Google has been tinkering in the field of medicine over the last few years, including developing a prototype electronic contact lens. The company’s latest health project involves detecting cardiovascular conditions by analyzing the vasculature of the retina. The researchers built a deep learning system that processed data from two datasets containing thousands of patients, each of which included images of a patient’s retina along with various risk factors and health conditions such smoking and high blood pressure. The system found correlations between various parameters measured within the retinal images and cardiovascular risk factors, as well as disease.

For example, the software was able to identify smokers just by looking at the retina 71% of the time. The software was also able to predict a patient’s systolic blood pressure on average within 11 mmHg, and this includes individuals with and without hypertension. Additionally, it was able to quite accurately guess the person’s age and gender.

In terms of clinical importance, Google’s software showed a pretty good ability (70%) to predict whether a cardiovascular event, such as a heart attack or stroke, would occur within five years after the retinal exam. The company claims that the algorithm is similar in its abilities at prediction to commonly used risk calculators that take into account measurements (cholesterol) obtained from blood samples.

Interestingly, the researchers that developed the software don’t know quite how it works, as it learned what to look for on its own. To better understand what it found, and to assuage skeptical doctors, the research team analyzed the system’s functionality and, for example, noted that it pays a lot of attention to blood vessels in the retina when evaluating one’s blood pressure.

Google says that it’ll need larger datasets containing a lot more cardiovascular events before it can have more confidence in the clinical utility of the software.

Larger datasets with more cardiovascular events would help improve the algorithm, as well as provide a more robust understanding of how it works and how it should be used in practice.