A hybrid model for improved nail disease classification using vision transformers and stable diffusion.

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Nail diseases are common and often diagnosed through visual inspection, which can lead to inaccuracies and treatment delays.
• Accurate and timely diagnosis is crucial for effective management of nail conditions such as fungal infections, paronychia, and psoriasis.

Purpose of the Study:

• To investigate the use of synthetic nail disease data generated via stable diffusion models to improve machine learning diagnostic accuracy.
• To enhance data transformation techniques using few-shot learning within a text-to-image stable diffusion model for generating diverse synthetic data.

Main Methods:

• Generation of synthetic nail disease data using stable diffusion models with few-shot learning.
• Application of synthetic data to augment a custom real-world nail disease dataset.
• Evaluation of pre-trained Convolutional Neural Network (CNN) MobileNetV2 and Vision Transformer models with the augmented dataset.

Main Results:

• Synthetic data improved the robustness of both MobileNetV2 and Vision Transformer models.
• MobileNetV2 achieved an accuracy increase of 3.26%.
• Vision Transformer achieved an accuracy increase of 3.02%.

Conclusions:

• Synthetic data generated by stable diffusion models is effective in enhancing the performance of machine learning models for nail disease classification.
• This approach offers a promising method to improve the accuracy and reliability of automated nail disease diagnosis, potentially reducing diagnostic errors and treatment delays.