Apps to ID melanoma from photos often inaccurate, risk delayed diagnosis: study
TORONTO - Smart phone apps that analyze photos of moles to assess whether they might be potentially deadly melanoma or benign lesions vary widely in accuracy, say researchers, who warn patients against using them for diagnosis.
A study of four such applications, available free or for a modest cost from the two most popular online app retailers, found they incorrectly classified at least 30 per cent of melanomas submitted for evaluation as being of no concern.
"One of them was so bad that it actually missed over 93 per cent (of the melanomas)," said principal author Dr. Laura Ferris, a dermatologist at the University of Pittsburgh Medical Center.
"That's kind of scary."
Melanoma, the most serious type of skin cancer, can develop in an existing mole or arise as a fresh pigmented or unusual-looking growth on the skin. They can develop anywhere, but are most common in areas previously exposed to sunlight, such as the back, legs, arms and face. However, melanomas can also appear on the palms of the hands and soles of the feet.
If caught early and treated, this skin cancer has a high cure rate, but the survival rate is low for patients whose melanomas have gone undetected and spread to other areas of the body.
To conduct the study, the researchers submitted 188 photos of lesions to each of the four smart phone apps, none of which was identified by the authors.
The researchers knew the status of the moles or lesions — 60 were melanomas and 128 benign — because they had been photographed before being removed or biopsied and sent to an in-house pathologist for testing.
"We knew what they were, either cancer or not," Ferris said from Pittsburgh. "And we went with the two ends of the spectrum: it's frankly melanoma or it is a really low-grade, not bad pigmented skin lesion."
The researchers wanted to see whether assessments by the four apps matched the previously confirmed diagnoses.
Ferris said three of the apps give answers within a minute after a photo is uploaded. The fourth, which forwards the photo to a U.S. board-certified dermatologist for analysis, takes about 24 hours to provide an online opinion as to whether a lesion is suspicious or not.
"And the main thing we were looking at was how often did they get melanoma right or how often did they miss it," she said.
"Interestingly, the results we got from the one where you had a dermatologist just looking at the picture ... was actually very good. It was 98.1 per cent.
"So there was a big difference between the performance of the dermatologist versus these (other three) applications."
Ferris said the overall performance of the apps in picking up melanomas is unacceptable, but there are currently no regulations governing such online health programs.
"And so the concern is that these are out there and they are available to the public and you can get this answer with no doctor involved," she said, adding that such apps have not been tested through a clinical trial nor do they meet any quality standard.
"Would I personally like to see some sort of regulation? Yes," she said.
"I'm afraid for the safety of my patients, that if they're using something like this and it's 30 per cent of the time going to tell them that their deadly cancer is fine, not to seek attention for it, I think that's a risk, a real public health risk."
