Automated retinal edema detection from fundus and optical coherence tomography scans

Abstract

Retinal edema is a retinal condition prominently found in diabetes patients, which is caused due to unusual deposit of cystoid fluid in the surrounding macula region. The risk of central vision loss and blindness greatly increases if this syndrome is not treated in time. Furthermore, the progression of the disease can also lead to blindness. The two widely used eye examination practices these days are Optical Coherence Tomography (OCT) and fundus photography. Both of these eye examination practices are non-invasive and can provide ophthalmologists with an early symptom of retinal or Macular Edema (ME). In literature, many researchers proposed automated algorithms for detecting ME from fundus or OCT scans. However, as per best of our knowledge, no automated system exists that incorporates both fundus and OCT images simultaneously for the diagnosis of ME. Therefore, in this research we made an effort towards devising a fully automated method to classify macular edema using both retina imaging modalities (fundus and OCT). The proposed system is based on first extracting the retina layer thickness and later segmenting the cystic spaces from the input OCT and fundus images. Then a 5D feature vector is formed from the extracted profiles which is passed to the supervised discriminant analysis (DA) classifier. We used 71 OCT and 71 fundus scans of 60 patients in this research out of which 15 patients were suffering from ME and 45 were healthy. Our proposed algorithm accurately detected 100% of ME cases and 93.33% of healthy subjects. Keywords: Retinal imaging, Macula Edema (ME), Machine learning, Optical Coherence Tomography (OCT), Fundus photography