Structure Tensor Graph Searches Based Fully Automated Grading and 3D Profiling of Maculopathy From Retinal OCT Images

Abstract

Maculopathy is a collective group of diseases that damages the central region of a retina known as macula. The major two forms of maculopathy are macular edema (ME) and central serous chorioretinopathy (CSCR). Different researchers have worked on the identification of these macular disorders using optical coherence tomography (OCT) images. However, to the best of our knowledge, no research is reported until now that can automatically extract retinal information for the grading of ME and CSCR as per clinically significant ME (CSME), non-clinically significant ME (non-CSME), Type-I CSCR, and Type-II CSCR clinical standards from OCT images. Therefore, this paper presents a novel structure tensor graph searches (ST-GS)-based segmentation framework that combines a structure tensor and graph theory to automatically extract retinal and choroidal layers along with fluid pores followed by automated reconstruction of 3-D retinal surfaces. The ST-GS can extract retinal information even from highly degraded OCT scans. Furthermore, the proposed system automatically grades ME and CSCR pathologies as CSME and non-CSME and Type-I CSCR and Type-II CSCR, respectively. The proposed system extracts seven distinct features, and it is trained on 30 (10 healthy, 10 ME, and 10 CSCR) labeled OCT volumes containing 3840 brightness scans (B-scans). After that, 90 (30 healthy, 30 CSCR, and 30 ME) unlabeled OCT volumes containing 11520 B-scans were used for testing the proposed system, where it correctly classified 88 out of 90 cases with the sensitivity, specificity, and accuracy ratings of 96.77%, 100%, and 97.78%, respectively. Furthermore, the proposed system has achieved a mean dice coefficient of 0.875±0.0342 and 0.92±0.0258 for extracting cyst and serous fluids, respectively.