Segmentation of Carotid Arteries From Three-Dimensional Black-Blood Magnetic Resonance Imaging With Sparse Annotation...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease Research

Background:

• Carotid atherosclerosis quantification is crucial for cardiovascular event risk monitoring and therapy evaluation.
• High-resolution 3D carotid MRI offers extended coverage but requires extensive manual segmentation of hundreds of 2D images.
• Current segmentation methods face challenges with the large datasets generated by extended 3D MRI coverage.

Purpose of the Study:

• To develop an automated, multi-dimensional hybrid framework for segmenting carotid arteries in 3D MRI.
• To reduce the need for extensive manual segmentation by generating surrogate ground truth data.
• To improve the accuracy and efficiency of carotid artery wall and lumen segmentation for clinical applications.

Main Methods:

• A multi-dimensional hybrid framework combining 3D and 2D Convolutional Neural Networks (CNNs).
• Automatic generation of surrogate ground truth using a Region of Interest (ROI) U-Net and Point U-Net.
• A 3D multiscale U-Net for initial ROI localization and rough segmentation, guiding a 2D ROI U-Net for detailed segmentation.
• Incorporation of a 3D inception module and novel loss functions for enhanced segmentation continuity and accuracy.

Main Results:

• The proposed framework significantly outperforms top-ranked solutions on the Carotid Artery Vessel Wall Segmentation challenge dataset.
• Achieved state-of-the-art performance in segmenting carotid artery outer walls and lumens from 3D MRI.
• Demonstrated effective automatic ROI localization and surrogate ground truth generation, reducing manual effort.

Conclusions:

• The developed hybrid framework offers an efficient and accurate solution for segmenting carotid arteries in 3D MRI.
• This approach has the potential to streamline patient monitoring and accelerate the evaluation of new therapies for cardiovascular diseases.
• The automated segmentation method advances the clinical utility of high-resolution 3D carotid MRI.