Multiclass plaque segmentation of coronary artery disease patients using an enhanced model

Background: The analysis of coronary artery plaques in coronary artery disease (CAD) patients is instrumental in enabling cardiologists to administer personalized treatments and improve patient prognosis. Intravascular optical coherence tomography (IVOCT) provides precise insights into plaque characteristics. Purpose: Enable cardiologists to provide personalized treatment for patients with CAD; the objective is to explore methods for swiftly and accurately automating the analysis of IVOCT data. Methods: The study employs a methodology that integrates two types of coordinate images and enhances the DeepLabV3+ model. This approach facilitates direct three-class segmentation results for fibrous, lipid, and calcified plaques. Our experimental data is split into training, validation, and test sets in a ratio of 7:2:1. Results: The primary findings of the research indicate that the enhanced DeepLabV3+ model achieves F1 scores of 0.855, 0.850, and 0.551 for fiber, lipid, and calcified plaque detection at the plaque level, respectively. Conclusions: This study lies in its potential to improve the accuracy and efficiency of plaque segmentation in CAD patients, ultimately benefiting cardiologists by enabling more personalized and effective treatments.