The abstract is read at the publisher; the summary is JClub's.
A transfer learning-based 3D full-resolution nnU-Net for intracranial artery segmentation on 7T 4D Flow MRI achieved Dice >0.85 and HD95 ~3 mm.
- Why it matters: Reliable hemodynamic quantification depends on accurate vessel masks, yet labeled data are scarce and it is unclear whether different automated segmentation pipelines change downstream hemodynamic estimates.
- What they did: The nnU-Net was pretrained on 355 TOF-MRA scans and fine-tuned on 11 7T 4D Flow MRI scans, then compared with two published U-Net variants using test sets with different resolutions.
- The result: nnU-Net segmentations matched manual WSS best (mean 1.57 ± 0.63 Pa, ICC 0.96; max 2.16 ± 1.05 Pa, ICC 0.97) with minimal bias (≤1.7%), while other models showed systematic under- or overestimation and some segment-specific flow differences.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.