“…to promote the rapid and accurate reading of … skin lesions.”
That was part of the answer Dr. Eric Topol gave in a recent New York Times interview when asked where Artificial Intelligence (AI) shows the most promise in medicine.
Dr. Topol expands on this in the first part of his new book Deep Medicine dealing with AI in medicine where he describes dermatologists, along with radiologists and pathologists, as “doctors with patterns” or those who rely heavily on images in their work. He said such doctors could potentially benefit from AI algorithmic support.
“Many,” he writes “were frankly surprised by what deep learning could accomplish: studies that claim AI’s ability to diagnose some types of skin cancer as well or perhaps even better than board-certified dermatologists”. However, he also noted that the work done with deep learning and AI to diagnose melanomas represent computer-based validation rather than using AI support in a real clinical environment.
The complexities involved in doing just that and the hope it held for dermatologists who fear losing out to technology formed the backbone of two standing-room only sessions at the recent American Academy of Dermatology(AAD) meeting in Washington DC in early March where AI was formally on the agenda for the first time. While the ability of algorithms to identify melanomas as well if not better than dermatologists has been grabbing headlines for a couple of years, the subject matter was still novel enough at the world’s premier gathering of dermatologists that a portion of each session was needed to discuss the basics of exactly what was being talked about.
Dermatologists were also chastised by one of their own for doing a lousy job of what is needed to fundamentally support the use of AI to support a pattern-based specialty – namely taking and a categorizing images of skin lesions.
Dr. Allan Halperin, is head of the dermatology service at Memorial Sloan Kettering Cancer Centre at president of the International Skin Imaging Collaboration, a group that has pioneered the collection of dermoscopy images to develop algorithms to diagnose melanoma. “Frankly its embarrassing that, as a specialty, we are far behind what can be done already,” he said, commenting on the image-gathering capabilities of dermatologists. “We’re not taking images of enough things in a standardized enough way to make them available in a pipeline to the computer scientists to build an AI.”
Dr. Roger Ho, assistant professor of dermatology at NYU Langone Health in New York, agreed. In a news release he noted “hundreds of thousands of photos that have been confirmed as benign or malignant are used to teach the technology to recognize skin cancer, but all of these images were captured in optimal conditions – they’re not just any old photos snapped with a smartphone.”
To address these issues, Dr. Curiel said community-based dermatologists need to be involved in building the databases that will inform machine learning for melanoma. It was also noted that how the databases are compiled will depend to a degree on who will be making use of the data – patients, primary care physicians or dermatologists.
Dr. Ho and others – including the incoming president of the AAD Dr. George Hruza feel that even if the use of deep learning and AI is optimized for use with skin lesions, dermatologists will always be needed to properly interpret the findings and counsel patients. In his address to the meeting, Dr. Hruza said he was excited to imagine that when it comes to AI, dermatologists will have a “seat at the table and will not be on the menu.”
“The future of augmented dermatology is an opportunity not a threat,” said Dr. Clara Curiel, director of the dermatology program at the University of Arizona. “AI should reinforce clinical skills.”
It remains to be seen whether this almost universally expressed optimism holds true for a specialty already facing inroads from numerous other pretenders in the art and science of diagnosing and managing skin conditions.
(Photo: Patterns of light and dark at AAD meeting, Washington DC)