Pulse by Maeil Business News Korea | September 24, 2020

South Korean startup Medi Whale has developed a new artificial intelligence (AI) technology to help detect not only physical characteristics but also health conditions such as high blood pressure, kidney disease and sarcopenia by using retinal blood vessel images.

Medi Whale’s AI algorithm can analyze retinal images to predict the person’s age, gender, height and weight, and find biomarkers related to specific diseases, the company’s CEO Choi Tae-geun said on Wednesday. Those biomarkers for disease detection include HbA1c for diabetes, creatinine levels for kidney disease, muscle mass for sarcopenia and systolic and diastolic blood pressure for cardio-cerebrovascular disease.

Medi Whale has sophisticated VGG-16, an AI model widely used in deep learning image classification problems, to make it learn more than 200,000 retinal images of healthy persons, patients with disorders and those with hypertension.

Duke-NUS Medical School professor Rim Hyung-taek, co-founder of Medi Whale, said that 236,257 retinal photographs were fed into the deep learning network and those images came from seven diverse Asian and European cohorts (two health screening centers in South Korea, the Beijing Eye Study, three cohorts in the Singapore Epidemiology of Eye Diseases study, and the UK Biobank).

The retina is the only organ that allows direct and non-invasive visualization of the microvasculature and neural tissues, affording a unique opportunity for the non-invasive detection of systemic vascular and neurological diseases, said Kim Sung-soo, an ophthalmologist at Severance Hospital.

The findings were published in the October edition of the Lancet Digital Health journal under the title of Prediction of systemic biomarkers from retinal photographs: development and validation of deep-learning algorithms.

The findings provide new insights into the potential clinical use of retinal photographs to predict systemic biomarkers for health screening with no blood sampling in populations with a similar ethnic background.

By Lee Sae-bom and Minu Kim