Automatic Segmentation and Measurement of Vascular Biomarkers in OCT-A Images ^†

Abstract

We propose an automatic methodology that identifies the vascularity zones in OCT-A images and their measurement for its use in clinical analysis and diagnostic processes. The segmentation and measurement contributes objectivity and repeatability in the results, desirable characteristics in any diagnosis and monitoring process. In the validation of the method, the correlation coefficient of Pearson and Jaccard index were used, obtaining satisfactory results.

1. Introduction

Optical Coherence Tomography Angiography (OCT-A) is a new non-invasive imaging modality that facilitates the analysis of the vascularity in the retina. The extraction of this vascular and avascular zones is useful for the analysis of several pathologies such as diabetic retinopathy, but their correct extraction requires objectivity, determinism and repeatability factors. Given the recent appearance of this image modality, there are few works, most of them are clinical proposals that study the repeatability and reproducibility of different biomarkers that are based on the OCT-A vascular properties in healthy patients, indicating the satisfactory impact of this analysis. For this reason, the automatic extraction of this zones is interesting, given the repeatability and objectivity that support its automation.

2. Methodology

We propose an automatic methodology that identifies the vascular and avascular zones in OCT-A images and their measurement for its use in clinical analysis and diagnostic processes [1]. We firstly intensify the vascular characteristics using morphological operators, facilitating the extraction in further steps. Then, a set of image processing techniques are combined to maximize their differences and, posteriorly, estimate their representative parameters, respectively. These biomarkers are based in the area of the Foveal Avascular Zone (FAZ) and the vascular density, features that can vary in healthy and pathological patients. In the case of the vascular density, four different ways of measurement were performed based on: the original image, the enhanced image, the thresholded image and the vascular skeletonization of the image. Figure 1, represents an example of input and outputs system.

3. Results

The proposed methodology was tested on a set of 144 non-pathological images labeled by an expert ophtalmologist, being used as reference in the validation of the method. The correlation coefficient of Pearson and the Jaccard index were used to validate the results between the expert and the system, validating the measurements and the coverage of the zone. In the correlation coefficient of Pearson, the areas of the expert and the areas of the system were compared, obtaining an average of 0.76, which represents a good correlation between both segmentations. With the Jaccard index, we obtained 0.73, also offering satisfactory results. Summarizing, the proposed methodology presented satisfactory results in both validation experiments.