Korea Biomedical Review | July 15, 2021
The application of deep learning to retinal photographs has yielded promising results in predicting age, sex, blood pressure, and hematological parameters. However, the broader applicability of retinal photograph-based deep learning for predicting other systemic biomarkers and the generalisability of this approach to various populations remains unexplored. With use of 236 257 retinal photographs from seven diverse Asian and European cohorts (two health screening centres in South Korea, the Beijing Eye Study, three cohorts in the Singapore Epidemiology of Eye Diseases study, and the UK Biobank), we evaluated the capacities of 47 deep-learning algorithms to predict 47 systemic biomarkers as outcome variables, including demographic factors (age and sex); body composition measurements; blood pressure; haematological parameters; lipid profiles; biochemical measures; biomarkers related to liver function, thyroid function, kidney function, and inflammation; and diabetes. The standard neural network architecture of VGG16 was adopted for model development.
A joint research team led by Severance Hospital professors has developed an algorithm that predicts the risk of new cardiovascular diseases by observing changes in blood vessels in the retina images, the hospital said in a news release Thursday.
A joint research team led by Severance Hospital has developed an algorithm that predicts the risk of new cardiovascular diseases by observing changes in blood vessels in the retina images.
The artificial intelligence (AI)-based algorithm uses coronary artery calcification index to predict heart diseases, including cardiac infarction or angina, more effectively without exposing patients to radiation.
The development was led by Professors Kim Hyun-chang at Yonsei University College of Medicine, Park Sung-ha and Kim Sung-soo at Severance Hospital, Lee Byung-kwon at Gangnam Severance Hospital, and Rim Hyung-taek at Duke-NUS Medical School in Singapore, and some Korean startups.
The prevalence of dyslipidemia is quite high, that two out of five Korean adults suffer from the disease.
Lipid-lowering therapy is used most often for dyslipidemia patients, and a blood test is necessary to measure the patient’s total cholesterol, low and high-density lipoprotein cholesterol, and triglycerides. If any one of the numbers is of the healthy values, health providers decide to go with lipid-lowering treatment.
When a definitive risk assessment is needed for the moderate-risk patient group, the future occurrence of coronary artery diseases such as myocardial infarction and angina pectoris is predicted based on the coronary artery calcification index through cardiac computed tomography (CT) imaging.
Coronary artery calcification index is the best predictor of cardiovascular disease risk, but it is not easy to receive a cardiac CT scan. In addition, CTs are expensive, and patients residing in countries with lower medical access have a hard time taking the scan.
Severance researchers explained that they had studied a non-invasive cardiovascular risk assessment model that allows doctors to predict without exposing patients to radiation.
The research team used digital retinal images from Severance Hospital and Gangnam Severance Hospital to develop the algorithm to determine the correlation between the coronary artery calcification index and the retina. They determined the presence or absence of coronary artery calcification index by applying deep learning technique of bio-imaging startup Medi Whale to 216,152 retinal images in five data sets from Korea, Singapore, and the U.K.
To verify the risk assessment algorithm developed afterward, the research team used the customized prevention data for the high-risk group of cardiovascular disease collected by Professor Park Sung-ha.
Patients determined to have high risk in the retinal examination and those with an increased risk in the coronary artery calcification index test had the same occurrence of cardiovascular disease and death caused by the disease.
“Retinal images can be easily taken by ophthalmology, so if a diagnostic solution is introduced, ophthalmology will become like a kind of health examination center,” Professor Kim Sung-soo said. “If the cardiology department or other primary care institutions can confirm the algorithm and detect patients with a high risk of cardiovascular disease at an early stage, it will become an essential test tool in the long term.
Professor Rim also said, “Application of deep learning to identify the relationship between the retina and systemic diseases is still in early stages, so we need to approach carefully considering several variables that could affect the study.”
Let’s build the future of digital healthcare together