Intelligent skin disease prediction system using transfer learning and explainable artificial intelligence.

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Skin diseases present a global health challenge, requiring laborious diagnostic processes.
• Accurate identification and prediction of skin conditions are complicated by disease complexity and numerous clinical features.
• Current diagnostic methods for skin diseases are time-consuming and require extensive clinical and histological data.

Purpose of the Study:

• To develop a rapid and accurate deep learning model for identifying common skin diseases.
• To leverage transfer learning for efficient disease prediction using VGG16 architecture.
• To enhance the interpretability of deep learning models in healthcare using layer-wise relevance propagation (LRP).

Main Methods:

• A deep learning model utilizing a pre-trained VGG16 for transfer learning was employed.
• A dataset comprising skin images of chickenpox, measles, monkeypox, and normal skin was curated and split for training and testing.
• Layer-wise relevance propagation (LRP) was applied to interpret the model's predictions and identify relevant visual features.

Main Results:

• The VGG16 transfer learning model achieved a testing accuracy of 93.29% in identifying skin diseases.
• The application of LRP provided relevance scores, highlighting specific visual symptoms crucial for diagnosis.
• The interpretable results from LRP offer valuable insights to support clinical decision-making.

Conclusions:

• Deep learning models, particularly VGG16 with transfer learning, show high accuracy in diagnosing skin diseases.
• Layer-wise relevance propagation (LRP) effectively addresses the 'black box' nature of deep learning, enhancing model transparency.
• The integration of interpretable AI can significantly aid healthcare professionals in timely and informed diagnosis and treatment of skin conditions.