Clinical AI Needs More Transparency on Uncertainty to Improve Patient Care

Clinical AI tools should provide measures of uncertainty for predictions to benefit individual patients.

Uncertainty measures like conformal prediction can improve AI clinical utility.

Predictive uncertainty is needed to judge AI reliability and determine appropriate actions.

Lack of uncertainty estimates can limit clinical implementation of AI.

Techniques exist to quantify uncertainty but are not widely adopted yet in clinical AI.