news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

In the future, we may be able to tell what disease you have by scanning your face...

I'm not making this up, a recent research result actually gives us some clues.

A research team at Peking University has developed an AI that can detect whether a person hasChronic diseases such as hypertension, diabetes and fatty liver。

It can also be used to predict our aging and health conditions, and it is not impossible to use it to detect cardiovascular diseases.

Like most of my friends, when I first saw this research, I thought that this kind of research was at most a laboratory result. Unexpectedly, after checking it out, I found that scientists have already been working on this kind of technology for a long time, andThere are already many practical application cases.

For example, many skin care companies have developed applications that recommend products by scanning the face, and can help users understand their skin conditions through facial recognition.

Moreover, the entire process is assessed based on the dermatological grading system, which is almost at the level of a dermatologist.

Even a few years ago, they developed a hardware based on this technology. When attached to a mobile phone, it allows the camera to see more facial details and give you a complete skin research report in just a few minutes.

In addition to identifying the most basic skin conditions, facial recognition can also help us diagnose many other diseases.

As early as 2014, a company called FDNA in the United States had already developed a "face scanning to see a doctor" system.Application Face2Gene 。

Basically, everyone who pays attention to this field knows their names, and there are many reports about them.

It can judge people just by looking at their faces.Genetic Screening, specifically for diagnosisVarious rare genetic syndromes, and can also detect some emotional disorders such as autism.

So far, it can accurately identifyThere are about 300 diseases。

For example, a few years ago, Face2Gene helped a little girl diagnose Wiedemann Steiner syndrome (a rare disease caused by a gene mutation) in advance, when various typical physical symptoms had not yet appeared.

Moreover, at the beginning of this year, Japan conducted a study on it and found that Face2Gene has an accuracy rate of up to 85.7% when screening congenital deformation syndrome.

This means that there is nothing wrong with doctors using it as an initial screening tool before diagnosis, and many hospitals do so.

According to the Times, a hospital in the UK uses Face2Gene to screen children for genetic diseases. It can identify more than 2,000 facial features of various diseases in a few seconds, saving doctors a lot of energy.

There are many similar technologies. Just last month, a research team from the University of Melbourne developed a tool that can determine whether a person has had a stroke by identifying facial muscles, with an accuracy rate of up to 82%.

In Japan, which has a serious aging population, they are also very interested in AI-assisted diagnosis technology.

At this year's CES, NEC's new product won the Artificial Intelligence Innovation Award. It only takes 10 seconds to estimate a person's pulse rate, oxygen level and breathing rate from his face.

NEC also plans toMultimodal biometric authenticationWe need to put in more effort, that is to say, we need to not only recognize faces, but also irises, etc., so that the accuracy of detection can be improved to a higher level.

Anyway, in recent years, there are more and more products that can recognize faces when seeing a doctor, but everyone may still have this question in their mind:How can so many diseases be detected with just facial recognition?

In fact, there is very rigorous scientific evidence that the growth of the human face is determined by genes.There are approximately 6,000 genes that are related to the appearance of the human face.

If a gene is missing or one of the genes mutates, it will be immediately reflected on the face, such as affecting the shape of the eyebrows, nose base or cheeks.

And ideally, facial recognition alone would be able to diagnose diseases like Down syndrome and progeria.Thousands ofGenetic diseases.

But in general, these subtle changes we mentioned,Even doctors may not be able to detect it with the naked eye, in comparison, AI is much more sensitive.

Take Face2Gene for example. It can perform millions of calculations on facial data in a very short period of time, including how slanted the eyes are, how narrow the palpebral fissure is, how low the ears are, etc. It can perform detailed quantitative analysis and then list possible diseases.

In addition, the temperature of different areas of the face will also have slight changes depending on different diseases. For example, the identification of hypertension and diabetes mentioned at the beginning is based on this principle.

It first scans a 3D image of a person's face with a thermal imager, and then predicts possible chronic diseases based on the temperature conditions in different areas. A cold nose and warm cheeks are obvious signs of elevated blood pressure.

Of course, using AI facial recognition to see a doctor does not mean that it will replace doctors. In essence, it can only play aHelp doctors to screen for specific diseasesrole.

Originally, to diagnose a disease, it might be necessary to screen more than 6,000 genes related to facial expressions. But with the use of AI, it can usually be done by checking one or two genes.

But then again, the premise for AI to achieve these goals is not only to have powerful algorithms, but also to haveSufficient patient data。

This is the biggest problem that this type of AI is trying to solve. Face2Gene, for example, has a high accuracy rate in identifying certain genetic diseases, but it is limited to white patients. If the patient has a different skin color or race, it may not be able to diagnose it.

The reason for such a big difference is very simple, because most of the database used for training are white patients in Europe and the United States, and there is very little data on other skin colors or races.

For example, in Europe and the United States, one of the criteria for determining whether a child has Down syndrome is to see whether the eyes are tilted upwards, while in Southeast Asia, many normal people's eyes also have this feature.

Therefore, the AI ​​algorithm still needs to set corresponding evaluation criteria for patients of different races and skin colors.

But no matter what, AI facial recognition is a technology that benefits you and me.

Shichao is really looking forward to launching this kind of app on mobile phones. It would be so convenient to see a doctor just by scanning your face...

Written by:squirrel

edit:Jiang Jiang

Art editor: Jessica Hsuan

Image, source：

Scitechdaily, yahoo, fortune, 36kr, etc.