Multi-scale deep feature fusion for automated classification of macular pathologies from OCT images

摘要

Identification of the macular pathologies at an early stage can prevent vision loss. Similarity in the pathological manifestations of common macular disorders like age related macular degeneration (AMD) and diabetic macular edema (DME) can make manual screening fallible. There is a growing interest among researchers for reliable automated detection of macular pathologies using computer methods. Therefore, in this paper we present a novel method for classification of DME and two stages of AMD namely the drusens (early stage) and the choroidal neo vascularization (CNV) (late stage) from healthy optical coherence tomography (OCT) images. The proposed method introduces a multi-scale deep feature fusion (MDFF) based classification approach using convolutional neural network (CNN) for reliable diagnosis. The MDFF captures the inter-scale variations in images to introduce discriminative and complementary information to the classifier. The proposed method is evaluated on an OCT dataset containing 84,484 images with different class distributions. The imbalance in the dataset is handled by introducing the cost sensitive loss function during the learning of the classifier. The proposed method achieves an average sensitivity, specificity and accuracy of 99.6%, 99.87% and 99.6% on the test set. The promising classification results make the proposed method highly suitable for preliminary automated diagnosis of macular pathologies in health care centres and eye clinics. (C) 2019 Elsevier Ltd. All rights reserved.