Chinese tuberculosis physicians outline explainability needs for clinical AI

Tuberculosis physicians in China identified five priorities for explainable artificial intelligence in clinical practice: global explanations, local explanations, case-based explanations, mixed visual-text presentation formats and implementation considerations. The findings emerged from interviews with 26 physicians and suggest that useful explainable AI should support differential diagnosis, fit clinical workflows and help physicians calibrate their trust in computer-generated assessments.

Researchers conducted the qualitative descriptive study between November 2025 and January 2026 in Hubei Province, China. Semi-structured interviews with 26 tuberculosis physicians examined what explanations physicians considered important when AI is used in an infectious disease setting marked by high uncertainty. Artificial intelligence has shown substantial potential in tuberculosis care, but physicians’ distrust of “black-box” algorithms remains a barrier to clinical adoption, according to the study’s rationale.

Five overarching themes emerged from the analysis. Physicians emphasized global, local and case-based explanations as complementary forms of explainability. They also highlighted exclusionary explanations, which they considered valuable for differential diagnosis, and expressed a preference for concise outputs combining visual and textual information. Access to more detailed information should remain optional, allowing physicians to obtain additional material without adding unnecessary cognitive burden to routine use.

The interviews indicate that transparency alone was not viewed as sufficient for meaningful clinical use. Instead, the physicians described explainable AI as a form of cognitive support for differential reasoning and trust calibration. Their preferences depended not only on the explanation offered but also on how it was presented and integrated into clinical work. The findings support user-centered, context-sensitive explanation design, but they do not establish that any particular explainable AI system improves diagnostic accuracy, reduces decision time, changes clinical decisions or improves patient outcomes.

Implementation emerged as a distinct theme. Physicians considered seamless integration with hospital systems, institutional financial support and robust data security essential for adoption. These findings indicate that technical design alone may not determine whether explainable AI can be incorporated into infectious disease practice. The study used an inductive-deductive thematic analysis, with two researchers coding the interview data independently and refining themes through iterative discussion until thematic saturation was reached.

Because the evidence comes from 26 tuberculosis physicians in a single Chinese province, it may not represent preferences among clinicians in other specialties, countries or health systems with different digital infrastructure and institutional arrangements. The study also does not provide clinical performance results for an explainable AI tool. Its contribution is a qualitative account of what physicians said they need from future systems: explanations suited to differential reasoning, concise multimodal presentation, optional detail, dependable integration, financial support and strong data protections. The authors conclude that these elements are important for responsible AI integration into tuberculosis practice, rather than evidence that routine deployment is ready or clinically proven.

Subject of Research: Tuberculosis physicians’ requirements for explainable artificial intelligence in clinical practice

Article Title: Opening the black box: a qualitative study of tuberculosis physicians’ needs for explainable AI in China

Article References: Zhang, J., Fu, Q., Du, S., Fang, L., Huang, L., & Wang, X. (2026). Opening the black box: a qualitative study of tuberculosis physicians’ needs for explainable AI in China. BMC Infectious Diseases. https://doi.org/10.1186/s12879-026-14355-7

Image Credits: AI Generated

DOI: 10.1186/s12879-026-14355-7

Keywords: tuberculosis, explainable AI, clinical decision support, physician interviews, infectious diseases, differential diagnosis, medical AI, trust calibration, healthcare implementation, China, Opening, black