Table 1.

Summary of the related work.

Paper	Key Findings
[12]	- Proposed the “Incremental Learning-based Cascaded Model” (ILCM) for TB identification in chest X-ray (CXR) photos. - Automated identification of infected areas and classification of TB cases. - Achieved an F1 score of 97.23% and overall accuracy of 93.20% on local data.
[13]	- Combined explainable Artificial Intelligence (XAI) techniques with CNN framework for TB detection in chest X-rays. - Accuracy between 98.7% and 99.1%. - Improved interpretability of CNN’s decision-making process.
[14]	- Utilized the Mayfly Algorithm (MA) and Dual Deep Learning Features for TB detection in CXR images. - Achieved accuracy rates of 97.8% using the KNN classifier. - Optimized feature selection and refinement using MA.
[15]	- Proposed an IoT-based healthcare application for early TB diagnosis using CXR images. - Used Adaptive Fuzzy C-means clustering and Deep Belief Network (DBN) for feature extraction and classification. - Performance improvement using Adaptive Monarch Butterfly Optimization (AMBO) method.
[16]	- Introduced a pipeline for automated TB screening in chest X-rays using deep learning. - Combined three deep learning architectures and applied various techniques for improved performance. - Achieved 97.1% classification accuracy and high evaluation metrics scores.
[17]	- Developed CBAMWDnet model for early TB identification using CXR images. - Combined Convolutional Block Attention Module (CBAM) and Wide DenseNet (WDnet) for improved feature extraction. - Achieved 98.80% accuracy and performed well on evaluation metrics.
[18]	- Proposed a deep learning model for joint diagnosis of lung disorders using chest X-rays. - Trained on publicly available Kaggle datasets and achieved 98.72% accuracy. - Outperformed existing methods in precision and diagnosis of specific disorders.
[19]	- Introduced a computer-aided diagnostic (CAD) method for automated chest X-ray-based TB identification. - Combined Gabor filters and deep features from pre-trained models for comprehensive detection. - Achieved high area under the curve (AUC) values on evaluation datasets.
[20]	- Addressed class disparity in the TBX11K dataset using the Synthetic Minority Over-sampling Technique (SMOTE). - Evaluated performance of Random Forest (RF) and XGBoost (XGB) models with and without SMOTE. - SMOTE improved the precision–recall trade-off, but slightly decreased overall accuracy.
[21]	- Explored deep learning models for chest X-ray quality control (QC) in TB testing. - Demonstrated exceptional performance in identifying anomalous X-rays and anomalies related to TB. - Good performance on external datasets.