Clinical Text Generation through Leveraging Medical Concept and Relations

TL;DR

This paper presents a neural sequence generation model that leverages medical concept embeddings to improve clinical text generation from brief medical histories, demonstrating reduced perplexity and positive human evaluation feedback.

Contribution

It introduces a novel approach integrating medical concept embeddings into a sequence-to-sequence model for clinical text generation, showing improved performance over baseline models.

Findings

Reduced perplexity compared to baseline models

Positive human evaluation results from medical doctors

Feasibility of using medical concept embeddings in clinical text generation

Abstract

With a neural sequence generation model, this study aims to develop a method of writing the patient clinical texts given a brief medical history. As a proof-of-a-concept, we have demonstrated that it can be workable to use medical concept embedding in clinical text generation. Our model was based on the Sequence-to-Sequence architecture and trained with a large set of de-identified clinical text data. The quantitative result shows that our concept embedding method decreased the perplexity of the baseline architecture. Also, we discuss the analyzed results from a human evaluation performed by medical doctors.