KWARA STATE UNIVERSITY, MALETE

The University for Community Development

Faculty of Information and Communication Technology
MACHINE LEARNING SYSTEM TO PREDICT TUBERCULOSIS IN
CHEST X-RAY IMAGES

BY
ADEBISI OMOIKHEFE VICTOR
20/7CS/00231

JULY 2024
 MACHINE LEARNING SYSTEM TO PREDICT TUBERCULOSIS
FROM CHEST X-RAY IMAGES
BY
ADEBISI OMOIKHEFE VICTOR

A RESEARCH PROJECT SUBMITTED TO THE DEPARTMENT OF

COMPUTER SCIENCE, FACULTY OF INFORMATION AND
COMMUNICATION TECHNOLOGY, KWARA STATE
UNIVERSITY, MALETE, IN PARTIAL FULFILMENT OF THE
REQUIREMENTS FOR THE AWARD OF BACHELOR OF
SCIENCE (B.Sc.) DEGREE IN COMPUTER SCIENCE.

JULY 2024

ii
 DECLARATION
I hereby declare that this research work titled “machine learning system to predict

tuberculosis from chest x-ray images” is my own work and has not been submitted
by any other person for any degree or qualification at any higher institution. I also declare
that the information provided therein are mine and those that are not mine are properly
acknowledged.

__________________________ ________________________

Name: Adebisi Omoikhefe Victor Signature and Date

iii
 CERTIFICATION
This is to certify that the research project titled “machine learning system to predict
tuberculosis from chest x-ray images” was carried out by “Name of the group
member”. The project has been read and approved as meeting the requirements for the
award of Bachelor of Science (B.Sc.) Degree in Computer Science in the Department of
Computer Science, Faculty of Information and Communication Technology, Kwara State
University, Malete.

______________________ ___________________
Dr. R.M Isiaka Signature/Date
Supervisor

_______________________ ____________________
Dr. (Mrs.) R.S. Babatunde Signature/Date
Head of Department

_______________________ _____________________
External Examiner Signature/Date

iv
 DEDICATION
This Project is dedicated to GOD Almighty, the beginning and the end who has been with
me since my birth till the moment and also, to my mum, my friends and supervisor for their
supports, guidance and prayers.

v
 ACKNOWLEDGEMENT
All praise and adoration belong to God for his mercy and protection over me throughout
my program in the university.
I acknowledge the efforts of my mum (Mrs. Veronica Titilayo Ikhide), may God spare her
life to reap the reward of her labor. My sincere appreciation also goes to my loving and
caring siblings starting from Adebisi Blessing for her leadership role and Samuel
Oyindasola for her courageous words towards the success of this program, and thanks to
entire family and its community in general, May God reward them all abundantly.
Furthermore, I acknowledge the support of my friends from (Solomon Odubanjo, Ayomide
Emmanuel, Olatunji Charles and Elelu Zainab). May Almighty God be with them and
crown their efforts with success.
I appreciate my colleagues in the university, my entire class mates. May God answer our
prayers and crown all our efforts with success. The school authority is also inclusive, for
creating an opportunity and avenue for us to be exposed to the outside world.
My profound gratitude goes to my supervisor, Dr. R.M. Isiaka, who did all he could to
make this report a successful one. My appreciation also goes to all lecturers in the
department.

vi
Contents
CHAPTER ONE .................................................................................................................................. 1
INTRODUCTION ............................................................................................................................ 1
1.1 Background of the Study .............................................................................................. 1
1.2 Statement of the Problem ........................................................................................... 3
1.3 Aim and Objectives ...................................................................................................... 4
1.4 Scope of the Study ....................................................................................................... 4
1.5 Significance and justification of the Study ................................................................... 4
1.6 Definition of terms ....................................................................................................... 4
CHAPTER TWO ................................................................................................................................. 7
LITERATURE REVIEW .................................................................................................................... 7
2.1 Review of related terms ............................................................................................... 7
2.2 Related Work ..................................................................................................................... 8
CHAPTER THREE ............................................................................................................................. 21
SYSTEM ANALYSIS AND DESIGN ................................................................................................. 21
3.1 Data Gathering and Analysis: ..................................................................................... 21
3.1.3 Algorithm -CNN (Convolutional Neural Network)- Development: ............................... 25
3.2 Desktop application development: ............................................................................ 27
CHAPTER FOUR .............................................................................................................................. 30
RESULTS AND DISCUSSIONS ....................................................................................................... 30
4.1 Results ........................................................................................................................ 30
Interfaces ............................................................................................................................... 30
4.2 Discussion......................................................................................................................... 36
CHAPTER FIVE ................................................................................................................................ 37
SUMMARY, CONCLUSION, RECOMMENDATION AND REFERENCES.......................................... 37
5.1 Summary .................................................................................................................... 37
5.2 Advantages of the system .......................................................................................... 37
5.3 Disadvantages ............................................................................................................ 37
5.4 Conclusion .................................................................................................................. 38

vii
 5.5 Recommendation ....................................................................................................... 38
APPENDIX ................................................................................................................................... 43

viii
 ABSTRACT
Tuberculosis (TB) remains a significant global health burden, claiming millions of lives
annually. Early and accurate diagnosis is critical for effective treatment and preventing
disease transmission. While chest X-rays (CXRs) are widely used as a screening tool for
TB, their diagnostic accuracy can be limited by factors such as inter-observer variability
and the expertise of the radiologist. This study investigates the potential of deep learning
to address these challenges and improve TB detection from CXR images. By leveraging
large-scale CXR datasets and advanced deep learning architectures, this research aims to
develop a highly accurate and efficient model capable of identifying TB-related
abnormalities. Successful implementation of such a model could significantly enhance TB
diagnosis, particularly in resource-limited settings where access to expert radiologists may
be scarce. However, addressing challenges such as data quality, model interpretability, and
ethical considerations is essential for the responsible and effective deployment of deep
learning technology in clinical practice. Ultimately, this research seeks to contribute to the
development of innovative tools for TB control and elimination.

ix
 CHAPTER ONE
INTRODUCTION

1.1 Background of the Study

Tuberculosis (TB) is a major cause of morbidity and mortality worldwide. It is estimated
that 25% of the world’s population are infected with Mycobacterium tuberculosis, with a
5–10% lifetime risk of progression into TB disease (Gill et al., 2022). Early recognition of
TB disease and prompt detection of drug resistance are essential to halting its global
burden. Culture, direct microscopy, biomolecular tests and whole genome sequencing are
approved methods of diagnosis; however, their widespread use is often curtailed owing to
costs, local resources, time constraints and operator efficiency. Methods of optimizing
these diagnostics, in addition to developing novel techniques, are under review (Gill et al.,
2022).

