Automated Segmentation of Bioresorbable Vascular Scaffold Struts in Intracoronary Optical Coherence Tomography Images

Abstract

Bioresorbable vascular scaffolds (BVS), the next step in the continuum of minimally invasive vascular interventions present new opportunities for patients and clinicians but challenges as well. As they are comprised of polymeric materials standard imaging is challenging. This is especially problematic as modalities like optical coherence tomography (OCT) become more prevalent in cardiology. OCT, a light-based intracoronary imaging technique, provides cross-sectional images of plaque and luminal morphology. Until recently segmentation of OCT images for BVS struts was performed manually by experts. However, this process is time consuming and not tractable for large amounts of patient data. Several automated methods exist to segment metallic stents, which do not apply to the newer BVS. Given this current limitation coupled with the emerging popularity of the BVS technology, it is crucial to develop an automated methodology to segment BVS struts in OCT images. The objective of this paper is to develop a novel BVS strut detection method in intracoronary OCT images. First, we preprocess the image to remove imaging artifacts. Then, we use a K-means clustering algorithm to automatically segment the image. Finally, we isolate the stent struts from the rest of the image. The accuracy of the proposed method was evaluated using expert estimations on 658 annotated images acquired from 7 patients at the time of coronary arterial interventions. Our proposed methodology has a positive predictive value of 0.93, a Pearson Correlation coefficient of 0.94, and a F1 score of 0.92. The proposed methodology allows for rapid, accurate, and fully automated segmentation of BVS struts in OCT images.

Polymeric endovascular strut and lumen detection algorithm for intracoronary optical coherence tomography images

“…To segment the image automatically, we use a K-means clustering algorithm with k ¼ 3 clusters; 31 [k ¼ 2, 3, and 4 cluster values were tested, but k ¼ 3 produced the highest positive predictive value (PPV)]. The K-means algorithm is an unsupervised machine learning algorithm that takes as an input a set of N observations fx 1 ; x 2 ; : : : ; x N g and classifies them into k clusters S ¼ fS 1 ; S 2 ; : : : ; S k g to minimize the sum of squares residual error within each cluster.…”