AI may be less accurate than clinicians in usage for management decisions
By Lori Solomon HealthDay Reporter
WEDNESDAY, Nov. 29, 2023 (HealthDay News) — A mobile phone-powered artificial intelligence (AI) technology is effective for diagnosing suspicious skin lesions, according to a study published in the October issue of The Lancet Digital Health.
Scott W. Menzies, Ph.D., from the University of Sydney, and colleagues compared the accuracy of AI algorithms and clinicians for the diagnosis and management of pigmented skin lesions. For the diagnostic study, the analysis included 172 suspicious pigmented lesions from 124 patients, while the management study included 5,696 pigmented lesions from 66 high-risk patients.
The researchers found that the diagnoses of the seven-class AI algorithm were similar to the specialists’ diagnoses (absolute accuracy difference, 1.2 percent) and significantly superior to the novice clinicians’ diagnoses (21.5 percent). Diagnoses based on the International Skin Imaging Collaboration AI algorithm were significantly inferior to the specialists’ diagnoses (–11.6 percent) but significantly superior to the novices’ diagnoses (8.7 percent). The best seven-class management AI was significantly worse than specialists’ management (absolute accuracy difference in correct management decision, –0.5 percent for dismissal and –0.4 percent for biopsy). For novice clinicians’ management, the seven-class management AI was significantly inferior (–0.4 percent for dismissal), but significantly superior (0.4 percent) for biopsy.
“An AI algorithm that was superior in experimental studies was significantly inferior to specialists in a real-world scenario, suggesting that caution is needed when extrapolating results of experimental studies to clinical practice,” the authors write.
The study was funded by MetaOptima Technology.
Copyright © 2023 HealthDay. All rights reserved.
