Abstract
Shared 2D nnU-Net lesion-probability maps can serve scan-level lesion detection and voxel-level lesion segmentation in moderate-to-severe traumatic brain injury, provided that each task uses its own decision rule and endpoint. A 2D nnU-Net trained on 572 T1-weighted MRI scans produced the shared maps. An 82-scan development cohort was used for checkpoint, model, and threshold selection and for cross-validated downstream analyses. On these development maps, a normalized predicted lesion-volume rule and a maximum-probability rule were at least as competitive as a logistic map-feature classifier for detection. In a retrospective paired analysis of 44 development cases, adaptive segmentation increased mean Dice from 0.5166 to 0.5541, a difference of +0.0375 (95% CI 0.0152-0.0611). On the hidden challenge set, the complete submitted detector achieved balanced accuracy 0.8762, with sensitivity 0.8313 and specificity 0.9211. The two blind segmentation aggregates were reported separately and were not paired; Dice was 0.4525 for standard segmentation and 0.1671 for adaptive segmentation. Shared probability maps therefore supported both tasks, but development analyses did not demonstrate an advantage for multifeature detection, and adaptive segmentation did not show external benefit in its separate blind evaluation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AIMS_TBI_001.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/AIMS_TBI_001_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/profile?id=~Nair_Ul_Islam1
BibTex
@InProceedings{IslNai_Shared_MICCAISAT2026,
author = { Islam, Nair Ul},
title = { { Shared nnU-Net Probability Maps for msTBI Lesion Detection and Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17254},
month = {pending},
page = {pending}
}