare existing and conventional methods
Copyright: © 2022 by the authors.
for medical image classification and, based on these findings, suggest a novel algorithm for
Licensee MDPI, Basel, Switzerland. medical image classification [3].
This article is an open access article The field of medical image processing and analysis has contributed to substantial
distributed under the terms and medical achievements. A correct diagnosis necessitates the precise identification of each
conditions of the Creative Commons disease by integrating methods and techniques that support more effective clinical diag-
Attribution (CC BY) license (https:// nosis depending on images obtained by various imaging modalities that have been used
creativecommons.org/licenses/by/ increasingly widely and successfully to detect illnesses [4]. This study aims to describe
4.0/). the process of medical image analysis, identify the techniques used in the analysis, and

Sensors 2022, 22, 7065. https://doi.org/10.3390/s22187065 https://www.mdpi.com/journal/sensors

Sensors 2022, 22, 7065 2 of 23

give a comprehensive literature review on illness identification based on medical imag-

ing across various diseases and diverse fields and applications in medical imaging. This
study searched for works related to the topic of the systematic literature review (SLR) and
provided information about the process applied to article selection. In the final step, we
kept only the forty most relevant articles that answered our research questions related to
medical image processing and analysis techniques for diagnosis.
The study is organized as follows. Section 2 describes the methodology of research
applied in the SLR. Section 3 provides a detailed explanation of medical imaging modalities
and medical image analysis processes in the surveyed studies. Here, the image processing
methods for disease diagnosis developed by the researcher are described. Various filtering
and image improvement methods are discussed. Popular segmentation methods are pre-
sented, feature extraction methods are introduced, and the classification methods utilized
for human disease diagnosis and evaluation metrics are discussed. Section 7 discusses the
outcomes and the future work.

2. Research Methodology
This section describes the protocol utilized to locate, collect, and assess the state-of-
the-art techniques under study. It is divided into four phases: research questions, research
strategy, article selection criteria, and research results.

2.1. Research Questions

The SLR (Systematic Literature Review) aims to address the research questions by
finding all relevant research outcomes from previous studies. The research questions are
divided into five sub-questions:
Q1: What are the modalities of medical imaging?
Q2: What is the task of medical image processing and analysis?
Q3: Which medical image processing methods are most used in diagnostic systems?
Q4: What diagnostic techniques have been adopted and developed?
Q5: Is the system that has been adopted or designed capable of producing good results?
We searched various databases such as Elsevier, IEEE Explorer, Springer, and Google
Scholar. We included the relevant studies that mainly focus on two or more questions based
on our research questions.

2.2. Research Strategy

Our systematic literature review collected many studies related to our search topic
over the last five years, between 2017 and 2021, from the following databases: Elsevier,
IEEE Xplorer, Springer, and Google Scholar. We used combinations of keywords and terms,
including “Medical Image Analysis”, “Medical Image Processing”, “Medical Image Pro-
cessing Techniques”, “Medical Image Processing Techniques” AND “Disease Diagnosis”,
and “Diagnostic Techniques” AND “Medical Image Processing”, and we obtained about
3204 articles meeting our searched keywords. After removing 206 duplicate articles, the
remaining set consisted of 2998 articles. After examining these studies by their titles and
abstracts, a new set of keywords were applied, including “Diseases Classification”, “Ma-
chine Learning”, “Deep Learning”, “Neural Networks”, and “Hybrid Diagnosis System”,
and 2900 articles were excluded. Of the 98 articles remaining, 58 were excluded after deep
reading and based on the exclusion criteria focused on articles that used machine learning
and deep learning methods to create a hybrid diagnostic system based on the merging of
two methods or the modification or improvement of a common method and its develop-
ment for proposing possible areas for future research. These articles reached the fourth
stage of our systematic exclusion technique and dealt with different topics and methods
related to machine learning (ML) methods found in [5–32], deep learning (DL) strategies
found in [33–54], and convolutional neural network (CNN) approaches found in [55–62].
Finally, 40 studies fulfilling our research criteria were obtained to be deeply analyzed. In
this way, the most suitable articles were selected based on the PRISMA (Preferred Reporting
Sensors 2022, 22, 7065 3 of 25

methods related to machine learning (ML) methods found in [5–32], deep learning (DL)
Sensors 2022, 22, 7065 strategies found in [33–54], and convolutional neural network (CNN) approaches3 offound 23
in [55–62]. Finally, 40 studies fulfilling our research criteria were obtained to be deeply
analyzed. In this way, the most suitable articles were selected based on the PRISMA (Pre-
ferred
ItemsReporting
for SystematicItemsReviews
for Systematic Reviews andtechnique,
and Meta-Analyses) Meta-Analyses)
as showntechnique,
in Figureas1.shown
By
inapplying
Figure 1.theByPRISMA
applying the PRISMA
technique, whichtechnique, which
is appropriate for is
anyappropriate for any systematic
systematic literature review,
literature review,
we kept only the we kept
most only the
relevant mostfrom
articles relevant
largearticles fromInlarge
databases. databases.
the final step, asInshown
the final
step, as shown
in Figure 1, theinlast
Figure 1, the
article last article
set was not only seta was not
result ofonly a result ofselection
the automatic the automatic
based selec-
on
tion basedcombinations
keyword on keywordbut combinations but also
also represented represented
the answers to ourthe answers
research to ourthat
questions research
are
questions
discussedthat are discussed
in Section 2.3. in Section 2.3.

Figure 1. Flowchart of the process for article selection.

Figure 1. Flowchart of the process for article selection.
2.3. Criteria for Article Selection
2.3. Criteria for Article
The following Selection
criteria were determined to choose articles:
1. The following
Articles usingcriteria
the mostwere determined
up-to-date to choose
techniques articles: medical images.
for analyzing
1.2. Articlesusing
Articles that were written
the most in the English
up-to-date language.
techniques for analyzing medical images.
2.3. Articlesthat
Articles published in the last
were written fiveEnglish
in the years (2017–2021).
language.
3.4. Studiespublished
Articles that were presented
in the lastatfive
a peer-reviewed conferences or journals.
years (2017–2021).
4. Studies that the
Following were presented
definition of at
thea peer-reviewed conferences
inclusion criteria, or journals.
the following exclusion criteria
were determined:
Following the definition of the inclusion criteria, the following exclusion criteria were
1. determined:
Duplicate references from the various electronic archives that were searched.
2. Articles with
1. Duplicate a page from
references countthe
of less than four.
various electronic archives that were searched.
3. Articles that fail to respond to any
2. Articles with a page count of less than of the research questions.
four.
4. Articles that were written in a language other
3. Articles that fail to respond to any of the research than questions.
English
5. Articles that did not address the study’s goals.
 2.4. Research Results
After searching the scientific databases and identifying the research results, the mo
Sensors 2022, 22, 7065 relevant articles related to the research aims were found, which included articles 4 of 23 that e
plored different medical image processing techniques and focused on the techniques
diagnosing diseases. The selected articles were carefully read, and the extracted resu
were 2.4.
analyzed
Researchand assessed to summarize the existing research, identify the most use
Results
techniques, and
After propose
searching thepossible areas forand
scientific databases future research.
identifying the research results, the most
relevant articles related to the research aims were found, which included articles that
explored different medical image processing techniques and focused on the techniques of
3. Results of Systematic Review
diagnosing diseases. The selected articles were carefully read, and the extracted results
This
were section
analyzedisand
divided into
assessed two parts:the
to summarize theexisting
first deals withidentify
research, medicalthe imaging modaliti
most useful
techniques, and propose possible areas for future research.
and the second deals with the analysis of medical images. Each revolves around the ma
objectives of the
3. Results systematic
of Systematic review.
Review
This section is divided into two parts: the first deals with medical imaging modalities,
3.1. Medical
and the Imaging Modalities
second deals with the analysis of medical images. Each revolves around the main
objectives images
Medical of the systematic
play a review.
critical role in assisting health care workers in reaching p
tients3.1.
forMedical
diagnosis
Imagingand treatment. Medical image processing is a set of procedures f
Modalities
extractingMedical
clinically
imagesuseful
play adata from
critical various
role in assistingimaging
health caremodalities for diagnosis
workers in reaching patients [63]. N
merous medical and
for diagnosis imaging modalities
treatment. include
Medical image ionizing
processing radiation,
is a set magnetic
of procedures resonance, n
for extracting
clear clinically
medicine, useful data from
optical various and
methods, imaging modalitiesas
ultrasound forthe
diagnosis
media.[63]. Numerous
Each modular medi-
media h
cal imaging modalities include ionizing radiation, magnetic resonance, nuclear medicine,
unique characteristics and responses to the human body’s structure [64]. These modalit
optical methods, and ultrasound as the media. Each modular media has unique character-
serveistics
various purposes,
and responses to such as obtaining
the human images
body’s structure inside
[64]. Thesethe humanserve
modalities bodyvarious
or image sa
ples of parts that
purposes, cannot
such as be images
obtaining seen with
insidethe nakedbody
the human eye or[65]. The
image classification
samples of medi
of parts that
imaging modalities
cannot be seen withand
thethe main
naked types
eye [65]. Theofclassification
imaging methodsof medicaladdressed in the survey
imaging modalities
and the main types of imaging
studies are illustrated in Figure 2. methods addressed in the surveyed studies are illustrated
in Figure 2.

FigureFigure
2. Classification ofofmedical
2. Classification imagingmodalities.
medical imaging modalities.

The The distribution of the forty chosen studies that used different modalities is illustrated
distribution of the forty chosen studies that used different modalities is illu
in Table 1; this table shows the detailed distribution of publication references, imaging
tratedmodality,
in Table 1; of
type this tableand
disease, shows thedatabases.
medical detailed distribution of publication references, i
aging modality, type of disease, and medical databases.
Sensors 2022, 22, 7065 5 of 23

Table 1. Distribution of studies for different medical imaging modalities.

