Deep learning algorithms that incorporate retinal fundus photographs may be able to predict high coronary artery calcium levels and, in turn, provide an inexpensive and radiation-free screening method, researchers from South Korea report.

Their study included individuals who underwent same-day bilateral retinal fundus imaging and coronary computed tomography scans that generated coronary artery calcium scores. The algorithm distinguished between high coronary artery calcium levels from those with a score rating of zero. The American Heart Association suggests artery calcium levels greater than 400, with other complications, indicates very high risk of cardiovascular failure, while  levels below 100 suggest low risk and anything in between equates to intermediate risk. Researchers also incorporated vessel and fovea-in-painted images to investigate areas of interest.

A total of 44,184 images from 20,130 individuals were included.

In differentiating between no coronary artery calcium from levels greater than 100, the artificial intelligence (AI) system performed well, achieving an area under receiver operating curve (AUROC) of approximately 82.3% and 83.2%, respectively, when using unilateral and bilateral fundus images.

Of note, the AI’s performance improved as the criteria for high coronary artery calcium was increased, showing a plateau at 100 and losing significant improvement thereafter.

“Interestingly, the performance increased rapidly when the threshold was changed from zero to 100, whereas above 100, the improvements were marginal,” the researchers wrote in their paper. This implies discriminative features related to high coronary artery calcium exist in retinal fundus images, and AI can perceive when the level is above 100, the study noted.

 The results further showed visual patterns that were mainly found in the temporal retinal vasculature and in some parts of the fovea. 

Ideally, a deep learning algorithm for predicting high coronary artery calcium could be run when a patient undergoes automatic retinal fundus photography, because only a few additional computations would be required, the study noted.

“We believe that our experiments are an early beginning in paving the way to implementing an additional screening method for the risk of CVD (cardiovascular disease) in an accessible and affordable way,” the investigators wrote.