Tongue Image Texture Classification Based on Image Inpainting and Convolutional Neural Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Traditional Chinese Medicine

Background:

• Tongue texture analysis is crucial for Traditional Chinese Medicine (TCM) diagnosis, particularly for classifying tongue toughness and tenderness.
• Texture discontinuity in tongue images, often due to tongue coating, negatively impacts classification accuracy.
• Existing methods struggle with the inherent complexities of tongue image texture.

Purpose of the Study:

• To develop a robust and accurate tongue image texture classification method.
• To address the challenge of texture discontinuity in tongue body images.
• To enhance the classification of tough and tender tongues using advanced AI techniques.

Main Methods:

• A novel method combining image inpainting and convolutional neural networks (CNNs) for tongue texture classification.
• Utilizing Gaussian mixture models to segment tongue coating from the tongue body.
• Employing a generative image inpainting model with contextual attention to restore tongue body continuity.
• Implementing a ResNet101 residual network for the final classification of inpainting images.

Main Results:

• The proposed method effectively separates tongue coating and body, ensuring continuity of the tongue body image.
• Image inpainting significantly reduces the interference of tongue coating on texture analysis.
• The ResNet101-based classifier achieved superior performance in tough and tender tongue classification compared to existing methods.
• Demonstrated improved accuracy and robustness in tongue texture analysis.

Conclusions:

• The integration of image inpainting with CNNs offers a promising approach for accurate tongue texture analysis in TCM.
• This method provides a new strategy for overcoming texture discontinuity challenges in tongue image classification.
• The findings highlight the potential of AI-driven image processing to advance TCM diagnostic capabilities.