Faster R-CNN, fourth-order partial differential equation and global-local active contour model (FPDE-GLACM) for plaque segmentation in IV-OCT image

The accurate segmentation of plaque in intravascular optical coherence tomography (IV-OCT) image plays an important role in coronary atherosclerotic heart disease (CAD) diagnosis. To effectively provide information of coronary artery stenosis, we propose a novel hybrid framework which includes the faster R-CNN, fourth-order partial differential equation and global-local active contour model (FPDE-GLACM). This framework can efficiently detect and segment the plaque area in Speckle noise-contaminated IV-OCT images. We first detect plaque area by faster R-CNN and set bounding-box as the initial contour for active contour model. And then we minimize the joint energy functional of PDE-GLACM part to achieve the segmentation and denoising of IV-OCT images by gradient descent and finite difference scheme. Specifically, by using the Gaussian image minus original image to get the edge guide image, GLACM part obtains accurate plaque segmentation results. We perform experiments on 5000 IV-OCT images and set clinical manual segmentation results as ground truth. As expected, the results illustrate that the proposed FPDE-GLACM can provide better performance on plaque detection and segmentation. And these results may assist doctor in CAD diagnosis and treatment.