NEW YORK — Smartphone technology can help improve the detection of melanoma, but several barriers must be overcome before it is used as more than a complementary tool, according to a presenter at HemOnc Today Melanoma and Cutaneous Malignancies.
“In the very short term, I think this technology has more value not as a standalone [approach] but as a way to augment what we do,” Allan C. Halpern, MD, chief of the dermatology service at Memorial Sloan Kettering and a HemOnc Today Editorial Board member, said during a presentation. “There are enormous opportunities to apply this to decision support and teaching.”
Melanoma mortality is increasing and likely will continue to rise as the baby boomer generation ages.
Allan C. Halpern
However, approximately 77% of all melanomas and 88% of lethal melanomas are detected by patients or their family members rather than physicians, and more than three-quarters of the U.S. population has never undergone a clinician-performed skin examination.
“This emphasizes how important public education and awareness is, but it raises the question: How can we do a better job of finding melanomas earlier?” Halpern said. “The answer cannot be to use more dermatologists [to conduct skin examinations] because, when the day is done, there just is not enough of them.”
A study by Esteva and colleagues — published in January in Nature — may have been “a watershed moment” in the application of artificial intelligence to melanoma diagnosis, Halpern said.
The researchers used a dataset of 129,450 clinical images of more than 2,000 diseases to train a deep convolutional neural network to recognize and distinguish between potentially malignant skin lesions and benign conditions. Results showed artificial intelligence performed comparably to a group of 21 board-certified dermatologists.
“Dermatologists have context, such as: Is there a personal history of melanoma? Is there a family history? What do the rest of the spots on your body look like?” Halpern said. “But, in this constrained environment, it did as well.”
Findings like these have generated excitement about the potential that smartphones could be used in melanoma detection.
Approximately 250 apps designed to help individuals learn about or identify melanoma, and consumers can purchase dermatoscopes that attach to their cell phones and take high-quality images of suspicious lesions.
However, several barriers must be overcome before melanoma diagnosis via smartphone is “ready for prime time,” Halpern said.
Regulatory hurdles are one factor.
Medical apps require FDA approval, but any that are billed as being capable of diagnosing cancer must have 100% sensitivity.
“Dermatologists don’t have that kind of sensitivity, so we are expecting more from machines,” Halpern said.
Second, most apps designed to improve melanoma detection are created to look at a single lesion.
“Context is really important, but nothing says that apps can’t begin to develop context,” Halpern said. “You can tell the app you have a history of melanoma, how old you are and how many times you have been in a tanning bed.”
The sociology of medicine also is a factor.
“One of the biggest determinants of whether smartphones can find their way into the critical chain of delivery of care is going to do with us,” Halpern said. “Does the profession embrace this an opportunity to bring care closer to patients, or is it considered a threat?”
“Artificial intelligence has enormous potential to make us better at what we do,” he added. “That doesn’t mean it won’t take some things we currently do away from us.”
The dramatic rate of change in the health care system — including the shift toward value-based reimbursement, consolidation, and the creation of alternative delivery systems — may accelerate the pace at which smartphone technology is adopted in practice, Halpern said.
He also provided an overview the International Skin Imaging Collaboration, a Memorial Sloan Kettering Cancer Center–led academia–industry partnership intended to facilitate digital skin imaging technologies to reduce melanoma mortality.
The collaboration created a public archive that includes approximately 13,000 high-quality dermoscopy images, including 7,500 images of melanoma.
In 2016, collaboration members conducted a challenge in which computers and eight international dermatologists reviewed 100 selected images.
Results revealed the dermatologists diagnosed melanoma with a mean sensitivity of 82% and a mean specificity of 59%. When computer sensitivity was 82%, its specificity was 64%.
When asked to determine whether a biopsy was warranted, dermatologists’ demonstrated a mean sensitivity of 89% and mean specificity of 47%. When computer sensitivity was 89%, its specificity was 50%.
The collaboration conducted a similar challenge earlier this year. The results — which Halpern described as “quite astounding” — will be released at the IEEE International Symposium on Biomedical Imaging, scheduled for April 18-21 in Australia. – by Mark Leiser
Halpern AC. ‘Smartphone’ diagnosis of melanoma: Feasible or folly? Presented at: HemOnc Today Melanoma and Cutaneous Malignancies; March 24-25, 2017; New York.
Disclosure: Halpern reports financial relationships with Canfield Scientific, Caliber ID and Scibase.