SML-Net: Semi-supervised multi-task learning network for carotid plaque segmentation and classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease Research

Background:

• Carotid plaque assessment is vital for predicting ischemic stroke risk.
• Current methods often treat plaque segmentation and classification as separate tasks, overlooking their interdependence.
• Acquiring extensive annotated data for segmentation is resource-intensive.

Purpose of the Study:

• To develop an integrated approach for simultaneous carotid plaque segmentation and classification.
• To address the limitations of separate task processing and high data annotation costs.
• To leverage information from both plaque and background regions for improved performance.

Main Methods:

• Proposed an end-to-end semi-supervised multi-task learning network (SML-Net).
• Implemented feature extraction and multi-scale feature fusion for enhanced semi-supervised segmentation.
• Utilized segmentation results to extract features from various dimensions for classification.

Main Results:

• SML-Net achieved 86.59% plaque classification accuracy and 82.36% Dice Similarity Coefficient (DSC).
• Outperformed leading single-task networks by improving DSC by 1.2% and accuracy by 1.84%.
• Surpassed the best multi-task network with a 1.05% increase in DSC and a 2.15% improvement in classification accuracy.

Conclusions:

• The proposed SML-Net effectively integrates segmentation and classification for carotid plaque analysis.
• This multi-task learning approach offers superior performance compared to single-task and existing multi-task methods.
• SML-Net provides a promising solution for efficient and accurate carotid plaque assessment, aiding in stroke risk evaluation.