Automatic eczema classification in clinical images based on hybrid deep neural network

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Dermatology

Background:

• Artificial neural networks (ANNs) show promise in dermatology for disease detection.
• Eczema is a common skin condition requiring accurate diagnosis.
• Automated diagnostic tools are needed to improve healthcare services.

Purpose of the Study:

• To develop and evaluate a hybrid deep learning model for automated multi-class eczema diagnosis.
• To introduce the Eczema Image Resource (EIR) dataset for research.
• To compare the performance of various deep learning models and techniques for eczema classification.

Main Methods:

• A hybrid model combining ReliefF optimized handcrafted features and deep activated features with a support vector machine was developed.
• The Eczema Image Resource (EIR) dataset, comprising 2039 labeled images across seven eczema categories, was utilized.
• Comparative analysis included ensemble models, attention mechanisms, and data augmentation techniques.

Main Results:

• The proposed Hybrid 6 network achieved the highest accuracy (88.29%), sensitivity (85.19%), and specificity (90.33%).
• The developed deep learning models demonstrated high accuracy in classifying eczema from images.
• Performance was found to be comparable to that of dermatologists.

Conclusions:

• Deep learning models offer a viable approach for accurate eczema classification.
• The Hybrid 6 model shows significant potential as a clinical decision support tool.
• Further research is needed to address factors affecting accuracy and improve model performance.