A cornerstone of radiological imaging for many decades, chest radiography (chest X-ray,
CXR) remains the most commonly performed radiological exam in the world with
industrialized countries reporting an average 238 erect-view chest X-ray images acquired
per 1000 of population annually (Çallı et al., 2021). Chest X-rays may be divided into three
principal types, according to the position and orientation of the patient relative to the X-
ray source and detector panel: posteroanterior, anteroposterior, lateral. The posteroanterior
(PA) and anteroposterior (AP) views are both considered as frontal, with the X-ray source
positioned to the rear or front of the patient respectively (Çallı et al., 2021). In recent years,
AI solutions have shown to be capable of assisting radiologists and clinicians in detecting
diseases, assessing severity, automatically localizing and quantifying disease features, or
providing an automated assessment of disease prognosis. AI for MI has received
extraordinary attention in 2020, as attested by a multitude of interdisciplinary projects
attempting to blend AI technologies with knowledge from MI in order to combat COVID-

1
19. A keyword search combining AI and MI revealed 2,563 papers in 2019, while 2020
has seen more than twice the number of such papers (5,401, cf. Figure 2).

Machine learning is a branch of computing that studies the design of algorithms with the
ability to “learn.” A subfield would be deep learning, which is a series of techniques that
make use of deep artificial neural networks, that is, with more than one hidden layer, to
computationally imitate the structure and functioning of the human organ and related
diseases (Alsaffar et al., 2021). Radiologists play a crucial role in interpreting medical
images for the diagnosis and prognosis of disease. Although AI technologies have recently
demonstrated performance that matches radiologists’ accuracy in a number of specific
tasks, it remains unclear whether radiologists who adopt AI assistance will replace those
who do not. As (Celi et al.) put it in 2019, ‘‘the question is not whether computers can
outperform human in specific tasks, but how humanity will embrace and adopt these
capabilities into the practice of medicine.’’ (Born et al., 2021). In recent years, deep
learning has become the technique of choice for image analysis tasks and made a
tremendous impact in the field of medical imaging (Litjens et al., 2017). Deep learning is
notoriously data-hungry and the CXR research community has benefited from the
publication of numerous large labeled databases in recent years, predominantly enabled by
the generation of labels through automatic parsing of radiology reports. This trend began
in 2017 with the release of 112,000 images from the NIH clinical center (Wang et al.,
2017b). In 2019 alone, more than 755,000 images were released in 3 labelled databases
(CheXpert (Irvin et al., 2019), MIMIC-CXR (Johnson et al., 2019), PadChest (Bustos et
al., 2020)). The integration of deep learning models into the diagnostic process of TB from
chest X-rays can also address the variability and subjectivity inherent in human
interpretation. Studies have shown that deep learning algorithms can match or even surpass
the performance of human radiologists in certain diagnostic tasks, providing consistent and
reproducible results (Esteva et al., 2017). Despite the potential benefits, the implementation
of deep learning models in TB diagnosis from chest X-rays faces several challenges. One

2
significant issue is the need for large and diverse datasets to train robust models that
generalize well across different populations and imaging conditions (Wang et al., 2017).
Additionally, there are concerns regarding the interpretability of deep learning models, as
these "black box" systems can be difficult to understand and trust by medical professionals.
Ensuring the ethical use of these technologies and addressing potential biases in the training
data are also crucial considerations (Amann et al., 2020).
In conclusion, the diagnosis of tuberculosis from chest X-rays is a critical area where the
application of deep learning models offers substantial promise. By leveraging large
datasets and advanced computational techniques, these models have the potential to
improve the accuracy, consistency, and accessibility of TB diagnosis. However, addressing
the challenges associated with data requirements, model interpretability, and ethical
considerations will be essential for the successful integration of these technologies into
clinical practice. Continued research and collaboration between technologists and
healthcare professionals are necessary to realize the full potential of deep learning in
combating the global TB epidemic.

1.2 Statement of the Problem

X-ray imaging although known for being --non-invasive, quick, and painless, able to
support medical and surgical treatment planning and Guide medical personnel as they
insert catheters or stents inside the body to treat tumors, or remove blood clots. (US FDA,
2023). The traditional methods used in the interpretation of an X-ray image rely on domain
knowledge on i.e Measuring CTR(cardio thoracic ratio), counting the number of ribs
showing, Costophrenic angles, diaphragm and checking if the lungs are clear. The
difficulty with this approach is the time it takes to get back results or interpretation from
radiologists (Ajiboye et al., personal communication, 2024).

3
1.3 Aim and Objectives
To simplify the detection of tuberculosis from chest X-ray images compared to
conventional approaches, with the following objectives:
i. Gathering of chest X-ray image data (from, Zenodo.org)
ii. Convolutional neural network machine learning model development with the
gathered dataset.
iii. Development of a graphical user interface

1.4 Scope of the Study

The system will be designed to enhance the efficiency and effectiveness of radiographers
and radiologists in a laboratory setting. It will streamline workflows by automating image
processing and analysis, improve diagnostic accuracy and communication.

1.5 Significance and justification of the Study

The implementation of the TB detection system has the potential to enhance patient
outcomes, prevent disease transmission and spread. Overall, the study's findings have the
potential to significantly impact TB control efforts and improve public health on a global
scale.

1.6 Definition of terms

Tuberculosis (TB): An infectious disease caused by the bacterium Mycobacterium
tuberculosis, typically affecting the lungs and characterized by coughing, fever, weight
loss, and fatigue.
Artificial Intelligence (AI): The simulation of human intelligence processes by machines,
especially computer systems, to perform tasks such as learning, problem-solving, and
decision-making.
Machine Learning: A subset of AI that enables computers to learn from data and improve
their performance on a task without being explicitly programmed.

4
Deep Learning: A specialized field of machine learning that utilizes artificial neural
networks with multiple layers (deep architectures) to extract high-level features from raw
data.
Convolutional Neural Networks (CNN): A class of deep neural networks, most commonly
applied to analyzing visual imagery, which uses a variation of multilayer perceptrons
designed to recognize patterns in pixel images.
Image Classification: The process of categorizing images into predefined classes or
categories based on their visual content.
Computer-Aided Diagnosis (CAD): A technology that assists healthcare professionals in
interpreting medical images, such as X-rays, by highlighting areas of interest or providing
automated diagnoses.
Radiology: The branch of medicine that deals with the study and interpretation of imaging
technologies, such as X-rays, CT scans, and MRI scans, for diagnosing and treating
diseases.
X-ray Imaging: A medical imaging technique that uses X-rays to create detailed images
of the inside of the body, particularly useful for visualizing bones and detecting
abnormalities such as lung infections.
Diagnostic Accuracy: The degree to which a diagnostic test or procedure correctly
identifies or excludes a disease or condition.
Sensitivity and Specificity: Performance measures of a diagnostic test; sensitivity measures
the proportion of true positive results, while specificity measures the proportion of true
negative results.
Feature Extraction: The process of selecting or extracting relevant features from raw data,
such as images, to represent the underlying patterns or characteristics of interest.
Model Training: The process of using labeled data to teach a machine learning model to
make predictions or classifications by adjusting its internal parameters.
Model Evaluation: The assessment of a trained machine learning model's performance
using metrics such as accuracy, precision, recall, and F1-score.

