Artificial intelligence will detect heart disease by four selfies

Scientists from Tsinghua University, in collaboration with clinicians from Fuwai Hospital (Beijing, China), have developed and tested a new diagnostic algorithm. Using only four photographs of a person from different angles, artificial intelligence can detect the presence of coronary heart disease. The effectiveness of the algorithm was proven in trials involving about eight thousand people: focusing only on selfies, the program correctly diagnosed 80% of cases. 

Coronary heart disease (CHD) - a disease whose essence is the violation of blood supply to the heart muscle, which leads to myocardial infarction. There are a number of signs, a kind of biomarkers of CHD, which are manifested in human appearance: thinning and fragility of the hair, the fold on the earlobe, yellow, slightly rising above the surface of the skin small spots around the eyelids, cloudy circles around the cornea.

At the same time, people find it difficult to notice these small signs and use them to diagnose heart disease. However, this can be done by an artificial intelligence trained to recognize the smallest features of appearance. In the course of the study, the experts took four photos of the participants' faces, as well as conducted angiography for volunteers - a procedure that allows assessing blood flow and the condition of blood vessels. The information obtained was used to create, train, and test the diagnostic algorithm. 

Tests have shown the effectiveness of the program: four selfies (in full-face, right and left profiles, the view of the head from above), the algorithm was able to detect disease in 80% of cases. At the same time, additional information related to various risk factors of disease development had practically no effect on the accuracy of diagnosis. This suggests that it was the analysis of photographs that allowed AI to diagnose the disease with high accuracy. At the same time, when tested on healthy people, the program gave a false positive result in 46% of cases. 

- The algorithm received the most useful information in the analysis of cheeks, nose, and forehead, - explained the head of the study, Professor Zhe Zheng. - However, we need to increase the specificity of the program, as 46% of "false positives" will increase anxiety in patients and overload the clinics with unnecessary analyses. 

Aug. 26, 2020, 10:22 a.m.

Similar articles:

Artificial intelligence trained to detect the DNA of cancer cells in the blood

The genetic material of lung cancer cells can be found ...

The symbiosis of EKG and Artificial Intelligence has taken diagnostics to the next level

Electrocardiogram (ECG) is one of the most known and affordable ...

Artificial intelligence detects 134 skin diseases based on photos on a smartphone

Scientists from Seoul National University (South Korea) have taught artificial ...

AI has taught people how to treat chronic pain

Scientists from the ATR Brain Information Research Laboratory (Japan, USA ...

Neural networks have learned to recognize 250 words uttered mentally

Scientists at the University of California have taken another step ...

AI is taught to imagine, and it outperforms doctors in diagnosis

Experts from Babylon Health, a British service specializing in online ...

Smartphone diagnoses diabetes 82% accurately

Diabetes mellitus is the seventh most common cause of death ...

The new app will help doctors avoid professional burnout and teach mindfulness

As coronavirus cases continue to rise in the world, health ...

COOKIE

Our site collects information using cookies to be more convenient and customized to your needs interests. The purposes of the use of cookies are defined in Policy the processing of personal data .If you agree to continue to receive cookies, please click the "Accept" button. If you don't agree or want to resolve this issue later, please change your browser cookie settings.