A Natural Language Processing-Based Virtual Patient Simulator and Intelligent Tutoring System for the Clinical...

Area of Science:

• Medical Education Technology
• Artificial Intelligence in Healthcare
• Clinical Diagnostic Reasoning Training

Background:

• Rising educational costs and social distancing measures necessitate innovative distance learning solutions in medical training.
• Virtual patient simulators (VPSs) offer a potential solution for remote clinical education.
• Integrating Natural Language Processing (NLP) and Intelligent Tutoring Systems (ITSs) can enhance the effectiveness of VPSs.

Purpose of the Study:

• To develop a VPS incorporating natural language interaction and an ITS for clinical diagnostic reasoning.
• To evaluate the preliminary short-term learning impact of this VPS on undergraduate medical students.

Main Methods:

• Developed a VPS named Hepius, utilizing Siamese long short-term memory networks for anamnesis and NLP with SNOMED ontology for diagnosis.
• Integrated an ITS based on knowledge, assessment, and learner models.
• Assessed short-term learning in 15 undergraduate students using pre- and post-simulation multiple-choice tests.

Main Results:

• Hepius enables students to collect clinical data and generate differential diagnoses via natural language.
• The ITS provides real-time feedback and identifies knowledge gaps for targeted review.
• Significant improvements were observed in overall test scores (P<.001) and core knowledge questions (P<.001) post-simulation.

Conclusions:

• The Hepius VPS, combining ITS and NLP, serves as a valuable tool for training diagnostic reasoning in medical undergraduates.
• This technology is particularly relevant for medical education facing restricted access to clinical settings, exacerbated by global health crises like COVID-19.