Where Does AI Stand in Dermatology?

In many specialties, it seems that artificial intelligence (AI) is gaining a lead on human physicians. Dermatology is the latest field to reckon with the technology, as a new international study has found that conventional neural networks (CNN) outperformed dermatologists at melanoma detection.

Researchers from Germany, France, and the Memorial Sloan Kettering Cancer Center in New York evaluated 58 dermatologists from across the world and their ability to diagnose skin lesions. Thirty of these physicians were considered experts with over five years of experience and 19 percent had around two to five years of experience.

In the study, the dermatologists looked at 100 dermoscopic images and were asked to classify the lesions as malignant melanoma or benign mole. They were also questioned about their clinical protocol and whether or not the lesion required “surgery, short term follow-up, or no action needed.” Overall, the dermatologists had a 73 percent accuracy rate at identifying nonmalignant lesions. A month later, the researchers provided the dermatologists with clinical data about the patients including age, gender, and lesion location, and gave them a second opportunity to assess the lesions. At the second go around, the physicians correctly diagnosed “89 percent of malignant melanomas and 76 percent of benign moles.”

The researchers compared the dermatologists’ evaluations against the CNN, a machine learning algorithm that teaches itself to recognize images the way brain cells transmit signals and react to what the eyes see. After feeding the CNN 300 images of skin lesions, the researchers found that the technology correctly identified 95 percent of melanomas.

The study demonstrates an opportunity for AI to assist in skin lesion diagnosis, and could serve as a life-saving technology in areas with dermatologist shortages. “Artificial intelligence (AI) promises a more standardized level of diagnostic accuracy,” said the researchers. “Such that all people, regardless of where they live or which doctor they see, will be able to access reliable diagnostic assessment."