Average Calibration Error: A Differentiable Loss for Improved Reliability in Image Segmentation

Theodore Barfoot*, Luis C.Garcia Peraza Herrera, Ben Glocker, Tom Vercauteren

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference paperpeer-review

Abstract

Deep neural networks for medical image segmentation often produce overconfident results misaligned with empirical observations. Such miscalibration challenges their clinical translation. We propose to use marginal L1 average calibration error (mL1-ACE) as a novel auxiliary loss function to improve pixel-wise calibration without compromising segmentation quality. We show that this loss, despite using hard binning, is directly differentiable, bypassing the need for approximate but differentiable surrogate or soft binning approaches. Our work also introduces the concept of dataset reliability histograms which generalises standard reliability diagrams for refined visual assessment of calibration in semantic segmentation aggregated at the dataset level. Using mL1-ACE, we reduce average and maximum calibration error by 45% and 55% respectively, maintaining a Dice score of 87% on the BraTS 2021 dataset.

Original languageEnglish
Title of host publicationMedical Image Computing and Computer Assisted Intervention – MICCAI 2024, 27th International Conference Proceedings
EditorsMarius George Linguraru, Qi Dou, Aasa Feragen, Stamatia Giannarou, Ben Glocker, Karim Lekadir, Julia A. Schnabel
PublisherSpringer Science and Business Media Deutschland GmbH
Pages139-149
Number of pages11
ISBN (Print)9783031721137
DOIs
Publication statusPublished - 2024
Event27th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2024 - Marrakesh, Morocco
Duration: 6 Oct 202410 Oct 2024

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume15009 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference27th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2024
Country/TerritoryMorocco
CityMarrakesh
Period6/10/202410/10/2024

Fingerprint

Dive into the research topics of 'Average Calibration Error: A Differentiable Loss for Improved Reliability in Image Segmentation'. Together they form a unique fingerprint.

Cite this