Transformer- and Generative Adversarial Network-Based Inpatient Traditional Chinese Medicine Prescription...

Area of Science:

• Artificial Intelligence in Medicine
• Computational Traditional Chinese Medicine
• Digital Health

Background:

• Traditional Chinese Medicine (TCM) diagnosis relies on complex physician reasoning from patient data in electronic health records (EHRs).
• Mastering TCM diagnostic reasoning requires extensive clinical experience, posing a challenge for new practitioners.
• Advancements in AI and computing power enable the development of systems to assist in TCM decision-making.

Purpose of the Study:

• To develop an AI-powered assistive tool for predicting Traditional Chinese Medicine (TCM) prescriptions.
• To leverage patient clinical EHR data for automated prescription recommendations.

Main Methods:

• A transformer-based neural network was trained on chronological patient EHR data, including medical history, symptoms, and treatments.
• Extracted EHR information encompassed current illness, medications, nursing care, vital signs, and lab results.
• Generative adversarial networks (GANs) were employed to augment training data and mitigate model overfitting.

Main Results:

• The study utilized 21,295 inpatient EHRs from Guang'anmen Hospital (2017-2018), covering 6352 medicine types.
• The transformer model demonstrated an average precision rate of 80.58% and an average recall rate of 68.49%.
• The model successfully predicted prescriptions based on comprehensive patient data.

Conclusions:

• The transformer-based TCM prescription recommendation model shows superior performance compared to conventional methods.
• Generative adversarial networks effectively addressed overfitting, enhancing the model's recall and precision.
• This AI tool has the potential to assist physicians in generating accurate TCM prescriptions.