Performance uncertainty in medical image analysis: a large-scale investigation of confidence intervals

Authors:Pascaline André (1), Charles Heitz (1), Evangelia Christodoulou (2, 5, 6), Annika Reinke (2, 4), Carole H. Sudre (3, 7, 8), Michela Antonelli (7, 8), Patrick Godau (2, 5), M. Jorge Cardoso (7), Antoine Gilson (1), Sophie Tezenas du Montcel (1), Gaël Varoquaux (9), Lena Maier-Hein (2, 4, 5, 10, 11), Olivier Colliot (1) ((1) Sorbonne Université, Institut du Cerveau - Paris Brain Institute - ICM, CNRS, Inria, Inserm, AP-HP, Hôpital de la Pitié-Salpêtrière, Paris, France (2) German Cancer Research Center (DKFZ) Heidelberg, Division of Intelligent Medical Systems, Germany (3) Unit for Lifelong Health and Ageing at UCL, Department of Population Science and Experimental Medicine and Hawkes InstituteCentre for Medical Image Computing, Department of Computer Science, University College London, UK (4) DKFZ Heidelberg, Helmholtz Imaging, Germany (5) National Center for Tumor Diseases (NCT), NCT Heidelberg, a partnership between DKFZ and Heidelberg University Hospital, Germany (6) AI Health Innovation Cluster, Germany (7) School of Biomedical Engineering and Imaging Science, King's College London, UK (8) Hawkes Institute, Department of Computer Science, University College London, UK (9) SODA project team, Inria Saclay-Île-de-France, France (10) Faculty of Mathematics and Computer Science, Heidelberg University, Germany (11) Medical Faculty, Heidelberg University, Germany)

Abstract:Performance uncertainty quantification is essential for reliable validation and eventual clinical translation of medical imaging artificial intelligence (AI). Confidence intervals (CIs) play a central role in this process by indicating how precise a reported performance estimate is. Yet, due to the limited amount of work examining CI behavior in medical imaging, the community remains largely unaware of how many diverse CI methods exist and how they behave in specific settings. The purpose of this study is to close this gap. To this end, we conducted a large-scale empirical analysis across a total of 24 segmentation and classification tasks, using 19 trained models per task group, a broad spectrum of commonly used performance metrics, multiple aggregation strategies, and several widely adopted CI methods. Reliability (coverage) and precision (width) of each CI method were estimated across all settings to characterize their dependence on study characteristics. Our analysis revealed five principal findings: 1) the sample size required for reliable CIs varies from a few dozens to several thousands of cases depending on study parameters; 2) CI behavior is strongly affected by the choice of performance metric; 3) aggregation strategy substantially influences the reliability of CIs, e.g. they require more observations for macro than for micro; 4) the machine learning problem (segmentation versus classification) modulates these effects; 5) different CI methods are not equally reliable and precise depending on the use case. Finally, we derived practical implications of this study in the form of a decision tree which shall prove useful to the community. This paves the way for future consensus guidelines on reporting performance uncertainty.

Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)

arXiv:2601.17103 [cs.CV]

(or arXiv:2601.17103v2 [cs.CV] for this version)