Retinal images in a healthy subject and those with diabetic retinopathy and stroke. An automated system under development in Hong Kong analyzes these images to predict stroke risk and speed the diagnosis of retinopathy.
An automatic retinal image analysis system can identify diabetes patients at risk of stroke, researchers at the Chinese University of Hong Kong report.
Standard retinal images are transmitted from the fundus camera via the Internet, uploaded into the system and read pixel-by-pixel using advanced biostatistics to process measurements on exudates, hemorrhages, new vessels and overall retinopathy.
“The retina is a window into cerebrovascular vessel health,” Dr. Benny Chung-ying Zee (PhD), a biostatistician and professor in the school of public health and primary care, told the Medical Post in an interview.
In a retrospective, case-control study, Dr. Zee and his colleagues evaluated 244 patients (156 males, 88 females, mean age 65 years), half of whom had suffered a stroke. Automated analysis of retinal images correctly identified 110 stroke patients (90%) and 104 controls (85%). The study, announced at a press conference, has not been published yet.
“There is room to improve,” Dr. Zee noted, adding that the quality of the photo can be enhanced and “we are continually improving the algorithm.”
He anticipated the system could be commercially available within three years once results are in from a three-year prospective study correlating computer-assessed retinal imaging with MRI findings in predicting stroke and dementia in a cohort of 1,200 seniors.
Initially, however, the system will be introduced into outpatient settings to automate the diagnosis of diabetic retinopathy. “At the moment, diabetes patients undergo an annual eye assessment and the process is manual. . . . Our system speeds up the process, reducing costs and waiting times,” he explained.
“Rural Canada could benefit from this technology that is cost-effective and makes use of the Internet,” said Dr. Zee, who was previously with the department of community health and epidemiology at Queen’s University and the National Cancer Institute of Canada Clinical Trials Group in Kingston, Ont.
Although there are other “projects out there with a similar direction . . . our methodology and technology are more advanced,” Dr. Zee said. “Our system is fully automated and our patented methodology uses advanced biostatistics targeted to solve the problems that others haven’t been able to solve for years.” M
