Dr. Roxana Daneshjou discusses AI in dermatology today, potential future applications, and why dermatologists should get involved to guide their safe, effective development.
Roxana Daneshjou, MD, PhD, FAAD, is Clinical Scholar, Dermatology, Stanford School of Medicine Stanford, California
“In health care, we have seen a proliferation of artificial intelligence (AI) applications. Currently there are no FDA-approved AI applications in dermatology but there are many different groups working on that pathway. In the next decade or so, we will see clinical applications likely coming down the pipeline to be applied to the field of dermatology,” said Roxana Daneshjou, MD, PhD, FAAD who presented “AI in Dermatology” at the Innovations in Dermatology conference in Scottsdale, Arizona.
Not only that, but there are apps, including phone-based apps, which have been released directly to consumers claiming to use AI to help diagnose skin and other conditions, according to Dr. Daneshjou.
“None of those, again, are approved in the United States. In other countries there have been some regulatory approvals for those applications. So, it is just very important for dermatologists to understand what the landscape of this technology is, what’s happening in the future, what to watch out for, how to counsel our patients, and how to play a part in the regulatory and policy decision-making processes involved with these technologies. It’s for all these different reasons that I think this is a really important topic for dermatologists to think about.”
Current AI in Dermatology
Direct-to-consumer facing apps are likely the most visible of AI being used in dermatology today, said Dr. Daneshjou.
“These are applications that you can download on the phone which claim that you can take an image of your lesion and it will provide some information about whether or not there is a concern about it being melanoma or some other malignancy.”
Patients have access to these applications independent of their physicians. Consumers may not know how these applications work, how well they work, and whether the information they provide is trustworthy, said Dr. Daneshjou.
“The important thing is to let patients know that there is no phone application that is approved yet to make any determination about whether a skin lesion is dangerous or not. So, if they have a concern about a skin lesion it’s very important that they consult their dermatologist.”
While educating patients, dermatologists should also pay attention to what different companies are doing and try and advocate for safe and effective AI apps, said Dr. Daneshjou.
“Dermatologists need to be the ones to stand up and ask the important questions. How does this application work? Has it been validated? What are the prospective randomized clinical trials that show that it’s safe for patients? What guardrails are in place to mitigate harm to our patients?”
“Dermatologists have an opportunity to be the advocates in this space; currently many of these technologies are being developed without consulting dermatologists.”
Future AI Applications
Companies and groups are working on a myriad of AI applications in health care and dermatology. Some AI tools help with scheduling and various other clinical workflow tasks, according to Dr. Daneshjou, who said that the focus of her research has been on AI tools that use images.
“I’m talking about all kinds of interesting applications like lesion monitoring, rash monitoring, automatically calculating body surface area involved, automation of severity scores from images …a whole range of applications that people are working on that really support the work of dermatologists.”
Among potential AI applications are decision support tools—human-in-the-loop augmented intelligence tools that will help, for example, non-specialist primary care physicians determine which patients might need to be referred to a dermatologist.
Dr. Daneshjou and colleagues have been working on developing AI tools to improve patient-submitted teledermatology photos to help dermatologists provide better, more efficient clinical care, she said.
“I think the most exciting things that are coming down the pipeline—and probably the things that will end up being most successful—are applications that help make our workflow easier, help patients access care better, and actually enhance and augment our abilities rather than trying to replace our abilities.”
As someone who has trained in and worked directly with AI algorithms, Dr. Daneshjou said these tools have a lot of “pain points,” where changes in such things as lighting, setting, and the distributions in the kinds of lesions can affect performance.
“As of now there is nothing that really comes anywhere close to being able to do a standalone diagnosis. As we know, the way that a dermatologist makes a diagnosis is so much more than just looking at an image. It’s patient history; it’s feeling the lesion. It’s the context of what else is going on with the patient.”
“I know there have been a lot of claims in the media about algorithms that can perform as well as dermatologists. I think subsequent research (including work we have done) has shown that when you take these algorithms and try and apply them to real-world data, their performance drops significantly compared to what was claimed and the initial research paper that got a lot of media buzz. So, I really don’t see that dermatologists are going to be replaced.”
Rather, it’s important for dermatologists to be involved in these conversations and not to be intimidated because many of these tools enhance the specialty, she said.
Disclosure: Dr. Daneshjou is a member of the scientific advisory board for MDalgorithms.
