Automated segmentation of tissue structures in optical coherence tomography data

Abstract

Segmentation of optical coherence tomography (OCT) images provides useful information, especially in medical imaging applications. Because OCT images are subject to speckle noise, the identification of structures is complicated. Addressing this issue, two methods for the automated segmentation of arbitrary structures in OCT images are proposed. The methods perform a seeded region growing, applying a model-based analysis of OCT A-scans for the seed's acquisition. The segmentation therefore avoids any user-intervention dependency. The first region-growing algorithm uses an adaptive neighborhood homogeneity criterion based on a model of an OCT intensity course in tissue and a model of speckle noise corruption. It can be applied to an unfiltered OCT image. The second performs region growing on a filtered OCT image applying the local median as a measure for homogeneity in the region. Performance is compared through the quantitative evaluation of artificial data, showing the capabilities of both in terms of structures detected and leakage. The proposed methods were tested on real OCT data in different scenarios and showed promising results for their application in OCT imaging.