A hybrid deep learning framework for skin disease localization and classification using wearable sensors

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate skin disease detection is crucial for effective patient treatment.
• Existing methods may lack interpretability and struggle with multimodal data integration.

Purpose of the Study:

• To develop a patch-based, interpretable deep learning framework for skin disease detection.
• To integrate wearable sensor data with clinical information for improved diagnostic accuracy.

Main Methods:

• Utilized a fully convolutional residual neural network (FCRN) for feature extraction from high-resolution skin images.
• Implemented pre-processing techniques for data standardization.
• Employed a convolutional neural network (CNN) for multimodal fusion of image and clinical data.
• Generated interpretable disease probability maps.

Main Results:

• The framework achieved state-of-the-art performance across key metrics like accuracy, sensitivity, and specificity.
• Interpretable disease probability maps effectively highlighted affected skin regions.
• Demonstrated improved classification performance through multimodal data integration.

Conclusions:

• The proposed framework offers an efficient, scalable, and explainable approach to skin disease detection.
• Combining wearable sensor technology with deep learning shows significant potential for clinical applications.
• The interpretable nature of the model enhances transparency and clinical utility.