Learning carotid vessel wall segmentation in black-blood MRI using sparsely sampled cross-sections from 3D data

Area of Science:

• Medical Imaging
• Cardiovascular Research
• Artificial Intelligence in Medicine

Background:

• Carotid artery atherosclerosis is a significant stroke risk factor.
• Accurate quantitative assessment of the carotid vessel wall is crucial for risk stratification.
• Reproducible and reliable automatic segmentation of vessel cross-sections is needed.

Purpose of the Study:

• To develop an automatic segmentation method for carotid artery cross-sections.
• To achieve orientation-invariant segmentation for accurate vessel wall thickness (VWT) assessment.
• To evaluate the model's performance on unseen data and diverse patient demographics.

Main Methods:

• A residual U-Net model was trained on sparsely sampled cross-sections of 3D black-blood MRI.
• The approach segments carotid arteries in cross-sections perpendicular to the centerline.
• The model was evaluated on 218 MRI datasets from 121 subjects with hypertension and plaque.

Main Results:

• Achieved high Dice coefficients (0.948 lumen, 0.859 wall) and low Hausdorff distances.
• Demonstrated robust performance on unrepresented regions and healthy subjects.
• Achieved a low median Hausdorff distance on a challenge test set.

Conclusions:

• The proposed method automates carotid artery vessel wall assessment, reducing manual effort.
• It enables reliable VWT measurement across various patient demographics and MRI settings.
• The method is suitable for clinical applications with human supervision.