Related Experiment Video
Updated: Oct 7, 2025
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
2.7K
An Improved Deep Learning Model: S-TextBLCNN for Traditional Chinese Medicine Formula Classification
Ning Cheng1, Yue Chen1, Wanqing Gao1
1School of Informatics, Hunan University of Chinese Medicine, Changsha, China.
Frontiers in Genetics
|January 10, 2022
Summary
This study introduces the S-TextBLCNN model for classifying traditional Chinese medicine (TCM) formula efficacy. The model effectively links herb properties to formula outcomes, improving classification accuracy and offering new insights into TCM formula compatibility.
Area of Science:
- Computational Medicine
- Pharmacology
- Artificial Intelligence
Background:
- Traditional Chinese Medicine (TCM) formula classification relies on understanding complex herb-formula relationships.
- Existing methods may not fully capture the intricate connections between individual herb efficacies and the overall efficacy of a TCM formula.
- The quantitative expression and analysis of TCM herb and formula data present a significant challenge.
Purpose of the Study:
- To propose and evaluate the S-TextBLCNN model for accurate classification of TCM formula efficacy.
- To investigate the relationship between herb efficacy and formula efficacy using deep learning.
- To provide a novel approach for exploring the internal rules of TCM formula combinations.
Main Methods:
- Natural language processing (NLP) was employed to quantitatively encode TCM herbs based on name, properties, and efficacy.
- A deep learning model, TextBLCNN (combining Bi-LSTM and CNN), was developed for formula classification using 2,664 stroke-related formulae.
- The Synthetic Minority Over-sampling Technique (SMOTE) was integrated to address data imbalance, resulting in the S-TextBLCNN model.
Main Results:
- Formula vectors derived from herb efficacy demonstrated the strongest performance in classification, highlighting a significant herb-formula efficacy link.
- The S-TextBLCNN model achieved an accuracy of 0.858 and an F1-score of 0.762, outperforming logistic regression, SVM, LSTM, and TextCNN.
- SMOTE significantly improved the F1-score by an average of 47.1% across 19 classifiers, effectively handling data imbalance.
Conclusions:
- The integration of TCM formula feature representation with the S-TextBLCNN model enhances formula efficacy classification accuracy.
- The study confirms a strong correlation between herb efficacy and overall formula efficacy.
- This research offers a new perspective for studying TCM formula compatibility and rules.
