Automatic A-line coronary plaque classification using combined deep learning and textural features in intravascular...

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Biomedical Engineering

Background:

• Coronary plaque characterization is crucial for cardiovascular disease management.
• Intravascular optical coherence tomography (OCT) provides high-resolution imaging of coronary arteries.
• Accurate and automated plaque classification remains a challenge.

Purpose of the Study:

• To develop and validate a fully automated method for classifying A-line coronary plaques.
• To combine deep learning and textural features for improved classification accuracy.
• To assess the clinical relevance of the automated classification for treatment planning and research.

Main Methods:

• Developed a method using combined deep learning and textural features for plaque classification.
• Employed preprocessing steps including guidewire/shadow removal and noise reduction.
• Utilized a convolutional neural network, feature selection (minimum redundancy maximum relevance), random forest classifier, and conditional random field (CRF) for noise cleaning.

Main Results:

• Achieved high sensitivities and specificities (e.g., 82.2%/90.8% for fibrolipidic, 82.4%/89.2% for fibrocalcific) with CRF noise cleaning.
• Demonstrated significant performance improvement (10-15%) after CRF noise cleaning.
• Showed favorable agreement between automated en face classification maps and manual labels, indicating clinical relevance.

Conclusions:

• The developed automated method accurately classifies coronary plaques in OCT images.
• The combination of deep learning and textural features, along with CRF noise cleaning, enhances classification performance.
• The method holds significant promise for clinical applications in treatment planning and for advancing cardiovascular research.