January 28, 2025

Explainable AI (XAI) in radiology refers to AI systems that assist physicians by providing transparent and understandable diagnostic advice, often accompanied by visual annotations. This helps physicians, especially those not specialized in radiology, to make more accurate diagnoses based on AI-generated insights.

Purpose of the Study

The study evaluated and compared how task experts (i.e., radiologists) and non-task expert physicians, such as emergency and internal medicine physicians, benefit from correct and explainable AI advice when reviewing X-rays. The goal was to understand the impact of AI-generated advice on diagnostic accuracy and physician confidence.

How the Study Was Done

Researchers conducted an experimental study where physicians reviewed X-rays accompanied by diagnostic advice. This advice came with or without visual annotations and was labelled as coming from either an AI system or a human radiologist. The study measured diagnostic accuracy, the perceived quality of advice, and physician confidence in their diagnostic decision to assess the effects of explainable AI.

Key Findings

The study found that non-task expert physicians significantly benefited from correct explainable AI advice, achieving higher diagnostic accuracy when the advice included visual annotations. Physicians rated AI advice as higher quality than human advice, although their confidence levels were not strongly affected by the source of the advice. The findings suggest that explainable AI can enhance diagnostic performance by providing clear and interpretable support, particularly for non-specialist physicians.

Who Performed the Study

This study was conducted by a collaborative team from institutions including LMU Munich, Massachusetts Institute of Technology, University Hospital Regensburg, University of Toronto, St. Michael’s Hospital, and Universidade Federal de São Paulo. Key researchers included Susanne Gaube, Harini Suresh, Martina Raue, Eva Lermer, Timo K. Koch, Matthias F. C. Hudecek, Alun D. Ackery, Samir C. Grover, Joseph F. Coughlin, Dieter Frey, Felipe C. Kitamura, Marzyeh Ghassemi, and Errol Colak.

What is Explainable AI in Radiology? - Toronto Radiology