5
Transfer Learning: A machine learning technique where a model trained on one task is
adapted or fine-tuned to perform another related task, often leading to improved
performance with limited training data

6
 CHAPTER TWO
LITERATURE REVIEW
2.1 Review of related terms

Artificial Intelligence: Artificial Intelligence (AI) in Healthcare: AI involves using

computers to perform tasks that typically require human intelligence. In healthcare, AI can
analyze vast amounts of data, like medical images, to assist in diagnosing diseases more
accurately and quickly than traditional methods.
Medical Imaging: Medical imaging refers to techniques used to create visual
representations of the inside of the body. Chest X-ray imaging is a common method to
view the lungs and can help detect conditions like tuberculosis by revealing abnormalities.
Disease Detection and Diagnosis: This involves identifying diseases from symptoms and
medical tests. Automated detection systems use AI to analyze medical images and provide
diagnoses, potentially improving accuracy and speed compared to manual analysis.
Tuberculosis (TB): TB is an infectious disease that primarily affects the lungs and is
caused by the bacterium Mycobacterium tuberculosis. Detecting TB early using imaging
techniques like chest X-rays is crucial for effective treatment and preventing its spread.
Image Preprocessing: Before AI can analyze medical images, they often need to be
cleaned and enhanced. This process, called image preprocessing, involves improving
image quality by reducing noise and highlighting important features.
Model Development: Creating AI models for medical image analysis involves using neural
networks, particularly Convolutional Neural Networks (CNNs), which are designed to
recognize patterns in images. Techniques like transfer learning and data augmentation help
improve these models.
Evaluation Metrics: To determine how well an AI model performs, we use metrics like
sensitivity (how well it identifies true cases), specificity (how well it avoids false alarms),
and the ROC curve, which helps visualize performance across different thresholds.

7
Data Quality and Quantity: AI models need lots of high-quality, labeled data to learn
effectively. Ensuring that this data is accurate and sufficiently comprehensive is a
significant challenge, as is maintaining patient privacy and data security.
Algorithm Performance: AI models must generalize well to new data without being too
tailored to the training data (overfitting) or too simplistic (underfitting). Ensuring robust
performance across different populations and real-world scenarios is crucial.
Clinical Integration: For AI tools to be useful, they must fit into existing clinical
workflows, be accepted by healthcare professionals, and comply with regulations and
ethical standards. This includes ensuring the tools are user-friendly and provide clear,
actionable insights.
Health Informatics: This field combines healthcare with information technology, focusing
on the management and analysis of health data. Electronic Health Records (EHRs) store
patient information digitally, which can be integrated with AI tools for comprehensive care.
Public Health Implications: Effective TB detection using AI can support public health
initiatives by enabling large-scale screening and early intervention, crucial for controlling
and reducing the spread of tuberculosis in populations.
Technological Advancements: Innovations like cloud computing provide the
infrastructure to store and process large medical datasets, while edge computing allows for
data processing closer to where it is collected, enhancing speed and efficiency in healthcare
applications.

2.2 Related Work

According to (Gill et al., 2022), in the journal New developments in tuberculosis diagnosis
and treatment, the research focused on Developing an AI algorithm for TB diagnosis using
chest X-rays. It Compared the ResNet, VGG, and AlexNet algorithms for TB diagnosis.
The U-Net algorithm was used for lung segmentation before TB classification and the
ResNet34, VGG, and AlexNet was used for the classification network. The following
results were obtained from the research, ResNet model had highest accuracy, sensitivity,

8
and specificity. AI algorithms showed AUC values above 0.99 in testing set. ResNet
algorithm recognized TB regions accurately on chest radiographs. AI models provided
accurate TB diagnoses across different subgroups. AI algorithm distinguished TB
radiographs from non-TB with high accuracy. The major limitations of the research was
that Negative sputum culture results may lead to misdiagnoses. Variability in images from
two different hospitals. Study population limited to people aged 15 years.

In conclusion I feel the research was able to meet up with its intended purpose despite its
limitations.
The research done by (Kik et al., 2022) which was documented in the research article,
“Diagnostic accuracy of chest X-ray interpretation for tuberculosis by three 2 artificial
intelligence-based software in a screening use-case: an individual 3 patient meta-analysis
of global data” aimed to Assess diagnostic accuracy of three CAD solutions for
tuberculosis screening, Evaluate performance of CAD systems against microbiological
reference standard and Compare CAD products with expert radiologists for sensitivity and
specificity. To do this the process of Collecting digital CXRs and demographic data from
6 source studies and analysis of images with CAD4TB, Lunit Insight CXR TB, and qXR
was done. The research highlighted that; three CAD products showed reasonable
diagnostic accuracy for microbiologically confirmed TB. CAD products had comparable
accuracy to experienced radiologists in TB identification and that the WHO recommends
CAD as an alternative to human CXR interpretation for TB screening. Although the study
lacks pre-validation for CAD systems and also didn’t take to account that Exclusion of
healthy individuals may affect sensitivity and specificity rates, the variability in culture
results across different studies. The Impact of certain risk factors on CAD accuracy was
not confirmed also. In conclusion AI algorithms for TB diagnosis in resource-limited
settings can be explored further and Enhanced for more accurate TB diagnosis in diverse
populations.

9
The research by (Qin et al., 2021), on the topic “Tuberculosis detection from chest x-rays
for triaging in a high tuberculosis-burden setting: an evaluation of five artificial intelligence
algorithms” focused on Evaluating five commercial AI algorithms for triaging tuberculosis
detection and to Present a new analytical framework for selecting AI software by using a
large dataset from tuberculosis screening centers in Dhaka and comparing AI algorithms
with radiologists and WHO's Target Product Profile. It discovered that AI algorithms
outperformed radiologists, reducing Xpert tests by 50%, AI algorithms had better
specificity compared to radiologists across classifications and that Performance varied with
age, history of tuberculosis, and patient source. Though there’s a high chance of Limited
generalization to other populations due to training strategies and Variability in performance
due to different training strategies and datasets.

(Geric et al., 2023), did a study “The rise of artificial intelligence reading of chest X-rays
for enhanced TB diagnosis and elimination” The study aimed to evaluate CAD software
for TB detection accuracy in different populations. Assess economic implications of CAD
as a triage tool for TB. Tools like CAD software was used for TB detection, challenges in
standardization, and validation. CAD refinement needed for non-TB lung diseases and sub-
populations. From the study it was discovered that CAD software shows high sensitivity
for TB detection in adults. CAD products need careful scrutiny and independent
performance validation. CAD needs refinement for detecting non-TB lung diseases and in
children. Economic evaluations show CAD as a cost-effective triage tool in TB. Certain
limitations were discovered from the study; CAD software needs local data calibration due
to accuracy heterogeneity. Performance varies with software versions, age, BMI, and
disease types. Limited evidence for CAD accuracy in children and non-TB abnormalities.
In conclusion, it enhances TB screening but needs refinement for non-TB diseases
detection and Validation for TB screening in children is essential for it.

