KIEGLFN: A unified acne grading framework on face images

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Acne severity grading is crucial for diagnosis and treatment.
• Current acne grading criteria lack standardization in the medical field.

Purpose of the Study:

• To develop a generalized acne diagnosis system applicable to various grading criteria.
• To create a system that mimics dermatologists' diagnostic process.

Main Methods:

• Developed a unified acne grading framework with adaptive image preprocessing.
• Employed a novel network structure fusing global and local features.
• Utilized a transfer fine-tuning strategy to adapt the model to different criteria.

Main Results:

• The preprocessing method effectively filters noise and enhances relevant features.
• The framework achieved accuracies of 84.52% and 59.35% on two distinct datasets.
• The system surpassed the state-of-the-art by 1.71%.

Conclusions:

• The developed framework achieves dermatologist-level diagnostic accuracy for acne.
• The transfer fine-tuning strategy significantly improves accuracy (6.5%) on small datasets.
• This system offers a standardized and generalized approach to acne grading.