Out-of-distribution Detection in Medical Image Analysis: A survey
It tackles the critical issue of ensuring trustworthy AI in medical diagnostics by summarizing methods to detect out-of-distribution samples, which is incremental as it compiles and organizes existing research rather than introducing new techniques.
This survey reviews recent advances in out-of-distribution detection for medical image analysis, addressing the problem of silent failures in deep learning models when encountering distributional shifts in clinical scenarios, and categorizes existing solutions and evaluation methods.
Computer-aided diagnostics has benefited from the development of deep learning-based computer vision techniques in these years. Traditional supervised deep learning methods assume that the test sample is drawn from the identical distribution as the training data. However, it is possible to encounter out-of-distribution samples in real-world clinical scenarios, which may cause silent failure in deep learning-based medical image analysis tasks. Recently, research has explored various out-of-distribution (OOD) detection situations and techniques to enable a trustworthy medical AI system. In this survey, we systematically review the recent advances in OOD detection in medical image analysis. We first explore several factors that may cause a distributional shift when using a deep-learning-based model in clinic scenarios, with three different types of distributional shift well defined on top of these factors. Then a framework is suggested to categorize and feature existing solutions, while the previous studies are reviewed based on the methodology taxonomy. Our discussion also includes evaluation protocols and metrics, as well as the challenge and a research direction lack of exploration.