Artificial intelligence tools improve skin cancer diagnostic accuracy

Artificial intelligence (AI)-driven tools can improve the skin cancer diagnostic accuracy of clinicians, nurse practitioners and medical students, according to a study published last week in npj Digital Medicine.

The researchers underscored that AI-based skin cancer diagnostic tools are developing rapidly, and these tools are likely to be deployed in clinical settings upon appropriate testing and successful validation.

However, the research team further noted that the significant promise of these models remains largely theoretical, as evidence to bolster the use of AI-enabled clinical decision support tools in skin cancer diagnosis is limited.

To address this, the researchers conducted a systematic review and meta-analysis to investigate the impact of AI assistance on skin cancer diagnostic accuracy.

Peer-reviewed articles evaluating AI-assisted skin cancer diagnosis published between January 1, 2017 and November 8, 2022 were pulled from PubMed, Embase, Institute of Electrical and Electronics Engineers Xplore (IEE Xplore) and Scopus for analysis. Of the 2,983 articles initially retrieved, only 12 were included in the systematic review and ten were included in the meta-analysis.

These studies contained over 67,000 assessments of potential skin cancers by a variety of practitioners – including medical students, primary care physicians and dermatologists – with and without AI assistance.

The researchers emphasized that AI tools play an assistive role for clinicians, rather than acting as a replacement for their expertise, leading the team to investigate how AI assistance impacts diagnostic performance.

“Previous studies have focused on how AI performs when compared with physicians,” explained Jiyeong Kim, PhD, a postdoctoral scholar at the Stanford Center for Digital Health, in a news release. “Our study compared physicians working without AI assistance with physicians using AI when diagnosing skin cancers.”

The research team indicated that previous studies have also shown that various factors – like a clinician’s degree of confidence in their own clinical decision, the degree of confidence of the AI tool and whether or not the clinician and the AI agree on the diagnosis – determine whether the clinician incorporates the algorithm’s advice into their clinical decision-making.

“We want to better understand how humans interact with and use AI to make clinical decisions,” Kim said.

The meta-analysis and review revealed that overall, healthcare practitioners across all training levels and specialties benefited from the use of AI tools.

Practitioners not using AI were able to accurately diagnose 74.8 percent of skin cancer cases and correctly flag 81.5 percent of patients with cancer-like skin conditions who did not have cancer. Those working with aid from AI correctly identified 81.1 percent of skin cancer cases and 86.1 percent of cancer-like skin conditions.

To gain more insights into which practitioners benefit most from the use of AI, the researchers performed subgroup analyses. These showed that all practitioners benefitted from these tools, but the largest improvements were seen among non-dermatologists.

Medical students, nurse practitioners and primary care doctors saw the biggest boost, improving about 13 points in sensitivity and 11 points in specificity, on average, with AI assistance. Dermatologists and dermatology residents performed better than their colleagues in general with and without AI, but their diagnostic performance also improved with AI-enabled clinical decision support.

The researchers noted that these findings highlight the potential of AI in imaging-heavy medical specialties like dermatology and radiology.

“This is a clear demonstration of how AI can be used in collaboration with a physician to improve patient care,” said Eleni Linos, MD, director of the Center for Digital Health and professor of dermatology and epidemiology at Stanford.

“If this technology can simultaneously improve a doctor’s diagnostic accuracy and save them time, it’s really a win-win. In addition to helping patients, it could help reduce physician burnout and improve the human interpersonal relationships between doctors and their patients,” Linos continued. “I have no doubt that AI assistance will eventually be used in all medical specialties. The key question is how we make sure it is used in a way that helps all patients regardless of their background and simultaneously supports physician well-being.”

This research is one of a host of efforts investigating how advanced analytics tools can enhance cancer care.

This week, a research team from the University of Pittsburgh Medical Center (UPMC) detailed how a predictive model can help forecast metastatic uveal melanoma patients’ response to adoptive therapy – a type of immunotherapy in which a patient’s T-cells are extracted, multiplied in a laboratory and reinfused.

Uveal melanoma is resistant to standard immunotherapies, resulting in poor prognoses for many patients once the cancer metastasizes. Previous research showed that adoptive therapy is successful in some patients, allowing tumor-infiltrating lymphocytes (TILs) to activate and attack tumor cells.

To identify which patients are likely to respond well to this type of therapy, the researchers designed the Uveal Melanoma Immunogenic Score (UMIS), which is designed to measure the activity of genes expressed by cells in the tumor microenvironment to forecast treatment success.