Warning about using the phone app to detect skin cancer

When it comes to smartphone technology, there are almost every app today. You can play chess, learn to speak Mandarin, or map the best driving route to the airport. Health is a big category for apps, but many of these programs lack the science to support their claims.

Experts say apps that scan moles and other lesions and claim to identify which ones are cancerous fall into areas that require more research and development.so paper Published in the April 2022 issue of British Journal of Dermatology (BJD)Memorial Sloan Kettering Cancer Center (MSK) A team supervised by a dermatologist Veronica Rotenberg When Alan Harpan The overall accuracy of these commercial apps has been reported to be low. Of the apps rated, the average accuracy was only 59%.

“We downloaded all the apps we found and tested them all to see how well they could be identified. melanoma“Dr. Rotenberg says. “On average, they found that they weren’t doing very well. Skin cancer is one of the areas where more data is really needed before it encourages patients to rely on the app.”

Disorders when using AI for cancer screening

Dr. Rotemberg and Halpern have long focused on developing artificial intelligence (AI) and other computer-based tools for detecting and diagnosing melanoma, the most deadly form of skin cancer. Metastatic Melanoma is expected to kill more than 7,500 people in the United States this year. However, if detected early, melanoma is almost always curable, so researchers are focusing on improving early detection.

However, there are many hurdles. Even digital tools created by top dermatologists and computer scientists have many flaws.another Recent papersPublished in the May 2022 issue Lancet Digital HealthOutlined why the development of these diagnostic algorithms is so difficult.

Researchers have shown that computers are superior to dermatologists in diagnosing skin lesions in previous studies of AI in dermatology, but these studies use extensive images. He states that he did not properly evaluate the actual clinical scenario. When a computer displays an image or skin condition that you have never seen before, it is less accurate than reading a more familiar image. The algorithm was only 6% accurate on average for the diagnoses not seen in training.

“One of the most important discoveries was that the system really needed to improve its algorithmic ability to identify untrained types of images,” says lead author Dr. Rotenberg. “For images in unknown categories, these tools perform almost as well as random chances.”

Dr. Rotemberg states that AI can also be stumbled upon by defective images. For example, hair and pen marks scattered in photos have reduced the accuracy of AI tools.

Back to top

Improving AI to detect skin cancer

The Lancet Digital Health The paper was the result of the 2019 Grand Challenge, which focused on developing algorithms for diagnosing skin cancer. In the ever-growing field of AI, the Grand Challenge event brings together a team of computer scientists from around the world to develop better machine learning systems for medical imaging and other disciplines.

“This is an exciting time and the research community is very interested in improving the detection of skin cancer,” says Dr. Rotenberg. “The building blocks are in place, but there are still many obstacles to overcome,” she adds, adding that the study represents an international collaboration with dermatologists and AI professionals around the world. .. “This kind of science is done by a large team.”

Another important issue to address is the fact that these may not work very well in people of color, researchers admit.

The BJD Studies show the danger of relying on AI-based tools that are not yet ready for prime time. “I was worried that these apps would classify everything as melanoma, leading to a lot of worry and unnecessary biopsies,” said Dr. Rotenberg. “But in reality, the opposite is true. The inability to identify melanoma as genuine may create a false sense of security.”

Experts agree: It’s too early to say “there’s an app for that” when it comes to identifying skin cancer.

Back to top