Hacia una mejor toma de decisiones de aneurismas intracraneales no rotos mediante segmentación automatizada a partir...

To enable accurate 3D morphological assessment and support clinical decision making, DIVA-seg: a Deep learning-based method for Intracranial Vessel and Aneurysm segmentation from MRA-TOF using a pseudo labeling approach was developed and validated.

Three MRA-TOF datasets were used: 1) labeled data for training (n=57) and testing (n=14), 2) unlabeled data for pseudo labels (n=518), and 3) labeled data for external validation (n=82). An nnU-Net (Model 1) was iteratively trained for creating pseudo labels for Dataset 2. Cases with stable segmentation performance across iterations were selected for further training. Stable cases (n=484) were combined with Dataset 1 to train a second nnU-Net (Model 2). Performance testing on Dataset 1 and 3 comprised of Dice similarity coefficients (DSC), 95%-Hausdorff distances, 3D morphological measures, and a blinded qualitative evaluation.

DIVA-seg achieved a mean (standard deviation) internal vessel and aneurysm DSC of 0.925 (±0.025) and 0.880 (±0.045), respectively. On the external test set the DSC were 0.899 (±0.028) and 0.861 (±0.114), respectively. Mean Hausdorff distances were 0.67mm for both test sets. Bland-Altman plots showed a high agreement between 3D morphological measures from ground truth and model segmentations; however, a proportional bias was observed for voxel volume, surface area, sphericity and shape index. The qualitative evaluation showed no clear preference for either ground truth or model segmentation.

The model achieved accurate and reliable segmentation of vessels and aneurysms internally and externally while also showing high agreement between 3D morphological measures from automatic and manual segmentations, indicating its potential clinical utility.

Accurate intracranial aneurysm assessment is essential for treatment planning and risk stratification. Manual aneurysm segmentation is labor-intensive and subject to substantial inter- and intra-observer variability. Although automated segmentation approaches have been proposed, many suffer from limited accuracy, lack of robustness across datasets, or insufficient validation on heterogeneous, real-world data. As a result, reliable and generalizable tools for aneurysm segmentation and morphological analysis remain an unmet need. DIVA-seg, an nnU-Net-based model, achieved high aneurysm segmentation accuracy (DSC >0.86; HD <0.7mm) and close agreement with expert annotations in clinically relevant 3D morphological measures, demonstrating consistent performance across internal and external datasets. This work demonstrates a robust and generalizable approach for automated intracranial aneurysm segmentation, enabling reliable morphological analysis. The proposed method has the potential to streamline aneurysm monitoring, reduce observer variability, and support future automated tools for risk predictions and clinical decision making.