Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Explainable Hybrid AI Framework for Enhanced Tuberculosis and Symptom Detection

Neel Patel, Alexander Wong, Ashkan Ebadi

TL;DR

This work tackles TB screening in resource-constrained settings by delivering an explainable, semi-supervised chest X-ray framework that jointly detects three-class disease (COVID-19, TB, normal) and seven radiographic symptoms. It combines a DISTL-style teacher–student architecture with a ViT-Small backbone, a DINO self-supervised head, and two supervised heads for disease and symptom detection, augmented by lung segmentation and CheXpert initialization. The model achieves $98.85\%$ accuracy and $98.89\%$ macro-F1 on disease classification, plus a $90.09\%$ macro-F1 on symptom multilabel detection, outperforming CNN baselines and providing Grad-CAM explanations that align with anatomical findings. This approach supports reliable screening and triage in real-world clinical workflows by delivering high sensitivity/specificity and interpretable radiographic evidence.

Abstract

Tuberculosis remains a critical global health issue, particularly in resource-limited and remote areas. Early detection is vital for treatment, yet the lack of skilled radiologists underscores the need for artificial intelligence (AI)-driven screening tools. Developing reliable AI models is challenging due to the necessity for large, high-quality datasets, which are costly to obtain. To tackle this, we propose a teacher--student framework which enhances both disease and symptom detection on chest X-rays by integrating two supervised heads and a self-supervised head. Our model achieves an accuracy of 98.85% for distinguishing between COVID-19, tuberculosis, and normal cases, and a macro-F1 score of 90.09% for multilabel symptom detection, significantly outperforming baselines. The explainability assessments also show the model bases its predictions on relevant anatomical features, demonstrating promise for deployment in clinical screening and triage settings.

An Explainable Hybrid AI Framework for Enhanced Tuberculosis and Symptom Detection

TL;DR

This work tackles TB screening in resource-constrained settings by delivering an explainable, semi-supervised chest X-ray framework that jointly detects three-class disease (COVID-19, TB, normal) and seven radiographic symptoms. It combines a DISTL-style teacher–student architecture with a ViT-Small backbone, a DINO self-supervised head, and two supervised heads for disease and symptom detection, augmented by lung segmentation and CheXpert initialization. The model achieves accuracy and macro-F1 on disease classification, plus a macro-F1 on symptom multilabel detection, outperforming CNN baselines and providing Grad-CAM explanations that align with anatomical findings. This approach supports reliable screening and triage in real-world clinical workflows by delivering high sensitivity/specificity and interpretable radiographic evidence.

Abstract

Tuberculosis remains a critical global health issue, particularly in resource-limited and remote areas. Early detection is vital for treatment, yet the lack of skilled radiologists underscores the need for artificial intelligence (AI)-driven screening tools. Developing reliable AI models is challenging due to the necessity for large, high-quality datasets, which are costly to obtain. To tackle this, we propose a teacher--student framework which enhances both disease and symptom detection on chest X-rays by integrating two supervised heads and a self-supervised head. Our model achieves an accuracy of 98.85% for distinguishing between COVID-19, tuberculosis, and normal cases, and a macro-F1 score of 90.09% for multilabel symptom detection, significantly outperforming baselines. The explainability assessments also show the model bases its predictions on relevant anatomical features, demonstrating promise for deployment in clinical screening and triage settings.
Paper Structure (19 sections, 1 equation, 3 figures, 5 tables)

This paper contains 19 sections, 1 equation, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Methods
  4. Lung Segmentation
  5. Data Split
  6. Network Architecture
  7. Training Strategy
  8. Network Training Process
  9. Phase 1: Supervised Pretraining
  10. Phase 2: Semi-Supervised DISTL Training
  11. Phase 3: Correction Phase
  12. Performance Evaluation
  13. Results
  14. Three‑class Disease Classification
  15. Symptom‑wise multilabel detection
  16. ...and 4 more sections

Figures (3)

  • Figure 1: High-level conceptual flow of the analysis.
  • Figure 2: Sample images in the dataset.
  • Figure 3: Symptom-level explainability examples. Each row represents a different symptom. The left image depicts the original chest X-ray, while the right image shows the Grad-CAM overlay highlighting areas of model focus. The green rectangle indicates the dataset-provided ground-truth bounding box for each symptom.