UMN Researchers Develop AI Framework to Identify Diagnostic Uncertainty

A new study published in npj Digital Medicine introduces ConfiDx, an AI framework developed by Dr. Rui Zhang’s team at the University of Minnesota to enhance diagnostic reliability by identifying and explaining uncertainty in clinical decision-making.

Diagnostic uncertainty is a persistent challenge in medicine, particularly in primary care and ICU settings where data may be incomplete. Recognizing and explaining diagnostic uncertainty helps providers address these information gaps and improve patient safety.

ConfiDx integrates medical guidelines into large language models, allowing the system to align with diagnostic standards, detect uncertain cases, and generate clear, evidence-based explanations.

In evaluations using real-world clinical datasets, ConfiDx outperformed existing models in diagnostic accuracy and uncertainty recognition. When used in collaboration with physicians, it significantly improved clinicians’ ability to identify and explain diagnostic uncertainty.

Lead author Dr. Shuang Zhou emphasized the collaborative value of the approach, noting that “this work demonstrates the importance of collaboration between AI and physicians and represents an advancement in trustworthy medical AI.”