Radiology report entry with automatic phrase completion driven by language modeling

John Eng, Jason M. Eisner

Research output: Contribution to journalArticle

Abstract

Keyboard entry or correction of radiology reports by radiologists and transcriptionists remains necessary in many settings despite advances in computerized speech recognition. A report entry system that implements an automated phrase completion feature based on language modeling was developed and tested. The special text editor uses context to predict the full word or phrase being typed, updating the displayed prediction after each keystroke. At any point, pressing the tab key inserts the predicted phrase without having to type the remaining characters of the phrase. Successive words of the phrase are predicted by a trigram language model. Phrase lengths are chosen to minimize the expected number of keystrokes as predicted by the language model. Operation is highly and automatically customized to each user. The language model was trained on 36,843 general radiography reports, which were consecutively generated and contained 1.48 million words. Performance was tested on 200 randomly selected reports outside of the training set. The phrase completion technique reduced the average number of keystrokes per report from 194 to 58; the average reduction factor was 3.3 (geometric mean) (95% confidence interval, 3.2-3.5). The algorithm significantly reduced the number of keystrokes required to generate a radiography report (P <.00005).

Original languageEnglish (US)
Pages (from-to)1493-1501
Number of pages9
JournalRadiographics : a review publication of the Radiological Society of North America, Inc
Volume24
Issue number5
StatePublished - Sep 2004

Fingerprint

Radiology
Language
Radiography
Confidence Intervals

Keywords

  • Computers
  • Information management
  • Information management, systems
  • Radiology reporting systems

ASJC Scopus subject areas

  • Radiology Nuclear Medicine and imaging
  • Radiological and Ultrasound Technology

Cite this

@article{50b847efaf0340a489cfe355c3d8b916,
title = "Radiology report entry with automatic phrase completion driven by language modeling",
abstract = "Keyboard entry or correction of radiology reports by radiologists and transcriptionists remains necessary in many settings despite advances in computerized speech recognition. A report entry system that implements an automated phrase completion feature based on language modeling was developed and tested. The special text editor uses context to predict the full word or phrase being typed, updating the displayed prediction after each keystroke. At any point, pressing the tab key inserts the predicted phrase without having to type the remaining characters of the phrase. Successive words of the phrase are predicted by a trigram language model. Phrase lengths are chosen to minimize the expected number of keystrokes as predicted by the language model. Operation is highly and automatically customized to each user. The language model was trained on 36,843 general radiography reports, which were consecutively generated and contained 1.48 million words. Performance was tested on 200 randomly selected reports outside of the training set. The phrase completion technique reduced the average number of keystrokes per report from 194 to 58; the average reduction factor was 3.3 (geometric mean) (95{\%} confidence interval, 3.2-3.5). The algorithm significantly reduced the number of keystrokes required to generate a radiography report (P <.00005).",
keywords = "Computers, Information management, Information management, systems, Radiology reporting systems",
author = "John Eng and Eisner, {Jason M.}",
year = "2004",
month = "9",
language = "English (US)",
volume = "24",
pages = "1493--1501",
journal = "Radiographics",
issn = "0271-5333",
publisher = "Radiological Society of North America Inc.",
number = "5",

}

TY - JOUR

T1 - Radiology report entry with automatic phrase completion driven by language modeling

AU - Eng, John

AU - Eisner, Jason M.

PY - 2004/9

Y1 - 2004/9

N2 - Keyboard entry or correction of radiology reports by radiologists and transcriptionists remains necessary in many settings despite advances in computerized speech recognition. A report entry system that implements an automated phrase completion feature based on language modeling was developed and tested. The special text editor uses context to predict the full word or phrase being typed, updating the displayed prediction after each keystroke. At any point, pressing the tab key inserts the predicted phrase without having to type the remaining characters of the phrase. Successive words of the phrase are predicted by a trigram language model. Phrase lengths are chosen to minimize the expected number of keystrokes as predicted by the language model. Operation is highly and automatically customized to each user. The language model was trained on 36,843 general radiography reports, which were consecutively generated and contained 1.48 million words. Performance was tested on 200 randomly selected reports outside of the training set. The phrase completion technique reduced the average number of keystrokes per report from 194 to 58; the average reduction factor was 3.3 (geometric mean) (95% confidence interval, 3.2-3.5). The algorithm significantly reduced the number of keystrokes required to generate a radiography report (P <.00005).

AB - Keyboard entry or correction of radiology reports by radiologists and transcriptionists remains necessary in many settings despite advances in computerized speech recognition. A report entry system that implements an automated phrase completion feature based on language modeling was developed and tested. The special text editor uses context to predict the full word or phrase being typed, updating the displayed prediction after each keystroke. At any point, pressing the tab key inserts the predicted phrase without having to type the remaining characters of the phrase. Successive words of the phrase are predicted by a trigram language model. Phrase lengths are chosen to minimize the expected number of keystrokes as predicted by the language model. Operation is highly and automatically customized to each user. The language model was trained on 36,843 general radiography reports, which were consecutively generated and contained 1.48 million words. Performance was tested on 200 randomly selected reports outside of the training set. The phrase completion technique reduced the average number of keystrokes per report from 194 to 58; the average reduction factor was 3.3 (geometric mean) (95% confidence interval, 3.2-3.5). The algorithm significantly reduced the number of keystrokes required to generate a radiography report (P <.00005).

KW - Computers

KW - Information management

KW - Information management, systems

KW - Radiology reporting systems

UR - http://www.scopus.com/inward/record.url?scp=4944260413&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=4944260413&partnerID=8YFLogxK

M3 - Article

C2 - 15371624

AN - SCOPUS:4944260413

VL - 24

SP - 1493

EP - 1501

JO - Radiographics

JF - Radiographics

SN - 0271-5333

IS - 5

ER -