A Hybrid Approach for Tuberculosis Detection using Convolutional Attention Networks and XGBoost

Tuberculosis remains a significant global challenge, with early diagnosis crucial for effective treatment and limiting its spread. Despite advancements in artificial intelligence for TB detection from chest X-rays, achieving human-level accuracy is still difficult. This study addresses this by integrating Convolutional Attention Networks with XGBoost, leveraging deep learning and gradient boosting. Using $\mathbf{7, 0 0 0}$ diverse chest X-rays, our hybrid model achieved 99.36% precision, surpassing state-of-the-art methods by 9.36 percentage points over the CAD4TB v6 baseline. Notably, it demonstrated $\mathbf{9 9 \%}$ precision and $\mathbf{9 8 \%}$ recall for TB-positive cases, indicating minimal false negatives. These findings highlight the potential for transforming TB diagnosis in clinical practice and global health efforts.