Studies (Author (Year) [Ref]) Imaging Modality Type of Disease Medical Database
Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (https://adni.loni.usc.edu/) (accessed on 18
Danni Cheng et al. (2017) [66] PET Alzheimer’s disease
June 2022). The dataset contained 339 brain images (93 AD, 146 MCI, 100 NC subjects).
Harvard Medical School website,
Syrine Neffati et al. (2017) [67] MRI Brain disease Open Access Series of Imaging Studies (OASIS) website. The dataset contained normal brains and seven types
of pathological brains with a total of 226 images (38 normal brains and 188 pathological brains).
Varun Jain et al. (2017) [68] MRI Brain tumor SICAS Medical Image Repository dataset contained 25 MRI brain images (20 benign, 5 malignant).
Travancore scan center, Thiruvananthapuram (www.liveratlas.org) (accessed on 19 June 2022). The dataset
Anjukrishna et al. (2017) [69] CT Liver cancer tumor
contained 80 abdominal CT images, 20 of a normal liver and the rest images of various liver diseases.
International Skin Imaging Collaboration (ISIC) dataset contained images of two classes (skin angioma and
Shouvik Chakraborty et al. (2017) [70] Dermoscopy skin imaging Skin cancers
basel cell carcinoma).
Dermatology Online Atlas (www.dermis.net), http://homepages.inf.ed.ac.uk/rbf/DERMOFIT/ (accessed on
Soumya Sourav et al. (2017) [71] Dermoscopy skin imaging Dermatological diseases
19 June 2022). The dataset contained 3000 images of four types (psoriasis, herpes, eczema, and melanoma).
MRI ADNI database (http://adni.loni.usc.edu) (accessed on 18 June 2022). The ADNI study was applied to people
Priyanka Lodha et al. (2018) [72] Alzheimer’s disease
PET between the age of 55 and 90.
Keerthana T K et al. (2018) [73] MRI Brain tumor Brain MRI medical image dataset which contained normal, benign, and malignant images.
CT scan image dataset collected from patients (age ranges from 35 to 75). The datasets contained 400 images
Latika A. Thamke et al. (2018) [74] CT Lung diseases
(100 normal, 100 pleural effusion, 100 bronchitis, and 100 emphysema).
The authors prepared a rich database that included two classes (diabetic and non-diabetic) of 400 people from
Abid Sarwar et al. (2018) [75] Medical data (non-image) Diabetes type-II
a large geographical area (age ranges from 5 to 75).
The authors gathered specimen images of the sickly greeneries, then trained and stored them in the database.
Pallavi. B et al. (2019) [76] Dermoscopy imaging Malignant melanoma skin cancer disease
This database contained normal and abnormal images.
The NCBI dataset associated with osteoporosis. The authenticated medical center Medpix NLM website. The
Neeraj Kumar et al. (2019) [77] CT Bone disease (osteoporosis)
database contained two classes with features (plane, modality, age, fracture, gender, weight, and history).
The KVASIR dataset consisted of 4000 images containing 8 classes of GI diseases. Some of the supplied image
Rabi et al. (2019) [78] Endoscopic images Gastrointestinal (GI) diseases
classes feature a green image depicting the location and form of the endoscope within the intestine.
Database gathered from the IMBA web page contained normal and abnormal CT images of cancers for both
Sakshi Sharma et al. (2019) [79] CT Lung cancer disease
males and females.
The dataset was based on previous literature. The dataset contained 5856 images (normal, bacteria,
Smir S. Yadav et al. (2019) [80] X-ray Pneumonia disease
and viruses).
Lung Image Database Consortium (LIDC). The dataset was composed of diagnostic and cancer screening
Sannasi Chakravarthy et al. (2019) [81] CT Lung cancer disease
thoracic CAT examinations with marked-up interpretations.
Aamir Bhat et al. (2019) [82] X-ray Osteoarthritis disease The datasets were gathered from numerous hospitals. The dataset contained 126 knee joint X-ray images.
Mohammed Aledhari et al. (2019) [83] X-ray chest radiographs Pneumonia disease National Institute of Health (NIH) dataset contained 1431 labeled X-ray images (normal and pneumonia).
Medical University of Bialystok (MUB) Clinic Hospital, publicly available DRIVE STARE, Kaggle.
Maciej Szymkowski et a l. (2020) [84] Retina color images Retina disease diagnosis
The database contained 500 images (250 healthy samples and 250 pathological samples).
Shashank Awasthi et al. (2020) [85] MRI Alzheimer’s disease The publicly available OASIS dataset contained MRI images of normal and AD patients.
Sensors 2022, 22, 7065 6 of 23

Table 1. Cont.

Studies (Author (Year) [Ref]) Imaging Modality Type of Disease Medical Database
The National Institute of Mental Health and Neurosciences (NIMHANS) dataset contained 800 images for
Halebeedu Suresha et al. (2020) [86] MRI Alzheimer’s disease 99 people (60 normal and 39 AD with age ranges from 55 to 87 years). The ADNI dataset contained 819 subjects
(229 normal, 192 AD, and 398 MCI).
Mammography Image Analysis Society (MIAS) dataset contained 320 mammogram images (51 malignant,
Chiranji Lal Chowdhary et al. (2020) [87] Breast X-ray (mammogram) Breast cancer disease
63 benign, and 206 normal).
MRI image dataset gathered by the authors. The number of samples considered for evaluation in this article was
Fateme Gholami et al. (2020) [88] MRI Brain tumors
30 gray images.
CT PET Center, Postgraduate Institute of Medical Education and Research (PGIMER), Chandigarh, India.
Jaspreet Kaur et al. (2020) [89] Detecting cancer
PET The dataset contained 200 medical images.
Broad Institute (http://portals.broadinstitute.org/cgi-bin/cancer/datasets) (accessed on 20 June 2022). The
Begüm Erkal et al. (2020) [90] Medical data (non-image) Brain cancer
dataset contained 42 samples (7129 features and 5 classes).
The dataset of patients from the York university contained the cardiac MRI DICOM images of patients suffering
G U Santosh Kumar et al. (2021) [91] MRI Cardiovascular diseases, cardiac attack
from various cardiovascular diseases.
Nasr Gharaibeh et al. (2021) [92] Retinal fundus images Diabetic retinopathy (DR) The Image-Ret database included two sub-databases (i.e., DIARETDB0 and DIARETDB1).
Laiba Zubair et al. (2021) [93] MRI Alzheimer’s disease The ADNI dataset contained 145 MRI images (39 AD, 45 CN, and 68 MCI).
The LC25000 dataset contained 25,000 color images of five types of lung and colon tissues (colon adenocarcinoma,
Mehedi Masud et al. (2021) [94] Histopathological image Lung cancer and colon cancer
benign colonic tissue, lung adenocarcinoma, benign lung tissue, and lung squamous cell carcinoma).
Harvard medical school database contained 70 images (25 normal and 45 abnormal, which comprised three
Muhammad Assam et al. (2021) [95] MRI MRI brain images classification
different kinds of diseases: brain tumor, acute stroke, and Alzheimer disease).
MRI images of the brain were collected from various hospitals and compiled in a Kaggle dataset that contained
Antor Hashan et al. (2021) [96] MRI Brain tumor
400 images (230 brain tumors and 170 normal)
The MIAS mammogram database (http://peipa.esex.ac.uk/info/mias.html) (accessed on 21 June 2022). The
Xiuzhen Cai et al. (2021) [97] Breast X-ray (mammogram) Breast cancer
dataset contained 322 mammography images that were taken from the UK National Breast Screening Program.
Makineni Kumar et al. (2021) [98] MRI Lung cancer disease MRI lung cancer image dataset, which contained normal and tumor images.
Md Riajuliislam et al. (2021) [99] Medical data (non-image) Thyroid disease (hypothyroid) The dataset from the registered diagnostic center Dhaka, Bangladesh, contained 519 data with 9 attributes.
The dataset was gathered by the authors and contained various ophthalmic diseases such as macular degeneration,
Aasawari M. Patankar et al.(2021) [100] Fundus images Ophthalmic diseases
retinopathy, myopia, cataract, and other abnormalities.
RIM-ONE dataset contained 455 fundus photographs.
Deepak R. Parashar et al. (2021) [101] Retinal fundus images Glaucoma classification Drishti-GS1 dataset contained 101 retinal images.
RIM-ONE released 1 dataset which contained 40 images.
Alzheimer’s dataset (available: https://www.kaggle.com/tourist55/alzheimers) (accessed on 22 June 2022)
Majdah Alshammari et al. (2021) [102] MRI Alzheimer’s disease contained 4 classes of diseases (896 mild demented, 64 moderate demented, 2240 very mild demented, and
3200 non-demented).
Kaggle dataset (available: https://www.kaggle.com/c/aptos2019-blindness-detection) (accessed on 22 June 2022)
Mohammed Al-Smadi et al. (2021) [103] Fundus images Diabetic retinopathy
contained 3562 images and was obtained from various clinics in India and represents real-world data.
X-ray Kaggle dataset contained 1400 healthy and pneumonia images.
Saeed Mohagheghi et al. (2021) [104] COVID-19 disease
CT J. Cohen’s COVID-19 dataset contained 210 COVID-19, normal, and pneumonia images.
CT, MRI, X-ray, PET, US, The authors downloaded 10,000 medical images for each modality from Open-i Biomedical Image Search Engine,
Sonit Singh et al. (2021) [105] Classifying medical image modalities
Microscopy images National Institute of Health, and U.S. National Laboratory of Medicine.
Sensors 2022, 22, 7065 7 of 23

3.2. Medical Image Analysis

This section is divided into subsections. These subsections introduce the major medical
image analysis methods used in the studies reviewed, including image preprocessing,
image segmentation, feature extraction to classification, and evaluation metrics.

3.2.1. Medical Image Preprocessing

Image processing is a method for enhancing the quality of an image after eliminating
irrelevant image data. Medical images include many irrelevant and unwanted segments.
To remove these segments in an image, some preprocessing methods are required. Image
preprocessing aims to improve the quality of the images contained in the dataset, which
improves the results of segmentation and feature extraction methods [106]. This section
presents the preprocessing methods of the studies surveyed. One of the most important
techniques to improve medical images and remove noise is filtering. One of the most
important filters used is the median filter [69,73,87,91,92,95,99], with its major benefit being
to preserve the edges and remove noise. In [71], the authors used the synthetic minority
over-sampling technique to generate synthetic samples from minor classes rather than
simply replicating them, and they used the changing perspective of images technique to
expand the dataset. To generate new images, computer vision techniques such as gray
scaling, blurring, enhancing contrast, changing the color channel, sharpening, minimizing
noise, and smoothing were used. In [73], the authors described in their study how to utilize
pixel-wise interpolation and the modified quadratic transform-based Radon transform
algorithm to improve the contrast of the images. In [73,85,100], the authors applied the
contrast limited adaptive histogram equalization (CLAHE) approach and morphologi-
cal operations to remove noises and improve the images. In [76,81], the authors used
a fast adaptive median filter to eliminate the noise from medical images and maintain
accurate information. In [77], binarization and thinning techniques were applied to the
selected images to produce better images. In [80,88,89], the authors improved the noise
removal process for medical images to obtain clear images using a Weiner filter for con-
trast adjustment. In [84], the authors developed two methods to convert the input color
images to grayscale, denoise, enhance contrast, and normalize the histogram. The first
method utilized only a green channel, while the second method depended on calculat-
ing the grayscale by taking an average of red, green, and blue channels computed using
certain weights.
In [86,96,102], the authors addressed the normalization method to enhance the visual
quality of images. In [87] the authors used noise removal techniques such as max-min
filter, midpoint filter, quantum noise filter, alpha-trimmed mean filter, impulse noise
filter, and wavelet thresholding methods for noise removal from images to smoothened
them to obtain better images. In [91], the authors employed linear and non-linear image
processing filters such as mean, Gaussian, Log, and fuzzy filters to eliminate noise from
images. In [92], the authors presented the BBHE approach (i.e., Brightness Preserving Bi-
Histogram Equalization), a method that decomposes the original image into two sub-images
to overcome the drawback of histogram equalization. This is accomplished using a gray-
level and histogram equalization approach for each sub-image. The Gradwrap algorithm,
addressed in [93], improves the image appearance by removing image distortions. The B1
non-uniformity algorithm corrects the image color and intensity information, and then the
N3 bias field correlation is applied after the Gradwrap and B1 non-uniformity algorithms
to correct intensity distortion. In [94], the authors used unsharp masking, a popular image
sharpening method, to improve the contrast of histopathological images. In [97] the Wang–
Mendel algorithm was employed to remove noise from the images; this algorithm is one
of the most effective approaches for eliminating noise from medical images due to its
high speed.
In [101], the authors analyzed fundus images for image scaling; R, G, and B channel
selection; and the preprocessed green channel components were analyzed by the 2D-VMD
technique, which has a non-stationary, non-recursive, and completely adaptive decomposi-
Sensors 2022, 22, 7065 8 of 23

tion technique for signal image analysis. In [103], the authors used image normalization
and data over-sampling techniques with data augmentation to enlarge the dataset so that
it could be used for deep learning tasks. Data augmentation approaches use a variety of
operations to images, including scaling, geometric deformation, noise addition, alterations
to the lighting, and image flipping. The authors employed the straightforward and efficient
semi-supervised learning technique of pseudo-label to improve the performance of deep
neural network models.

