AI can be taught to flag possible skin cancers on photos taken with smartphone cameras—and the images can be ordinary “people shots” rather than closeups of suspicious lesions.
The concept passed its audition at MIT, where biomedical engineers trained, tested and validated deep convolutional networks with more than 38,000 images.
Along with dermatology-grade images, the dataset included more than 15,000 wide-field shots captured with consumer cameras.
Using lesion classification by experienced dermatologists as the ground truth, the researchers found their system achieved sensitivity and specificity of right around 90% each when tasked with separating suspicious lesions from benign skin discolorations and busy backgrounds.
In clinical use, the authors note, such automated screening prowess could help patients head to the dermatologist for early diagnosis—or at least avoid burdensome individual lesion imaging.
“Rather than evaluate a single lesion at a time looking for predetermined signs of neoplasia, the algorithm identifies lesions that differ from most of the other marks on that patient’s skin, flagging them for further examination and ranking them in order of suspiciousness,” the authors explain in their study summary. “The algorithm performed similarly to board-certified dermatologists and could potentially be used at primary care visits to help clinicians triage suspicious lesions for follow-up.”
Science Translational Medicine published the study Feb. 17. Lead and senior authors are, respectively, Luis Soenksen, PhD, and Martha Gray, PhD.