Revealing the Structure of Medical Dictations with Conditional Random Fields

Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 1–10, Honolulu, October 2008. c⃝2008 Association for Computational Linguistics Revealing the Structure of Medical Dictations with Conditional Random Fields Jeremy Jancsary and Johannes Matiasek Austrian Research Institute for Artificial Intelligence A-1010 Vienna, Freyung 6/6 firstname.lastname@ofai.at Harald Trost Department of Medical Cybernetics and Artificial Intelligence of the Center for Brain Research, Medical University Vienna, Austria harald.trost@meduniwien.ac.at Abstract Automatic processing of medical dictations poses a significant challenge. We approach the problem by introducing a statistical frame- work capable of identifying types and bound- aries of sections, lists and other structures occurring in a dictation, thereby gaining ex- plicit knowledge about the function of such elements. Training data is created semi- automatically by aligning a parallel corpus of corrected medical reports and correspond- ing transcripts generated via automatic speech recognition. We highlight the properties of our statistical framework, which is based on conditional random fields (CRFs) and im- plemented as an efficient, publicly available toolkit. Finally, we show that our approach is effective both under ideal conditions and for real-life dictation involving speech recog- nition errors and speech-related phenomena such as hesitation and repetitions. 1 Introduction It is quite common to dictate reports and leave the typing to typists – especially for the medical domain, where every consultation or treatment has to be doc- umented. Automatic Speech Recognition (ASR) can support professional typists in their work by provid- ing a transcript of what has been dictated. However, manual corrections are still needed. In particular, speech recognition errors have to be corrected. Fur- thermore, speaker errors, such as hesitations or rep- etitions, and instructions to the transcriptionist have to be removed. Finally, and most notably, proper structuring and formatting of the report has to be performed. For the medical domain, fairly clear guidelines exist with regard to what has to be dic- tated, and how it should be arranged. Thus, missing headings may have to be inserted, sentences must be grouped into paragraphs in a meaningful way, enu- meration lists may have to be introduced, and so on. The goal of the work presented here was to ease the job of the typist by formatting the dictation ac- cording to its structure and the formatting guide- lines. The prerequisite for this task is the identifi- cation of the various structural elements in the dic- tation which will be be described in this paper. complaint dehydration weakness and diarrhea full stop Mr. Will Shawn is a 81-year-old cold Asian gentleman who came in with fever and Persian diaper was sent to the emergency department by his primary care physician due him being dehydrated period . . . neck physical exam general alert and oriented times three known acute distress vital signs are stable . . . diagnosis is one chronic diarrhea with hydration he also has hypokalemia neck number thromboctopenia probably duty liver cirrhosis . . . a plan was discussed with patient in detail will transfer him to a nurse and …

Content truncated to 500 words