Multi-Feature Fusion with Deep Learning for Enhanced Tuberculosis Detection from Chest X-rays

Early and accurate diagnosis of tuberculosis (TB) remains a critical global health challenge, particularly in resource-limited settings. A novel multi-feature fusion approach combining handcrafted features with deep learning for TB detection from chest X-rays. Our proposed methodology integrates Scale-Invariant Feature Transform (SIFT), Oriented FAST and Rotated BRIEF (ORB), Gabor filter responses, and Local Binary Patterns (LBP) features in a unified framework coupled with neural network classification. The system achieves 86.76% accuracy on our test dataset, outperforming conventional single-feature methods. The further analysis of feature, demonstrate the robustness of our approach across varied image qualities. This work addresses significant research gaps in multi-feature fusion strategies for medical image analysis and provides a pragmatic solution for TB screening that balances computational efficiency with diagnostic accuracy.