News Release

Developed an AI-based classification system for facial pigmented lesions

Supporting laser treatment decisions: Potential for more optimized treatment

Peer-Reviewed Publication

Kindai University

White balance correction and ROI extraction for CNN training

image: 

A typical image preprocessing workflow. The facial image is corrected for white balance with Python's OpenCV. A square region of interest (ROI) containing the main lesion is cropped. ROIs from five lesion types (ADM, Eph, Mel, Sen, LM/LMM) still show natural variation in skin tone and pigmentation. Only images with sufficient clarity (Laplacian variance ≥ 10) proceed for CNN analysis.

view more 

Credit: Professor Atsushi Otsuka from Kindai University Faculty of Medicine, Japan

A research team led by Haruyo Yamamoto, Chisa Nakashima, and Atsushi Otsuka from Department of Dermatology, Kindai University Faculty of Medicine, in collaboration with the Faculty of Engineering at Kindai University and other institutions, has developed a diagnostic system that uses artificial intelligence (AI) to accurately identify the type of facial pigmented lesions and support laser treatment decisions. A paper on this study was published online in Cureus, an international medical journal on June 5, 2025.

 

1. Key Points

  • Demonstrated superior diagnostic accuracy compared to dermatologists, when identifying five types of lesions, confirming the usefulness of the system in early detection and treatment decisions.
  • Developed a system that uses AI to accurately classify and assist in the diagnosis of five types of facial pigmented lesions that are difficult to diagnose.
  • Contributes to establishing a method that accurately identifies pigmented lesions, reduces the risk of mistreatment, and supports appropriate treatment decisions.

 

2. Research Background

Facial pigmented lesions, come in many different types, such as melasma, ephelides, acquired dermal melanocytosis, solar lentigo, and lentigo maligna melanoma, but they are often visually similar, which makes differential diagnosis challenging. On the other hand, an appropriate treatment for these lesions varies greatly depending on the type, and accurate diagnosis is essential as this directly affects the feasibility and selection of laser treatment. For example, inappropriate laser use can exacerbate melasma, and delaying necessary surgical excision for lentigo maligna melanoma due to misdiagnosis can have severe consequences. In recent years, imaging diagnostic technology using deep learning models has attracted attention, and research findings indicated that it has accuracy equal to or superior to that of dermatologists in differentiating among skin lesions. While deep learning-based image diagnosis has been successful in detecting melanoma, there has been insufficient research into benign and malignant pigmented lesions on the face that are directly related to laser treatment planning, and therefore there is a need for the development of a diagnostic support system.

 

3. Content

The research team developed a classification system using deep learning models, InceptionResNetV2 and DenseNet121, to identify five types of facial pigmented lesions (melasma, ephelides, acquired dermal melanocytosis, solar lentigo, and and lentigo maligna/lentigo maligna melanoma). Training and validation were performed using 432 clinical images, and their diagnostic accuracies were compared to the diagnoses of 9 board-certified dermatologists (experts) and 11 noncertified dermatologists (non-experts). Both models demonstrated diagnostic accuracies of 87% and 86%, respectively. Both models overperformed the median diagnostic accuracy of 80% for experts and 63% for non-experts. Especially in identifying LM/LMM, both models achieved 100% sensitivity, suggesting its potential use as a diagnostic support tool in clinical practice.

Based on these results, the developed deep learning models far surpass the accuracy of dermatologists in diagnosing facial pigmented lesions and may contribute to diagnostic support and determining appropriate treatment plans.


Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.