10
The research (Application of artificial intelligence in digital chest radiography reading for
pulmonary tuberculosis screening) done by (Cao et al., 2020) focused on developing CAD
systems for TB screening using deep learning technology and to evaluate AI software
platforms for TB-associated abnormalities in CXRs. DCNNs were used for TB detection
with high accuracy. Training was done on TB-specific dataset, testing was also done on
non-TB dataset. Results showed that CAD systems detect abnormalities in CXRs
comparable to radiologists. DL systems outperform human radiologists in detecting TB-
related abnormalities. AI-based CAD systems can be used for TB screening and triage.
Challenges included generalizability, overdiagnosis, and dataset distribution shifts. CXR
has modest specificity and high inter-reader variability. Shortage of qualified radiologists
in TB prevalent settings. High cost of CXR hardware limits availability at primary care
level. It can be concluded that AI in CXR reading enhances TB screening, overcoming
radiologist shortages. CAD systems improve TB detection accuracy, aiding in early
diagnosis

According to (Codlin et al., 2021), in the research “Independent evaluation of 12 artifcial

intelligence solutions for the detection of tuberculosis” aimed to evaluate 12 AI solutions
for tuberculosis detection performance. Assess CAD software performance against human
readers in TB screening. Investigate the impact of radiography equipment on CAD
software performance. Address the gap between TB treatment and incidence for
elimination. Explore the potential of AI in improving TB screening initiatives. Developed
test library of CXR images for independent CAD evaluation. Evaluated 12 CAD software
solutions using TB clinicians as reference. The results showed that Six CAD systems
performed on par with the Expert Reader. Qure.ai, Delft Imaging, and Lunit showed
significantly better performance. Xpert positivity was higher with the DRTECH
radiography system. CAD software performance can be affected by radiography equipment
used. Some of the highlighted limitations were; High CAD score with negative Xpert test
affects software performance. False positive Xpert results in patients with past TB

11
treatment. Performance impairment based on radiography system used for CXR capture.
Inter-reader variability in CXR images poses challenges in interpretation. In conclusion,
CAD software enhances TB screening, matching human reader performance. AI
technology improves TB screening, aiding in early disease detection.

(Nafisah & Muhammad, 2024), performed research on the topic “Tuberculosis detection
in chest radiograph using convolutional neural network architecture and explainable
artificial intelligence” Develop automatic TB detection system using deep learning models.
Compare CNN models for TB detection performance on CXR datasets. Utilized
explainable AI to visualize TB-infected lung regions. The methodology involved studying
DL methods for TB detection in patient radiography. Propose automatic TB detection
system for CXR images. Evaluate system on multiple datasets and cross-dataset scenarios.
The results showed that, EfficientNetB3 achieved highest accuracy of 99.1% in TB
detection. ResNet50 and EfficientNetB3 showed smooth loss graphs during training.
Pretrained CNN models demonstrated high performance in TB classification. It can be
concluded that the proposed system achieves 99.1% accuracy using EfficientNetB3 CNN
model. Segmented lung CXR images enhance TB detection performance significantly. U-
Net architecture aids in semantic segmentation for medical image analysis.

(Alsaffar et al., 2021) focused on Detecting tuberculosis in medical X-ray imaging using
classification methods and also applying artificial intelligence for automatic classification
of chest X-ray images in the study “Detection of Tuberculosis Disease Using Image
Processing Technique” the following methods were employed; Three classification
methods used: logistic regression, support vector machines, KNN. Preprocessing involved
padding, resizing images to 224x224 dimensions. From the research; results include
comparison of classification scenarios using different machine learning methods. The best
performance was observed with SVM as the learning method. Results show the
effectiveness of the method for tuberculosis detection. The limitations of the study

12
comprised; Previous studies required medical exams and trained professionals for input
parameters. Proposed approach only uses radiographic images of the lungs. It’s concluded
that the study could do, Automatic classification of medical images with and without
tuberculosis. Deep learning features extraction using RESNET50 neural network

The research “Deep and Hybrid Learning Technique for Early Detection of Tuberculosis
Based on X-ray Images Using Feature Fusion” done by (Fati et al., 2022) aimed to Develop
AI techniques for early tuberculosis detection using X-ray images. Enhance X-ray images
and increase contrast in the region of interest. The various algorithms were used; Hybrid
method combines CNN, PCA, SVM for TB detection. ANN uses ResNet-50, GLCM,
DWT, LBP for high accuracy diagnosis. The results showed that; Hybrid systems achieved
high accuracy in diagnosing tuberculosis from X-ray images. ANN with fused features
showed superior results for both datasets. Improved images and contrast in X-ray datasets
for early TB detection. One limitation of the research was lack of sufficient images for the
TB detection but it was addressed using image augmentation. In conclusion, When Early
TB detection is crucial; CNN and ANN models show high accuracy. Proposed methods
enhance TB diagnosis, aiding doctors and radiologists.

(Acharya et al., 2022)aimed to develop AI model for TB detection using deep learning
techniques, Utilize normalization-free networks for X-ray image classification and
Achieve high accuracy and AUC for TB diagnosis in radiology. in the research “AI-
Assisted Tuberculosis Detection and Classification from Chest X-Rays Using a Deep
Learning Normalization-Free Network Model”. It involved Progressive resizing for
training models using chest X-ray images. ImageNet fine-tuned Normalization-Free
Networks (NFNets) for classification. Score-Cam algorithm highlights regions in chest X-
Rays for detailed inference. The results showed; Proposed method achieved 96.91%
accuracy, 99.38% AUC. Score-CAM highlighted TB regions in chest X-rays effectively.
Models outperformed other deep learning architectures in TB classification. The

13
limitations of the developed models were due to; Gradient clipping sensitivity and weight
factor evaluation challenges. Discrepancies in network behavior during training and
testing. It can be concluded from the research that; Deep learning-assisted TB diagnosis
with normalization-free network achieved high accuracy and that the Proposed method can
be a secondary decision tool in clinical settings.

(Alghamdi et al., 2021), performed a survey “Deep Learning Approaches for Detecting
COVID-19 From Chest X-Ray Images: A Survey” The research aims to review and
critically assess the use of deep learning methods for diagnosing COVID-19 from chest X-
ray (CXR) images. It focuses on exploring the potential of convolutional neural networks
(CNNs) and other deep learning architectures for automatic diagnosis of COVID-19 using
visual information from CXR images. The study also highlights the need for
comprehensive and diverse datasets, explainable decisions, and justifiable predictions in
this domain. Various deep learning architectures, particularly CNNs, were analyzed for
detecting COVID-19 from CXR images. The study reviewed preprint and published reports
from March to May 2020, focusing on CNN-based transfer learning and the datasets used
by different studies. Challenges such as dataset imbalance, model transparency, and the
need for domain knowledge integration were discussed. The Results showed that Most
studies utilized CNN-based transfer learning with publicly available datasets, but there is
room for improvement due to dataset limitations and imbalance issues. The research
highlighted the varying dataset sizes and the need for data augmentation to address the
scarcity of COVID-19 CXR images. The main limitation identified was the class imbalance
problem in COVID-19 datasets, affecting the robustness of machine learning algorithms.
In conclusion, The study emphasizes the significant potential of deep learning methods for
automatic COVID-19 diagnosis from CXR images but stresses the importance of
collaboration between medical personnel and computer scientists to validate the
effectiveness of these techniques

