Just send a selfie to the cardiologist and receive a heart disease risk assessment. Sounds futuristic? It could actually become routine, according to a recent study.
The study shows for the first time that it is possible to have a self-learning artificial intelligence (AI) to assess selfies for clues on whether people have underlying heart disease. Although the AI needs further adjustments to be able to assess people of different ethnicities with certainty, the results on a small group of test people are very promising, according to the study authors.
It is conceivable, for example, that the algorithm could be used in the future to identify patients or risk groups and then use them in further clinical trials.
It has long been known that cardiovascular risk is "literally written on your face". For example, early greying of the hair, the formation of wrinkles, xanthelasma, or an arcus cornea are indications of disorders in the fat metabolism and the associated risks for the heart.
To participate in the study, nearly 5,800 test persons were thoroughly examined for signs of cardiovascular diseases. In addition, imaging of the vascular area was carried out. Trained nurses then took four facial photos of each subject. This data was then used to train the AI.
The AI was finally tested on photos of 1,000 other subjects. The algorithm detected patients with heart disease correctly in 80% of cases (= sensitivity). People without heart disease were detected with only 54% (= specificity). The false-positive rate was therefore 46%. This is too high to use the test responsibly at the moment. Overdiagnosis and subjecting many patients to undergo unnecessary tests do not yet justify a possible benefit of this AI test.
Lin S et al., Feasibility of using deep learning to detect coronary artery disease based on facial photo. European Heart Journal 2020; DOI:10.1093/eurheartj/ehaa640