Computational Diagnosis of Diabetic Macular Edema using Deep Learning Techniques on Optical Coherence Tomography

In the developed world, diabetic retinopathy is the leading cause of blindness among the working age population. Of an estimated 425 million people with diabetes worldwide, nearly 10% are afflicted with a vision-threatening disease, Diabetic Macular Edema (DME) being the leading etiology. Optical Coherence Tomography (OCT) is essential in the diagnosis and management of DME. Currently, OCT-derived measures are the standard of care in the diagnosis of DME and in monitoring of therapeutic effects, with thickness measurements of the macula serving as the primary clinical measure. However, the morphology of DME extends to additional characteristics such as the presence of intraretinal fluid and retinal layers disorganization. The objective of the research is to promote the clinical care of DME, by providing computational-based quantitative and qualitative analysis of OCT. We present a deep learning method developed for 2D OCT images. The proposed method will thereafter be extended to OCT volumetric data and subsequently, to volumetric OCT data longitudinally (4D data) to track the disease course and to assess for possible prognostic factors. Two convolutional neural networks were implemented for 2D OCT analysis. The first is for the classification of OCT scans of normal versus DME and the second is an encoder-decoder architecture that segments anomalies present on 2D OCT scans. We achieved high accuracy results on classification and segmentation of 2D OCT scans. These 2D results encourage the extension of the analysis to volumetric and longitudinal data.