Automatic Plaque Detection in IVOCT Pullbacks Using Convolutional Neural Networks.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Coronary heart disease (CHD) is a leading preventable cause of death.
• Intravascular optical coherence tomography (IVOCT) is crucial for detecting arterial plaque but generates large datasets.
• Current automatic plaque detection methods primarily use traditional machine learning, with limited exploration of deep learning.

Purpose of the Study:

• To investigate the efficacy of deep learning models for automatic plaque detection and differentiation in IVOCT images.
• To develop and evaluate a deep learning-based clinical decision support system for CHD.
• To establish a new benchmark dataset of labeled in vivo IVOCT patient images.

Main Methods:

• Utilized a novel dataset of in vivo IVOCT images, expertly labeled for plaque.
• Employed state-of-the-art deep learning models, including transfer learning, for direct plaque classification.
• Investigated Cartesian and polar image representations, data augmentation, and a multi-path architecture fusing both representations.
• Addressed plaque differentiation alongside detection.

Main Results:

• The combined deep learning model achieved high performance metrics: 91.7% accuracy, 90.9% sensitivity, and 92.4% specificity.
• Transfer learning and tailored data augmentation strategies enhanced model performance.
• The multi-path architecture effectively exploited features from both image representations.

Conclusions:

• Deep learning approaches are highly feasible for developing effective clinical decision support systems for IVOCT-based plaque detection.
• The developed model shows significant potential for improving the speed and accuracy of CHD diagnosis.
• Further research into deep learning for IVOCT image analysis can advance cardiovascular disease management.