Using retinal imaging, a newly developed ‘deep learning’ algorithm has demonstrated considerable accuracy in identifying patients who may be at risk for major cardiac events.
Researchers from Google and Verily Life Sciences developed the deep learning model to test whether a number of cardiovascular risk factors can be identified using only retinal fundus images. “Because blood vessels can be non-invasively visualized from retinal fundus images, various features in the retina, such as vessel caliber, bifurcation or tortuosity, microvascular changes and vascular fractal dimensions, may reflect the systemic health of the cardiovascular system as well as future risk,” the authors write.
Data from over 280,000 patients was used to create the models; images from over 13,000 of these patients were used to validate them. Due to a lack of established baselines for predicting cardiac features from retinal images, the researchers used the average value as the baseline for continuous predictions.
Results showed that the deep learning model was able to predict age ±5 years with 78% accuracy, systolic blood pressure ±15mmHg with 72% accuracy, and BMI ±5 with 80% accuracy. With regard to major adverse cardiovascular events (MACE) within 5 years, the area under the receiver operating characteristic curve (AUC) was found to be 0.70 (95% CI: 0.648-0.740) with the model compared to 0.72 (95% CI: 0.67-0.76) with the composite European SCORE risk calculator.
The researchers conclude that their results “indicate that the application of deep learning to retinal fundus images alone can be used to predict multiple cardiovascular risk factors.” The study identifies retinal imaging as an avenue for better understanding cardiovascular risk stratification, however, the authors note, more research is needed with much larger sample sizes.
For more information visit Nature.com.