AI technology not fully explored, optimized in skin cancer detection
Click Here to Manage Email Alerts
Despite advances in artificial intelligence and machine learning, such technologies have not been optimized by dermatologists and other professionals for skin cancer detection, according to a study.
“Recent advances in artificial intelligence (AI) in dermatology have demonstrated the potential to improve the accuracy of skin cancer detection,” Eric J. Beltrami, BS, of the University of Connecticut School of Medicine, and colleagues wrote.
Specifically, the technology may allow for improved diagnostics and, consequently, skin cancer management.
Beltrami and colleagues aimed to define the fundamental terminology of AI in dermatology, along with potential benefits, limitations and relevant commercial applications.
They define AI as “the use of a computer to perform tasks that would otherwise require human intelligence and decision making.”
Machine learning is a type of AI which is marked by computers programmed to develop algorithms that can then be used to complete human tasks. Importantly, machine learning can be supervised by human programmers.
The next advance in machine learning are deep neural networks, or DNNs, according to the authors.
“In a DNN, nodes are arranged in multiple, hidden layers with machine learning occurring at each level, like layers of neurons in a brain,” they wrote.
The data are then analyzed, allowing for a trained algorithm to generate probabilities for a diagnosis for any individual skin lesion. DNNs were first applied in research into skin lesions in 2016.
While a number of smartphone apps and other technologies using DNNs have been investigated, there are only two products currently approved by the FDA. MelaFind (MELA Sciences) was approved in 2011 but discontinued for sale or clinical use in 2017. The product was associated with low specificity, resulting in excessive biopsies.
Another product, Nevisense (SciBase) has received premarket approval for melanoma detection for use only by dermatologists.
Beyond the lack of approved devices, dermatologists have expressed concern that AI technology may disrupt the dermatologist-patient relationship.
More research is needed to investigate the many potential products that could ultimately become FDA-approved. With more products on the market, non-dermatologists may be able to triage benign lesions and increase detection of potentially cancerous lesions, thereby easing the workload of dermatologists.
“A clearer understanding of the technology may help to reduce physician concerns about AI and promote its use in the clinical setting,” Beltrami and colleagues wrote. “Ultimately, the development and validation of AI technologies, their approval by regulatory agencies and widespread adoption by both dermatologists and other clinicians may enhance patient care. Technology-augmented detection of skin cancer has the potential to improve quality of life, reduce health care costs by reducing unnecessary procedures, and promote greater access to high quality skin assessment.”