Deep feature learning for automatic tissue classification of coronary artery using optical coherence tomography

Area of Science:

• Medical imaging
• Artificial intelligence in medicine
• Cardiovascular pathology

Background:

• Kawasaki disease (KD) can lead to coronary artery complications like aneurysms and stenosis.
• Accurate classification of coronary artery layers (intima, media, scar) is crucial for analyzing OCT images in pediatric patients.
• Optical coherence tomography (OCT) provides high-resolution intracoronary imaging.

Purpose of the Study:

• To develop a fully automated method for classifying coronary artery tissues from OCT images.
• To compare the performance of convolutional neural networks (CNNs), random forest (RF), and support vector machine (SVM) classifiers.
• To enhance the characterization of the challenging-to-identify media layer in pediatric coronary arteries.

Main Methods:

• Utilized convolutional neural networks (CNNs) as a feature extractor.
• Compared CNNs with random forest (RF) and support vector machine (SVM) for tissue classification.
• Applied the methods to optical coherence tomography (OCT) images of pediatric coronary arteries.

Main Results:

• The combination of CNN as a feature extractor and RF as a classifier demonstrated high robustness.
• Achieved a classification rate of up to 96% for coronary artery layers.
• Successfully characterized the thin media layer, often difficult to distinguish.

Conclusions:

• The developed automated method, particularly CNN-RF, is effective for classifying coronary artery layers in OCT images.
• This approach aids in the analysis of Kawasaki disease-related vascular changes in children.
• The method shows promise for improving diagnostic accuracy and patient management.