Abstract
Medical images are acquired at different spatial resolutions, but segmentation models and evaluation metrics do not always account for these differences. We investigate how variation in CT slice thickness affects both the evaluation of segmentation boundaries and the uncertainty produced by a segmentation model. Using the KiTS23 dataset, we show that a boundary metric defined in voxel space can give systematically different results across different slice thicknesses, whereas defining the same neighbourhood in physical space removes this dependence more consistently. We also find that reported Dice scores depend on how results are aggregated across cases. Finally, model uncertainty at through-plane boundaries increases with slice thickness, although its relationship to annotation uncertainty remains to be established. These findings highlight several ways in which voxel size can influence both the measurement and interpretation of segmentation performance.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/UNSURE2026_058.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/profile?id=~Khadijah_Mohammed2
BibTex
@InProceedings{MohKha_When_MICCAISAT2026,
author = { Mohammed, Khadijah AND Voiculescu, Irina},
title = { { When Voxels Are Not Equal: Spacing-Driven Metric Bias and Metric Uncertainty in Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17260},
month = {pending},
page = {pending}
}