The new tools by the Singapore Eye Research Institute use artificial intelligence-based deep learning algorithms to scan photos of patients’ retina.

Scientists from the Singapore Eye Research Institute (SERI) have developed two novel screening tools for detecting chronic kidney disease (CKD) and predicting one’s biological age. These tools, dubbed RetiKid and RetiAge respectively, use artificial intelligence-based deep learning algorithms to scan photos of patients’ retina – the light-sensitive back part of the eye. The new innovations offer a non-invasive and effective approach to screening, which can alert patients to systemic health concerns that do not always present with obvious symptoms in early stages.

 

Chronic kidney disease screening

RetiKid complements existing chronic kidney disease screening strategies. Currently, to screen for CKD, a patient has to undergo a blood test to measure the creatinine level (a type of waste product made by muscles) and urine test to check for albumin. Through RetiKid’s non-invasive approach, individuals with a fear of needles may find the screening to be more bearable, potentially leading to a higher take-up rate and preliminary screening compliance.

“Drawing blood might not be practical in all settings. Since retinal images are acquired non-invasively, RetiKid can be applied as an effective first-level case finding tool for detecting CKD in general populations and high-risk groups such as those with diabetes, complementing existing CKD screening strategies,” explained Associate Professor Charumathi Sabanayagam, deputy head, Ocular Epidemiology Research Group, SERI, and principal investigator of the RetiKid study.

“The retina and kidneys share a close biological relationship. While blood vessels in the kidneys cannot be examined readily, blood vessels in the retina can be visualised directly using digital retinal photography. Thus, problems with blood vessels in the retina could provide clues to changes in kidney blood vessels in people with or without diabetes,” she added.

By introducing RetiKid as a preliminary screening test, patients can be counselled at the same time to proceed with the routine confirmatory tests with blood and urine. Those who test negative by RetiKid can be scheduled for a follow-up screening again at an interval appropriate for their health condition and future risk.

Manpower for CKD screening can also be further optimised as RetiKid automates the screening process, enabling a larger number of at-risk patients to be screened more effectively. It will also allow screening to be done in a few minutes at any healthcare site or community site in the future, therefore increasing the scope of kidney screening to beyond the usual blood testing sites.

 

Predicting biological age quickly

 Compared to chronological age, measures of biological age may better capture individuals’ physiological changes associated with the ageing process. Hence, biological age can be used to assess the general health status of individuals.

The current standard of measuring biological age involves examining the DNA of a patient, which can be invasive, laboratory-intensive, and takes at least a day. RetiAge provides a novel, relatively non-invasive and time-saving tool, which can be easily adopted in clinics. Once a person is identified to have “older” retinas, they can take steps to improve their lifestyle habits and behaviours, such as physical activity, proper nutrition, and avoidance of smoking and other health-harming behaviour.

“The blood vessels in the retina are also indicative of the ageing process and overall health of one’s circulatory system and even the brain. By using digital technology on retinal images, we can predict a person’s biological age, and in turn, their risk of systemic diseases and lifespan,” said Professor Cheng Ching-Yu, head of the Ocular Epidemiology Research Group and Data Science Research Platform at SERI, and principal investigator of the RetiAge study.

He added: “Biological age and its biomarkers are of greater interest than chronological age to researchers, especially with a global ageing population and rising incidences of chronic diseases. The retina is a non-invasive window into one’s biological age and systemic health status, and can tell us many things about a person’s morbidity and mortality risks.”

Developed by SERI and the National University of Singapore’s School of Computing in 2019, RetiKid was tested with over 23,000 retinal images from close to 12,000 study participants from Singapore and China. Results of the study were published in The Lancet Digital Health in May 2020. The study revealed that the RetiKid algorithm identified CKD with 91 percent accuracy in internal test, and with 73 percent and 83 percent accuracy in two external test sets.

The RetiAge algorithm, developed by SERI and South Korean healthcare start-up Medi Whale in 2021, was trained using more than 129,000 retina photos from over 40,000 participants from South Korea to predict the probability of a person having an “older” retina. The researchers then further evaluated RetiAge’s ability to predict a person’s 10-year risk of systemic disease and death, among some 56,000 participants in the UK Biobank, a large-scale biomedical databased and research resource.

Results of the study showed that compared to people with the “youngest” retinas (the 1st quartile), those with “oldest” retinas (the 4th quartile) had double the risk of 10-year all-cause mortality, triple the risk of cardiovascular disease mortality and 1.7 times higher risk of cancer mortality, even though the two groups of people have the same chronological age.

Both RetiKid and RetiAge can potentially be integrated with the Singapore Eye Lesion Analyser Plus (SELENA+) – a retinal image-based deep learning system also developed by SERI and licensed to EyRIS – which is currently available at all polyclinics for patients to screen for diabetic eye diseases, glaucoma and age-related macular degeneration. This enables patients to be screened for more diseases with one image.

As imaging, cloud computing and mobile technologies advance, RetiKid also has the potential to be integrated into smartphones in the future. A point-of-care diagnosis can be given using a cloud-based RetiKid algorithm, from which healthcare providers can receive reports instantly.

“Tools like RetiKid have potential to be widely used in primary care to improve the current rate of CKD screening. Individuals who screen positive may be more convinced to undergo further assessments such as blood and urine tests, which is the current gold standard, to confirm the diagnosis of CKD. Timely detection of CKD allows clinicians and patients the opportunity to intervene early and slow down CKD progression,” said Dr Cynthia Lim, senior consultant, Department of Renal Medicine, Singapore General Hospital, who is also a key investigator on the RetiKid project.

Researchers from the RetiAge project are also currently working on refining the algorithm to optimise its prediction performance in the local population. They are also studying if RetiAge can be used to predict other age-related diseases, in order to allow people to have a better understanding of their overall health status.

As part of the validation phase of RetiKid, SERI and The National Kidney Foundation (NKF) have partnered for a community outreach programme to recruit 1,200 participants from February 2022 to January 2024. Eligible participants are those at high risk of developing CKD; family members of patients who have been diagnosed with CKD or kidney failure; patients diagnosed with diabetes and hypertension; and Malay adults. Participants will be screened for CKD using blood and urine tests, and additionally using RetiKid. Interested parties can contact 1800-543-6397, or e-mail – ckd.clinic@nkfs.org.

 

 

(** PHOTO CREDIT: Singapore Eye Research Institute)