Search papers, labs, and topics across Lattice.
This study addresses the challenge of segmenting ischemic stroke lesions on T1-weighted MRI scans acquired from diverse scanners and protocols, where standard deep learning models struggle due to subtle lesion characteristics. By integrating a MedNeXt-L backbone with an innovative 3D CarveMix augmentation technique that dynamically generates synthetic lesion placements during training, the authors enhance the model's exposure to diverse lesion patterns. The approach yields a mean Dice score of 0.648 across 1,453 scans, marking a significant improvement over previous methods while maintaining the same training budget.
Dynamic synthetic lesion generation boosts stroke lesion segmentation performance, achieving a notable Dice score improvement without additional data.
Segmenting ischemic stroke lesions on T1-weighted (T1w) MRI acquired across different scanners and protocols without intensity standardization is difficult because lesions are subtle and share intensity characteristics with cerebrospinal fluid. Standard deep learning architectures trained across multiple centers plateau around Dice 0.66, with acute lesions ($\le 7$ days post-stroke) performing substantially worse due to severe sample scarcity. We combine a MedNeXt-L ($k=5$) backbone with on-the-fly 3D CarveMix augmentation that pastes real lesion patches into healthy brain regions during training. By generating synthetic lesion placements dynamically within each fold with subject-level split isolation, the model sees more diverse lesion patterns without requiring pre-generated copies on disk. We evaluate on 1,453 native T1w scans from 55 clinical centers in the ISLES 2026 challenge. Our method achieves a mean 5-fold cross-validation Dice of 0.648 at 500 epochs, a +0.018 improvement over the MedNeXt-L backbone at a matched training budget (0.630)