3.2.2. Segmentation Techniques

Image segmentation is responsible for recognizing and outlining items of interest in
input images. Automatic medical image segmentation aids in the diagnosis of diseases
and the identification of pathogens. Image segmentation methods can be divided into
machine learning methods such as supervised and unsupervised machine learning and
classical segmentation methods such as threshold-based, edge-based, and region-based
methods [107]. This section presents the segmentation methods used by the studies sur-
veyed. In [69,73,74,76,77,81], the authors used thresholding. In this method, the image
is partitioned into its foreground and background depending on the threshold value.
In [69,96], the authors applied the contour technique; active contour is an active segmenta-
tion model that separates the pixels of interest from an image using energy forces and limits.
In [79,91], the authors suggested using watershed segmentation techniques; this approach
is used to separate different objects. The study [82,91] employed Otsu’s threshold approach
to extract the object; this approach finds an optimum threshold automatically. Because of
its simple calculation, Otsu’s is the most successful technique for image thresholding. The
authors in [84] used the Gaussian matched filter approach with binarization realized with
local entropy thresholding. This method allowed them to acquire more exact results than
using only one binarization threshold. Additionally, the authors employed the crossing
number algorithm or minutiae detection to extract minutiae and then automatically count
their number.
The study [95] adopted a color space-based method for mammography image segmen-
tation, followed by mathematical morphology. Post-processing mathematical morphology
improved performance, including filling, closure, and other operations.

3.2.3. Feature Extraction Techniques

Feature extraction is the process of extracting meaningful data from raw data. It is
crucial in image processing as it enhances the image’s quality by reducing the dimension-
ality of the image, extracting the unique features, and transforming the input data into a
set of features that are used for classification purposes. There are various features such
as color, texture, and shape, and each type has several methods to extract features from
medical images [108]. The distribution of the chosen studies that used different feature
extraction and reduction methods is illustrated in Table 2; this table comprises the detailed
distribution of publication references, the type of features, and the method used.

3.2.4. Classification Techniques

Image classification is a difficult task in image analysis. The primary purpose of
medical image classification is to accurately establish which parts of the human body are
infected with the disease [109]. This study reviews the chosen studies that used the newest
classification techniques for medical image classification. The study [67] proposed a new
classifier depending on KPCA and SVM. The support vector machine with kernel principal
component analysis (SVM-KPCA) approach was designed to classify the images, and the
proposed method achieved effective results, which gave 100% accuracy, 100% sensitivity,
and 100% specificity. In the study [68], the authors suggested a hybrid approach consisting
of a radial basis function neural network (RBFNN) to classify brain MRI images, the
accuracy of this model varied from 80% to 90%.
Sensors 2022, 22, 7065 9 of 23

Table 2. Distribution of studies for different feature extraction methods.

Studies (Author (Year) [Ref]) Type of Features Method Used

DWT transforms to extract features.
Syrine Neffati et al. (2017) [67] Texture features
The kernel PCA (KPCA) technique for feature reduction.
DWT transform for feature extraction.
Varun Jain et al. (2017) [68] Texture features
PCA technique for diminishing the number of features.
AnjKrishna M et al. (2017) [69] Texture features SFTA and modified SFTA algorithms to extract features.
Several key points such as features from SIFT to detect key (interest) points and feature extraction.
Shouvik Chakraborty et al. (2017) [70]
which it creates a descriptor for that point Bag-of-features concept to decrease the number of key points.
The full volume of the brain is extracted from MRI images and other
Priyanka Lodha, et al. (2018) [72] Cognitive and biological features
cognitive and biological features.
Keerthana T K et al. (2018) [73] Texture features GLCM for features extraction.
GLCM for features extraction.
Texture and shape features
Latika A. Thamke et al. (2018) [74] Moment Invariant (MI).
Pixel coefficient value
WHT transforms to calculate the pixel value of the image.
Nine features are mean, standard deviation, entropy, RMS, variance,
Pallavi. B et al. (2019) [76] Combination of texture and color features
smoothness, kurtosis, skewness, and inverse difference momentum.
DWT transforms for feature extraction.
Bahaa Rabi et al. (2019) [78] Texture and shape features HOG transforms for feature extraction.
PCA and SVD methods for feature reduction.
GLCM for feature extraction.
Sannasi Chakravarthy et al. (2019) [81] Texture features
CCSA for feature selection manually.
Aamir Bhat et al. (2019) [82] Texture features HOG and DWT transform for feature extraction.
Extract the characteristic points (so-called minutiae) on retina images
Maciej Szymkowsk et al. (2020) [84] Biometric features
and count the number of minutiae on the resulting image.
The features such as mean of zero crossing, mean of IMF, standard
Shashank Awasthi et al. (2020) [85] Combination of fractal and statistical features deviation, etc.
PCA for feature reduction.
Halebeedu Suresha et al. (2020) [86] Texture features HOG transforms for feature extraction.
Vital features for segmentation The vital features are texture, shape, margin, and intensity.
Chiranji Lal Chowdhary et al. (2020) [87]
Texture features for classification The gray-level histogram computations compute the texture features.
Jaspreet Kaur et al. (2020) [89] Texture features GLCM technique for feature extraction.
Haralick and shape-based features.
Nasr Gharaibeh et al. (2021) [92] Texture and shape features
US-PSO-RR algorithm for feature reduction.
2D Fourier Features (2D-FFT) and
Mehedi Masud et al. (2021) [94] Texture features
2D Wavelet Features (2D-DWT).
DWT transforms for feature extraction.
Muhammad Assam et al. (2021) [95] Color features
Color Moments (CMs) to reduce the number of features.
Xiuzhen Cai et al. (2021) [97] Texture features Combination of GLCM and DWT.
Makineni Kumar et al. (2021) [98] Shape features Diameter, Perimeter, Entropy, Intensity, and Eccentricity.
Age, sex, ID, etc.
Md Riajuliislam et al. (2021) [99] Data of patients
PCA for feature selection.
Aasawari M. Patankar et al. (2021) [100] Texture, color, and edges features Wavelet transform, DCT approach, and color information.
Texture-based Zernike moment, chip histogram, and
Deepak R. Parashar et al. (2021) [101] Texture features
Haralick descriptors.
Sonit Singh el at. (2021) [105] Statistical and texture features Local binary pattern (LBP) and GLCM.

The study [70] developed a neural-based detection system by employing skin imaging
for two different skin disorders. The ANN was trained using the non-dominated sorting
genetic algorithm-II, a well-known multi-objective optimization technique (NN-NSGA-II).
The proposed model was compared to two well-known metaheuristic-based classifiers,
NN-PSO (ANN trained with PSO) and NN-CS (ANN trained with Cuckoo Search). The
proposed bag-of-features enabled the N-NSGA-II model and obtained 90.56% accuracy,
88.26% precision, 93.64% recall, and 90.87% F-measure with the experimental data, indicat-
ing its superiority over other models. The study [71] applied multiple artificial intelligence
(AI) techniques, such as the convolutional neural network and support vector machine,
which were combined with image processing tools to construct a superior structure; the
accuracy attained after training with CNN alone was approximately 91%, which was raised
to approximately 95.3% when combined with the SVM. In [79], the authors developed
a new SVM-FA (support vector machine optimized with firefly technique) classifier for
Sensors 2022, 22, 7065 10 of 23

diagnosing lung cancer in CT images where the SVM classifier, optimized with the firefly
technique, was applied to the preprocessed data. A comparative analysis was conducted
between the proposed work, traditional work, and the SVM classifier to assess the compe-
tence level of the proposed SVM-FA technique. The suggested work was successful and
efficient, achieving an accuracy of 96 % and specificity of 83.3%.
The study [86] applied a deep neural network (DNN) with the rectified Adam opti-
mizer to detect Alzheimer’s disease in MRI images. The experimental outcomes showed
that DNN with the rectified Adam optimizer outperformed the existing work, landmark-
based features with the SVM classifier, with a 16% classification accuracy. In a study [92],
the authors addressed a combination of MLC (maximum likelihood classifier) and SVM
(support vector machine) classifiers for the classification and diagnosis of DR (diabetic
retinopathy) disease in the fundus images. The researchers utilized these classifiers to raise
the performance level, and the suggested approach demonstrated high accuracy (98.60%),
sensitivity (99%), and specificity (99%). In [95], the authors introduced an individual clas-
sifier called: feed-forward ANN (FF-ANN) and two hybrid classifiers, namely: random
subspace with random forest (RSwithRF) and random subspace with Bayesian network
(RSwithBN), for the classification of MRI brain images. The proposed system showed
that ANN and hybrid classification approaches are the most appropriate for classification
because of their high accuracy rates and achieved an accurate classification of 95.83%,
97.14%, and 95.71%, respectively.
In [100], the authors used various techniques, such as machine learning and different
deep learning models, to predict various ophthalmic diseases; the study showed that the
efficiency of each strategy varied depending on the input dataset and based on the many
symptoms of each disease.

3.2.5. Metric Evaluation

Accuracy, specificity, and sensitivity are the most prevalent metrics utilized in most
disease diagnosis applications for humans. The number of true positive (TP), true negative
(TN), false positive (FP), and false-negative (FN) samples are used to calculate these
measures [110]. The sensitivity determines how many positive samples were identified
(TP), whereas the specificity determines how many negative samples were identified (TN).
The ratio of correctly recognized samples to the total number of samples is used to calculate
classification accuracy [110]. Other metrics such as precision, recall, F1 score, and AUC
must be used in conjunction with these metrics [111]. Table 3 illustrates the number of
times each performance metric was utilized in each study.

Table 3. Frequency count of performance metrics utilized in every chosen study.

Studies [Ref] Performance Metrics No. of Studies

[68,69,71,73,75–77,84,85,89,95,100,102,103,105] Accuracy 15
[67,79,81,88,92,93,97,101] Accuracy, Specificity, Sensitivity 8
[70,82,86,94] Accuracy, Precision, Recall, F1 score 4
Accuracy, Specificity, Sensitivity, Precision, Recall,
[72,74,98,104] 4
F1 score
[90,96] Accuracy, F1 score 2
[66] Accuracy, Specificity, Sensitivity, AUC 1
[80] Accuracy, Specificity, Recall 1
[78] Accuracy, Specificity, Precision, Recall, F1 score 1
[87] Accuracy, Sensitivity 1
[83] Accuracy, Specificity 1
[99] Accuracy, Specificity, Sensitivity, F1 score 1
Sensors 2022, 22, 7065 11 of 23

4. Machine Learning Techniques

This study classified machine learning (ML) segmentation and classification tech-
niques as supervised and unsupervised machine learning. Supervised learning algorithms
generate mathematical models using a set of labeled data (images) which are utilized for
training, examples of supervised machine learning algorithms are K-nearest neighbors
(K-NN), support vector machines (SVM), decision trees (DT), linear regression, logistic
regression, random forest (RF), artificial neural networks (ANN), gradient boosting, and
naïve Bayes models. Unsupervised learning algorithms create mathematical models based
on a set of data that only contains inputs and no required output labels. The algorithms
identify patterns in the data and classify them. Examples of unsupervised machine learning
algorithms are K-mean clustering, hierarchical clustering, Apriori algorithm, principal
component analysis (PCA), and fuzzy C-means (FCM) [112].
In the surveyed studies, several techniques of machine learning (ML) in segmenta-
tion and classification were deployed. The study [85] compared the IPFCM (intuitionist
possibilistic fuzzy C-mean) method with the hybridization of the negative function of the
intuitionistic and the negative function of the possibilistic for mammogram image segmen-
tation with conventional segmentation methods such as the Otsu algorithm, FCM (fuzzy
C-mean) clustering, IFCM (intuitionist fuzzy C-mean) clustering, and PFCM (possibilistic
fuzzy C-mean). Moreover, they found that this proposed method was the best approach. In
studies [88,91,98], the authors used the fuzzy clustering method. Fuzzy clustering is the
most extensively used for image segmentation because of its benefits compared with tradi-
tional clustering methods, which includes: making regions more homogeneous, decreasing
the number of erroneous spots, reducing noise sensitivity, and removing noisy regions [88].
In [92], the authors introduced the MSW-FCM technique (modified spatial weighted fuzzy
C-means) for accurately segmenting blood vessels in retinal fundus images.
The distribution of the chosen studies that used the most popular machine learning
techniques to diagnose human body disease is illustrated in Table 4. This table comprises
the detailed distribution of publication references, techniques used, classified tasks, and
classification accuracy results.

