Diagnostic Method of Diabetes Based on Support Vector Machine and Tongue Images

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Traditional Chinese Medicine

Background:

• Diabetes mellitus is a global health concern requiring novel diagnostic tools.
• Traditional Chinese Medicine (TCM) utilizes tongue diagnosis for health assessment.
• Digitalization of TCM diagnostic methods offers potential for objective evaluation.

Purpose of the Study:

• To develop and validate a diabetes diagnostic method using standardized tongue images.
• To employ Support Vector Machine (SVM) for classification based on tongue features.
• To assess the feasibility of digitalized tongue diagnosis for diabetes.

Main Methods:

• Collected tongue images from 296 diabetic and 531 non-diabetic subjects using a TDA-1 digital tongue instrument.
• Separated tongue body and coating using division-merging and chrominance-threshold methods.
• Extracted color and texture features, trained SVM models, and optimized parameters using genetic algorithms (GA) and principal component analysis (PCA).

Main Results:

• Normalized tongue image parameters improved diabetes prediction accuracy from 77.83% to 78.77%.
• Principal Component Analysis (PCA) reduced feature dimensions without compromising accuracy (AUC) while saving training time.
• Genetic Algorithm (GA) optimization increased cross-validation accuracy to 83.06%, outperforming other algorithms.

Conclusions:

• A diabetes diagnostic method based on standardized tongue images is valuable and feasible.
• Digitalized tongue diagnosis shows significant potential for objective and accurate health assessment.
• This approach integrates TCM principles with modern machine learning for improved healthcare outcomes.