Conversational System for Clinical Communication Training Supporting User-defined Tasks

Abstract

Effective clinical communication is essential for delivering safe and high-quality patient care, especially in emergent cases. Standard communication protocols have been developed to improve communication accuracy and efficiency. However, traditional training and evaluation require substantial manpower and time, which can be infeasible during public crises when training is most needed. This research aims to facilitate autonomous, low-cost, adaptive clinical communication training via artificial intelligence (AI)-powered techniques. We propose a conversational system for clinical communication training supporting user-defined tasks. Two data augmentation (DA) methods, term replacement and context expansion, are proposed to allow non-professional users to create Al models with a small number of samples. Equipped with biomedical ontology and pre-trained language models, our system is able to simulate clinical communication scenarios, provide timely evaluation, and adapt to new tasks with minimal editing. Various experiments demonstrate that our proposed algorithms can achieve satisfactory performance using a small amount of training data. Real-world practice in local hospitals shows that our system can provide expert-level evaluation and deliver effective clinical communication training.