Abstract
Automatic stroke lesion segmentation in T1-weighted magnetic resonance imaging remains challenging due to variation in acquisition conditions, disease stage, lesion size, and appearance. This paper describes our segmentation pipeline for the ISLES 2026 challenge, combining the medium residual-encoder nnU-Net preset (ResEnc M) with connected-component filtering based on physical volume. The pipeline was evaluated using stratified 5-fold cross-validation on 1450 scans from 55 sites, with the Default 3D full-resolution nnU-Net included as a reference. The results revealed that both configurations performed similarly, with ResEnc M showing modest improvements across most metrics. We then evaluated minimum connected-component volume thresholds ranging from 0 to 0.5 mL and observed similar trends for both configurations. For ResEnc M, a 0.05 mL threshold increased the F1-score from 0.575 to 0.609 while slightly reducing the mean Dice score from 0.663 to 0.658; the absolute lesion-count difference remained nearly unchanged at 1.86. F1-score improvements occurred mainly in medium- and large-lesion cases, with no improvement in small-lesion cases. Overall, small-component filtering improved lesion detection at a minor cost to voxel overlap, underscoring the need to assess both voxel-wise and lesion-wise performance.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ISLES_030.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/forum?id=qz7EKFRSf4
BibTex
@InProceedings{AmaKim_ResidualEncoder_MICCAISAT2026,
author = { Amador, Kimberly AND Jeong, Sumin AND Kang, Chris AND Martinez, David A. AND Vigneshwaran, Vibujithan AND Forkert, Nils D.},
title = { { Residual-Encoder nnU-Net with Small-Component Filtering for Multi-Site Stroke Lesion Segmentation in T1-weighted MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17255},
month = {pending},
page = {pending}
}