AML-Net: A Preliminary Screening Model for Mild Hypertension

Abstract

Hypertension, often known as high blood pressure, is a severe medical condition that is a prevalent risk factor responsible for many medical accidents. During the development of hypertension, there lies a critical window that makes hypertension curable. In order to reverse hypertensive features and to prevent aggravation of hypertension, the critical window to treat and maintain blood pressure happens in pre-hypertension or early hypertension stages when the symptoms are mild. This makes preliminary screening essential to prevent unwanted and unnecessary damages caused by hypertension. Inspired by an ophthalmologist's traditional approach to using fundus images, this paper collects and analyzes fundus images as an imaging modality to identify cases of mild hypertension. In this paper, we propose the Attention-based metric learning network (AML-Net) to discriminate between mild hypertension and healthy images to address preliminary screening. Experiment results stated the proposed end-to-end algorithm is able to achieve a 93.75% accuracy. To the best of our knowledge, this is the first automated approach towards preliminary hypertension screening and the first deep learning-based work on mild hypertensive fundus images.