How We Won the ISLES'24 Challenge by Preprocessing

TL;DR

The paper tackles CT-only ischemic stroke lesion segmentation in the ISLES'24 setting, where ground-truth comes from follow-up MRI. It demonstrates that a preprocessing pipeline with SynthStrip skull-stripping and modality-specific intensity windowing, paired with a standard nnU-Net ResEnc L, yields meaningful gains over baseline CT preprocessing and achieves top performance on the ISLES'24 test. Cross-validation shows Dice improvements from $21.8\%$ to $31.8\%$ using the proposed preprocessing, though substantial test-time variability remains. The work highlights the critical role of CT-specific preprocessing for stroke lesion segmentation and points to robustness and clinical priors as key avenues for future improvements.

Abstract

Stroke is among the top three causes of death worldwide, and accurate identification of stroke lesion boundaries is critical for diagnosis and treatment. Supervised deep learning methods have emerged as the leading solution for stroke lesion segmentation but require large, diverse, and annotated datasets. The ISLES'24 challenge addresses this need by providing longitudinal stroke imaging data, including CT scans taken on arrival to the hospital and follow-up MRI taken 2-9 days from initial arrival, with annotations derived from follow-up MRI. Importantly, models submitted to the ISLES'24 challenge are evaluated using only CT inputs, requiring prediction of lesion progression that may not be visible in CT scans for segmentation. Our winning solution shows that a carefully designed preprocessing pipeline including deep-learning-based skull stripping and custom intensity windowing is beneficial for accurate segmentation. Combined with a standard large residual nnU-Net architecture for segmentation, this approach achieves a mean test Dice of 28.5 with a standard deviation of 21.27.

Figures (1)

• Figure 1: Data (a) before and (b) after preprocessing, with windowing bounds shown below b).