MULTIMODAL EXPLAINABLE ARTIFICIAL INTELLIGENCE FOR EARLY DIAGNOSIS OF TUBERCULOSIS FROM CHEST IMAGING AND CLINICAL DATA

Abstract

Tuberculosis (TB) remains a major public-health challenge, particularly in resource-constrained settings. This study investigated a multimodal explainable artificial intelligence (AI) framework for early TB diagnosis using chest X-ray images and clinical data. Three approaches were compared: imaging-only, clinical-data-only, and multimodal AI. The proposed framework combined deep-learning image analysis, clinical feature processing, multimodal fusion, and explainable AI (XAI). The research model achieved 95.1% accuracy, 95.6% sensitivity, 94.6% specificity, 95.1% F1-score, and 0.982 AUROC. Explainability analysis identified clinically relevant radiographic regions and patient-level variables contributing to predictions. The findings suggest that multimodal explainable AI can improve TB screening accuracy and transparency. However, external validation, fairness assessment, and integration with confirmatory diagnostic procedures are required before clinical deployment.