Automatic analysis of medical dialogue in the home hemodialysis domain : structure induction and summarization

Abstract

Spoken medical dialogue is a valuable source of information, and it forms a foundation for diagnosis, prevention and therapeutic management. However, understanding even a perfect transcript of spoken dialogue is challenging for humans because of the lack of structure and the verbosity of dialogues. This work presents a first step towards automatic analysis of spoken medical dialogue. The backbone of our approach is an abstraction of a dialogue into a sequence of semantic categories. This abstraction uncovers structure in informal, verbose conversation between a caregiver and a patient, thereby facilitating automatic processing of dialogue content. Our method induces this structure based on a range of linguistic and contextual features that are integrated in a supervised machine-learning framework. Our model has a classification accuracy of 73%, compared to 33% achieved by a majority baseline (p<0.01). We demonstrate the utility of this structural abstraction by incorporating it into an automatic dialogue summarizer. Our evaluation results indicate that automatically generated summaries exhibit high resemblance to summaries written by humans and significantly outperform random selections (p<0.0001) in precision and recall.

(cont.) In addition, task-based evaluation shows that physicians can reasonably answer questions related to patient care by looking at the automatically-generated summaries alone, in contrast to the physicians' performance when they were given summaries from a naive summarizer (p<0.05). This is a significant result because it spares the physician from the need to wade through irrelevant material ample in dialogue transcripts. This work demonstrates the feasibility of automatically structuring and summarizing spoken medical dialogue.