Automated comprehensive evaluation of coronary artery plaque in IVOCT using deep learning

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Manual characterization and quantification of coronary artery plaque in intravascular optical coherence tomography (IVOCT) images are time-consuming and subjective.
• Accurate plaque assessment is crucial for predicting cardiovascular events.

Purpose of the Study:

• To develop and validate a deep learning-based semantic segmentation model (EDA-UNet) for automated characterization and quantification of coronary artery plaque in IVOCT images.
• To assess the model's performance in external validation datasets.

Main Methods:

• A deep learning semantic segmentation model (EDA-UNet) was developed using IVOCT images from two centers for training and internal testing.
• The model was externally validated on IVOCT images from an independent center.
• Performance was evaluated using Dice coefficients for plaque types and correlation analysis for quantitative metrics.

Main Results:

• External testing yielded Dice coefficients of 0.8282 for fibrous plaque, 0.7408 for calcified plaque, and 0.7052 for lipid plaque.
• The EDA-UNet model showed strong correlation and consistency with ground truth for calcification scoring and thin-cap fibroatheroma (TCFA) identification.
• The median analysis time per case was reduced to 18 seconds.

Conclusions:

• The EDA-UNet model provides an efficient and accurate automated tool for coronary artery plaque characterization and quantification using IVOCT.
• This deep learning approach has the potential to significantly improve the clinical workflow for cardiovascular disease assessment.