Table 4. Distribution of studies for the most common ML classification methods.

Studies (Author (Year) [Ref]) Techniques Task Accuracy Results

Classify liver tumor: Normal,
Naïve Bayes 78%
Anju krishna M et al. (2017) [69] Cirrhosis, HCC, and
and SVM 92.5%
Hemangioma
SVM, 97.56%
Gradient boosting, 97.25%
Priyanka Lodha, et al. (2018) [72] NN, Alzheimer’s disease 98.36%
K-NN, 95%
and RF 97.86%
The system provided better accuracy
Diagnosis and classification of brain
Keerthana T K et al. (2018) [73] SVM with the
tumor disease
genetic algorithm GA-SVM.
Classify lung disease kinds: Normal, The K-NN classifier gave better
Latika A. Thamke et al. (2018) [74] K-NN, Multiclass-SVM, and DT Bronchitis, Pleural, outcomes (97.5%) than other
Emphysema, and Effusion classifiers.
The results showed that the ensemble
ANN, SVM, K-NN, Naïve Bayes, and
Abid Sarwar et al. (2018) [75] Diabetes type-II technique provided a superior
Ensemble
accuracy of 98.60%.
A combination of texture and color
features outcomes in the highest
Pallavi. B et al. (2019) [76] Multi-level SVM Classify skin disease
classification accuracy using multi
SVM.
SVM Bone disease prediction of Both classifiers gave efficient
Neeraj Kumar et al. (2019) [77]
and NN osteoporosis outcomes.
The highest accuracy of classification
SVM, K-NN,
Bahaa Rabi et al. (2019) [78] Classify eight GI classes was 99.8%,
LD, and DT
using the decision tree.
Sannasi Chakravarthy et al. (2019) [81] PNN Lung cancer at the early stage 90%
Sensors 2022, 22, 7065 12 of 23

Table 4. Cont.

Studies (Author (Year) [Ref]) Techniques Task Accuracy Results

SVM 85.33%
Aamir Bhat et al. (2019) [82] Knee osteoarthritis in early stage
and ANN 73.82%
SVM (linear),
95.25%
SVM (3rd-degree
96.45%
Maciej Szymkowski et al. (2020) [84] polynomial), The healthy and unhealthy
73.96%
K-NN,
80.42%
and K-Means
81%
Logistic regression, Alzheimer’s disease classification in
Shashank Awasthi et al. (2020) [85] 79.88%
Naïve Bayes, and SVM MRI image
92.34%
DT, 82.5%
RSDA, 96.1%
Chiranji Lal Chowdhary et al. (2020) [87] Detecting breast cancer
SVM, 88.13%
and FSVM 98.85%
The accuracy of SVM
varied from 95.5–98%.
SVM Detecting cancer (cancerous and
Jaspreet Kaur et al. (2020) [89] Accuracy of t h e K-
and K-NN non-cancerous)
NN classifier varied
from 69.5–95.5%.
The experimental outcomes suggest
Random Forest, K-NN, Bayes, LMT, that the Multilayer Perceptron
Begüm Erkal et al. (2020) [90] Detecting brain cancer
DT, and Multilayer Perceptron. approach outperforms other machine
learning methods in accuracy.
SVM, 99.35%
DT, 99.35%
Md Riajuliislam et al. (2021) [99] RF, Hypothyroid at the early stage 99.35%
Naïve Bayes, 94.23%
and Logistic egression 99.35%
Deepak R. Parashar et al. (2021) [101] Multi-stage classifier (SVM) Glaucoma classification 91%

5. Deep Learning
Deep learning (DL) is well-known for its performance in image segmentation and
classification models [113]. Convolution neural networks (CNN) are the most extensively
utilized deep learning approach in the articles reviewed. CNN is an important image pro-
cessing approach that allows for accurately classifying aberrant and normal samples [114].
CNN employs a layered perceptron-driven architecture composed of fully connected net-
works in which every neuron in one layer is coupled to all neurons in the subsequent
layers. The input images, an in-depth feature extractor, and a classifier are the three main
components of a CNN [111]. There are three kinds of layers in CNN, each of which per-
forms a different function: (1) convolutional, (2) pooling, and (3) fully connected. The
convolutional layer extracts the characteristics of the structure. The fully connected layer
then decides which class the current input belongs to, depending on the retrieved features.
Then, the pooling layer is responsible for shrinking feature maps and network parameters.
Transfer learning algorithms can increase CNN performance in the case of limited input
data [115]. A CNN can be created from scratch using an existing pre-trained network
without retraining or fine-tuning a pre-trained network on a target dataset [111].
According to this review, some research publications used distinct deep neural network
architectures; Table 5 summarizes studies that used deep learning in disease diagnosis.
The study [66] suggested a new classification structure that depends on a combination
of 2D CNN and recurrent neural networks (RNN) that learn the properties of 3D PET
images by decomposing them into a series of 2D slices. The intra-slice characteristics are
captured using hierarchical 2D CNNs, while the inter-slice features are extracted using the
gated recurrent unit (GRU) of an RNN for final classification. The experimental outcomes
showed that the suggested method has promising performance for Alzheimer’s disease
(AD) diagnosis. The authors in [80] applied CNN-based algorithm on a chest X-ray dataset
to detect pneumonia. Three approaches were examined, a linear support vector machine
classifier with local rotation and orientation-free features, transfer learning on two CNN
models: Visual Geometry Group, i.e., VGG16 and InceptionV3, and a capsule network
training from scratch. Data augmentation is a data preprocessing technique applied to all
Sensors 2022, 22, 7065 13 of 23

three approaches, the outcomes revealed that data augmentation is an effective technique
for all three algorithms to improve performance and efficiency. In [86], the authors proposed
a successful approach for predicting the probability of brain cancers in MRI images using
CNNs and the (Adam) optimizer algorithm. An adaptive moment estimation (Adam)
optimizer has been introduced to expedite training the network and evaluate the model
to attain maximum accuracy. The study [93] introduced a new deep learning-based CNN
model created by the Bayesian optimization algorithm for classifying Alzheimer’s disease,
mild cognitive impairment (MCI), and cognitively normal (CN) in MRI images. The
proposed method gives extraordinary outcomes compared to the existing techniques. The
study [97] compared a new optimized version of CNN and a new, improved metaheuristic,
named the advanced thermal exchange optimizer for the detection of breast cancers with
three different techniques including multilayer perceptron (MLP), multiple instances (MI),
and transfer learning (TL), which were applied in the MIAS mammography database to
demonstrate the superiority of proposed method.

Table 5. Summary of articles that included deep learning in detecting diseases.

Studies (Author (Year) [Ref]) Architecture Task Results

91.19% (Accuracy),
91.40% (Sensitivity),
91.00% (Specificity), and
Classify AD vs. NC
Danni Cheng et al. (2017) [66] CNN + RNN 95.28% (AUC). 78.86% (Accuracy),
Classify MCI vs. NC
78.08% (Sensitivity),
80.00% (Specificity), and
83.90% (AUC).
0.923 (Accuracy), 0.926 (Specificity),
and 0.923
VGG16 Classify pneumonia
Smir S. Yadav et al. (2019) [80] (Recall)
InceptionV3 Capsule Net vs. normal
0.824 (Accuracy), 0.846 (Specificity),
and 0.824 (Recall)
VGG16 68% (Accuracy)
ResNet-50 58% (Accuracy)
Mohammed Aledhari et al. (2019) [83] Pneumonia vs. normal
Inception v3 53% (Accuracy)
(fine-tuned VGG16) 75% (Accuracy)
Maciej Szymkowski et al. (2020) [84] ResNet50 Retina diagnosis 86% (Accuracy)
Halebeedu Subbaraya et al. (2020) [86] CNN + Adam Optimizer Classify brain tumors 90% (Accuracy) and 0.89 (F1 score)
CNN+ AD vs. CN
Laiba Zubair et al. (2021) [93] 99.3% (Accuracy)
Bayesian optimization vs. MCI
96.33% (Accuracy), 96.39%
Mehedi Masud et al. (2021) [94] CNN Five types of lungs and colon cancers (Precision), 96.37% (Recall),
96.38% (F-Measure)
CNN + Thermal Exchange Breast cancer 93.79 (Accuracy), 96.89 (Sensitivity),
Xiuzhen Cai et al. (2021) [97]
Optimizer diagnosis and 67.7 (Specificity).
96% (Accuracy), 96% (Sensitivity),
Makineni Kumar et al. (2021) [98] LTD-CNN Lung tumor detection 93% (Specificity), and
94% (Precision).
AD vs. mild demented vs. moderate
98% accuracy for testing and 97% in
Majdah Alshammari et al. (2021) [102] CNN + ML+ Adam Optimizer demented vs. very mild demented,
training.
vs. non-demented
77.6% (QWK)
ResNet,
82% (QWK)
Inception V3
79.6% (QWK)
Mohammed Al-Smadi et al. (2021) [103] Inception V4 DenseNet Diabetic retinopathy diagnosis
81.8% (QWK)
Xception
80.9% (QWK)
EfficientNet
80% (QWK)
97% (Accuracy), 99% (Sensitivity),
Saeed Mohagheghi et al. (2021) [104] CNN with CBMIR COVID-19 disease diagnosis 99% (Specificity 97% (Precision),
99% (Recall), and 98% (F-Measure).
VGG-16 62% (Accuracy)
VGG-19 98.18% (Accuracy)
ResNet-50 90% (Accuracy)
Sonit Singh el at. (2021) [105] Inception-v3 Classifying medical image modalities 99% (Accuracy)
Xception 98.36% (Accuracy)
MobileNet 98.73% (Accuracy)
Inception-ResNet v2 98.18% (Accuracy)
Sensors 2022, 22, 7065 14 of 23

In a study [102], the authors suggested using the ML approach in the training process
to create a prediction model by training the CNN algorithm in addition to using the “Adam”
optimizer from the Python Keras optimization library that has an initial learning rate of
0.001. The evaluation was carried out after partitioning the data into 80% and 20% for
training and evaluation, respectively, to compute the accuracy of classification and loss
of model over a set number of 10 epochs, the proposed algorithm gave good results. In
the study [103], the authors developed a transfer learning-based technique to determine
the severity of diabetic retinopathy; the proposed model was a deep learning model that
combined multiple pre-trained image classification CNN models with the global average
pooling (GAP) technique. The accuracy of the model attained 82.4% quadratic weighted
transfer learning kappa (QWK). In [104], the authors proposed a diagnosis system based on
deep neural networks and image retrieval method. Transfer learning and hashing functions
increased the CNN performance and image retrieval algorithms. The proposed system
attained an accuracy of 97% for CNN and a content-based medical image retrieval (CBMIR)
method. The study [105] compared the effectiveness of modern CNN models for the task
of modality classification and reported the superiority of a deep learning-based method
over classic feature engineering approaches based on multi-label learning algorithms. The
experimental results demonstrated that deep learning is more efficient than traditional
methods and produced better and more robust feature representations when compared to
handcrafted feature extraction approaches. The findings showed that deep transfer learning
techniques work well in the medical field, where data is scarce. The Google Inception-v3
model performed the best when it came to classifying medical picture modalities. Except
for VGG-16 and Res-Net-50, the other models behaved similarly to Inception-v3.
Generally, the studies utilized accuracy, specificity, sensitivity, precision, recall, f1
score, and AUC as evaluation metrics. According to the conclusions of the studies, deep
learning algorithms achieved good results in most of the evaluation metrics (as in Table 5).

