EyeWeS: Weakly Supervised Pre-Trained Convolutional Neural Networks for Diabetic Retinopathy Detection

Pedro Costa, Teresa Araújo, Guilherme Aresta, Adrian Galdran, Ana Maria Mendonça, Asim Smailagic, Aurélio Campilho

Abstract: Diabetic Retinopathy (DR) is one of the leading causes of preventable blindness in the developed world. With the increasing number of diabetic patients there is a growing need of an automated system for DR detection. We propose EyeWeS, a general methodology that enables the conversion of any pre-trained convolutional neural network into a weakly-supervised model while at the same time achieving an increased performance and efficiency. Via EyeWeS, we are able to design a new family of methods that can not only automatically detect DR in eye fundus images, but also pinpoint the regions of the image that contain lesions, while being trained exclusively with image labels. EyeWeS improved the results of Inception V3 from 94.9% Area Under the Receiver Operating Curve (AUC) to 95.8% AUC while maintaining only approximately 5% of the Inception V3’s number of parameters. The same model is able to achieve 97.1% AUC in a cross-dataset experiment. In the same cross-dataset experiment we also show that EyeWeS Inception V3 is effectively detecting microaneurysms and small hemorrhages as the indication of DR.

Keywords: Diabetic Retinopathy Detection, Weakly-Supervised, Multiple Instance Learning, Image Classification