Home - Emerging Tech - Diagnosing Skin Cancer via Smartphone
Research News

Diagnosing Skin Cancer via Smartphone

What are the issues raised by smartphone apps that detect skin cancer?

By
Mary Bates, Contributor
Wed, 05/09/2018

B2D-AusJournDerm-app-resized-WEB.jpg

Shutterstock/Gordana Sermek

SHARE: 

Technological advancements in artificial intelligence have made smartphones capable of performing myriad tasks for us -- but can they, and should they, diagnose disease?

Some studies suggest that deep convolutional neural networks can identify skin cancer lesions with a high degree of accuracy. However, these algorithms and the apps developed around them do not exist in a vacuum.

“There are critical ethical and medico-legal considerations that need to be addressed for quality assurance and patient safety,” said Saxon Smith, a dermatologist and associate professor at the University of Sydney.

In a recent opinion paper for The Australian Journal of Dermatology, Smith and his colleague Lisa Abbott explored some of the issues raised by integrating skin cancer-detecting apps into health systems. They say this technology is poised to change the diagnosis and management of skin cancer for both physicians and patients, but whether those changes are beneficial will depend on how they are regulated and implemented.

The Changing Landscape of Skin Cancer Diagnosis

At first glance, diagnostic apps might be appealing to consumers for their convenience, accessibility and affordability. However, there are significant perils to using such apps as a substitute for seeing a doctor. A trained dermatologist can take a medical history, conduct risk assessment, provide patient education, perform a total body skin check, and ensure adequate follow-up when they see patients in person.

Diagnostic apps also rely on the patient to identify suspicious skin lesions, something most patients are ill-equipped to do. Benign diagnoses for poorly selected lesions may offer the patient false reassurance.

“Previous research has shown that up to 60 percent of melanomas detected in private dermatology practice are in fact detected by the dermatologist without the patient having been aware of the life-threatening lesion,” said Smith.

Smith sees this technology developing as an additional diagnostic tool to assist the decision-making of doctors, rather than being the sole diagnostic tool used by patients themselves.

Since dermatology is a field that is heavily reliant on image interpretation, it seems inevitable that deep learning software will impact it. But Smith and Abbott emphasize that such software should be viewed as a supplement to clinical examination skills and judgment, rather than a replacement. Even if apps are developed that can perform as well as dermatologists in identifying skin cancer, the information they provide will still need to be interpreted and managed by doctors.

Legal and Regulatory Issues

Smith and Abbott also urge doctors to be involved in supporting accurate software and making sure dangerous and inaccurate products are removed from the market.

In a study from five years ago, scientists reported that the sensitivities of four commercial diagnostic skin cancer apps ranged from 7 to 98 percent.

“The Federal Trade Commission stepped in to say you can’t make claims about skin cancer detection without any data that shows you can do it,” said Laura Ferris, an associate professor of dermatology at the University of Pittsburgh and one of that study’s authors.

Ferris said that due to the hurdles an app company must go through, there are not really any apps that just give diagnoses any more.

“If you have something that’s a true diagnostic, that makes it a medical device and it has to go through an FDA approval process that is time-consuming and expensive,” she said.

So what can a skin cancer app do for patients? Ferris points to existing apps for patients that she sees as beneficial, such as those that provide educational information about what melanoma looks like or how to protect one’s skin from the sun.

“There are also apps that help patients take pictures of their own moles so they can track them,” she said. “It’s good because it’s not giving them a medical decision, but rather a tool to follow their own moles or show their doctor how a mole has changed.”

Ferris shares Smith and Abbott’s conservative optimism about applying artificial intelligence to dermatology.

“We need to figure out how to best use these technologies to help us do better for patients and make sure we’re not, in the process, causing a bunch of new problems,” Ferris said.