ORCID
- Joan Mnyambo: 0000-0002-4365-7067
- Amir Aly: 0000-0001-5169-0679
- Emmanuel Ifeachor: 0000-0001-8362-6292
- Yinghui Wei: 0000-0002-7873-0009
Document Type
Article
Abstract
Drug-resistant tuberculosis (DR-TB) continues to pose a major global health challenge, with hundreds of thousands of new cases reported annually, resulting in a substantial mortality burden. Delayed or inaccurate diagnosis contributes to ongoing transmission, treatment failure, and increased deaths, highlighting the need for timely and dependable detection, particularly in low-resource healthcare settings. Although chest X-rays (CXRs) are widely available and commonly used for tuberculosis screening, reliable and interpretable approaches for distinguishing drug-resistant from drug-sensitive tuberculosis (DS-TB) remain limited. Despite recent advances in deep learning, explainable artificial intelligence (XAI) methods for DR-TB identification are still underexplored, thereby reducing clinical trust and adoption. In this study, we propose a data-efficient image transformer (DEiT) framework for classifying DR-TB and DS-TB from chest X-rays, integrating multiple XAI techniques. The proposed model outperformed contemporary convolutional neural networks (CNNs) and transformer baselines, achieving high diagnostic accuracy and robust feature representations. Explainability was provided using Raw Attention, Gradient-weighted Class Activation Mapping (Grad-CAM), and Local Interpretable Model-agnostic Explanations (LIME), which consistently highlighted clinically relevant DR-TB manifestations, including consolidations, infiltrates, collapse, cavitation, and nodular patterns across lung regions. Quantitative analyses revealed a trade-off between explanation faithfulness and interpretability, with strong agreement observed between Raw Attention and Grad-CAM, and complementary insights provided by LIME. Cross-dataset evaluation on the Shenzhen chest X-ray set demonstrated effective transferability to heterogeneous imaging conditions, while ablation studies confirmed the contribution of key architectural components. These findings establish DEiT as an interpretable solution for DR-TB classification, supporting AI-assisted triage, early diagnosis, and decision support.
DOI Link
Publication Date
2026-10-15
Publication Title
Biomedical Signal Processing and Control
Volume
126
Issue
B
ISSN
1746-8094
Acceptance Date
2026-07-02
Deposit Date
2026-10-09
Funding
This research was funded by the Commonwealth Scholarship Commission (CSC), United Kingdom, under the Foreign, Commonwealth and Development Office (FCDO) scholarship programme (Scholarship No. TZCS-2023-543), in collaboration with the University of Plymouth, United Kingdom. It also received support from the Government of Tanzania through the Ministry of Education, Science and Technology .
Keywords
Tuberculosis, Drug-resistant tuberculosis, Chest X-ray, Vision transformers, XAI, Interpretability
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Mnyambo, J., Aly, A., Ifeachor, E., & Wei, Y. (2026) 'Explainable AI for drug-resistant tuberculosis detection from chest X-rays: Multi-level interpretability and cross-dataset transfer learning: Multi-level interpretability and cross-dataset transfer learning', Biomedical Signal Processing and Control, 126(B). Available at: 10.1016/j.bspc.2026.110989
