Skin Parameter Scoring Using Artificial Intelligence: A Case Series

Evaluation of facial skin parameters is important for accurate diagnosis and computerized systems can reduce subjectivity. We evaluated the feasibility of an application-based facial analysis artificial intelligence (AI) system by comparing its output to a human dermatologist evaluator.

Both an application-based AI and human dermatologist assessed the severity of 14 parameters (age spots, eye bags, moisture, oiliness, redness, texture, wrinkles, acne, droopy lower eyelid, droopy upper eyelid, firmness, dark circles, radiance, pores). The AI generated a raw score between 0 and 100 (100 = flawless) that was converted to match the human rating between 0 and 4 (4 = very severe). Differences in ratings were calculated to determine the degree of agreement. A 0-point difference is defined as complete agreement, a 1-point difference is good, a 2-point difference is poor, a 3-point difference is very poor, and a 4-point difference is worst.

In total, 46 facial clinical photos (13 male, 33 female) taken indoors under fluorescent lighting with and without flash were assessed. Parameters with high rates (> 90%) of complete or good agreement were redness (100%), eye bags (96%), and wrinkles (96%). Parameters with moderate rates (80-90%) of agreement were radiance (89%), dark circles (84%), droopy lower eyelid (82%), oiliness (80%), acne (80%), and pores (80%). Parameters with low rates (< 80%) of agreement were texture (78%), age spots (78%), droopy upper eyelid (77%), moisture (72%), and firmness (65%).

In a preliminary study of one neural-network AI, its evaluation of skin parameters had an 83% overall complete or good agreement with a human rater. While software-based AI facial analysis systems may lack hardware optimization, they are simple to use and convenient to implement on a variety of hand-held devices. Low start-up costs, rapid data collection, and easy use can be advantages over higher cost, desktop-based facial analysis devices.