Carotid plaque segmentation and classification using MRI-based plaque texture analysis and convolutional neural...

Area of Science:

• Cerebrovascular disease research
• Artificial intelligence in medical imaging
• Deep learning for medical diagnostics

Background:

• Accurate carotid plaque segmentation and classification are vital for stroke risk assessment.
• Current methods face challenges with manual intervention, variability, and limited generalizability.
• These limitations hinder the clinical utility of existing approaches.

Purpose of the Study:

• To develop a fully automated hybrid deep learning framework for carotid plaque analysis.
• To improve the accuracy and generalizability of plaque segmentation and classification.
• To enhance stroke risk stratification and cerebrovascular disease management.

Main Methods:

• A hybrid framework combining Mask R-CNN for segmentation and a dual-path CNN for classification was proposed.
• Expert-annotated MRI scans were utilized for training and validation.
• Performance was evaluated using metrics like Dice Similarity Coefficient (DSC), Intersection over Union (IoU), accuracy, and ROC-AUC.

Main Results:

• Mask R-CNN achieved robust segmentation (mean DSC/IoU of 0.34) despite complex anatomy.
• The custom CNN demonstrated high classification accuracy (86.17%) and ROC-AUC (0.86), outperforming Inception V3.
• The hybrid model significantly surpassed conventional methods in plaque characterization and high-risk plaque identification.

Conclusions:

• The developed framework offers a fully automated, reproducible, and generalizable solution for carotid plaque analysis.
• This AI-driven approach reduces manual dependency and inter-observer variability.
• The findings support its potential as a diagnostic tool for standardized cerebrovascular disease management.