AI Diagnostic Accuracy Comparable to Experts in Diagnosing Skin Lesions, but Human Expertise Outperforms in Treatment Recommendations
Artificial intelligence (AI) has made significant strides in medical diagnostics, and dermatology is no exception. Researchers led by dermatologist Harald Kittler from MedUni Vienna conducted a study to evaluate the practicality of AI in diagnosing and recommending treatment for pigmented skin lesions. Published in The Lancet Digital Health, the study compared the performance of AI algorithms in smartphone applications with that of medical professionals in a clinical setting.
The study focused on assessing the diagnostic accuracy of two AI algorithms in smartphone applications and comparing them to the expertise of doctors. The results revealed that AI applications generally performed well in diagnosing pigmented skin lesions. However, when it came to treatment recommendations, doctors outperformed the AI.
To conduct the study, the AI applications were tested in realistic clinical conditions at the University Department of Dermatology at MedUni Vienna and the Sydney Melanoma Diagnostic Center in Australia. Two distinct scenarios were utilized, each employing a different AI-based smartphone application: a novel 7-class AI algorithm and a previously used ISIC algorithm in retrospective preliminary studies.
In Scenario A, the 7-class AI algorithm exhibited diagnostic accuracy equivalent to that of medical experts, surpassing less experienced physicians. On the other hand, the ISIC algorithm performed worse than experts but better than inexperienced users. These findings indicate that AI-assisted smartphone applications can make diagnostic decisions comparable to experts in real clinical settings. However, in terms of treatment decisions, the 7-class algorithm was significantly less accurate than the experts but performed better than the less experienced physicians.
Despite the promising diagnostic capabilities of AI, Harald Kittler cautioned against relying solely on AI applications for treatment recommendations. He noted that AI applications often tend to suggest removing more benign lesions than experts would, potentially leading to unnecessary evaluation and treatment. While AI can be a valuable tool in skin lesion diagnosis, it should be used critically and in conjunction with human expertise.
In conclusion, this study underscores the potential of AI in diagnosing pigmented skin lesions, demonstrating its effectiveness in a clinical setting. However, it emphasizes the need for a cautious approach, particularly in treatment recommendations where human expertise currently outperforms AI. Integrating AI as a supplementary diagnostic tool alongside experienced medical professionals may offer the best approach to benefit patients while minimizing unnecessary procedures.
For individuals concerned about skin cancer, it is worth exploring other studies that examine the link between eating fish and a higher risk of skin cancer, as well as how the Mediterranean diet could potentially lower the risk of skin cancers. These studies provide further insights into maintaining good skin health.
In summary, AI algorithms have demonstrated comparable diagnostic accuracy to experts in diagnosing skin lesions. However, treatment recommendations are an area where human expertise currently outperforms AI. The study highlights the need for cautious use of AI in dermatology, ensuring that it is employed alongside the expertise of medical professionals. By doing so, patients can receive the best possible care while minimizing unnecessary procedures and evaluations.
