Skip to navigation Skip to main content Skip to footer

Approved Research

Weakly-supervised Deep Learning for the Improvement of Cardiovascular Disease (CVD) Prediction using Retinal Fundus Images

Principal Investigator: Professor Patrick Then
Approved Research ID: 67263
Approval date: April 20th 2021

Lay summary

The purpose of this research project is to explore new methods of predicting Cardiovascular Disease (CVD) from retinal images and patients' health records using Deep Learning algorithms. Over the past decades, there is increasing recognition that conventional CVD risk-scoring tools such as Framingham risk score (FRS), the Systematic COronary Risk Evaluation (SCORE) and the QRISK; which uses traditional CVD risk factors, such as age, gender, blood pressure, smoking status; cannot accurately predict the risk of CVD. This is because these traditional risk factors do not fully reflect the changes in the human circulation system. Since the retina is the only part of the human circulation that can be imaged directly and non-invasively, retinal images have the potential to replace invasive tests such as glycated hemoglobin (HbA1c) and cholesterol profile in the prediction of CVD risks.

The project is estimated to take around 2 years, in which the evaluation of its performance will be mainly based on the added value over conventional CVD risk-scoring tools, such as FRS, SCORE and QRISK. The main benefit of this research is the translational value of deep learning in the prediction of CVD risk, which subsequently will enable the implementation of tele-retinal screening to predict CVD risks in underserved rural areas through remote medicine.