14
(Jain et al., 2021) performed a research on “Deep learning based detection and analysis of
COVID-19 on chest X-ray images” The research aimed to utilize deep learning techniques
to detect and analyze COVID-19 on chest X-ray images. The methodology involved
collecting 6432 chest X-ray scans from the Kaggle repository, cleaning the data, and using
data augmentation techniques to increase the size of the dataset. Three deep learning
models, Inception V3, Xception, and ResNeXt, were compared for their performance, with
Xception achieving the highest accuracy of 97.97% in detecting COVID-19 cases. The
study acknowledged the limitations of a potentially small dataset for training, which could
lead to overfitting, emphasizing the need for validation with new data. The research
concluded that Xception net showed the best performance in classifying COVID-19
affected patients and suggested the potential automation of diagnosis tasks in the future. It
also highlighted the importance of consulting medical professionals for practical
applications and emphasized the economic feasibility of using deep learning for disease
detection.
(Benmalek et al., 2021) The research paper aimed to evaluate the performance of different
convolutional neural networks (CNN) in diagnosing COVID-19 using chest imaging
techniques like chest X-ray (CXR) and computed tomography (CT) scan images. The study
compares ResNet-18, InceptionV3, and MobileNetV2 models, with ResNet-18 showing
the highest precision and sensitivity for COVID-19 cases. The methodology involves
analyzing confusion matrices for each model, showcasing their precision, sensitivity,
specificity, and F1-score. The results indicate high accuracy in classifying COVID-19
cases, outperforming other diagnostic methods like saliva sample RT-PCR. However,
limitations include misclassification of viral pneumonia cases as COVID-19 due to image
quality issues and dataset biases. Despite the promising outcomes, challenges in image
quality and dataset merging may affect the models' clinical applicability. In conclusion,
combining deep learning models with chest imaging offers an efficient method for COVID-
19 detection, with CT scans demonstrating superior performance over CXR images,

15
although challenges related to image quality and dataset integration need to be addressed
for reliable clinical use.
(Zhou et al., 2021) “A review of deep learning in medical imaging: Imaging traits,
technology trends, case studies with progress highlights, and future promises” The research
paper aims to review the application of deep learning in medical imaging, highlighting
technological trends, case studies, and future promises in the field. It emphasizes the
challenges unique to medical imaging that deep learning must address, such as the need for
big data with annotations and advancements in high-performance computing. The paper
reviews various technological challenges in medical imaging domains and tasks, focusing
on solutions to the data challenge. It discusses the trend of making neural networks deeper,
starting from AlexNet to more advanced models like VGGNet, Inception Net, and ResNet.
Techniques like skip connections, deep supervision, attention mechanisms, and squeeze
and excitation are explored to enhance network performance. The study showcases
progress highlights in selected medical imaging cases, including thoracic, neuro,
cardiovascular, abdominal, and microscopy imaging. It discusses the advancements made
in these areas using deep learning techniques, demonstrating the potential of AI in
improving diagnostic accuracy and efficiency in clinical practices. One limitation
highlighted is the long-tailed distribution of disease patterns in medical images, where most
diseases are infrequent, posing challenges for model training and generalization.
Additionally, the "black box" nature of deep learning models is acknowledged, as they lack
explicit interpretability based on known physical or mathematical principles. It can be
concluded that by emphasizing the transformative potential of deep learning in medical
imaging, particularly in identifying complex imaging patterns imperceptible to visual
radiologic evaluation. It suggests that AI-based image analysis can redefine diseases on a
neurobiological basis, enable early detection, and enhance personalized risk estimates,
thereby revolutionizing clinical protocols and contributing to precision medicine

16
(Panayides et al., 2020) The research paper aims to review cutting-edge solutions in
medical imaging informatics, discuss clinical translation, and propose future directions for
enhancing clinical practice. The methodology involves exploring advances in medical
imaging acquisition technologies for various modalities and emphasizing the importance
of efficient medical data management strategies. The results highlight the significance of
transfer learning approaches in popular frameworks to widen the deep learning research
base for new applications. However, the study acknowledges limitations, such as the need
for efficient enterprise-wide clinical data repositories to aggregate diverse medical
information systematically. In conclusion, the paper underscores the pivotal role of medical
imaging informatics in driving clinical research and practice, with AI-based approaches
gaining FDA approval for tasks like image enhancement, segmentation, and abnormality
detection. The future direction foresees substantial growth in FDA-approved AI-based
solutions for radiology and digital pathology images, revolutionizing healthcare practices.

(Ma et al., 2024) The research aimed to evaluate the effectiveness of MedSAM, a
foundation model for medical image segmentation, compared to other models like SAM7,
U-Net, and DeepLabV3. The methodology involved internal and external validation, where
MedSAM showed superior generalization abilities across various medical image
segmentation tasks. Results indicated that MedSAM outperformed SAM7 and specialist
models in most tasks, showcasing its potential for diverse segmentation challenges.
However, SAM7 excelled in specific RGB image segmentation tasks due to distinct target
appearances. The study's limitations include the focus on segmentation tasks only and the
need for further validation on a broader range of medical imaging tasks. In conclusion,
MedSAM's consistent performance on internal and external validation sets highlights its
versatility and potential to advance diagnostic and therapeutic tools in medical imaging,
ultimately enhancing patient care.

17
(Willemink et al., 2020) in the research “Preparing Medical Imaging Data for Machine
Learning” The research aimed to evaluate the effectiveness of MedSAM, a foundation
model for medical image segmentation, compared to other models like SAM7, U-Net, and
DeepLabV3. The methodology involved internal and external validation, where MedSAM
showed superior generalization abilities across various medical image segmentation tasks.
Results indicated that MedSAM outperformed SAM7 and specialist models in most tasks,
showcasing its potential for diverse segmentation challenges. However, SAM7 excelled in
specific RGB image segmentation tasks due to distinct target appearances. The study's
limitations include the focus on segmentation tasks only and the need for further validation
on a broader range of medical imaging tasks. In conclusion, MedSAM's consistent
performance on internal and external validation sets highlights its versatility and potential
to advance diagnostic and therapeutic tools in medical imaging, ultimately enhancing
patient care.

(Sarvamangala & Kulkarni, 2022), in the study “Convolutional neural networks in medical
image understanding: a survey” The research paper aims to provide a comprehensive
survey of the applications of Convolutional Neural Networks (CNNs) in medical image
understanding, with the underlying objective of motivating researchers in this field to
extensively utilize CNNs in their work. The methodology involves discussing CNNs, their
award-winning frameworks, and major medical image understanding tasks like
classification, segmentation, localization, and detection. The results highlight the critical
and comprehensive survey of CNN applications in medical image understanding for
various ailments like brain, breast, and lung conditions. However, the paper acknowledges
some limitations, such as not covering all aspects of medical image understanding due to
its vast scope. In conclusion, the paper emphasizes the effectiveness of CNNs in surpassing
human experts in image understanding tasks and encourages further research in applying
CNNs to enhance medical diagnosis and treatment planning.

