Characterization of coronary artery pathological formations from OCT imaging using deep learning

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Coronary artery disease poses significant health risks, potentially leading to myocardial infarction and sudden death.
• Optical Coherence Tomography (OCT) offers high-resolution imaging (10-20 µm) for characterizing coronary artery tissues.
• Kawasaki disease (KD) can cause pathological formations within coronary arteries, necessitating accurate diagnostic methods.

Purpose of the Study:

• To investigate deep learning models for robust characterization of intracoronary pathological formations in pediatric Kawasaki disease using OCT imaging.
• To evaluate the performance of different pre-trained convolutional neural networks for tissue classification.
• To assess the efficacy of a majority voting approach for improving classification accuracy.

Main Methods:

• Experiments utilized OCT imaging data from 33 retrospective pediatric cases with KD.
• Deep features were extracted using three different pre-trained convolutional networks.
• A majority voting ensemble method was applied to combine predictions from individual models for final classification.

Main Results:

• The deep learning models achieved high accuracy, sensitivity, and specificity for tissue classification, with values up to 0.99 ± 0.01.
• The majority voting approach demonstrated robustness in interpreting OCT images.
• The study successfully identified various intracoronary pathological formations associated with KD.

Conclusions:

• Deep learning models are effective for automatic interpretation of OCT images in the context of Kawasaki disease.
• The majority voting method significantly enhances the robustness and accuracy of OCT image analysis.
• This AI-driven approach holds promise for improved diagnosis and management of coronary artery pathologies in KD patients.