Automatic Acne Severity Grading with a Small and Imbalanced Data Set of Low-Resolution Images.

Area of Science:

• Dermatology
• Machine Learning
• Computer Vision

Background:

• Developing automated acne vulgaris grading systems is data-intensive and costly.
• Traditional methods require high-resolution images, balanced datasets, and extensive labeling.
• Acne grading systems face challenges in data acquisition and labeling efficiency.

Purpose of the Study:

• To develop a deep learning model for acne severity grading using the Investigator's Global Assessment (IGA) scale.
• To train a model on low-resolution images with minimal data and imbalanced severity grades.
• To reduce the cost and complexity of creating automated acne grading tools.

Main Methods:

• A deep learning model was trained and validated using 1374 image triplets from 391 acne patients.
• Images included frontal and lateral views, labeled with IGA severity grades by a dermatologist.
• The model was designed to handle low-resolution images and imbalanced data distributions.

Main Results:

• The model achieved 66.67% accuracy on the test set.
• Performance was consistent across all acne severity grades, despite data imbalance.
• The developed method matched the performance of more data-intensive approaches.

Conclusions:

• The deep learning model shows promising accuracy with limited data.
• It has potential as an assistive tool for medical practitioners.
• The model can provide patients with a standardized, readily available acne grading solution.