TCMPR: TCM Prescription Recommendation Based on Subnetwork Term Mapping and Deep Learning

Area of Science:

• Computational medicine
• Artificial intelligence in healthcare
• Traditional Chinese Medicine (TCM)

Background:

• Traditional Chinese Medicine (TCM) is vital for clinical diagnosis and treatment.
• Current AI-driven prescription recommendations struggle with complex and individualized patient symptoms.
• Representing unrecorded symptom terms in knowledge bases remains a significant challenge.

Purpose of the Study:

• To propose a subnetwork-based symptom term mapping (SSTM) method for improved TCM prescription recommendation.
• To develop an SSTM-based TCM prescription recommendation (TCMPR) system.
• To enhance the representation of clinical symptom embeddings, particularly for unrecorded terms.

Main Methods:

• Developed a subnetwork-based symptom term mapping (SSTM) method.
• Extracted subnetwork structures from knowledge networks to represent symptom embeddings.
• Constructed an SSTM-based TCM prescription recommendation (TCMPR) method.

Main Results:

• The proposed SSTM method effectively represents clinical symptom terms, including unrecorded ones.
• The TCMPR method demonstrated superior performance compared to state-of-the-art approaches.
• Comprehensive experiments confirmed the high performance and robustness of the TCMPR method across various hyperparameters.

Conclusions:

• The SSTM method offers a robust way to represent complex clinical symptom data for TCM.
• The TCMPR system significantly improves personalized TCM prescription recommendations.
• This approach has the potential to advance precision medicine in TCM clinical practice.