Fluid Region Analysis and Identification via Optical Coherence Tomography Image Samples ^†

Abstract

The work herein proposed presents a methodology which aims to identify cystoid regions using OCT scans. This method obtained satisfactory results detecting cystoid regions with different levels of complexity without needing any preprocessing nor candidate filtering steps.

1. Introduction

Optical Coherence Tomography (OCT) is a non-invasive medical imaging modality that provides morphological information about the retinal tissues. This information is commonly used in the early diagnosis and analysis of patients with potential eye and systemic diseases as, for reference, Diabetic Retinopathy (DR), Diabetic Macular Edema (DME) and Age-related Macular Degeneration (AMD), three of the leading causes of blindness in adults of working age in developed countries. Given the global relevance of this topic, an accurate identification of any present cystoid region is crucial to perform an adequate diagnosis, treatment, prevention and rehabilitation.

2. Methodology

In this work, we propose a novel methodology for the automatic identification and characterization of the intraretinal cystoid fluid regions using OCT images [1]. To achieve this, we analyzed a complete and heterogeneous set of 326 intensity and texture descriptors. The most relevant features were selected using the Relief-F and L0 feature selectors and tested with the Linear Discriminant Classifier, the Quadratic Discriminant Classifier and k Nearest Neighbors with k = 5.

3. Results

The proposed methodology was tested using 51 OCT images obtained with a CIRRUS© HD-OCT confocal laser ophthalmoscope (Carl Zeiss Meditec, Inc., Dublin, California). From these images, a total of 723 samples from cystoid and non-cystoid regions were extracted, using a sample size of 51 × 51. This sample size was empirically determined to be large enough to detect both big and small cystoid bodies in the retinal layers.

The methodology correctly identified the intraretinal cystoid fluid regions with a satisfactory accuracy of 90.6%. As shown in Figure 1, both fluid regions and non-cystoid regions are detected by the system despite the multiple complications.