Some might view the use of non-human intelligence in dermatology as a threat. But this expert believes it's up to dermatologists to transform this technology into an opportunity to ‘augment' clinical practice.
Imagine this: A patient calls for her annual skin cancer check. A voice-activated artificial intelligence (AI) tied to your back-office operations answers, sets up the appointment or triages the call. The patient arrives to an appointment for total body imaging with an automated system, coupled with a machine learning system designed to identify suspicious skin lesions that need further single lesion assessment. Following individual lesion assessment, the decision to reassure the patient, monitor the lesion with close-up and dermoscopic images or perform a skin biopsy is rendered.
“This initial process of total body evaluation is performed in parallel with a dermatologist, followed by individual lesion examination by both AI-based devices and by clinical inspection, which are ultimately complementary,” says Clara N. Curiel-Lewandrowski, M.D., professor of dermatology at the University of Arizona (UA) and director of the Multidisciplinary Cutaneous Oncology Program, UA Skin Cancer Institute.
This, Dr. Lewandrowski says, is a dynamic process upon which the human brain and computer systems continue to learn from one another.
Some might view the use of non-human intelligence in dermatology as a threat.
“However, it is up to us as dermatologists to transform this technological shift into an opportunity to ‘augment’ our practice with dermatologists at the helm of this paradigm,” says Dr. Lewandrowski.
“With the expected AI revolution in front of us, it is of paradigm importance for dermatologists to familiarize themselves with the ABCDs of machine learning to help guide further development and implementation,” she says.
People might use machine learning, artificial intelligence and augmented intelligence interchangeably, but each is different and important for the future of patient care.
Artificial intelligence is a large group of image-based data to guide a computer or network to make decisions, according to Pooja Sodha, M.D., assistant professor of dermatology at Duke University Medical Center, and medical advisor and vice president of research and development for FaceMD+, which uses artificial intelligence and machine learning for management of benign skin conditions.
“Machine learning is a subset within artificial intelligence in which large computational networks to try to learn through trial and error,” Dr. Sodha says. “In dermatology, the data is image-based. You basically train that algorithm to recognize certain features of the data set images so it can deduce how to go from an image to an actual diagnosis. The ability of the algorithm is dictated by how much data there is. The more data and the more training you offer it, the better it is able to pick up nuances between images, which may vary in quality and resolution.
“If it incorrectly recognizes an image, the great thing about machine learning is that it has the ability to take feedback and correct itself. So, you can train these networks and machine learning programs to identify nuances and be able to overlook the variability between images but pick up on the true identifying diagnostic features. That’s what makes it a powerful tool for diagnosis of nevi, melanoma, skin cancers and more.”
Augmented intelligence brings the different components of artificial intelligence together to enhance the way dermatologists and other clinicians practice. It complements dermatologists’ decision making, Dr. Lewandrowski says.
Studies suggest artificial intelligence and machine learning have the potential to be at least as accurate as dermatologists for diagnosing melanoma, says Art Papier, M.D., associate professor of dermatology and medical informatics at the University of Rochester.
Using large image datasets to train software algorithms to pick up on features of skin lesions through pattern recognition isn’t entirely new to the specialty, but it’s a rapidly evolving field in dermatology and other visual specialties, says Dr. Papier, who is CEO of VisualDx, a healthcare informatics company that develops digital health products to enhance diagnostic accuracy.
The technologies will need more than proven efficacy to penetrate dermatology and other specialties, he says.
“I don’t know if dermatologists and the medical-legal system are ready for software to help diagnose melanoma, for example. If you have software that’s aiding the diagnosis of melanoma, and the software makes a mistake, who is responsible? It’s not really clear. There isn’t a comfort level yet, though the technologies are getting better and better,” Dr. Papier says.
“I believe looking out five or 10 years into the future you could easily imagine arrays of cameras taking pictures of patients, tracking their changes in their nevi and detecting melanoma much more accurately than people. That’s how fast technology is evolving, but the healthcare system-the legal system-both on the reimbursement side and the medical-legal side need to catch up with these technologies once they’re really proven.”
However, new generation 2D and 3D total body scanning technologies are currently changing the ways in which dermatologists monitor melanoma, according to a handout presented by Josep Malvehy, M.D., from the Melanoma Unit at the Hospital Clinic Barcelona, Spain, during the “Melanoma: The Future is Now” panel at this year’s AAD annual meeting.
“In monitoring melanocytic lesions, dermatologists have traditionally [worked] on naked eye examination and dermoscopy assessment, clinical memory recall and, when available, a 2D digital camera,” Dr. Malvehy says.
One example of how scanners have evolved: Vectra (Canfield) is a 3D whole body photoimaging technology. The device captures images via 92 cameras with white or cross-polarized lighting.
“The cameras capture the images simultaneously and then construct a digital 3D avatar of the patient. The image capture happens in just a matter of seconds – the 3D representation facilitates a 360-degree rotation to view all body angles, including curved surfaces. We can use florescence and regular photography to track lesions, giving us the ability to see every single lesion on the body,” says Miami, Fla., dermatologist Jill Waibel, M.D., who is on staff at the Miami Cancer Institute. “The Vectra also gives you the ability to analyze all nevi on a patient’s body from darkest to lightest, and the most irregular shape. It gives dermatologists one more analytical tool. The Vectra also has dermoscopy capabilities, which enables us to track individual lesions every three months.”
Dr. Waibel says she uses Vectra for patients who are at high risk for skin cancer.
“We live in South Florida, which is one of the worst markets in the world for skin cancer, so we’re looking mainly for melanoma, but we’re also looking for non-melanoma skin cancers, such as basal cell and squamous cell,” she says.
High-risk patients are those with a history of cutaneous melanoma and dysplastic nevi, have more than 50 nevi or have more than five non-melanoma skin cancers. The Vectra and other technologies like it do not replace the dermatologist’s visual total body exam. Rather it augments what the dermatologist does and gives the specialist another objective image to view, according to Dr. Waibel.
“Sometimes early melanoma has little globules around the edge that you are not going to be able to see with the naked eye, so this technology is really going to give us an edge in early detection, which is early survival for melanoma, so that’s why it’s such a huge win and advancement in detection,” she says.
While technology appears to be penetrating every aspect of the dermatology practice, it is not taking over. AI can act as a partner or support, freeing a dermatologist up to concentrate more on the art of medicine that a machine can’t master, Eric Topol, director and founder of the Scripps Research Translational Institute, and author of Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again, tells TIME.1
This is why dermatologists need to be involved in AI development. If dermatologists don’t help provide enough good data for machine learning, for example, the resulting algorithm will not be capable of robust diagnostic capabilities, according to Dr. Sodha.
“I always hope that dermatologists can get comfortable and engaged in the field because the more we are able to help our non-dermatology colleagues develop great programs, the more comfortable we will be in utilizing these algorithms and devices in the future for our benefit and our patients’ benefit,” Dr. Sodha says. Â
Dr. Waibel conducted a clinical trial for Michaelson Diagnostics, maker of VivoSight, a medical scanner to help clinicians detect non-melanoma skin cancer.
1. Park A. Machines treating patients? It’s already happening. Time. March 21, 2019