Visual explanations for polyp detection: How medical doctors assess intrinsic versus extrinsic explanations

Area of Science:

• Medical Artificial Intelligence
• Computer Vision in Healthcare
• Explainable AI (XAI)

Background:

• Deep learning (DL) shows promise for analyzing medical images to detect diseases.
• The 'black box' nature of DL hinders clinical adoption due to physician skepticism.
• Existing explainable AI (XAI) methods lack consensus among medical professionals.

Purpose of the Study:

• To assess medical professionals' opinions on state-of-the-art explainable AI (XAI) methods.
• To evaluate XAI methods in a gastrointestinal disease detection use case.
• To compare intrinsic versus extrinsic explanation methods for clinical utility.

Main Methods:

• Conducted a study involving medical professionals evaluating XAI methods.
• Applied intrinsic and extrinsic explanation categories to a gastrointestinal disease detection task.
• Collected physician feedback on the perceived value and usability of AI explanations.

Main Results:

• Physicians showed a preference for intrinsic explanation methods over extrinsic ones.
• The study participants recognized the value of AI-driven explanations in a clinical context.
• Feedback indicated a need for tailored explanations to meet physician expectations.

Conclusions:

• Intrinsic explainable AI (XAI) methods are more favored by physicians for medical applications.
• Addressing physician needs can facilitate the integration of AI into clinical workflows.
• Tailored XAI can help overcome the 'black box' problem, promoting AI adoption in healthcare.