18
(Varoquaux & Cheplygina, 2022) in the study “Machine learning for medical imaging:
methodological failures and recommendations for the future” aimed to Identify systematic
challenges in medical imaging analysis to provide recommendations for future
improvements in the field, it highlights the challenges in medical imaging analysis and
recommendations for future improvements, The Importance of robust validation methods
using multiple datasets. Impact of incorrectly chosen baselines on algorithm evaluation.
Best practices for medical machine learning evaluation and benchmarking standards. It also
points out the criteria for accepting methods based on significant improvement over
existing solutions. The research was able to identify challenges in medical imaging analysis
with biases and data limitations. It also discusses efforts to counteract problems and
provides recommendations for the future. Some of the limitations of the study were Dataset
biases, hidden subgroups, and mislabeled instances. Lack of reproducibility due to omitted
details and misinterpreted tests. Overfitting, lack of robust validation, and reuse of datasets,
it can be concluded from the research that challenges in clinical ML research impact
diagnostic accuracy and model evaluation. Strategies for improvement require procedural,
normative, and goal changes.

Singh et al., (2020). “3D Deep Learning on Medical Images: A Review” The research
objectives of this paper are to trace the development of three-dimensional convolutional
neural networks (3D CNNs) from their machine learning origins, provide a mathematical
description, and outline the preprocessing steps required for medical images before they
are analyzed using the model. The methodology involves a historical overview of the
evolution of 3D CNNs, starting from the foundational advancements in machine learning
and CNN architectures, particularly since the breakthrough success of AlexNet in 2012.
The paper includes a mathematical exposition and describes the necessary preprocessing
techniques for medical imaging data. The results of the research highlighted the
advancements and efficacy of 3D CNNs in improving the accuracy and efficiency of
medical image analysis across different tasks. The use of 3D CNNs has shown significant

19
potential in enhancing disease diagnosis and assisting clinicians in their work. However,
the paper acknowledges several limitations. The challenges associated with the use of the
model in medical imaging include computational complexity, the need for large and diverse
datasets, potential biases in training data, and the generalizability of models across.

20
 CHAPTER THREE
SYSTEM ANALYSIS AND DESIGN

3.1 Data Gathering and Analysis:

This is the systematic process of gathering observations or measurements. Data being the
most important factor for accuracy in Machine Learning will be gathered from relevant
sources. Medical imaging data, specifically chest X-ray images with and without the
tuberculosis condition will be gathered form the sources - Mendeley. Data, the open-source
community Kaggle or zenodo.org. These sites are chosen because of their reliability in
dataset quality and the recommendations by other machine learning experts.

The present dataset contains both healthy images and tuberculosis affected images, they
are from "Tuberculosis (TB) Chest X-Ray Database'' collected by researchers from Qatar
and Dhaka University, Doha, Qatar, and collaboration with doctors from Hamad Medical
Corporation and Bangladesh. All the images are CRX in PNG format and a size 512x512.

21
Figure 3.1: Chest x-ray images without tuberculosis.

Figure 3.2: Chest x-ray images with tuberculosis.

22
CITE: Visuña, L. (2022). Computer-Aided diagnostic for classifying Chest X-Ray Images
Using Deep Ensemble Learning [Data set]. Zenodo.
https://doi.org/10.5281/zenodo.6637854

3.1.1 Data preprocessing:

Image preprocessing is the steps taken to format images before they are used by model
training and inference. This includes, but is not limited to, resizing, orienting, and color
corrections. Preprocessing is required to clean image data for model input. For example,
fully connected layers in convolutional neural networks required that all images are the
same sized arrays. Image preprocessing may also decrease model training time and increase
model inference speed. If input images are particularly large, reducing the size of these
images will dramatically improve model training time without significantly reducing
model performance.
The images were preprocessed in jupyter notebook (from the anaconda software) see figure
3.3 below.

23
 Figure 3.3 image preprocessing in Jupyter notebook

3.1.2 Feature Extraction:

Features are characteristics of the objects of interest, which represent the maximum
relevant information that the image has to offer for the complete characterization of a
tumor. Feature extraction methodologies analyze objects and images to extract the most
prominent features that are representative of the various classes of objects. Features are
used as inputs to classifiers that assign them to the class that they represent. In this research,
the CNN algorithm is proposed to be used. Features used for tuberculosis detection are
widely divided into two main categories: global features and local features.

24
 Figure 3.4 feature extraction parameters using Jupyter notebook

Global features are feasible features that are classified as general features and domain-
specific features. This leads to the case of images where general features are considered
for the detection process (Gemescu et al., 2019). Local features are features extracted from
a different section of the chest image which depends on image partitioning, there is a need
for features to be at different levels, for instance at the chest level (Eweje et al. 2021).

3.1.3 Algorithm -CNN (Convolutional Neural Network)- Development:

A Convolutional Neural Network (CNN) is a type of deep learning algorithm that is
particularly well-suited for image recognition and processing tasks. It is made up of
multiple layers, including convolutional layers, pooling layers, and fully connected layers.

25
The Convolutional Neural Network will be used because they are efficient in automatically
extracting meaningful spatial features from images, enabling accurate object recognition.
Can learn hierarchical features like edges, textures, and shapes, which are important for
image data. Can reduce the number of weights and computational cost compared to fully
connected layers.

Figure 3.5 CNN algorithm setup

Furthermore, they reduce overfitting (Overfitting is when a model fits too closely to its
training data and cannot generalize well to new data) by using pooling and dropout layers
the algorithm can effectively handle translation invariant structures in the data, meaning
they can recognize objects regardless of their position or scale.

26
3.2 Desktop application development:
For ease of use and access to the software, an offline application will be developed. An
offline application has some advantages compared to web-based application; 1. Faster
than web apps 2. Greater functionality as they have access to system resources 3. Can
work offline 3. Easier to build due to the availability of developer tools, interface
elements, and SDKs. The graphical user interface was developed in using the pycharm
IDE. See figure 3.5 below

Figure 3.6 Pycharm integrated development environment

The Python Kivy/KivyMD framework will be used to create the user interface. KivyMD is
an extension of the Kivy framework. KivyMD is a collection of Material Design widgets
for use with Kivy, a GUI framework for making mobile applications. It is similar to the
Kivy framework but provides a more attractive GUI.

27
 Figure 3.7 Kivy file snippet used for the user interface development

Python is a high-level, general-purpose programming language renowned for its readability

and simplicity. Its syntax, characterized by significant indentation, enhances code clarity
and maintainability. Unlike many languages, Python is dynamically typed, meaning
variable types are determined at runtime, providing flexibility in development. As a
garbage-collected language, Python automatically manages memory allocation and
deallocation, relieving programmers from manual memory management.
Python's versatility is evident in its support for multiple programming paradigms. It excels
in structured programming, offering a clear and sequential approach to problem-solving.
Additionally, Python's object-oriented capabilities enable the creation of complex and
modular applications through classes and objects. Functional programming concepts are
also embraced, allowing for elegant and concise code through functions as first-class
citizens. This adaptability makes Python suitable for a wide range of applications, from
web development and data analysis to scientific computing and machine learning.

