South Korean researchers have developed an artificial intelligence tool designed to help detect skin cancer that arises in the nails at an early stage.
Severance Hospital said on the 14th that a joint research team led by Oh Byung-ho, a professor of dermatology, and Chu Yu-sung, a doctor in the department of precision medicine at Yonsei University's Wonju College of Medicine, had developed a diagnostic support system that uses deep-learning-based AI to distinguish nail melanoma from benign melanonychia.
Melanonychia is a condition in which black or brown vertical streaks appear on the nails. It occurs relatively commonly, caused by melanocytes producing excess pigment when stimulated, or by trauma, pregnancy, fungal infection and other factors. Benign cases require no special treatment. Some, however, can be an early sign of melanoma, a deadly form of skin cancer, so the two must be told apart. The problem is that early nail melanoma closely resembles benign lesions in appearance. Even experienced dermatologists sometimes struggle to distinguish them by sight alone.
Accurately diagnosing melanoma in the nails requires a biopsy of the matrix, the tissue at the root of the nail. Because this involves cutting the nail deep near the root, patients endure considerable pain during the procedure, which can leave lasting complications such as permanent nail deformity. For these reasons, biopsies were often postponed in favor of watchful waiting.
The team collected clinical data from a total of 294 patients — 122 with nail melanoma and 172 with benign melanonychia — and trained the AI on the data to complete the diagnostic support model. To reduce diagnostic confusion, the researchers excluded pediatric melanonychia from the AI training, and they separated the data so that images of the same patient were not used in both training and evaluation, improving the model's reliability.
According to the analysis, the model showed high accuracy of more than 95% in distinguishing nail melanoma from benign melanonychia. It also performed well in external validation using patient data from other medical institutions.
The team had dermatology specialists and residents diagnose the same cases without AII, then compared those results with diagnoses that drew on the AI's output. Their diagnostic accuracy rose from 70% to 80.8%. The gain was especially large among dermatology residents, and agreement between clinicians also improved. The findings demonstrated that AI can help diagnose nail melanoma in actual clinical practice.
"This study points to the potential of AI diagnostic support technology that can be used in real clinical settings," Oh said. "Going forward, early diagnosis of nail melanoma could help improve patient survival rates and reduce unnecessary biopsies."
The findings were published in the latest issue of the international journal of the German dermatological society.
