A practical approach to Sasang constitutional diagnosis using vocal features.

Area of Science:

• Integrative and Complementary Medicine
• Biomedical Engineering
• Speech Science

Background:

• Sasang constitutional medicine (SCM) classifies individuals into four types (SC) for personalized treatment.
• Voice characteristics are key indicators for SC type diagnosis.
• Previous SCM voice analysis models suffered from limited data, long recording times, and low accuracy.

Purpose of the Study:

• To develop a practical and simplified SC classification model for SCM.
• To reduce model complexity and improve clinical applicability.
• To enhance the generalization performance of SC type diagnosis.

Main Methods:

• Utilized voice recordings from 2,341 participants.
• Extracted 21 vocal features from sentence recordings.
• Applied Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection and multinomial logistic regression for model development.

Main Results:

• The proposed model achieved classification accuracies of 47.9% for males and 40.4% for females on the test set.
• Demonstrated superior generalization performance compared to previous methods using more features.
• Required shorter voice recordings, enhancing practical usability.

Conclusions:

• A simplified SC classification model using fewer variables was developed and validated.
• The proposed method, incorporating LASSO, is suitable for clinical settings.
• The model's simplicity is expected to yield more stable and reliable SCM diagnostic results.