28
Beyond its core features, Python's extensive standard library provides a rich set of modules
and functions for various tasks, often described as a "batteries included" philosophy. This
comprehensive library, combined with its readability and ease of use, has contributed to
Python's widespread adoption across industries and among programmers of all levels.

Figure 3.8 Python code snippet used for the user interface development

29
 CHAPTER FOUR
RESULTS AND DISCUSSIONS

System requirements, Screenshots, Explanation how the application works and its
output
Minimum requirements
Intel core duo
2 GB of RAM
Pycharm IDE
Required Python Libraries
Windows operating system
Best requirements:
2.6 GHz Processor
A graphics processing unit
8 GB or more of RAM
Anaconda IDE
Required Python Libraries
Windows operating system

4.1 Results
This section shows the resulting system’s functionality and usage;

Interfaces
Opening the application will give you an interface that allows you to either upload a single
image for prediction or multiple images coupled with a toggle button that makes the
interface more appealing depending on the time of day (light mode or dark mode), two alert
buttons (the one at the left of the reload button shows the information of the project’s owner

30
and the one at the right side of the screen shows a warning sign signifying that the software
is not 100% accurate). See figure 4.1 below

Figure 4.1: Starting of the system, light mode to the left and dark mode to the right.

Figure 4.2 below is the interface designed for user interaction, featuring two separate text
boxes where users can enter the patient’s information, such as name and relevant medical

31
details. This organization helps in accurately capturing and managing patient data, ensuring
that all necessary information is recorded for effective processing.

Figure 4.2: Image verification interface

The interface also includes a back button positioned at the top left of the screen, which
allows users to easily navigate back and make changes to image uploads or other details if
needed. The verify button is used to submit the uploaded image to a trained machine
learning model for analysis and prediction. Additionally, the image box prominently
displays the uploaded image, enabling users to visually confirm its accuracy and
completeness before proceeding with the verification process.

32
 Figure 4.3: Without tuberculosis, not certain, with tuberculosis

The interfaces shown in figure 4.3 shows the prediction gotten from the machine learning
model, the predicted state of the is listed in the order left to right (not infected, may be
infected and infected).
From this screen when the save record button is clicked the details entered previously and
the prediction will be saved to an excel file for retrieval in a later time. See figure 4.4 below

33
 Figure 4.4 excel sheet used for storing patient’s information

From the main screen when the upload multiple image button is clicked the user is directed
to select up to 10 different images which will be given to the predictive machine learning
model and the infected images will be displayed in red while the non-infected images will
be displayed in green (see figure 4.4 below).

34
Figure 4.4 multiple image prediction screen

35
4.2 Discussion
The application interface is designed with user-friendliness and functionality in mind. The
option to upload either a single image or multiple images, combined with a toggle for light
and dark modes, ensures that the application is adaptable to various user preferences and
environmental conditions, making it more accessible and comfortable for prolonged use.
The clear layout, featuring separate text boxes for patient information and a visually
distinct image box for confirmation, helps users maintain organized data entry and
verification processes. The inclusion of alert buttons for project ownership and accuracy
warnings adds transparency, fostering user trust and ensuring they are aware of the
software's limitations.
Despite these strengths, the design offers opportunities for further enhancement. The
manual image upload feature and interaction with various interface elements are beneficial
for providing detailed control and customization, catering to users who appreciate a hands-
on approach. The application’s capability to handle batch processing is a significant
advantage, enabling efficient analysis of multiple images. To further optimize user
experience, especially for those who may be less experienced, simplifying batch processing
and refining the interface could make the application even more intuitive and responsive.
These improvements would enhance user satisfaction and ensure the system performs
efficiently, even with increased data volume.

36
 CHAPTER FIVE

SUMMARY, CONCLUSION, RECOMMENDATION AND

REFERENCES

This chapter presents the summary of the whole research study and the conclusion to the
research. This section also presents related topics or areas for future researches.

5.1 Summary
In this project on tuberculosis prediction, a tuberculosis prediction system was developed
using a chest x-ray image dataset for predicting the disease. The work focused on the
development of machine learning enhancements for accurately predicting tuberculosis. In
the later stages, discussion on various tools and techniques to bring the idea into practical
application was carried out. After developing the predictive model, it was rigorously tested,
and the results were analyzed, highlighting a few deficiencies. Following the testing phase,
the advantages of the system were described and suggestions were offered for further
enhancements and improvements.

5.2 Advantages of the system

i. Able to predict tuberculosis with a high accuracy

ii. Runs fast and consumes low power
iii. Provide significant help for the people with low orientation about prediction.
iv. Lower operational costs.
v. Doesn’t require machine learning experts to use

5.3 Disadvantages
i. It is limited only to checking if an image has tuberculosis or not

37
 ii. It is only available for desktop at the moment
iii. Might mistake some other diseases for tuberculosis i.e pneumonia

5.4 Conclusion
This project detects ML-based framework for intrusion detection uptake classification and
correlates its underlying factors. The whole research has been succeeded based on the
newly proposed dataset gathered from zenodo.org. The recommended framework reveals
that a weighted ensemble of ML models successfully enhances the prediction results, as it
weighted aggregates the output probabilities of the ensemble candidates' model.

5.5 Recommendation
In future works, this work can be taken into more detail and more work can be done on the
project in order to bring modifications and additional features. The current system can only
accurately predict tuberculosis infected and non-infected images, and also a larger dataset
can be used to retrain the model so as to improve the efficiency of the software. The current
version of the software supports only few areas of the system application software which
is only on machine learning environment.

