Severance Hospital Research Team Develops Deep Learning-Based AI Diagnostic Support Model
Achieves Over 95% Accuracy in Distinguishing Melanoma from Benign Melanonychia

Technology has been developed that enables artificial intelligence (AI) to distinguish between melanoma, a form of skin cancer, and benign lesions presenting as black streaks on the nails with an accuracy of over 95%.


“Is It Cancer?” Black Streaks on Nails Easily Distinguished as Melanoma or Not View original image

On August 11, Professor Oh Byoung-ho of the Department of Dermatology at Severance Hospital, Yonsei University, and Dr. Chu Yoosung of the Department of Precision Medicine at Yonsei University Wonju College of Medicine announced that their research team has developed a deep learning-based AI-assisted diagnostic model capable of discriminating between nail melanoma and benign melanonychia.


Melanonychia, which appears as black streaks on the nails, is a relatively common condition, but in some cases, it may be an early sign of the potentially fatal skin cancer, nail melanoma. However, early-stage nail melanoma closely resembles benign lesions in appearance, making it difficult even for experienced dermatologists to distinguish between the two by visual inspection alone.


For an accurate diagnosis, a nail matrix tissue biopsy must be performed. However, the procedure may cause permanent nail deformity or lead to aesthetic and functional complications, creating obstacles to its implementation. As a result, in actual clinical practice, tissue biopsies are often postponed, with doctors instead monitoring changes over time, thus limiting opportunities for early diagnosis.


The research team developed a deep learning-based AI-assisted diagnostic model using clinical data from patients with nail melanoma and benign melanonychia. Clinical information was collected from a total of 294 patients, including 122 with nail melanoma and 172 with benign melanonychia. To reduce diagnostic confusion, childhood melanonychia cases were excluded from AI training. Furthermore, to improve the reliability of the AI model, patient data were split at the patient level so that photos of the same patient would not be used for both training and evaluation.


“Is It Cancer?” Black Streaks on Nails Easily Distinguished as Melanoma or Not View original image

As a result, the AI model demonstrated high diagnostic accuracy, achieving over 95% accuracy (AUROC 0.954) in distinguishing between nail melanoma and benign melanonychia. The model also maintained excellent performance during external validation using patient data from other medical institutions.


The research team also confirmed that the AI could be helpful in actual clinical settings. When dermatologists and residents first diagnosed the same cases without AI assistance and then re-diagnosed after referencing the AI results, overall diagnostic accuracy improved from 70% to 80.8%. The greatest improvement in accuracy was observed among residents, and consistency between clinicians' diagnoses also improved.


Professor Oh Byoung-ho stated, "It is significant that this research demonstrated the potential of diagnostic support technology that can be practically applied in clinical settings," and added, "We hope that early diagnosis of nail melanoma using this technology will improve patient survival rates and reduce unnecessary biopsies."



The results of this study were published in the latest issue of the international journal 'Journal der Deutschen Dermatologischen Gesellschaft.'


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing