Conformer-Based Dental AI Patient Clinical Diagnosis Simulation Using Korean Synthetic Data Generator for Multiple...

Area of Science:

• Medical Education Technology
• Artificial Intelligence in Healthcare
• Natural Language Processing

Background:

• Standardized Patients (SP) are crucial for clinical practice education, simulating real patient interactions.
• Current SP methods face challenges including high costs and a shortage of trained educators.
• Deep learning offers a potential solution to automate and enhance SP simulations.

Purpose of the Study:

• To develop an AI patient using deep learning to overcome limitations in traditional SP education.
• To create a Korean SP scenario data generator for training AI patient responses.
• To evaluate the performance of a Conformer-based AI patient against a Transformer model.

Main Methods:

• Utilized the Conformer deep learning model for AI patient implementation.
• Developed a Korean SP scenario data generator for diagnostic question response training.
• Trained the AI patient using common conversational data and personalized SP scenario data.
• Compared Conformer model performance against Transformer using BLEU score and Word Error Rate (WER).

Main Results:

• The Conformer-based AI patient demonstrated a 3.92% improvement in BLEU score.
• The Conformer-based AI patient showed a 6.74% improvement in WER compared to the Transformer model.
• The AI patient successfully learned specific clinical information from personalized data.

Conclusions:

• The Conformer model offers a more effective approach for developing AI patients in SP simulations.
• This AI patient technology can potentially be adapted for various medical and nursing fields.
• The developed system provides a scalable and cost-effective alternative for clinical skills training.