DeepISLES sets new benchmark for stroke MRI segmentation
Researchers, including contributions from the DQBM Menze Lab, have developed DeepISLES, an ensemble deep learning algorithm for segmenting ischemic stroke lesions in MRI scans. Built from top-performing entries in the ISLES’22 challenge, DeepISLES combines diverse algorithmic strengths to achieve superior accuracy, robustness to varied imaging conditions, and performance comparable to expert neuroradiologists.
The model was validated on the largest available external stroke MRI dataset (N = 1685), outperforming state-of-the-art solutions in both lesion detection and segmentation. Its results correlate strongly with clinical stroke scores, highlighting its value for diagnosis, treatment planning, and research.
DeepISLES is freely available in multiple formats, supporting immediate use in both clinical and academic settings.
Read the publication here: 10.1038/s41467-025-62373-x