Abstract
Diffusion MRI (dMRI) is central to investigating microstructural tissue properties in neurological diseases. In temporal lobe epilepsy (TLE), dMRI of surgically resected tissue can be compared with histological findings to validate models of tissue microstructure. Reliable segmentation of gray matter (GM) and white matter (WM) is a fundamental prerequisite for dMRI-based analyses, yet conventional segmentation methods rely on whole-brain anatomy and spatial priors. This limits their applicability to incomplete specimens such as surgically resected tissue or acquisitions with limited field of view. Ex vivo settings introduce an additional challenge because post-mortem tissue changes alter MRI parameters, reducing the reliability of standard segmentation approaches. We investigate whether dMRI alone can support region-agnostic tissue segmentation of brain fragments, and whether structural imaging can contribute to an improved segmentation. The nnU-Net is pretrained on diverse neocortical subregions from in vivo whole human brain scans and fine-tuned on ex vivo temporal lobe specimens. Nine models are evaluated across dMRI-derived inputs (b0, FA, MD) and structural inputs ($\mathrm{T}_1$-weighted images, quantitative $\mathrm{R}_1$ and $\mathrm{R}_2^*$ maps), individually and in combination. On atlas-extracted fragments, dMRI-derived contrasts alone show limited and variable performance, but multi-contrast models combining dMRI-derived and structural inputs achieve the strongest results. On ex vivo TLE specimens, dMRI-derived models perform competitively despite post-mortem tissue changes, and multi-contrast models again achieve the highest performance. These results demonstrate that region-agnostic multi-tissue segmentation of fragmented brain dMRI is feasible. Since structural images consistently improve performance, they should be acquired alongside dMRI in fragmented-brain imaging protocols.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/cdmri_010.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=DOqFSjcSXc
BibTex
@InProceedings{BogLau_Towards_MICCAISAT2026,
author = { Bogs, Laura AND Lüthi, Nina AND Fritz, Francisco J. AND Sura, Noémie AND Oeschger, Jan Malte AND Mordhorst, Laurin AND Sauvigny, Thomas AND Heinrich, Mattias P. AND Mohammadi, Siawoosh},
title = { { Towards Region-Agnostic Multi-Tissue Segmentation of Brain MRI: Application to Temporal Lobe Epilepsy Resections } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17278},
month = {pending},
page = {pending}
}