The design shown in the demo requires the person to take three pictures of their impurities from different angles and distances. The user can optionally add information such as which body part is affected and how long the problem has occurred. Clicking Submit will turn off photos on Google. Then the application displays “suggested conditions”, showing possible conditions illustrated with pictures. Clicking one of them displays a list of basic information such as symptoms, infection, and treatment options. Google says the app has been trained on “hundreds of thousands of images of skin” and can identify 288 conditions, including skin cancers, which cover nearly 90 percent of popular web searches for skin conditions.
The Food and Drug Administration exempts some health programs it considers “less dangerous” — such as “wellness” advice such as diabetes management or information about health symptoms — from medical device approvals. It requires approval for others, such as those providing specific diagnoses, or applications that act as medical devices such as a stethoscope. The line between applications that require a permit and those that do not are difficult to draw because the medical programs and the rules governing them are relatively new.
Bradley Thompson, regulatory attorney with Epstein Becker Green, asks clients a few key questions when trying to determine whether they will need FDA approval. They include how the program’s output is presented to a person and whether the company makes certain medical claims.
The Google app doesn’t highlight a single potential skin condition in response to the person’s photos, and displays a warning that “the suggested conditions listed here are not a medical diagnosis.” A company spokesperson likened the app to a search engine that displays results for someone to look at and draw their own conclusions about.
However, Google has also emphasized medicinal pieces for skin application. Health chief DeSalvo said Google developed the app because there aren’t enough skin specialists to help everyone with skin problems. The Google Blog post The app links to peer-reviewed studies in which the company’s technology has been compared to doctors, saying, “Our AI system can achieve accuracy that’s on par with US board-certified dermatologists.”
This boast caught the eye of Thompson, the attorney. “It really suggests that this is at least comparable to what a human doctor would do,” he says, the kind of claim that might interest the FDA.
Daneshjou, a dermatologist and researcher at Stanford, believes the Google app can appear to consumers and regulators as providing medical expertise, not just search results. It says the app can be considered a “high-risk” device, and requires FDA approval, since certain skin conditions such as skin cancer can be dangerous.
Daneshjou contributed to a recent study It raises concerns about how thoroughly the US Food and Drug Administration is scrutinizing health programs for artificial intelligence, and says it may be too early to bring open dermatology tools to consumers. “If a patient believes that this algorithm works as well as a board-certified dermatologist, they may have more confidence in it,” she says. This may cause people to require unnecessary biopsies or treatment from a doctor, or not to make an important visit.
Daneshjou says Google should also reveal more about how it tests its technology on different skin tones. The company’s dermatological AI studies to date have included relatively few people with darker skin.
Google says these posts do not represent its latest data or improved image recognition models. The spokesperson said the dermatology app design and disclaimer were informed by user experience studies; Additional studies are underway, and the company will also be researching how people use the service after it becomes available in Europe.
Google has faced practical challenges when deploying other promising AI-powered health programs outside the lab. In 2018, the company began testing a system capable of detecting eye diseases in people with diabetes in clinics in Thailand. In 2020, the company published a study about the offering that said the system was كان Rejected more than 20 percent of patients’ photographs due to issues such as variable lighting and nurses’ practical limitations.
Updated, 6-23-21, 11:30 AM ET: This article has been updated with comment from the Food and Drug Administration.
More great wired stories