On Neural-Network-Based Detection for Hypertensive Subjects Using Classification of Retinal Fundus Photographs

Abstract

In this paper, a method of detecting hypertensive subjects is proposed, using retinal fundus photographs classified by convolutional neural networks (CNNs for short). The proposed method employs Inception-v3 model as a CNN. The data to be presented to the proposed model are prepared from retinal fundus photographs. The scheme of fine tuning is conducted to construct a discrimination model. The model specifies the photographs corresponding to the subjects being in hypertensive states. Experimental results establish that the proposed method can achieves favorable metric values. Especially, Accuracy value achieved by the proposed method is comparatively high for the small scale of dataset, compared with the previously proposed method.