Automatic Clinical Note Generation from Doctor-Patient Conversations Using Medical Event Extraction and Term Normalization

ACL ARR 2025 May Submission2953 Authors

Abstract:

Currently, it still remains a time-consuming and error-prone work for doctors to manually write clinical notes in both online medical consultation and offline clinic visit, highlighting the demand for automation. Recent studies of automatic clinical note generation attempt to generate end-to-end from conversation to clinical note by leveraging a large language model (LLM), but the generated content is deficient due to irrelevant information and informal language use. To address these issues, this paper breaks down the end-to-end generation and introduces a tripartite framework including medical event extraction, term normalization and clinical note generation.
The proposed method improves the quality of generation by blocking the irrelevant chats and emphasizing on critical medical events, as well as ``translating'' patient's informal language into doctor's formal language with external knowledge base. The experimental evaluation on the benchmarking data sets demonstrates that the proposed method outperforms all baselines and achieves the state-of-the-art performance. Besides, the real-world study further brings a more promising result that over 50\% of doctors' time spent on manually writing clinical notes could be saved under the assistance of the proposed method, leaving more time for patient care.

Paper Type: Long

Research Area: NLP Applications

Research Area Keywords: Large Language Model, Clinical Note Generation

Languages Studied: English

Keywords: Large Language Model; Clinical Note Generation

Submission Number: 2953