Development and accuracy of an artificial intelligence algorithm for acne grading from smartphone photographs

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Facial acne affects a significant patient population.
• Accurate assessment of acne severity and lesion type is crucial for effective treatment.
• Current methods for acne evaluation can be subjective and time-consuming.

Purpose of the Study:

• To develop and validate an artificial intelligence algorithm (AIA) for smartphones.
• To enable objective determination of facial acne severity using the GEA scale.
• To identify and classify acne lesions (comedonal, inflammatory) and postinflammatory hyperpigmentation (PIHP).

Main Methods:

• Collected 5972 smartphone images from 1072 acne patients.
• Trained the AIA using expert dermatologist assessments of acne severity (GEA scale) and lesion identification from tagged images.
• Iteratively refined the algorithm (6 versions) based on sensibility, specificity, and correlation with expert evaluations.

Main Results:

• The final AIA version achieved 68% accuracy in GEA grading, comparable to dermatologists.
• Significant improvements in precision and recall were observed between AIA versions 4 and 6 for inflammatory lesions, non-inflammatory lesions, and PIHP.
• The F1 score reached 84% for inflammatory lesions, 61% for non-inflammatory lesions, and 72% for PIHP.

Conclusions:

• The developed AIA demonstrates a reliable and objective method for assessing facial acne severity and characteristics via smartphone.
• This AI-powered tool has the potential to assist dermatologists and improve acne management.
• Further validation and integration into clinical practice could enhance patient care for acne.