6. Diseases Diagnosis System

Image processing has been extensively employed in various illness diagnosis proce-
dures (human, animal, and plant), helping professionals select the appropriate treatment.
In diagnosing human diseases, image processing techniques play an essential role. They
can be used to identify disease signs (on the skin, for example) or in molecular research
using microscope images that show the anatomy of the tissues [116]. The disease diag-
nosis systems comprise stages and methods for diagnosis. It starts from the first stage,
which is the stage of collecting images from different sources, either through a database
available online or collecting the images from different sources, i.e., images of patients
available on the internet or from a specific hospital. Then the preprocessing image stage as
different filtering methods are applied to improve the image, and then the segmentation
stage follows by isolating the regions of interest and extracting the important features in
the feature extraction stage. At the end of the process, the input image is classified and
metrics for evaluating the effectiveness of disease diagnosis techniques are applied. The
general system stages for the diagnosis of any disease in image processing are illustrated
in Figure 3. This figure shows a conceptual map that explains concepts linked to disease
diagnosis steps and a descriptive brief of the major evaluation metrics.
In this study, several disease diagnosis methods were studied that used different
image treatment and classification strategies, which dealt with diagnosing human diseases,
to examine the new and important methods that addressed in the studies.
Sensors 2022, 22, 7065 Sensors 2022, 22, 7065 15 of 23 17 o

Figure
Figure 3. Conceptual map3.ofConceptual map of disease
disease diagnosis system.diagnosis system.

7. Discussion and Future

In thisDirections
study, several disease diagnosis methods were studied that used different
The growing age treatment
interest and classification
in employing strategies,
medical image which dealt with
processing diagnosing
approaches andhuman
AI diseas
to examine the new and important methods that addressed in
techniques, such as ML and DL, may reduce the doctor’s workload and repetitive and the studies.
monotonous procedures to diagnose and analyze patient data and images [117]. In this
7. Discussion
section, we discuss and Future
the state-of-the-art Directionsand datasets used to diagnose human
approaches
The growing
diseases through an answer interest
to research in employing
questions medical
and future image processing approaches and
directions.
techniques, such as ML and DL, may reduce the doctor’s workload and repetitive a
7.1. Answer to Research Questions
monotonous procedures to diagnose and analyze patient data and images [117]. In t
The data were section, we discuss
extracted from the thestudied
state-of-the-art
articles approaches
as they were and datasets and
explained usedclarified
to diagnose hum
diseases through
to answer the research questions. an answer to research questions and future directions.
“Q1: What are the modalities of medical imaging?” The methods of medical imaging
7.1. Answer to Research Questions
were identified in the articles, where many modern imaging methods for diseases have been
The data were
used, which help in diagnosing extracted
diseases in thefrom
earlythestages.
studied articles
The MRI as they were
imaging explained
modality wasand clarif
to answer the research questions.
used in fourteen articles in the diagnosis of brain diseases, lung diseases, cardiovascular
diseases, and cardiac“Q1: Whatthe
attacks; areCTtheimaging
modalities of medical
modality wasimaging?” The methods
used in eight articlesofinmedical
the imag
were identified in the articles, where many modern
diagnosis of lung diseases, liver tumors, and bone diseases. The X-Ray modality was imaging methods for diseases ha
been used, which help in diagnosing diseases in the early stages. The MRI imaging m
utilized in seven articles to diagnose osteoarthritis disease, pneumonia disease, and breast
dality was used in fourteen articles in the diagnosis of brain diseases, lung diseases, c
cancer. Five articles used retina fundus images to diagnose eye illnesses for retina disease
diovascular diseases, and cardiac attacks; the CT imaging modality was used in eight
diagnosis. PET imaging was utilized in four articles to diagnose Alzheimer’s disease, and
ticles in the diagnosis of lung diseases, liver tumors, and bone diseases. The X-Ray m
dermoscopy skin imaging was used to diagnose skin problems.
dality was utilized in seven articles to diagnose osteoarthritis disease, pneumonia disea
“Q2: What is the task of medical image processing and analysis?“ In the analysis of
and breast cancer. Five articles used retina fundus images to diagnose eye illnesses
the articles, it wasretina
identified
disease that the taskPET
diagnosis. of the analysis
imaging wasof medical
utilized images
in four andto
articles processing
diagnose Alzheime
it to diagnose several diseases in different parts of the human body to
disease, and dermoscopy skin imaging was used to diagnose skin problems. assist the patient in
detecting the disease “Q2:and treating
What is it theintask
the of
early stage.image processing and analysis?“ In the analysis
medical
“Q3: Which themedical image
articles, it wasprocessing
identified methods
that the taskare of
most used in diagnostic
the analysis systems?”
of medical images and process
Through the analysis of the articles,
it to diagnose severalwe discovered
diseases that by
in different using
parts filtering
of the humantechniques,
body to assistwethe patien
may improve thedetecting
initial image and get
the disease anda treating
more exact it in detection of the ROI borders. Noise
the early stage.
can be reduced by smoothing with a low pass or median filtering, while edge sharpening
and increased contrast can help to keep the ROIs’ borders. Color normalization, such as
histogram equalization or specification, can boost contrast. A variety of transforms might
be used to extract important features. Fourier and wavelet transform can be utilized to
Sensors 2022, 22, 7065 16 of 23

locate conditions in a domain other than the spatial one. Because the borders of the object
or regions containing the crucial information for the diagnosis must be identified correctly,
image segmentation is one of the most essential phases in a disease diagnosis procedure.
The simplest but effective method is gray-level segmentation utilizing a single, multiple,
or more advanced (Otsu) criteria. Several human illness diagnosis approaches have used
active contour detection and modifications, such as a snake. Several clustering approaches
were applied in the segmentation process, such as K-mean and fuzzy C-means, to separate
the ROIs. The articles used many methods to extract the important features from objects.
We found that the texture features were the most used in analyzing the articles. It was
obvious from the analysis of the articles that the most popular classification method used
in articles was a neural network with all types and SVM, where it was used alone or with
another technique as a hybrid system for the classification of diseases. The rest of the
classification techniques examined in this review such as K-means, K-NN, decision trees,
naïve Bayes, random forest, logistic regression, and gradient boosting. All the classification
or clustering methods can be exploited in the last stage of the diagnosis application to
create a new diagnosis system to identify disease.
“Q4: What diagnostic techniques have been adopted and developed?” Analyzing the
articles, we found that some of the articles adopted new hybrid methods for classification
and diagnosis of diseases, such as using the SVM method with KPSA, SVM with FA, SVM
with MLC, as well as using it with fuzzy. Moreover, the combination of random forest (RF)
with random subspace (RS) and the networks CNN with RNN and with optimizers were
used to create systems for disease diagnosis.
“Q5: Is the system that has been adopted or designed capable of producing good
results?” The main classification methods discussed in the articles achieved efficient and
good results in diagnosing diseases. The accuracy of SVM ranged between 88% and
100% in the several human disease diagnosis applications examined, naïve Bayes achieved
accuracy between 78% and 94%, decision trees ranged between 82% and 99%, and random
forest ranged between 80% and 99%. The different kinds of neural networks achieved
accuracy between 73% and 97%, and the accuracy of CNN ranged between 86% and 99.3%.
Finally, the accuracy of the K-nearest neighbor ranged between 73% and 95.5% in the
referenced approaches.

7.2. Future Directions

It takes great effort to improve the performance of classification or diagnosis of diseases
using multiple methods of medical images. In future work, we will try to present new
research directions that can be further exploited in disease diagnosis through medical
image processing techniques. We will compare the most common methods that are used in
the diagnosis systems and select the most effective methods that introduce higher accuracy
to the diagnosis for the medical image database that we use to build a new diagnostic
system for diseases. The future research directions are discussed briefly as follows:
A. Study and analysis of the best and most common classification and diagnosis methods.
B. Experimenting with a set of medical data to compute the classification accuracy of
the methods used.
C. Comparison of classification accuracy of these methods to identify the methods that
are highly accurate.
D. Building a new diagnostic method by deriving a hybrid method, modifying a
previous method, or combining previous methods.
The current study improved our knowledge of finding the best techniques that can be
beneficial to our future research. We hope that this systematic literature review will also be
useful to other researchers in their endeavor to improve disease diagnosis through medical
image processing techniques. Due to our comparisons, the most common methods used in
diagnosis systems have been discussed, making it easier to select effective methods that
introduce higher accuracy to diagnosis methods using medical databases.
Sensors 2022, 22, 7065 17 of 23

8. Conclusions
Medical imaging plays a critical role in inspecting and diagnosing human diseases.
For diagnosis, many algorithms based on diverse methodologies have been created. As a
result, disease detection has become an important topic in medical image processing and
medical imaging research.
This review studied the articles on disease diagnosis published between 2017 and
2021. Overall, forty articles were analyzed from specialized academic repositories. The
review focused on six factors: datasets used, various medical imaging modalities, image
preprocessing techniques, image segmentation techniques, feature extraction techniques,
classification approaches, and performance metrics, which were used to build and evaluate
the disease diagnostic models. In addition, this systematic review highlighted a variety of
AI techniques and presented a comprehensive study by exploring new diagnosis techniques,
disease diagnosis issues, provided a variety of insightful information (such as the use of
ML and DL), and an evaluation for each study. We aimed to address our study questions
about effective diagnosis methodologies and discover the solutions proposed by many
researchers in the diagnosis of diseases based on this. It was discovered that developing a
new disease diagnosis method is quite important.
According to the findings of this SLR, researchers have adopted various methods
to classify medical images associated with multiple disease diagnoses. These methods
have shown promising results in terms of accuracy, cost, and detection speed. We found
after analyzing the forty articles that the best preprocessing technique was the median
filter, which was used in many studies, as has proven its ability to reduce noise and
preserve the boundaries of the object. Regarding segmentation approaches, threshold
techniques were the most used to extract the lesion from an image. Threshold-based
approaches are the most often utilized among all the traditional methods, according to
classic review publications, due to their applicability for numerous segmentation issues in
medical images [118]. The methods for extracting features from medical images depend on
the images used by analyzing the articles, we found that extracting texture features gave
the best results. As for the applied classification methods, we found that the support vector
machine method gave the best result in classification, as its accuracy in [67] reached 100%
when it was used with KPC (support vector machine with kernel principal component
analysis (SVM- KPCA)) approach. Thus, this method achieved the best accuracy and the
best performance in diagnosis. Among the machine learning-based approaches whose
associated works and analyses are presented, the supervised learning methods, notably the
neural network-based methods, were the most widely employed, with different kinds of
neural networks used to identify various diseases.
We also discovered that deep learning utilizing the CNN network has unique skills
and advances in recognizing and classifying medical images, particularly those connected
to breast, lung, and brain cancer. Other common classification approaches comprise fuzzy
clustering, K-NN, K-means, decision trees, random forests, and other prominent classifica-
tion algorithms. The most utilized measures were accuracy, sensitivity, and specificity.
This study aimed to propose future research directions by focusing on imaging modali-
ties, techniques, and procedures used in the reviewed articles. Furthermore, this publication
will aid in the development of new research that assesses and compares various medical
image processing and analysis techniques. The findings provided in this SRL reveal that
tremendous progress has been made in medical image processing over the last five years. In
addition, the goal of this SRL was to undertake a detailed analysis of research achievements
linked to the usage of medical image processing techniques applied to medical databases
to know the current state-of-the-art techniques.

