RSNA 2004 

Abstract Archives of the RSNA, 2004


SSA21-01

Recommendation Rates in Unstructured Radiology Reports Database of a Tertiary Healthcare Center: Analysis with an Automatic Search Engine Based on the Principles of Information Theory and Natural Language processing

Scientific Papers

Presented on November 28, 2004
Presented as part of SSA21: Radiology Informatics (Workflow and Practice Management, Emphasizing Reporting)

Participants

Mannudeep K. Kalra MD, Presenter: Nothing to Disclose
Keith J. Dreyer DO, Abstract Co-Author: Nothing to Disclose
Michael M Maher MD, Abstract Co-Author: Nothing to Disclose
Autumn M Hurier, Abstract Co-Author: Nothing to Disclose
Thomas Schultz, Abstract Co-Author: Nothing to Disclose

PURPOSE

Recent increase in radiologic exams has raised concern for high rates of recommendations for subsequent investigations by radiologists. Previous study has validated accuracy of LEXIMER (Lexicon Mediated Entropy Reduction) search engine for automatic classification of unstructured radiology reports. The purpose of our study was to classify three million unstructured radiology reports with LEXIMER into those with recommendations (RT) and those without any recommendations (RT0).

METHOD AND MATERIALS

An electronic database of 3, 026, 637 unstructured radiology reports from 01/01/1995 to 12/31/2002 was created. The database comprised of CT (n=367930), MRI (n= 188765), ultrasound (n=241344), radiograph (n=1804030), fluoroscopic investigations (n=41309), mammography (n= 193697), nuclear medicine (n=110731), PET (n=8511), and angiography (n=70320) reports. To determine patterns of recommendation rates, the reports database were classified automatically for each calendar year from 1995 to 2002 into two categories, RT and RT0, using LEXIMER program. Data analysis was performed to assess frequency and trends of recommendation rates RT and RT0 in reports in various imaging modalities and regions of interest. In addition, RT rates, defined as number of reports RT divided by total number of reports analyzed, were calculated.

RESULTS

In entire database, overall RT rate was 8.2% (247485/3026637) and RT0 rate was 91.8% (2779152/3026637). Range of RT rates from year 1995 to 2002 for different imaging modalities were 11-20% (CT), 5-13% (MRI), 7-15% (ultrasound), 4-7% (radiograph), 3-4% (fluoroscopy), 5-12% (mammography), 4-5% (nuclear medicine), 3-28% (PET), and 11-18% (angiography). Significant increase in RT rates has been noted in CT, MRI, ultrasound and PET in 2002 compared to 1995 (p<0.05). Significant differences in RT rates between imaging modalities, notably CT, ultrasound and radiography, in different regions of the body were noted (p<0.05).

CONCLUSIONS

Although recommendations for subsequent action in radiology reports show wide variation between different imaging modalities and region of interest, LEXIMER program revealed that overall recommendation rate in radiology reports is less than 10%.

Cite This Abstract

Kalra, M, Dreyer, K, Maher, M, Hurier, A, Schultz, T, Recommendation Rates in Unstructured Radiology Reports Database of a Tertiary Healthcare Center: Analysis with an Automatic Search Engine Based on the Principles of Information Theory and Natural Language processing.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4406026.html