Look at the “face” to recognize the “disease”

In ancient times, Bian Que, a divine doctor, could know Cai Huangong’s disease status by “looking at the color and listening to the sound”, and now with the help of advanced artificial intelligence technology, he can also see the “face” and recognize the “disease” without asking about the medical history or examining the body.

Recently, Liu Xuenan, a doctoral student at the School of Computer and Information Science of Hefei University of Technology, and his collaborators proposed a new method for non-contact atrial fibrillation screening with facial video.

“People just open the phone app and record a 20-second video of their face in front of the screen. The software then feeds back parameters such as pulse wave, heart rate, and atrial fibrillation risk to the user. Liu Xuenan introduced.

The results were published in the IEEE Journal of Biomedical and Health Informatics, a well-known biomedical journal. The reviewers commented on the work, “This is an interesting new method for screening for atrial fibrillation, for which the authors conducted a comprehensive and rigorous experimental validation.” The study makes a significant contribution to home atrial fibrillation monitoring.”

Video atrial fibrillation screening and signal-level anti-interference model Photo courtesy of interviewee

The “invisible killer” hidden in the heart

Atrial fibrillation is the most common arrhythmic disease, it can cause a series of complications, multiplying the risk of stroke, heart failure and other cardiovascular diseases, known as the “invisible killer”.

According to statistics, the number of patients with atrial fibrillation in China has reached 20 million, and with the further deepening of the aging of the population, the prevalence of the elderly population will continue to rise. The prevalence of young people has also increased in recent years.

However, atrial fibrillation is difficult to diagnose. A recent study showed that about one-third of people with atrial fibrillation are unaware of their condition.

ECG is one of the gold standards for clinical diagnosis of cardiovascular diseases such as atrial fibrillation. However, ECG relies on professional physicians, manual testing is inefficient, and it is difficult to apply to atrial fibrillation screening in large populations.

Wearable devices represented by smart watches and wristbands provide a feasible way for atrial fibrillation screening. “However, as the focus of screening, the elderly group has difficulties in using smart wearable devices, coupled with the obstacles formed by the purchase cost to low-income groups, resulting in the low penetration rate of smart watches in the elderly population, and the application of atrial fibrillation screening has limitations.” Liu Xuenan said frankly.

With the rapid development of artificial intelligence and computer vision technology, photoplethysmography (VPPG) based on facial video came into being. However, this technology is too sensitive to various types of motion interference in the real-life environment, which has become a bottleneck for its practical application. “For example, speaking, shaking the head, changing expressions, etc., will affect the test results of VPPG technology.”

To this end, Liu Xuenan proposed two anti-interference screening models of atrial fibrillation, VidAF and PFDNet, to solve the problem of motor interference.

Video atrial fibrillation screening and feature-level anti-interference model Photo courtesy of interviewee

Artificial intelligence analyzes atrial fibrillation risk

So, what is the principle of condensing a test that can only be done in a hospital into a video?

Liu Xuenan explains, “The color of our facial skin changes with the pulse that is invisible to the naked eye. This change in skin tone can be captured with a normal camera, enabling non-contact detection of pulse waves. According to the unique rhythm and strength of atrial fibrillation patients’ short pulse, combined with artificial intelligence, the risk of atrial fibrillation can be assessed. ”

At present, the research on VPPG technology at home and abroad mostly stays at the level of pulse wave detection. In this study, Liu Xuenan was the first to link this technology to disease screening.

“The difficulty in studying this is that the skin tone changes caused by the pulse in the video are very weak, which makes it difficult to extract the pulse signal. In addition, the diversity of pulse signals in different subjects and the complexity of motion disturbances in the actual environment further complicate atrial fibrillation screening.” Liu Xuenan said.

To this end, he studied the behavior of weak pulse signals in different facial regions, chromaticity space, signal phases and frequency bands, and “tailored” a set of extraction methods for pulse signals, namely the VidAF model, which only “favors” pulse signals.

On this basis, Liu Xuenan collected a large number of data of atrial fibrillation patients in the First Affiliated Hospital of University of Science and Technology of China (Anhui Provincial Hospital) in the past three years, and trained a stable atrial fibrillation screening model, namely PFDNet model, by comparing and studying the similar characteristics of pulse signals of different patients and the common attributes of various motion interferences.

In the experiment, Liu Xuenan tested 1200 sample datasets of 100 atrial fibrillation patients and 100 non-atrial fibrillation subjects. The sensitivity (measuring the detection rate of the model for patients with atrial fibrillation) and specificity (measuring the non-false alarm rate of the model for normal people) were above 0.950 when the subject’s face remained stationary, and the Kappa coefficient (measuring the consistency of the model’s detection results and true results) was 0.931; when the subject’s facial movements occurred, the sensitivity of the model was 0.975, the specificity was 0.900, and the Kappa coefficient was 0.875.

The experimental results show that the two models show significant robustness to motion interference, and the detection results of atrial fibrillation are basically consistent with the clinical diagnosis results.

Liu Xuenan introduced that the two models can be combined with cameras to provide special atrial fibrillation detection instruments, and can also be installed in various devices such as mobile phones, personal computers and home monitoring, so as to achieve atrial fibrillation screening in a low-cost, flexible and diverse way, which is more suitable for the elderly with high incidence of atrial fibrillation.

Establish a video contactless heart health assessment system

With the improvement of people’s health awareness, the popularization of electronic devices such as smartphones and computers, and the development of information technology such as artificial intelligence and big data, the era of digital health has arrived.

TCM emphasizes “working to cure diseases before they occur”, that is, smart doctors want to prevent the occurrence of diseases. Liu Xuenan said, “Inspired by this idea, I plan to combine the four diagnoses of traditional Chinese medicine with modern digital health technology to develop a video contactless heart health assessment system for the home environment.” ”

“Looking” is to observe the patient’s complexion, tongue, expression and other characteristics; “Smell” is to listen to the patient’s voice, cough, wheezing and other characteristics; “Q&A” is to interact with the patient’s Q&A to collect information such as symptoms and disease history; “Cutting” is to judge the function of the body by taking the pulse. At present, all four types of information can be collected by cameras.

In addition to atrial fibrillation, Liu Xuenan also carried out research on the screening of coronary heart disease, hypertension, arteriosclerosis and other diseases, and relied on the engineering project of major scientific and technological achievements in Anhui Province, and jointly developed with the team members of the “Physiological Computing and Smart Health Laboratory” of Hefei University of Technology, and created a “Bang Health” mobile APP with pulse, heart rate and atrial fibrillation detection functions.

Liu Xuenan said, “As researchers, we should face the actual needs of the public, seize the window of opportunity of the current era, accelerate the breakthrough of the technical difficulties of video physiological measurement, and establish an intelligent, convenient and low-cost new way for the public’s daily disease screening and prevention.” Set up a line of defense outside the hospital to promote early detection and timely treatment of cardiovascular disease. (Source: Wang Min, China Science News)

Related paper information:

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button