Author Contributions: The concept of the article was proposed by N.P., the data resources and
validation were contributed by B.M.R., the formal analysis, investigation, and draft preparation were
performed by B.M.R. The supervision and review of the study were headed by N.P. The final writing
Sensors 2022, 22, 7065 18 of 23

was critically revised by N.P. and finally approved by the authors. All authors have read and agreed
to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The study does not report any data.
Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations
CT—Computed Tomography; MRI—Magnetic Resonance Imaging; US—Ultrasound;
PET–Positron Emission Tomography; GI—Gastrointestinal Diseases; FCM—Fuzzy C-Mean;
IFCM—Intuitionist Fuzzy C-Mean; PFCM—Possibilistic Fuzzy C-Mean; IPFCM—Intuitionist
Possibilistic Fuzzy C-Mean; MSW-FCM—Modified Spatial Weighted Fuzzy C-mean;
DWT—Discrete Wavelet Transform; PCA—Principal Component Analysis; KPCA—Kernel
Principal Component Analysis; SFTA—Segmentation-based Fractal Textural Analysis;
SIFT—Scale Invariant Feature Transform; GLCM—Gray Level Co-occurrence Matrix MI-
Moment Invariant; WHT—Walsh Hadamard Transform; HOG—Histogram of Oriented
Gradients; SVD—Single Value Decomposition; CCSA—Chaotic Crow Search Algorithm;
2D-FFT—Two Dimension Fourier Features Transform; 2D-DWT—Two Dimension Wavelet
Features Transform CM—Color Moment; SVM—Support Vector Machine; RBFNN—Radial
Basis Function Neural Network; ANN—Artificial Neural Network; SVM-FA—Support Vector
Machine with Firefly Technique; DNN—Deep Neural Network; MLC—Maximum Likelihood
Classifier; FF-ANN—Feed-Forward Artificial Neural Network; K-NN—K Nearest Neighbor;
LDA—Linear Discriminate Analysis; PNN—Probabilistic Neural Network; RSDA—Rough
Set DATA Analysis; FSVM—Fuzzy Support Vector Machine; CNN—Convolutional Neu-
ral Network; RF—Random Forest; DT—Decision Trees; ROI—Region Of Interest; RNN—
Recurrent Neural Network.

References
1. Ansari, Z.; Mateenuddin, Q.; Abdullah, A. Performance research on medical data classification using traditional and soft
computing techniques. Int. J. Recent Technol. Eng. (IJRTE) 2019, 8, 990–995.
2. Kermany, D.S.; Goldbaum, M.; Cai, W.; Valentim, C.C.S.; Liang, H.; Baxter, S.L.; McKeown, A.; Yang, G.; Wu, X.; Yan, F.; et al.
Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning. Cell 2018, 172, 1122–1131.e9. [CrossRef]
[PubMed]
3. Sudheer Kumar, E.; Shoba Bindu, C. Medical image analysis using deep learning: A systematic literature review. In Proceedings
of the International Conference on Emerging Technologies in Computer Engineering, Jaipur, India, 4–5 February 2019; pp. 81–97.
4. Murtaza, G.; Shuib, L.; Abdul Wahab, A.W.; Mujtaba, G.; Nweke, H.F.; Al-garadi, M.A.; Zulfiqar, F.; Raza, G.; Azmi, N.A. Deep
learning-based breast cancer classification through medical imaging modalities: State of the art and research challenges. Artif.
Intell. Rev. 2020, 53, 1655–1720. [CrossRef]
5. Myszczynska, M.A.; Ojamies, P.N.; Lacoste, A.; Neil, D.; Saffari, A.; Mead, R.; Hautbergue, G.M.; Holbrook, J.D.; Ferraiuolo, L.
Applications of machine learning to diagnosis and treatment of neurodegenerative diseases. Nat. Rev. Neurol. 2020, 16, 440–456.
[CrossRef]
6. Ker, J.; Bai, Y.; Lee, H.Y.; Rao, J.; Wang, L. Automated brain histology classification using machine learning. J. Clin. Neurosci. 2019,
66, 239–245. [CrossRef]
7. Amrane, M.; Oukid, S.; Gagaoua, I.; Ensari, T. Breast cancer classification using machine learning. In Proceedings of the 2018
Electric Electronics, Computer Science, Biomedical Engineerings’ Meeting (EBBT), Istanbul, Turkey, 18–19 April 2018; pp. 1–4.
8. Vijayvargiya, A.; Kumar, R.; Dey, N.; Tavares, J.M.R. Comparative analysis of machine learning techniques for the classification of
knee abnormality. In Proceedings of the 2020 IEEE 5th International Conference on Computing Communication and Automation
(ICCCA), Greater Noida, India, 30–31 October 2020; pp. 1–6.
9. Barstugan, M.; Ozkaya, U.; Ozturk, S. Coronavirus (COVID-19) classification using ct images by machine learning methods. arXiv
2020, arXiv:2003.09424.
10. Kwekha-Rashid, A.S.; Abduljabbar, H.N.; Alhayani, B. Coronavirus disease (COVID-19) cases analysis using machine-learning
applications. Appl. Nanosci. 2021, 11, 1–13. [CrossRef]
Sensors 2022, 22, 7065 19 of 23

11. Pahar, M.; Klopper, M.; Warren, R.; Niesler, T. COVID-19 cough classification using machine learning and global smartphone
recordings. Comput. Biol. Med. 2021, 135, 104572. [CrossRef]
12. Abdulkareem, N.M.; Abdulazeez, A.M.; Zeebaree, D.Q.; Hasan, D.A. COVID-19 world vaccination progress using machine
learning classification algorithms. Qubahan Acad. J. 2021, 1, 100–105. [CrossRef]
13. Ballı, S. Data analysis of Covid-19 pandemic and short-term cumulative case forecasting using machine learning time series
methods. Chaos Solitons Fractals 2021, 142, 110512. [CrossRef]
14. Sivaranjani, S.; Ananya, S.; Aravinth, J.; Karthika, R. Diabetes prediction using machine learning algorithms with feature
selection and dimensionality reduction. In Proceedings of the 2021 7th International Conference on Advanced Computing and
Communication Systems (ICACCS), Coimbatore, India, 19–20 March 2021; pp. 141–146.
15. Jindal, H.; Agrawal, S.; Khera, R.; Jain, R.; Nagrath, P. Heart disease prediction using machine learning algorithms. In Proceedings
of the IOP conference series: Materials science and engineering, Gorakhpur, India, 14–15 February 2020; p. 012072.
16. Fernandez Escamez, C.S.; Martin Giral, E.; Perucho Martinez, S.; Toledano Fernandez, N. High interpretable machine learning
classifier for early glaucoma diagnosis. Int. J. Ophthalmol. 2021, 14, 393–398. [CrossRef] [PubMed]
17. Mijwil, M.M. Implementation of Machine Learning Techniques for the Classification of Lung X-Ray Images Used to Detect
COVID-19 in Humans. Iraqi J. Sci. 2021, 62, 2099–2109. [CrossRef]
18. Devi, R.L.; Kalaivani, V. Machine learning and IoT-based cardiac arrhythmia diagnosis using statistical and dynamic features of
ECG. J. Supercomput. 2020, 76, 6533–6544. [CrossRef]
19. Khanday, A.M.U.D.; Rabani, S.T.; Khan, Q.R.; Rouf, N.; Mohi Ud Din, M. Machine learning based approaches for detecting
COVID-19 using clinical text data. Int. J. Inf. Technol. 2020, 12, 731–739. [CrossRef] [PubMed]
20. Erickson, B.J.; Korfiatis, P.; Akkus, Z.; Kline, T.L. Machine learning for medical imaging. Radiographics 2017, 37, 505. [CrossRef]
21. Palimkar, P.; Shaw, R.N.; Ghosh, A. Machine learning technique to prognosis diabetes disease: Random forest classifier approach.
In Advanced Computing and Intelligent Technologies; Springer: Berlin/Heidelberg, Germany, 2022; pp. 219–244.
22. Zoabi, Y.; Deri-Rozov, S.; Shomron, N. Machine learning-based prediction of COVID-19 diagnosis based on symptoms. Npj Digit.
Med. 2021, 4, 3. [CrossRef]
23. Raihan, M.M.S.; Shams, A.B.; Preo, R.B. Multi-class electrogastrogram (EGG) signal classification using machine learning
algorithms. In Proceedings of the 2020 23rd International conference on computer and information technology (ICCIT), DHAKA,
Bangladesh, 19–21 December 2020; pp. 1–6.
24. Arumugam, K.; Naved, M.; Shinde, P.P.; Leiva-Chauca, O.; Huaman-Osorio, A.; Gonzales-Yanac, T. Multiple disease prediction
using Machine learning algorithms. Mater. Today Proc. 2021, in press. [CrossRef]
25. Saygılı, A. A new approach for computer-aided detection of coronavirus (COVID-19) from CT and X-ray images using machine
learning methods. Appl. Soft Comput. 2021, 105, 107323. [CrossRef]
26. Ahmad, A.; Garhwal, S.; Ray, S.K.; Kumar, G.; Malebary, S.J.; Barukab, O.M. The number of confirmed cases of covid-19 by using
machine learning: Methods and challenges. Arch. Comput. Methods Eng. 2021, 28, 2645–2653. [CrossRef]
27. Wroge, T.J.; Özkanca, Y.; Demiroglu, C.; Si, D.; Atkins, D.C.; Ghomi, R.H. Parkinson’s Disease Diagnosis Using Machine Learning
and Voice. In Proceedings of the 2018 IEEE Signal Processing in Medicine and Biology Symposium (SPMB), Philadelphia, PA,
USA, 1 December 2018; pp. 1–7.
28. Sun, W.; Zhang, P.; Wang, Z.; Li, D. Prediction of cardiovascular diseases based on machine learning. ASP Trans. Internet Things
2021, 1, 30–35. [CrossRef]
29. Chittora, P.; Chaurasia, S.; Chakrabarti, P.; Kumawat, G.; Chakrabarti, T.; Leonowicz, Z.; Jasiński, M.; Jasiński, Ł.; Gono, R.;
Jasińska, E. Prediction of chronic kidney disease-a machine learning perspective. IEEE Access 2021, 9, 17312–17334. [CrossRef]
30. Wu, C.-C.; Yeh, W.-C.; Hsu, W.-D.; Islam, M.M.; Nguyen, P.A.A.; Poly, T.N.; Wang, Y.-C.; Yang, H.-C.; Li, Y.-C.J. Prediction of fatty
liver disease using machine learning algorithms. Comput. Methods Programs Biomed. 2019, 170, 23–29. [CrossRef] [PubMed]
31. Subramani, P.; BD, P. Prediction of muscular paralysis disease based on hybrid feature extraction with machine learning technique
for COVID-19 and post-COVID-19 patients. Pers. Ubiquitous Comput. 2021, 25, 1–14. [CrossRef] [PubMed]
32. Jena, L.; Patra, B.; Nayak, S.; Mishra, S.; Tripathy, S. Risk prediction of kidney disease using machine learning strategies. In
Intelligent and Cloud Computing; Springer: Berlin/Heidelberg, Germany, 2021; pp. 485–494.
33. Ardakani, A.A.; Kanafi, A.R.; Acharya, U.R.; Khadem, N.; Mohammadi, A. Application of deep learning technique to manage
COVID-19 in routine clinical practice using CT images: Results of 10 convolutional neural networks. Comput. Biol. Med. 2020,
121, 103795. [CrossRef] [PubMed]
34. Han, S.S.; Kim, M.S.; Lim, W.; Park, G.H.; Park, I.; Chang, S.E. Classification of the Clinical Images for Benign and Malignant
Cutaneous Tumors Using a Deep Learning Algorithm. J. Investig. Dermatol. 2018, 138, 1529–1538. [CrossRef]
35. Kavitha, M.; Jayasankar, T.; Venkatesh, P.M.; Mani, G.; Bharatiraja, C.; Twala, B. COVID-19 disease diagnosis using smart deep
learning techniques. J. Appl. Sci. Eng. 2021, 24, 271–277.
36. Klang, E.; Barash, Y.; Margalit, R.Y.; Soffer, S.; Shimon, O.; Albshesh, A.; Ben-Horin, S.; Amitai, M.M.; Eliakim, R.; Kopylov, U.
Deep learning algorithms for automated detection of Crohn’s disease ulcers by video capsule endoscopy. Gastrointest. Endosc.
2020, 91, 606–613.e602. [CrossRef]
37. Oh, S.L.; Hagiwara, Y.; Raghavendra, U.; Yuvaraj, R.; Arunkumar, N.; Murugappan, M.; Acharya, U.R. A deep learning approach
for Parkinson’s disease diagnosis from EEG signals. Neural Comput. Appl. 2020, 32, 10927–10933. [CrossRef]
Sensors 2022, 22, 7065 20 of 23

