A landmark study shows that AI-powered eye scans can play an effective role in identifying risk factors for cardiovascular disease, including within primary care. The research shows that the rapid, non-invasive retinal scan can be easily integrated into general practices to identify early risks of heart attack and stroke.
“The simplicity of using the retinal camera and the high acceptance by both doctors and patients prove that this technology can be incorporated into GPs' daily practice, enabling screenings prior to a GP visit,” said Wenyi Hu, lead author of the study. Wu stressed that much work remains to be done, especially in terms of the technology's accuracy with specific patient groups, such as men over 60. The study was recently published in npj Digital Medicine,
Design and methodology
The study was conducted by Hu as part of her doctoral research at the University of Melbourne, in collaboration with CERA's Ophthalmic Epidemiology team. A total of 361 participants, aged 45 to 70 years, participated. All participants were patients at two general practices and had recently undergone a cardiovascular risk assessment, such as a blood pressure or cholesterol test.
Each participant underwent an eye scan with a retinal camera that mapped blood vessels at the back of the eye. AI technology then generated a real-time report on the patient's cardiovascular risk profile. In addition, an assessment was performed using the World Health Organization's (WHO) CVD risk chart, which measures various factors such as age, gender, smoking habits, blood pressure and cholesterol.
Results and findings
The results of the retinal scans were compared with the WHO risk scores, and the correlation between the two methods was thoroughly analyzed. The outcomes were also validated using data from the UK Biobank, with the result that the retinal scan showed similar accuracy in predicting the risk of heart attack or stroke within 10 years. Key findings of the study:
- The retinal scan showed moderate correlation with WHO risk scores: 67.4% of the results matched, 17.1% of the scans overestimated the risk, while 19.5% underestimated the risk.
- The predictive value of the retinal scan was similar to that of the WHO method for estimating the 10-year risk of coronary heart disease or stroke, compared with UK Biobank data.
- The success rate for obtaining usable images was 93.9%, indicating the reliability of the technology in clinical practice.
- Both 92.5% of patients and 87.5% of GPs expressed satisfaction with the technology.
Future use in general practices
Dr. Malcolm Clark, family physician and co-author of the study, highlights the potential of retinal scans for increasing cardiovascular risk assessment in general practice. According to Clark, the technology could be a valuable tool in the future for early identification of patients who may need additional testing.
“In the future, I envision patients automatically receiving a reminder via SMS to have an eye scan, which then sends a risk report to their physician. This could be incorporated into regular health check-ups, such as cervical or colon cancer screenings,” Dr. Clark said.
The future of AI in healthcare
Associate Professor Lisa Zhuoting Zhu, promoter of the research, sees the integration of AI-driven eye scans as an important step toward improving public health care. “We are working toward a future where cheaper, scalable and accessible cardiovascular screening is possible for everyone, including people in remote or vulnerable communities,” Zhu said.
She adds that AI-driven eye checks can provide valuable insights not only into cardiovascular health, but also brain and kidney health. This makes the technology a potential pillar in routine preventive healthcare beyond traditional screening methods.
Diagnostics in the eye of the beholder
Researchers at Sharjah's Skoltech University and the AIRI Institute recently developed an AI toolto automate the analysis of retinal images for diagnosing diabetic retinopathy. This innovation accelerates the diagnostic process and enhances accuracy. In initial tests using the DRIVE database, the tool achieved a 97% accuracy rate and an 84% sensitivity rate, effectively identifying microvessels—a task challenging for previous models. Despite limitations in training data size, the researchers improved performance through data augmentation and adaptive threshold algorithms, indicating the tool's potential for clinical application.
Last year, researchers evaluated deep-learning AI models for diagnosing infectious keratitis (IK), a corneal inflammation leading to blindness, particularly in regions with limited access to specialized eye care. Analyzing over 136,000 corneal images across 35 studies, the AI models demonstrated diagnostic accuracy comparable to, and sometimes exceeding, that of ophthalmologists, with a sensitivity of 89.2% and specificity of 93.2%. The models effectively distinguished between healthy eyes, infected corneas, and various IK causes. Further validation with diverse data is needed to enhance their clinical reliability.
