A mobile app can help detect skin cancer in older patients using AI

As a collateral effect of the isolation caused by the Covid-19 pandemic, a lower rate of visits to routine or preventive medical check-ups has been recorded in recent years.

As a solution so that these exams are not totally relegated, a team of researchers from Stanford University developed a system that allows preventive skin cancer controls to be carried out through a mobile app.

Early detection of skin cancer with a mobile

To help older patients get the care they needed, Kavita Sarin, an associate professor of dermatology at Stanford, and her team searched for new solutions. “This study was born from a clinical responsibility with our patients”, he pointed the professor of the Faculty of Medicine of that house of studies.

To develop this solution, they turned to the mobile application SkinIO, which allows not only doctors, but any user, to capture high-quality photos of potentially cancerous lesions, without the need to go to a clinic in the first instance. Using a secure data transmission system, the images are sent to a dermatologist for review.

To test the effectiveness of this system, it was implemented in a pilot plan, extended from November 2020 to July 2021, with 27 residents in a retirement community for the elderly in the San Francisco Bay Area, United States.

Assuming that for older adults, the target audience for this initiative, smartphones may seem somewhat alien, the project team provided patients with a step-by-step guide on taking pictures that they had to take, in this case, from a Tablet.

The app uses machine learning to analyze photos for skin lesions or abnormalities, flagging any that look suspicious. By itself, this is not a diagnostic tool, as it is intended solely for imaging and monitoring a patient’s skin. “The software just says, ‘Hey, this could use a closer look’”Sarin said, noting that later, the doctors are the ones in charge of issuing the diagnosis.

Using this application, the coordinators of this investigation shared the collected photos with Sarin, who reviewed them and marked those with lesions that appeared to be cancerous.

Later, Sarin asked the research coordinators who were with the patient to take new photographs of the suspicious lesions, this time using a dermatoscope, a specialized imaging device for this purpose, which takes detailed images of the outer layer of skin that it is not visible to the naked eye. These more detailed images allowed Sarin to accurately diagnose the injuries.

Next, the research coordinators organized a virtual visit for all patients with the principal investigator, to review the findings. Sarin then scheduled a face-to-face visit with the patient for a more detailed examination, recommending treatments if necessary, or issuing a clean bill of health, if applicable.

During this trial, the app flagged 63% of observed injuries for further extensive investigation. Of this sample, most of the lesions detected ended up being benign.

“The lesion detection algorithm is not perfect and there were some issues with lesion detection, but for a testing situation it is a good tool”Sarin said.

Of the 27 people who participated in the study, the app and Sarin identified skin cancers in three patients. 11 of them were scheduled for an in-person follow-up review and four began treatment at home. Some of the skin cancers detected corresponded to lesions that the patient had not previously identified, which means that without the application’s full-body images, they could have been missed, according to what the researcher pointed out.

Although the solution offered by this project does not in any way replace the attention of a professional, its use as a preventive control tool could facilitate the monitoring of this disease that, with early detection, can be better addressed clinically.