38. Bychkov, D.; Linder, N.; Turkki, R.; Nordling, S.; Kovanen, P.E.; Verrill, C.; Walliander, M.; Lundin, M.; Haglund, C.; Lundin, J.
Deep learning based tissue analysis predicts outcome in colorectal cancer. Sci. Rep. 2018, 8, 3395. [CrossRef]
39. Vorontsov, E.; Cerny, M.; Régnier, P.; Di Jorio, L.; Pal, C.J.; Lapointe, R.; Vandenbroucke-Menu, F.; Turcotte, S.; Kadoury, S.; Tang,
A. Deep Learning for Automated Segmentation of Liver Lesions at CT in Patients with Colorectal Cancer Liver Metastases. Radiol.
Artif. Intell. 2019, 1, 180014. [CrossRef]
40. Morris, S.A.; Lopez, K.N. Deep learning for detecting congenital heart disease in the fetus. Nat. Med. 2021, 27, 764–765. [CrossRef]
[PubMed]
41. Cong, L.; Feng, W.; Yao, Z.; Zhou, X.; Xiao, W. Deep Learning Model as a New Trend in Computer-aided Diagnosis of Tumor
Pathology for Lung Cancer. J. Cancer 2020, 11, 3615–3622. [CrossRef] [PubMed]
42. Noreen, N.; Palaniappan, S.; Qayyum, A.; Ahmad, I.; Imran, M.; Shoaib, M. A Deep Learning Model Based on Concatenation
Approach for the Diagnosis of Brain Tumor. IEEE Access 2020, 8, 55135–55144. [CrossRef]
43. Liu, Y.; Jain, A.; Eng, C.; Way, D.H.; Lee, K.; Bui, P.; Kanada, K.; de Oliveira Marinho, G.; Gallegos, J.; Gabriele, S.; et al. A deep
learning system for differential diagnosis of skin diseases. Nat. Med. 2020, 26, 900–908. [CrossRef]
44. Xu, X.; Jiang, X.; Ma, C.; Du, P.; Li, X.; Lv, S.; Yu, L.; Ni, Q.; Chen, Y.; Su, J.; et al. A Deep Learning System to Screen Novel
Coronavirus Disease 2019 Pneumonia. Engineering 2020, 6, 1122–1129. [CrossRef]
45. Sun, C.; Xu, A.; Liu, D.; Xiong, Z.; Zhao, F.; Ding, W. Deep Learning-Based Classification of Liver Cancer Histopathology Images
Using Only Global Labels. IEEE J. Biomed. Health Inform. 2020, 24, 1643–1651. [CrossRef]
46. Ayan, E.; Ünver, H.M. Diagnosis of Pneumonia from Chest X-Ray Images Using Deep Learning. In Proceedings of the 2019
Scientific Meeting on Electrical-Electronics & Biomedical Engineering and Computer Science (EBBT), Istanbul, Turkey, 24–26
April 2019; pp. 1–5.
47. Goceri, E. Diagnosis of skin diseases in the era of deep learning and mobile technology. Comput. Biol. Med. 2021, 134, 104458.
[CrossRef]
48. Doppalapudi, S.; Qiu, R.G.; Badr, Y. Lung cancer survival period prediction and understanding: Deep learning approaches. Int. J.
Med. Inform. 2021, 148, 104371. [CrossRef]
49. Jojoa Acosta, M.F.; Caballero Tovar, L.Y.; Garcia-Zapirain, M.B.; Percybrooks, W.S. Melanoma diagnosis using deep learning
techniques on dermatoscopic images. BMC Med. Imaging 2021, 21, 6. [CrossRef]
50. Saratxaga, C.L.; Moya, I.; Picón, A.; Acosta, M.; Moreno-Fernandez-de-Leceta, A.; Garrote, E.; Bereciartua-Perez, A. MRI Deep
Learning-Based Solution for Alzheimer’s Disease Prediction. J. Pers. Med. 2021, 11, 902. [CrossRef]
51. Placido, D.; Yuan, B.; Hjaltelin, J.X.; Haue, A.D.; Chmura, P.J.; Yuan, C.; Kim, J.; Umeton, R.; Antell, G.; Chowdhury, A. Pancreatic
cancer risk predicted from disease trajectories using deep learning. bioRxiv 2022. [CrossRef]
52. He, X.; Yang, X.; Zhang, S.; Zhao, J.; Zhang, Y.; Xing, E.; Xie, P. Sample-efficient deep learning for COVID-19 diagnosis based on
CT scans. medRxiv 2020. [CrossRef]
53. Hosny, K.M.; Kassem, M.A.; Foaud, M.M. Skin Cancer Classification using Deep Learning and Transfer Learning. In Proceedings
of the 2018 9th Cairo International Biomedical Engineering Conference (CIBEC), Cairo, Egypt, 20–22 December 2018; pp. 90–93.
54. Mohammed, M.; Mwambi, H.; Mboya, I.B.; Elbashir, M.K.; Omolo, B. A stacking ensemble deep learning approach to cancer type
classification based on TCGA data. Sci. Rep. 2021, 11, 15626. [CrossRef] [PubMed]
55. Avanzato, R.; Beritelli, F. Automatic ECG Diagnosis Using Convolutional Neural Network. Electronics 2020, 9, 951. [CrossRef]
56. Alanazi, S.A.; Kamruzzaman, M.M.; Islam Sarker, M.N.; Alruwaili, M.; Alhwaiti, Y.; Alshammari, N.; Siddiqi, M.H. Boosting
Breast Cancer Detection Using Convolutional Neural Network. J. Healthc. Eng. 2021, 2021, 5528622. [CrossRef]
57. Saranya, N.; Karthika Renuka, D.; Kanthan, J.N. Brain Tumor Classification Using Convolution Neural Network. J. Phys. Conf.
Ser. 2021, 1916, 012206. [CrossRef]
58. Badža, M.M.; Barjaktarović, M.Č. Classification of Brain Tumors from MRI Images Using a Convolutional Neural Network. Appl.
Sci. 2020, 10, 1999. [CrossRef]
59. Li, L.; Chen, Y.; Shen, Z.; Zhang, X.; Sang, J.; Ding, Y.; Yang, X.; Li, J.; Chen, M.; Jin, C.; et al. Convolutional neural network for the
diagnosis of early gastric cancer based on magnifying narrow band imaging. Gastric Cancer 2020, 23, 126–132. [CrossRef]
60. Sathiyamoorthi, V.; Ilavarasi, A.K.; Murugeswari, K.; Thouheed Ahmed, S.; Aruna Devi, B.; Kalipindi, M. A deep convolutional
neural network based computer aided diagnosis system for the prediction of Alzheimer’s disease in MRI images. Measurement
2021, 171, 108838. [CrossRef]
61. Sekaran, K.; Chandana, P.; Krishna, N.M.; Kadry, S. Deep learning convolutional neural network (CNN) With Gaussian mixture
model for predicting pancreatic cancer. Multimed. Tools Appl. 2020, 79, 10233–10247. [CrossRef]
62. Subramanian, R.R.; Achuth, D.; Kumar, P.S.; Reddy, K.N.k.; Amara, S.; Chowdary, A.S. Skin cancer classification using Con-
volutional neural networks. In Proceedings of the 2021 11th International Conference on Cloud Computing, Data Science &
Engineering (Confluence), Noida, India, 28–29 January 2021; pp. 13–19.
63. Patil, D.D.; Deore, S.G. Medical image segmentation: A review. Int. J. Comput. Sci. Mob. Comput. 2013, 2, 22–27.
64. Miranda, E.; Aryuni, M.; Irwansyah, E. A Survey of Medical Image Classification Techniques. In Proceedings of the 2016
international conference on information management and technology (ICIMTech), Bandung, Indonesia, 16–18 November 2016.
[CrossRef]
Sensors 2022, 22, 7065 21 of 23

65. Nisa, S.Q.; Ismail, A.R.; Ali, M.A.B.M.; Khan, M.S. Medical Image Analysis using Deep Learning: A Review. In Proceedings of the
2020 IEEE 7th International Conference on Engineering Technologies and Applied Sciences (ICETAS), Kuala Lumpur, Malaysia,
18–20 December 2020; pp. 1–3.
66. Cheng, D.; Liu, M. Combining convolutional and recurrent neural networks for Alzheimer’s disease diagnosis using PET images.
In Proceedings of the 2017 IEEE International Conference on Imaging Systems and Techniques (IST), Beijing, China, 18–20 October
2017; pp. 1–5.
67. Neffati, S.; Taouali, O. An MR brain images classification technique via the Gaussian radial basis kernel and SVM. In Proceedings
of the 2017 18th International Conference on Sciences and Techniques of Automatic Control and Computer Engineering (STA),
Monastir, Tunisia, 21–23 December 2017; pp. 611–616.
68. Varun Jain, S.G. Analysis of Brain MRI Tumor Detection and Classification using Hybrid Approach. Int. J. Comput. Sci. Commun.
IJCSC 2017, 8, 42–47.
69. Edwin, D.; Hariharan, S. Classification of Liver Tumor using Modified SFTAbased Multi Class Support Vector Machine. In
Proceedings of the 2017 International Conference on Current Trends in Computer, Electrical, Electronics and Communication
(CTCEEC), Mysore, India, 8–9 September 2017; pp. 854–859.
70. Chakraborty, S.; Mali, K.; Chatterjee, S.; Anand, S.; Basu, A.; Banerjee, S.; Das, M.; Bhattacharya, A. Image based skin dis-
ease detection using hybrid neural network coupled bag-of-features. In Proceedings of the 2017 IEEE 8th Annual Ubiqui-
tous Computing, Electronics and Mobile Communication Conference (UEMCON), New York, NY, USA, 19–21 October 2017;
pp. 242–246.
71. Hasija, Y.; Garg, N.; Sourav, S. Automated detection of dermatological disorders through image-processing and machine learning.
In Proceedings of the 2017 International Conference on Intelligent Sustainable Systems (ICISS), Palladam, India, 7–8 December
2017; pp. 1047–1051.
72. Lodha, P.; Talele, A.; Degaonkar, K. Diagnosis of Alzheimer’s Disease Using Machine Learning. In Proceedings of the 2018 Fourth
International Conference on Computing Communication Control and Automation (ICCUBEA), Pune, India, 16–18 August 2018;
pp. 1–4.
73. Keerthana, T.; Xavier, S.B. An Intelligent System for Early Assessment and Classification of Brain Tumor. In Proceedings of the
2018 Second International Conference on Inventive Communication and Computational Technologies (ICICCT), Coimbatore,
India, 20–21 April 2018; pp. 1265–1268.
74. Thamke, L.A.; Vaidya, M.V. Classification of Lung Diseases Using a Combination of Texture, Shape and Pixel Value by K-NN
Classifier. In Proceedings of the 2018 2nd International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud)
(I-SMAC)I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC), Palladam, India, 30–31 August 2018; pp. 235–240.
75. Sarwar, A.; Ali, M.; Manhas, J.; Sharma, V. Diagnosis of diabetes type-II using hybrid machine learning based ensemble model.
Int. J. Inf. Technol. 2020, 12, 419–428. [CrossRef]
76. Pallavi, B.; Keshvamurthy. A Hybrid Diagnosis System for Malignant Melanoma Detection in Dermoscopic Images. In
Proceedings of the 2019 4th International Conference on Recent Trends on Electronics, Information, Communication & Technology
(RTEICT), Bangalore, India, 17–18 May 2019; pp. 1471–1476.
77. Neeraj Kumar, V.S. A Hybrid Classification and Prediction Methodology for the Diagonosis of Osteoporosis. Int. J. Innov. Technol.
Explor. Eng. 2019, 8, 4648–4653. [CrossRef]
78. Rabi, B.; Attallah, O.; Zaghlool, M.S.; Sharkas, M.A. Automatic Classification of Gastrointestinal Diseases Based on Machine
Learning Techniques. In Proceedings of the 2019 29th International Conference on Computer Theory and Applications (ICCTA),
Alexandria, Egypt, 29–31 October 2019; pp. 85–89.
79. Sakshi Sharma, M.S.; Baljeet, N. CDCT: CT Scan Images based on Mechanism for Lung Cancer Detection. Int. J. Recent Technol.
Eng. IJRTE 2019, 8, 931–935.
80. Yadav, S.S.; Jadhav, S.M. Deep convolutional neural network based medical image classification for disease diagnosis. J. Big Data
2019, 6, 113. [CrossRef]
81. Sannasi Chakravarthy, S.R.; Rajaguru, H. Lung Cancer Detection using Probabilistic Neural Network with modified Crow-Search
Algorithm. Asian Pac. J. Cancer Prev. 2019, 20, 2159–2166. [CrossRef]
82. Aamir Yousuf Bhat, A.S. Normal And Abnormal Detection For Knee Osteoarthritis Using Machine Learning Techniques. Int. J.
Recent Technol. Eng. 2019, 8, 6026–6033. [CrossRef]
83. Aledhari, M.; Joji, S.; Hefeida, M.; Saeed, F. Optimized CNN-based diagnosis system to detect the pneumonia from chest
radiographs. In Proceedings of the 2019 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), San Diego,
CA, USA, 18–21 November 2019; pp. 2405–2412.
84. Szymkowski, M.; Saeed, E.; Omieljanowicz, M.; Omieljanowicz, A.; Saeed, K.; Mariak, Z. A Novelty Approach to Retina
Diagnosing Using Biometric Techniques With SVM and Clustering Algorithms. IEEE Access 2020, 8, 125849–125862. [CrossRef]
85. Awasthi, S.; Kapoor, E.; Srivastava, A.P.; Sanyal, G. A New Alzheimer’s Disease Classification Technique from Brain MRI images.
In Proceedings of the 2020 International Conference on Computation, Automation and Knowledge Management (ICCAKM),
Dubai, United Arab Emirates, 9–10 January 2020; pp. 515–520.
86. Suresha, H.S.; Parthasarathy, S.S. Alzheimer Disease Detection Based on Deep Neural Network with Rectified Adam Optimization
Technique using MRI Analysis. In Proceedings of the 2020 Third International Conference on Advances in Electronics, Computers
and Communications (ICAECC), Bengaluru, India, 11–12 December 2020; pp. 1–6.
Sensors 2022, 22, 7065 22 of 23

