A Robust Tuberculosis Diagnosis Using Chest X-Rays Based on a Hybrid Vision Transformer and Principal Component...

Area of Science:

• Medical Imaging and Artificial Intelligence
• Computer-Aided Diagnosis (CAD) Systems
• Machine Learning in Healthcare

Background:

• Tuberculosis (TB) is a significant global health challenge caused by Mycobacterium tuberculosis, primarily affecting the lungs but also other organs.
• Active TB presents symptoms, is transmissible, and faces challenges like drug resistance, co-infections, and limited resources, hindering eradication efforts.
• Accurate and timely TB diagnosis via chest X-rays (CXRs) is crucial but complicated by atypical findings and radiologist shortages in high-burden areas.

Purpose of the Study:

• To develop and evaluate an advanced computer-aided diagnosis (CAD) system for the accurate identification of Tuberculosis (TB) from chest X-ray (CXR) images.
• To address diagnostic delays and improve patient outcomes, especially in resource-limited settings.
• To enhance the efficiency and reduce the cost of the TB diagnostic process.

Main Methods:

• A hybrid CAD system integrating Vision Transformer (ViT) for deep feature extraction, Principal Component Analysis (PCA) for dimensionality reduction, and Machine Learning (ML) for classification.
• The ViT model served as the base for extracting intricate features from CXR images.
• Pre-processing techniques including resizing, scaling, and noise removal were applied to the TB CXR dataset to optimize diagnostic accuracy.

Main Results:

• The proposed hybrid CAD model demonstrated superior performance compared to existing classifiers.
• Achieved exceptional metrics: 99.90% precision, 99.52% recall, 99.71% F1-score, 99.84% accuracy, 0.48% false negative rate (FNR), 99.52% specificity, and 99.90% negative predictive value (NPV).
• The system effectively identified TB in CXR images, enabling rapid medical intervention.

Conclusions:

• The developed ViT-based hybrid CAD system offers a highly accurate and efficient solution for TB detection in CXR images.
• The model's performance surpasses current state-of-the-art classifiers, indicating its potential for clinical application.
• This AI-driven approach can significantly aid healthcare professionals in diagnosing TB, improving patient management and public health outcomes.