Transfer learning for Diabetic Macular Edema (DME) detection on Optical Coherence Tomography (OCT) images

Abstract

Diabetic Macular Edema (DME) is a common eye disease that causes irreversible vision loss for diabetic patients, if left untreated. Thus, early diagnosis of DME could help in early treatment and prevent blindness. This paper aims to create a framework based on deep learning for DME recognition on Spectral Domain Optical Coherence Tomography (SD-OCT) images through transfer learning. First, images are pre-processed: denoised using Block-Matching and 3-Dimension (BM3D) filtering and cropped through image boundary extraction. Later, features are extracted using CNN of AlexNet and finally images are classified using SVM classifier. The results are evaluated using 8-fold cross-validation. The experiments show that denoised and cropped images lead to better classification performances, exceeding previous other recent published works of 96% accuracy.