38
REFERENCE
Acharya, V., Dhiman, G., Prakasha, K., Bahadur, P., Choraria, A., M, S., J, S., Prabhu, S.,
Chadaga, K., Viriyasitavat, W., & Kautish, S. (2022). AI-Assisted Tuberculosis
Detection and Classification from Chest X-Rays Using a Deep Learning
Normalization-Free Network Model. Computational Intelligence and
Neuroscience, 2022, 1–19. https://doi.org/10.1155/2022/2399428
Alghamdi, H. S., Amoudi, G., Elhag, S., Saeedi, K., & Nasser, J. (2021). Deep Learning
Approaches for Detecting COVID-19 From Chest X-Ray Images: A Survey. IEEE
Access, 9, 20235–20254. IEEE Access.
https://doi.org/10.1109/ACCESS.2021.3054484
Alsaffar, M., Alshammari, G., Alshammari, A., Aljaloud, S., Almurayziq, T. S., Hamad,
A. A., Kumar, V., & Belay, A. (2021). Detection of Tuberculosis Disease Using
Image Processing Technique. Mobile Information Systems, 2021, e7424836.
https://doi.org/10.1155/2021/7424836
Benmalek, E., Elmhamdi, J., & Jilbab, A. (2021). Comparing CT scan and chest X-ray
imaging for COVID-19 diagnosis. Biomedical Engineering Advances, 1, 100003.
https://doi.org/10.1016/j.bea.2021.100003
Born, J., Beymer, D., Rajan, D., Coy, A., Mukherjee, V. V., Manica, M., Prasanna, P.,
Ballah, D., Guindy, M., Shaham, D., Shah, P. L., Karteris, E., Robertus, J. L.,
Gabrani, M., & Rosen-Zvi, M. (2021). On the role of artificial intelligence in
medical imaging of COVID-19. Patterns, 2(6), 100269.
https://doi.org/10.1016/j.patter.2021.100269
Çallı, E., Sogancioglu, E., van Ginneken, B., van Leeuwen, K. G., & Murphy, K. (2021).
Deep learning for chest X-ray analysis: A survey. Medical Image Analysis, 72,
102125. https://doi.org/10.1016/j.media.2021.102125
Codlin, A. J., Dao, T. P., Vo, L. N. Q., Forse, R. J., Van Truong, V., Dang, H. M., Nguyen,
L. H., Nguyen, H. B., Nguyen, N. V., Sidney-Annerstedt, K., Squire, B., Lönnroth,

39
 K., & Caws, M. (2021). Independent evaluation of 12 artificial intelligence
solutions for the detection of tuberculosis. Scientific Reports, 11(1), 23895.
https://doi.org/10.1038/s41598-021-03265-0
Fati, S. M., Senan, E. M., & ElHakim, N. (2022). Deep and hybrid learning technique for
early detection of tuberculosis based on X-ray images using feature fusion. Applied
Sciences, 12(14), 7092.
Gemescu, I. N., Thierfelder, K. M., Rehnitz, C., & Weber, M.-A. (2019). Imaging Features
of Bone Tumors. Magnetic Resonance Imaging Clinics of North America, 27(4),
753–767. https://doi.org/10.1016/j.mric.2019.07.008
Geric, C., Qin, Z. Z., Denkinger, C. M., Kik, S. V., Marais, B., Anjos, A., David, P.-M.,
Ahmad Khan, F., & Trajman, A. (2023). The rise of artificial intelligence reading
of chest X-rays for enhanced TB diagnosis and elimination. The International
Journal of Tuberculosis and Lung Disease, 27(5), 367–372.
https://doi.org/10.5588/ijtld.22.0687
Gill, C. M., Dolan, L., Piggott, L. M., & McLaughlin, A. M. (2022). New developments in
tuberculosis diagnosis and treatment. Breathe, 18(1), 210149.
https://doi.org/10.1183/20734735.0149-2021
Jain, R., Gupta, M., Taneja, S., & Hemanth, D. J. (2021). Deep learning based detection
and analysis of COVID-19 on chest X-ray images. Applied Intelligence, 51(3),
1690–1700. https://doi.org/10.1007/s10489-020-01902-1
Kasban, H., El-Bendary, M. A. M., & Salama, D. H. (2015). A Comparative Study of
Medical Imaging Techniques.
Kik, S. V., Gelaw, S. M., Ruhwald, M., Song, R., Khan, F. A., Van Hest, R., Chihota, V.,
Nhung, N. V., Esmail, A., Celina Garfin, A. M., Marks, G. B., Gorbacheva, O.,
Akkerman, O. W., Moropane, K., Ngoc Anh, L. T., Dheda, K., Fox, G. J., Marano,
N., Lönnroth, K., … Denkinger, C. M. (2022). Diagnostic accuracy of chest X-ray
interpretation for tuberculosis by three artificial intelligence-based software in a

40
 screening use-case: An individual patient meta-analysis of global data.
https://doi.org/10.1101/2022.01.24.22269730
Ma, J., He, Y., Li, F., Han, L., You, C., & Wang, B. (2024). Segment anything in medical
images. Nature Communications, 15(1), 654. https://doi.org/10.1038/s41467-024-
44824-z
Nafisah, S. I., & Muhammad, G. (2024). Tuberculosis detection in chest radiograph using
convolutional neural network architecture and explainable artificial intelligence.
Neural Computing and Applications, 36(1), 111–131.
https://doi.org/10.1007/s00521-022-07258-6
Panayides, A. S., Amini, A., Filipovic, N. D., Sharma, A., Tsaftaris, S. A., Young, A.,
Foran, D., Do, N., Golemati, S., Kurc, T., Huang, K., Nikita, K. S., Veasey, B. P.,
Zervakis, M., Saltz, J. H., & Pattichis, C. S. (2020). AI in Medical Imaging
Informatics: Current Challenges and Future Directions. IEEE Journal of
Biomedical and Health Informatics, 24(7), 1837–1857. IEEE Journal of
Biomedical and Health Informatics. https://doi.org/10.1109/JBHI.2020.2991043
Qin, Z. Z., Ahmed, S., Sarker, M. S., Paul, K., Adel, A. S. S., Naheyan, T., Barrett, R.,
Banu, S., & Creswell, J. (2021). Tuberculosis detection from chest x-rays for
triaging in a high tuberculosis-burden setting: An evaluation of five artificial
intelligence algorithms. The Lancet Digital Health, 3(9), e543–e554.
https://doi.org/10.1016/S2589-7500(21)00116-3
Sarvamangala, D. R., & Kulkarni, R. V. (2022). Convolutional neural networks in medical
image understanding: A survey. Evolutionary Intelligence, 15(1), 1–22.
https://doi.org/10.1007/s12065-020-00540-3
Singh, S. P., Wang, L., Gupta, S., Goli, H., Padmanabhan, P., & Gulyás, B. (2020). 3D
Deep Learning on Medical Images: A Review. Sensors, 20(18), Article 18.
https://doi.org/10.3390/s20185097

41
Varoquaux, G., & Cheplygina, V. (2022). Machine learning for medical imaging:
Methodological failures and recommendations for the future. Npj Digital Medicine,
5(1), 1–8. https://doi.org/10.1038/s41746-022-00592-y
Willemink, M. J., Koszek, W. A., Hardell, C., Wu, J., Fleischmann, D., Harvey, H., Folio,
L. R., Summers, R. M., Rubin, D. L., & Lungren, M. P. (2020). Preparing Medical
Imaging Data for Machine Learning. Radiology, 295(1), 4–15.
https://doi.org/10.1148/radiol.2020192224
Zhou, S. K., Greenspan, H., Davatzikos, C., Duncan, J. S., Van Ginneken, B., Madabhushi,
A., Prince, J. L., Rueckert, D., & Summers, R. M. (2021). A Review of Deep
Learning in Medical Imaging: Imaging Traits, Technology Trends, Case Studies
With Progress Highlights, and Future Promises. Proceedings of the IEEE, 109(5),
820–838. Proceedings of the IEEE. https://doi.org/10.1109/JPROC.2021.3054390

42
 APPENDIX
Appendix A Project Libraries Codes
Appendix B
Project Code from Jupyter

43