Development of a Novel Scar Screening System with Machine Learning.

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Hypertrophic scars and keloids cause significant functional and cosmetic issues.
• Patient and physician education on scar management is crucial due to low treatment seeking.
• A need exists for accessible and accurate scar screening tools.

Purpose of the Study:

• To develop a computer vision algorithm for scar screening.
• To compare the diagnostic accuracy of the algorithm with that of physicians.
• To create a system involving healthcare providers and patients for scar assessment.

Main Methods:

• Utilized 3768 digital scar images from various sources.
• Employed Google Cloud AutoML Vision for image analysis and labeling.
• Compared algorithm-generated diagnoses against physician consensus and individual physician diagnoses.

Main Results:

• The algorithm achieved an average precision of 80.7% and recall of 71%.
• The algorithm demonstrated 77% accuracy, outperforming the average physician accuracy of 68.7%.
• The algorithm's Cohen kappa coefficient (0.69) indicated better agreement than physicians (0.59).

Conclusions:

• A machine learning algorithm can effectively diagnose four scar types.
• Future iterations can be integrated into telehealth and digital imaging platforms.
• This AI-powered scar screening system offers valuable support for physicians and patients.