Currently, the algorithm is still being developed, and it needs to be tested in larger groups of people from different ethnic backgrounds, but according to the researchers this algorithm has the potential to be used as a screening tool that could help to identify possible heart disease in the general population as well as in those who are in high-risk groups, who could be referred to follow up with further clinical investigations.
“To our knowledge, this is the first work demonstrating that artificial intelligence can be used to analyze faces to detect heart disease. It is a step towards the development of a deep learning-based tool that could be used to assess the risk of heart disease, either in outpatient clinics or by means of patients taking ‘selfies’ to perform their own screening. This could guide further diagnostic testing or a clinical visit,” said Professor Zhe Zheng, who led the research and is vice director of the National Center for Cardiovascular Diseases and vice president of Fuwai Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, People’s Republic of China.
“Our ultimate goal is to develop a self-reported application for high risk communities to assess heart disease risk in advance of visiting a clinic. This could be a cheap, simple and effective of identifying patients who need further investigation. However, the algorithm requires further refinement and external validation in other populations and ethnicities,” said Zheng.
While it is difficult for humans to use facial features and successfully to predict and quantify the risk of heart disease, certain facial features have already been associated with an increased risk of heart disease, this includes but is not limited to ear lobe crease, wrinkles, thinning or gray hair, xanthelasmata, and arcus corneae.
5,796 patients from 8 hospitals in China undergoing imaging procedures to investigate their blood vessels were enrolled in this study who were divided at random into groups: 5,216 were in the training group and 580 were put in the validation group. 4 facial photos were taken with a digital camera by trained research nurses: one frontal, 2 profile, and one view of the top of the head. Participants were interviewed to collect data on medical history, lifestyle, and socioeconomic status, and radiologists reviewed angiograms to assess the degree of heart disease depending on how many blood vessels were narrowed by 50% or more, and their location, all to create, train, and validate the deep learning algorithm.
The algorithm was tested on a further 1,013 patients from 9 hospitals in China, the majority of whom in all groups were of Han Chinese ethnicity. The algorithm was found to outperform existing methods of predicting the risk of heart disease. Within the validation group, the algorithm correctly detected heart disease in 80% of cases and correctly detected that heart disease was not present in 61% of cases. In the test group, the algorithm has a sensitivity of 80% and a specificity of 54%.
“The algorithm had a moderate performance, and additional clinical information did not improve its performance, which means it could be used easily to predict potential heart disease based on facial photos alone. The cheek, forehead and nose contributed more information to the algorithm than other facial areas. However, we need to improve the specificity as a false positive rate of as much as 46% may cause anxiety and inconvenience to patients, as well as potentially overloading clinics with patients requiring unnecessary tests.”
However, this study was limited by the low specificity of the test, the fact that only one center in the test group was different to those centers which provided the participants for developing the algorithm, and the algorithm needs to be trained and tested in other ethnic groups, which further limits the algorithm’s generalisability to other populations.
“Overall, the study by Lin et al. highlights a new potential in medical diagnostics……The robustness of the approach of Lin et al. lies in the fact that their deep learning algorithm requires simply a facial image as the sole data input, rendering it highly and easily applicable at large scale,” writes Charalambos Antoniades, professor of cardiovascular medicine at the University of Oxford, UK, and Dr. Christos Kotanidis, a DPhil student working under Prof. Antoniades at Oxford in an editorial.
“Using selfies as a screening method can enable a simple yet efficient way to filter the general population towards more comprehensive clinical evaluation. Such an approach can also be highly relevant to regions of the globe that are underfunded and have weak screening programs for cardiovascular disease. A selection process that can be done as easily as taking a selfie will allow for a stratified flow of people that are fed into healthcare systems for first-line diagnostic testing with CCTA. Indeed, the ‘high risk’ individuals could have a CCTA, which would allow reliable risk stratification with the use of the new, AI-powered methodologies for CCTA image analysis.“
The editorial also says that this technology also raises some ethical questions about the “misuse of information for discriminatory purposes. Unwanted dissemination of sensitive health record data, that can easily be extracted from a facial photo, renders technologies such as that discussed here a significant threat to personal data protection, potentially affecting insurance options. Such fears have already been expressed over misuse of genetic data, and should be extensively revisited regarding the use of AI in medicine.”
The authors of the report also agree on this point noting that, “Ethical issues in developing and applying these novel technologies is of key importance. We believe that future research on clinical tools should pay attention to the privacy, insurance and other social implications to ensure that the tool is used only for medical purposes.“
Prof. Antoniades and Dr. Kotanidis also write in their editorial that defining CAD as ≥ 50% stenosis in one major coronary artery “may be a simplistic and rather crude classification as it pools in the non-CAD group individuals that are truly healthy, but also people who have already developed the disease but are still at early stages (which might explain the low specificity observed)“.
