Quantification of atherosclerotic plaque components using in vivo MRI and supervised classifiers

Area of Science:

• Biomedical Imaging
• Cardiovascular Research
• Artificial Intelligence in Medicine

Background:

• Carotid atherosclerotic plaque characterization is crucial for stroke risk assessment.
• Accurate in vivo quantification of plaque components is challenging with current methods.

Purpose of the Study:

• To evaluate the efficacy of supervised classifiers in quantifying carotid atherosclerotic plaque components using multisequence MRI.
• To compare the performance of algorithmic classifiers against human MRI readers.

Main Methods:

• Multisequence MRI data acquired from 25 symptomatic subjects.
• Histological micrographs of endarterectomy specimens used as ground truth.
• Four supervised classifiers and two human readers quantified plaque components (calcified tissue, fibrous tissue, lipid core, intraplaque hemorrhage).

Main Results:

• Classifiers and human readers could not significantly quantify calcified tissue.
• A simple Bayesian classifier demonstrated superior performance for other tissue types compared to other classifiers and human readers.
• Classifiers outperformed human readers in quantifying the combined lipid and hemorrhage proportions.

Conclusions:

• Algorithmic classifiers show promise for accurate in vivo quantification of carotid plaque components.
• Supervised machine learning offers benefits over manual interpretation for specific plaque constituents.
• Further research is needed to improve calcium quantification in atherosclerotic plaques.