Multi-Task and Few-Shot Learning-Based Fully Automatic Deep Learning Platform for Mobile Diagnosis of Skin Diseases

Area of Science:

• Biomedical Imaging
• Artificial Intelligence in Dermatology
• Mobile Health Technology

Background:

• Fluorescence imaging offers superior molecular detail for skin disease diagnosis compared to RGB imaging.
• Smartphone advancements enable portable optical imaging systems for mobile healthcare.
• Current deep learning algorithms for skin disease diagnosis using RGB images show limited performance.

Purpose of the Study:

• To develop an advanced deep learning network, the fluorescence-aided amplifying network (FAA-Net), for improved skin disease diagnosis.
• To integrate RGB and fluorescence imaging capabilities into a multi-modal smartphone system.
• To enhance diagnostic accuracy by addressing challenges like limited image data and identifying disease-specific regions.

Main Methods:

• Developed a multi-modal smartphone imaging system capturing both RGB and fluorescence images.
• Created FAA-Net, incorporating a meta-learning algorithm to handle insufficient image data.
• Integrated a novel attention-based module within FAA-Net to automatically locate and emphasize potential skin disease areas.
• Conducted a clinical trial to evaluate FAA-Net against state-of-the-art models.

Main Results:

• The developed FAA-Net model demonstrated an 8.61% improvement in mean accuracy for skin disease classification.
• The area under the curve (AUC) for skin disease classification showed a 9.83% improvement with FAA-Net compared to other advanced models.
• Clinical trial results validated the superior performance of the multi-modal system and FAA-Net.

Conclusions:

• The FAA-Net, combined with a multi-modal smartphone imaging system, significantly enhances the accuracy of skin disease diagnosis.
• Meta-learning and attention mechanisms in FAA-Net effectively address data limitations and improve disease localization.
• This approach represents a promising advancement for AI-powered mobile healthcare solutions in dermatology.