OCT in dermatology: a process for determining whether a fully diversified dataset is needed for AI model-building

Area of Science:

• Dermatology
• Biomedical Imaging
• Artificial Intelligence

Background:

• Optical coherence tomography (OCT) offers depth penetration for skin pathology detection.
• Artificial intelligence (AI) model development requires extensive datasets of healthy and diseased skin.
• Variability in healthy skin across age and skin type poses challenges for AI model training.

Purpose of the Study:

• To analyze the variation of the attenuation coefficient in healthy skin across different skin types and ages.
• To establish a baseline understanding of healthy skin parameters for AI model development.
• To streamline the data acquisition process for AI-based OCT skin imaging.

Main Methods:

• Selected 100 subjects with Fitzpatrick skin types I-V and ages 13-83.
• Analyzed the attenuation coefficient in the dermis and epidermis, and in sun-exposed and sun-protected areas.
• Performed statistical analysis to determine significance of variations across age, skin type, and location.
• Calculated 95% confidence intervals for the attenuation coefficient.

Main Results:

• The study quantified variations in the skin attenuation coefficient across diverse demographics and anatomical locations.
• Statistical significance was determined for comparisons between skin layers and sun-exposed/protected areas.
• Confidence intervals were established for the attenuation coefficient across different ages and skin types.
• Identified key parameters and their variations in healthy skin relevant for AI model training.

Conclusions:

• Pre-analyzing healthy skin features using parameters like the attenuation coefficient is crucial for effective AI model building.
• This approach provides a roadmap for optimizing subject recruitment and image acquisition for AI-based OCT.
• The findings are expected to accelerate the clinical translation of AI-powered OCT for skin diagnostics.