Abstract
Accurate delineation of ischemic stroke lesions on T1-weighted (T1w) MRI is important for quantifying lesion burden across acute, sub-acute, and chronic disease stages, but automated methods have struggled to generalize across the heterogeneity of multi-center, multi-scanner data. We describe our submission to the ISLES’26 challenge, which targets infarct segmentation in native-space T1w MRI. Our approach is a self-configuring residual-encoder U-Net (nnU-Net ResEnc-M) trained with five-fold cross-validation stratified by acquisition site, lesion size, and lesion maturity, and deployed as a five-fold ensemble. On the 1,453-scan ISLES’26 training set we obtain a cross-validated mean Dice of 0.66, improving over a plain-convolution baseline through the residual encoder and extended training. Augmenting training with an external, publicly available chronic-lesion cohort, screened for overlap with the provided data, further improves cross-validated Dice to 0.67, with the benefit growing at longer training budgets. We report internal cross-validation performance and discuss the practical trade-offs of external-data augmentation for this task.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ISLES_053.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=DnK8HXyK7h
BibTex
@InProceedings{DraChr_ResidualEncoder_MICCAISAT2026,
author = { Drake, Christopher A.},
title = { { Residual-Encoder nnU-Net for Multi-Phase Ischemic Stroke Lesion Segmentation in Native-Space 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}
}