Tongue color clustering and visual application based on 2D information.

Area of Science:

• Medical Imaging
• Computational Biology
• Machine Learning

Background:

• Tongue color variations are associated with various diseases.
• Objective and rapid clinical decision-making requires reliable diagnostic tools.
• Current methods for tongue color analysis may lack objectivity and speed.

Purpose of the Study:

• To apply unsupervised learning for the 2D clustering of tongue colors.
• To develop a method for objective and accurate tongue color-based diagnosis.
• To assist clinicians in making faster and more precise diagnostic decisions.

Main Methods:

• Analysis of 595 typical tongue images.
• Transformation of 3D tongue image information into 2D using Principal Component Analysis (PCA).
• Clustering of tongue images into four diagnostic groups using K-Means algorithm.
• Evaluation of clustering performance using Clustering Accuracy (CA), Jaccard Similarity Coefficient (JSC), and Adjusted Rand Index (ARI).

Main Results:

• The 2D transformation retained 89.63% of the original information in the L*a*b* color space.
• Successful classification of tongue images into four distinct clusters.
• Achieved high performance metrics: CA of 89.04%, ARI of 0.721, and JSC of 0.890.

Conclusions:

• 2D tongue color information is effective for clustering and enhancing visualization.
• The combination of K-Means and PCA provides a viable approach for tongue color classification and diagnosis.
• The proposed methods offer a potential reference for image-based diagnosis technologies.