Opening the black box: a qualitative study of tuberculosis physicians’ needs for explainable AI in China
Abstract
Abstract Background Artificial intelligence (AI) has shown substantial potential in tuberculosis (TB). However, physicians’ distrust of “black-box” algorithms remains a major barrier to clinical adoption. Although explainable AI (XAI) has emerged to enhance transparency, limited evidence exists regarding TB physicians’ specific needs for AI explainability in high-uncertainty infectious disease contexts. Method A qualitative descriptive study was conducted between November 2025 and January 2026 in Hubei Province, China. 26 TB physicians participated in semi-structured interviews. Data were analyzed using an inductive-deductive thematic analysis approach. Coding was conducted independently by two researchers and refined through iterative discussion until thematic saturation was reached. Results Five overarching themes emerged, including global explanations, local explanations, case-based explanations, mixed visual-text presentation formats, and implementation considerations. Physicians emphasized the importance of exclusionary explanations to support differential diagnosis and preferred concise, multimodal outputs aligned with clinical workflows. Optional access to detailed information was favored to reduce cognitive burden. Seamless system integration, institutional financial support, and robust data security were considered essential for AI adoption. Conclusion XAI in TB extends beyond algorithmic transparency and functions as cognitive support for differential reasoning and trust calibration. User-centered, context-sensitive AI explanation design and supportive institutional policies are critical to ensuring responsible and effective integration of AI into infectious disease practice.