87. Chowdhary, C.L.; Mittal, M.; Pattanaik, P.A.; Marszalek, Z. An Efficient Segmentation and Classification System in Medical
Images Using Intuitionist Possibilistic Fuzzy C-Mean Clustering and Fuzzy SVM Algorithm. Sensors 2020, 20, 3903. [CrossRef]
88. Gholami, F. Improved fuzzy clustering with swarm intelligence for medical image analysis. In Proceedings of the 2020 6th Iranian
Conference on Signal Processing and Intelligent Systems (ICSPIS), Mashhad, Iran, 23–24 December 2020; pp. 1–5.
89. Kaur, J.; Singh, M.; Singla, S.K. Study of Fusion of medical images and classification comparison using different kernels of SVM
and K-NN classifiers. In Proceedings of the 2020 First IEEE International Conference on Measurement, Instrumentation, Control
and Automation (ICMICA), Kurukshetra, India, 24–26 June 2020; pp. 1–6.
90. Erkal, B.; Başak, S.; Çiloğlu, A.; Şener, D.D. Multiclass Classification of Brain Cancer with Machine Learning Algorithms. In
Proceedings of the 2020 Medical Technologies Congress (TIPTEKNO), Antalya, Turkey, 19–20 November 2020; pp. 1–4.
91. Kumar, G.U.S.; Kanth, T.V.R.; Raju, S.V.; Malyala, S. Advanced Analysis of Cardiac Image Processing Using Hybrid Approach.
In Proceedings of the 2021 International Conference on Advances in Electrical, Computing, Communication and Sustainable
Technologies (ICAECT), Bhilai, India, 19–20 February 2021; pp. 1–6.
92. Gharaibeh, N.; Al-hazaimeh, O.M.; Abu-Ein, A.; Nahar, K.M. A hybrid svm naïve-bayes classifier for bright lesions recognition in
eye fundus images. Int. J. Electr. Eng. Inf. 2021, 13, 530–545. [CrossRef]
93. Zubair, L.; Irtaza, S.A.; Nida, N.; Haq, N.U. Alzheimer and Mild Cognitive disease Recognition Using Automated Deep Learning
Techniques. In Proceedings of the 2021 International Bhurban Conference on Applied Sciences and Technologies (IBCAST),
Islamabad, Pakistan, 12–16 January 2021; pp. 310–315.
94. Masud, M.; Sikder, N.; Nahid, A.-A.; Bairagi, A.K.; AlZain, M.A. A Machine Learning Approach to Diagnosing Lung and Colon
Cancer Using a Deep Learning-Based Classification Framework. Sensors 2021, 21, 748. [CrossRef] [PubMed]
95. Assam, M.; Kanwal, H.; Farooq, U.; Shah, S.K.; Mehmood, A.; Choi, G.S. An Efficient Classification of MRI Brain Images. IEEE
Access 2021, 9, 33313–33322. [CrossRef]
96. Hashan, A.M.; Agbozo, E.; Al-Saeedi, A.A.K.; Saha, S.; Haidari, A.; Rabi, M.N.F. Brain Tumor Detection in MRI Images Using
Image Processing Techniques. In Proceedings of the 2021 4th International Symposium on Agents, Multi-Agent Systems and
Robotics (ISAMSR), Batu Pahat, Malaysia, 6–8 September 2021; IEE; pp. 24–28.
97. Cai, X.; Li, X.; Razmjooy, N.; Ghadimi, N. Breast Cancer Diagnosis by Convolutional Neural Network and Advanced Thermal
Exchange Optimization Algorithm. Comput. Math. Methods Med. 2021, 2021, 5595180. [CrossRef] [PubMed]
98. Kumar, M.S.; Rao, K.V.; Kumar, G.A. MRI Image Based Classification Model for Lung Tumor Detection Using Convolutional
Neural Networks. Traitement Du Signal 2021, 38, 1837–1842. [CrossRef]
99. Riajuliislam, M.; Rahim, K.Z.; Mahmud, A. Prediction of Thyroid Disease(Hypothyroid) in Early Stage Using Feature Selection and
Classification Techniques. In Proceedings of the 2021 International Conference on Information and Communication Technology
for Sustainable Development (ICICT4SD), Dhaka, Bangladesh, 27–28 February 2021; pp. 60–64.
100. Patankar, A.M.; Thorat, S.S. Diagnosis of Ophthalmic Diseases in Fundus Image Using various Machine Learning Techniques. In
Proceedings of the 2021 6th International Conference on Communication and Electronics Systems (ICCES), Coimbatre, India,
8–10 July 2021; pp. 1114–1118.
101. Parashar, D.R.; Agarwal, D.K. SVM based Supervised Machine Learning Framework for Glaucoma Classification using Retinal
Fundus Images. In Proceedings of the 2021 10th IEEE International Conference on Communication Systems and Network
Technologies (CSNT), Bhopal, India, 18–19 June 2021; pp. 660–663.
102. Alshammari, M.; Mezher, M. A Modified Convolutional Neural Networks For MRI-based Images For Detection and Stage
Classification Of Alzheimer Disease. In Proceedings of the 2021 National Computing Colleges Conference (NCCC), Taif, Saudi
Arabia, 27–28 March 2021; pp. 1–7.
103. Al-Smadi, M.; Hammad, M.; Baker, Q.B.; Sa’ad, A. A transfer learning with deep neural network approach for diabetic retinopathy
classification. Int. J. Electr. Comput. Eng. 2021, 11, 3492. [CrossRef]
104. Mohagheghi, S.; Alizadeh, M.; Safavi, S.M.; Foruzan, A.H.; Chen, Y.-W. Integration of CNN, CBMIR, and visualization techniques
for diagnosis and quantification of covid-19 disease. IEEE J. Biomed. Health Inform. 2021, 25, 1873–1880. [CrossRef]
105. Singh, S.; Ho-Shon, K.; Karimi, S.; Hamey, L. Modality classification and concept detection in medical images using deep transfer
learning. In Proceedings of the 2018 International conference on image and vision computing New Zealand (IVCNZ), Auckland,
New Zealand, 19–21 November 2018; pp. 1–9.
106. Perumal, S.; Thambusamy, V. Preprocessing by contrast enhancement techniques for medical images. Int. J. Pure Appl. Math. 2018,
118, 3681–3688.
107. Kulwa, F.; Li, C.; Zhao, X.; Cai, B.; Xu, N.; Qi, S.; Chen, S.; Teng, Y. A State-of-the-Art Survey for Microorganism Image
Segmentation Methods and Future Potential. IEEE Access 2019, 7, 100243–100269. [CrossRef]
108. Kumar, K.K.; Chaduvula, K.; Markapudi, B. A Detailed Survey On Feature Extraction Techniques In Image Processing For
Medical Image Analysis. Clin. Med. 2020, 7, 2020.
109. Lai, Z.; Deng, H. Medical Image Classification Based on Deep Features Extracted by Deep Model and Statistic Feature Fusion
with Multilayer Perceptron. Comput. Intell. Neurosci. 2018, 2018, 2061516. [CrossRef] [PubMed]
110. Elaziz, M.A.; Hosny, K.M.; Salah, A.; Darwish, M.M.; Lu, S.; Sahlol, A.T. New machine learning method for image-based diagnosis
of COVID-19. PLoS ONE 2020, 15, e0235187. [CrossRef] [PubMed]
111. Khan, W.; Zaki, N.; Ali, L. Intelligent Pneumonia Identification From Chest X-Rays: A Systematic Literature Review. IEEE Access
2021, 9, 51747–51771. [CrossRef]
Sensors 2022, 22, 7065 23 of 23

112. Alloghani, M.; Al-Jumeily, D.; Mustafina, J.; Hussain, A.; Aljaaf, A.J. A Systematic Review on Supervised and unsupervised
Machine learning Algorithms for Data Science. In Supervised and Unsupervised Learning for Data Science; Berry, M., Mohamed, A.,
Yap, B., Eds.; Springer: Cham, Switzerland, 2019; pp. 3–21. [CrossRef]
113. Houssein, E.H.; Emam, M.M.; Ali, A.A.; Suganthan, P.N. Deep and machine learning techniques for medical imaging-based
breast cancer: A comprehensive review. Expert Syst. Appl. 2021, 167, 114161. [CrossRef]
114. Yap, M.H.; Pons, G.; Marti, J.; Ganau, S.; Sentis, M.; Zwiggelaar, R.; Davison, A.K.; Marti, R. Automated breast ultrasound lesions
detection using convolutional neural networks. IEEE J. Biomed. Health Inform. 2017, 22, 1218–1226. [CrossRef]
115. Krizhevsky, A.; Sutskever, I.; Hinton, G.E. ImageNet classification with deep convolutional neural networks. Commun. ACM 2017,
60, 84–90. [CrossRef]
116. Petrellis, N. A Review of Image Processing Techniques Common in Human and Plant Disease Diagnosis. Symmetry 2018, 10, 270.
[CrossRef]
117. D’hooge, J.; Fraser, A.G. Learning About Machine Learning to Create a Self-Driving Echocardiographic Laboratory: Technical
Considerations. Circulation 2018, 138, 1636–1638. [CrossRef]
118. Buettner, R.; Bilo, M.; Bay, N.; Zubac, T. A Systematic Literature Review of Medical Image Analysis Using Deep Learning. In
Proceedings of the 2020 IEEE Symposium on Industrial Electronics & Applications (ISIEA), TBD, Malaysia, 17–18 July 2020;
pp. 1–4.