arXiv 2412.03318

Domain-Agnostic Stroke Lesion Segmentation Using Physics-Constrained Synthetic Data

By Liam Chalcroft, Jenny Crinion, et al.

Published 2024-12-04

Mindmap

Browse the paper's core ideas, clusters, and relationships in a structured outline.

Segmenting stroke lesions in MRI is challenging due to diverse acquisition protocols that limit model generalisability. In this work, we introduce two physics-constrained approaches to generate synthetic quantitative MRI (qMRI) images that improve segmentation robustness across heterogeneous domains. Our first method, , trains a neural network to estimate qMRI maps from standard MPRAGE images, enabling the simulatio…

View the original paper on arXiv