An efficient deep learning-based framework for tuberculosis detection using chest X-ray images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Early tuberculosis (TB) diagnosis is critical for disease control and reducing mortality.
• Chest X-rays (CXRs) are widely used for lung disease screening but manual interpretation is time-consuming and prone to inter-observer variability.
• Developing accurate and cost-effective computer-aided diagnosis (CAD) systems for TB is a significant research challenge.

Purpose of the Study:

• To propose an efficient and accurate deep learning network, TBXNet, for the classification of TB from CXR images.
• To address the limitations of manual CXR screening in TB diagnosis.
• To provide a reliable CAD system for early TB detection.

Main Methods:

• Development of TBXNet, a deep learning network featuring five dual convolution blocks with varying filter sizes (32, 64, 128, 256, 512).
• Fusion of dual convolution blocks with a pre-trained layer to leverage transfer learning.
• Validation of the network's performance on multiple datasets (Dataset A, B, and C).

Main Results:

• TBXNet achieved high classification accuracies of 98.98% on Dataset A and 99.17% on Dataset B.
• On Dataset C (normal, TB, pneumonia, COVID-19), TBXNet demonstrated superior performance with Precision (95.67%), Recall (95.10%), F1-score (95.38%), and Accuracy (95.10%).
• The proposed model outperformed existing state-of-the-art methods in TB diagnosis using CXRs.

Conclusions:

• The proposed TBXNet is an efficient and accurate deep learning model for TB diagnosis from CXRs.
• TBXNet offers a promising computer-aided diagnosis solution, improving upon the limitations of manual interpretation.
• The model's generalizability and high performance indicate its potential for widespread clinical application in TB screening.