By Erik L. Ridley, AuntMinnie staff writer
March 2, 2012

VIENNA - A data-mining software platform can analyze text in German radiology reports and provide semantic links between images and textual descriptions, according to a presentation on Friday at the European Congress of Radiology (ECR).

The platform, developed via a collaboration between image software developer Averbis, the University of Freiburg, and the University of Leipzig, has a range of clinical, research, and teaching applications, said Dr. Philipp Daumke of Averbis. Daumke described the system during the Friday scientific session.

To improve access to images and free-text data in radiology reports, the developers wanted to create a tool that could provide semantic searching and data mining of radiology reports, diagnostic support through detection of similar cases, semantic links between reports and images, and quality assurance by cross-checking a diagnosis with external data, Daumke said.

The system employs a scalable text analysis platform that can be used to extract information from the German radiology reports. An image reference and context recognition component also facilitates the creation of semantic links between text and image data, according to the developers. In addition, radiologists can access key images and relevant text passages in a search engine that can be used for ad-hoc queries.

The system extracts information related to size, volume, time and date, image references, anatomical terms, descriptions, relations, negations, and abbreviations, Daumke said.

Its semantic analysis capabilities are such that a search for "lung metastasis" will also find reports with related terms such as lung metastases, metastases in the lung, metastasis in the lower part of the lung, pulmonary metastases, pulmonary metastatic relapse, pulmonary filiae, and lung filiae, he said.

"There is a huge variation [in terminology], and we are able to find all of these variations," he said.

The team evaluated the system in 133,325 radiology reports. The system was able to identify 32,533 image references and successfully retrieve 32,157 images from the local PACS server. Based on randomized validation of the identified image references, the researchers concluded that the correct reference was extracted with a mean average precision of 0.97.

Clinically, the system is used by radiologists to validate diagnoses and for looking for "best practice" reports, he said. It can also be used as a mobile application for clinicians and be integrated into portals for referring physicians.

For research, the system is used to collect patient cohorts, Daumke said. It can also be useful for teaching applications, such as creating a database for teaching cases and for presentation of lectures and talks, he said.

Data mining of image analysis tools expands informatics applications
VIENNA - Data mining of quantitative measurements made by radiologists when interpreting images is expected to become increasingly important, thanks to...
Data-mining tool identifies overexposed chest x-rays
CHICAGO - Taking a high-quality portable chest x-ray can be difficult, leading to an overexposed image when digital radiography equipment is used. One...
Just-in-time data mining addresses uncertainty in radiology reports
Tuesday, November 29 | 3:50 p.m.-4:00 p.m. | SSJ11-06 | Room S102D
Imagine sitting at a diagnostic...
Data mining with natural language processing identifies mammo trends
Thursday, December 2 | 11:50 a.m.-12:00 p.m. | SSQ10-09 | Room S402AB
Researchers at Massachusetts...
Data mining techniques simplify PET report evaluations
Thursday, December 2 | 12:45 p.m.-1:15 p.m. | LL-INE1153-THB | Lakeside Learning Center
In this poster...

Copyright © 2012

Last Updated np 3/9/2012 5:37:23 PM