Artificial Intelligence-Based Traditional Chinese Medicine Assistive Diagnostic System: Validation Study

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Healthcare
• Traditional Chinese Medicine Diagnostics

Background:

• AI diagnostic systems show promise but are underutilized in Traditional Chinese Medicine (TCM) due to complex syndrome differentiation.
• TCM syndrome prediction is challenging due to many-to-many relationships between diseases and syndromes.
• Existing AI models for TCM are limited to single disease types.

Purpose of the Study:

• Develop an AI-assisted diagnostic system for multiple common TCM diseases using electronic health record (EHR) notes.
• Simultaneously diagnose diseases and predict corresponding TCM syndromes.
• Overcome limitations of previous AI applications in TCM diagnostics.

Main Methods:

• Processed unstructured EHR notes using Natural Language Processing (NLP) with a bidirectional long short-term memory network-conditional random forest model.
• Employed a Convolutional Neural Network (CNN) for predicting one of 187 TCM disease types.
• Utilized a novel integrated learning model for syndrome prediction, combining four existing methods via majority-rule voting.

Main Results:

• The system was trained and tested on 22,984 EHRs from Guanganmen Hospital (2017-2018).
• Achieved high diagnostic accuracy for 187 TCM diseases, with top-1, top-3, and top-5 accuracies of 80.5%, 91.6%, and 94.2%, respectively.
• Demonstrated strong generalization capability on the test dataset.

Conclusions:

• An AI-based assistive diagnostic system for 187 TCM diseases was successfully developed.
• A novel integrated learning model for TCM syndrome prediction demonstrated superior performance over individual methods.
• Future work includes algorithm refinement and increased data for broader TCM disease coverage and improved accuracy.