Dual-student ensemble boosts semi-supervised ischemic stroke segmentation on follow-up NCCT

A parameter-free dual-student ensemble semi-supervised method achieved the highest Dice score for ischemic stroke segmentation on follow-up non-contrast CT, outperforming other methods especially when labeled data were scarce.

• The ensemble framework used strong data augmentation and a parameter-free dual-student module to improve pseudo-label reliability on unlabeled data.
• The method was evaluated on a manually annotated follow-up NCCT dataset of 9,020 image slices across four labeled-data ratios.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.