arXiv 2412.03318
Domain-Agnostic Stroke Lesion Segmentation Using Physics-Constrained Synthetic Data
By Liam Chalcroft, Jenny Crinion, et al.
Published 2024-12-04
Wiki summary
Explore the paper's summary, context, and related research on Papiers.
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…