At present, the application software has been approved by the European Union, but if it is to be listed in the United States, it needs to pass the strict examination of the FDA. Daneshjou, a dermatologist at Stanford University, believes that it may be too early to disclose artificial intelligence dermatology tools to consumers.
Using image recognition detection, covering 288 kinds of skin diseases.
If you don't feel well, first search the internet for what disease it may be. This is what many people will do. Although many people know that "Google sees a doctor" is not reliable, they still can't help but open the search engine.
Google is about to launch a self-proclaimed "reliable" skin disease diagnosis medical application. At the Google I/O Google Developers Conference last month, Karen DeSalvo, Google's chief health officer, said that Google would launch dermatological AIDS in the EU as early as the end of this year.
202 1 Google I/O developer conference site
He showed the specific usage of the application. Users need to take three photos of their skin defects from different angles and distances, and can note text information, such as physical feelings and the time when symptoms appear, and then click "Submit" to upload photos. The application will automatically analyze these pictures and give "suggested results", that is, tell the user what kinds of skin diseases this may be, and list the symptoms, infectivity and treatment scheme of the disease in each possible result.
However, when the application displays the results, it will warn the user that "the suggested results listed here are not medical diagnosis".
Google said that the application has been trained with "hundreds of thousands of skin images" and can identify 288 skin diseases, including skin cancer, covering about 90% of common dermatology online searches.
According to statistics, users search for skin problems in Google for billions of times every year. Google claims that the application was developed because there are not enough dermatologists to help all people with skin diseases. Google's blog post mentioned a comparative study, saying that "our artificial intelligence system can achieve the same accuracy as dermatologists certified by the US board of directors."
But Roxana Daneshjou, a dermatologist at Stanford University, is skeptical. She said that this technology has little evidence of clinical experiments. In the United States, no AI image analysis tool has been approved for dermatologists.
Choose Europe as the market, not the United States, because the FDA has a strict review of health software.
It is worth noting that Google did not choose the US domestic market to promote this skin disease diagnosis software, but chose Europe as the first stop.
Google said that its skin application has been approved by the European Union-"CE marks it as a Class I medical device" (note: CE is the EU market safety certification mark and a mandatory certification mark), which means that it can be sold in the EU and other countries or regions that recognize this standard.
This surprised many people, because the supervision of artificial intelligence applications in the EU has always been known for its strictness. Harvey, managing director of Hardian Health, a British digital health consulting company, said that last month, the European Union introduced stricter artificial intelligence regulations, requiring health-related applications to prove that the application is effective. By 2025, pre-existing applications must comply with the new regulations.
Google showed the application at the Google Developers Conference a week before the new EU regulations came into effect.
In the United States, the application may still have a long way to go, waiting for FDA approval. FDA, the US Food and Drug Administration, authorized by the federal government, is the highest law enforcement agency specializing in food and drug management. An FDA spokesman declined to comment on the product, but said that software claimed to be used for "diagnosing, curing, preventing or treating human beings" may be regarded as medical equipment and need to be reviewed and approved.
It is understood that the FDA adopts a hierarchical approach in the examination and approval of health software. For those software that only provides health advice or health information, FDA considers it as "low risk" and does not need to obtain medical device approval. More stringent approval is required for software that provides specific diagnosis or is used as medical equipment. But with the continuous updating of medical software, this boundary is sometimes difficult to determine.
However, Google's dermatological assistance should be regarded as "requiring approval" at present. An FDA spokesperson said that a guide has been issued for the software to encourage data collection from different groups of people.
Dermatologist Daneshjou also believes that Google should disclose more information about the testing of the software in people with different skin colors. There is evidence that Google's artificial intelligence dermatology research has little research on people with darker skin color.
In fact, the application of image recognition in search engines is very mature. For example, when people see a strange plant on the road, they can get information about the plant immediately by taking photos and uploading them to search engines.
However, the application of image recognition in the diagnosis of dermatosis is obviously different from that of general object recognition. Some skin diseases are very dangerous, such as melanoma, which can be fatal. If the software goes wrong, it will bring serious consequences.
"It may be too early to disclose artificial intelligence dermatology tools to consumers." Daneshjou said.