Exploring tuberculosis physicians' preferences for AI explainability in China: a protocol for a discrete choice...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Health Services Research

Background:

• Artificial intelligence (AI) shows promise in tuberculosis (TB) diagnosis, but clinical uptake hinges on physician trust and understanding.
• Explainable AI (XAI) aims to enhance trust, yet specific physician preferences for XAI features in TB diagnosis remain underexplored.
• Understanding these preferences is crucial for developing user-centered AI diagnostic tools.

Purpose of the Study:

• To investigate tuberculosis physicians' preferences for explainable AI (XAI) attributes in diagnostic settings.
• To quantify the trade-offs physicians are willing to make among different AI explainability features.
• To inform the development and implementation of effective XAI systems for TB diagnosis.

Main Methods:

• A discrete choice experiment (DCE) was designed using a D-efficient approach.
• Six key attributes of AI explainability were identified through literature review, interviews, and expert consultation.
• Preference data from TB physicians in Hubei Province, China, will be analyzed using multinomial and mixed logit models.

Main Results:

• The study protocol outlines the methodology for analyzing physician preferences for AI explainability attributes.
• The analysis will estimate the relative importance of different explainability features.
• Heterogeneity in preferences across physician subgroups will be explored.

Conclusions:

• This research protocol details a DCE to elicit TB physicians' preferences for AI explainability.
• The findings will guide the design of user-centered XAI systems, fostering greater adoption of AI in TB diagnostics.
• Dissemination aims to improve AI implementation in clinical practice.