A Study on Tuberculosis Classification in Chest X-ray Using Deep Residual Attention Networks^.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Detection

Background:

• Deep learning is increasingly integrated into clinical workflows for medical image analysis.
• Accurate detection of tuberculosis (TB) from chest X-rays remains a critical challenge.
• Attention mechanisms in neural networks can improve feature representation and model performance.

Purpose of the Study:

• To investigate the impact of attention mechanisms on the classification of tuberculosis in X-ray images.
• To propose and evaluate a novel Convolutional Block Attention Module (CBAM) for convolutional neural networks.
• To enhance the performance of deep learning models for tuberculosis detection.

Main Methods:

• Development of a Convolutional Block Attention Module (CBAM) for feed-forward convolutional neural networks.
• Implementation of CBAM to refine feature maps adaptively through inferred attention maps.
• Validation of the proposed method on a dataset of 4990 chest X-ray radiographs from three hospitals.

Main Results:

• The CBAM module achieved high precision and recall in classifying tuberculosis from X-ray images.
• The attention mechanism facilitated object localization within the X-ray images.
• The proposed model demonstrated superior performance compared to existing methods on the validation dataset.

Conclusions:

• Attention mechanisms, specifically the proposed CBAM, significantly improve the performance of deep learning models for tuberculosis detection in X-ray images.
• CBAM enhances both classification accuracy and localization capabilities, making it suitable for clinical applications.
• The findings suggest that attention-enhanced deep learning models offer a promising approach for computer-aided detection of tuberculosis.