Knowledge-point classification using simple LSTM-based and siamese-based networks for virtual patient simulation.

Area of Science:

• Medical Education Technology
• Artificial Intelligence in Healthcare
• Computational Linguistics

Background:

• Medical education emphasizes developing critical thinking skills.
• Virtual Diagnosis and Treatment Platform (VP) uses simulated patient interactions to assess medical students' diagnostic abilities.
• Analyzing student questions during interviews provides insights into their medical history inquiry skills.

Purpose of the Study:

• To extract insights from case summaries and patient interviews for improved evaluation and feedback in medical education.
• To develop and evaluate advanced computational methods for analyzing student performance in virtual clinical settings.

Main Methods:

• Employed a systematic approach using Long Short-Term Memory (LSTM) and Siamese-based neural networks for knowledge-point classification.
• Utilized a dataset from "Clinical Diagnosis and Treatment Skills Competitions" (Years 1-3) in Taiwan.
• Generated knowledge points from sequential questions in case summaries and patient interviews for classification.

Main Results:

• The Siamese-based network achieved over 93% accuracy in knowledge-point classification.
• Stratified 10-fold cross-validation demonstrated high performance with a standard deviation below 0.007.
• Results confirm the effectiveness of proposed neural network methodologies for virtual clinical diagnosis systems.

Conclusions:

• Advanced neural networks, especially Siamese-based networks, are viable for knowledge-point classification in virtual clinical diagnosis.
• Effective classification of knowledge points offers valuable insights into students' thinking capabilities.
• The validated methodologies support the development of enhanced virtual clinical training platforms.