Abstract Archives of the RSNA, 2011
LL-INS-TU6A
Improved Radiation Dose Estimates Using the PARSE Open Source Toolkit: Patient-centric Data Extraction for More Accurate Radiation Exposure Histories and Quality Assurance in Nuclear Medicine
Scientific Informal (Poster) Presentations
Presented on November 29, 2011
Presented as part of LL-INS-TU: Informatics
Ichiro Ikuta MD, Presenter: Nothing to Disclose
Aaron D. Sodickson MD, PhD, Abstract Co-Author: Consultant, Siemens AG
Consultant, Bayer AG
Elliot Joseph Wasser MD, Abstract Co-Author: Nothing to Disclose
Graham Ingersoll Warden MD, Abstract Co-Author: Nothing to Disclose
Victor H. Gerbaudo PhD, Abstract Co-Author: Nothing to Disclose
Ramin Khorasani MD, Abstract Co-Author: Stockholder, Medicalis Corp
Advisory Board, General Electric Company
Advisory Board, EMC Corp
Nuclear medicine accounts for approximately 26% of all medical radiation exposure. Our aim was to validate a new tool that extracts radiation exposure information from nuclear medicine reports, and to demonstrate how the resulting data can enhance radiation monitoring of patients and quality assurance efforts.
We developed PARSE (Perl Automation for Radiopharmaceutical Selection and Extraction), an open-source toolkit designed to automate the extraction of radiopharmaceutical (tracer) and administered activity from nuclear medicine reports. Following iterative code training, manual validation was performed on 2512 nuclear medicine reports randomly selected from the time period 09/17/1985 to 02/28/2011. Recall (sensitivity) is the proportion of reports containing the required fields (tracer and activity) that were correctly retrieved. Precision (positive predictive value) is the proportion of retrieved reports that had all exposure information correctly assigned. Validation was performed for three distinct categories of nuclear medicine exams involving single or multiple injections of single or multiple tracers. We demonstrate two use-case examples: 1) For a patient with multiple exams, we used standard tracer-specific lookup tables to calculate growth in cumulative organ doses from extracted tracers and activities, then assigned effective dose using ICRP-103 organ dose weighting factors. 2) For three common tracers (F-18 FDG, Tc-99m MDP, and I-131 NaI), distributions of administered activity were extracted from 204561 reports.
PARSE validation yielded an overall recall of 97.7 ± 0.6% and precision of 95.2 ± 0.9% for all exam types combined. For exams utilizing single tracer injections, multiple injections of the same tracer, and multiple injections of different tracers, recall was 97.2 ± 0.8%, 99.5 ± 0.7%, and 98.7 ± 1.8%, respectively, and precision was 94.6% ± 1.1%, 98.7 ± 1.1%, and 92.5 ± 4.1%, respectively. The figure contains the two use-case examples of patient-centric dose monitoring, and departmental quality assurance.
PARSE can reliably perform large-scale automated extraction of radiation data from nuclear medicine reports.
PARSE processing allows accurate assessment of nuclear medicine doses, enhancing patient-centric radiation dose monitoring and large-scale quality assurance.
Ikuta, I,
Sodickson, A,
Wasser, E,
Warden, G,
Gerbaudo, V,
Khorasani, R,
Improved Radiation Dose Estimates Using the PARSE Open Source Toolkit: Patient-centric Data Extraction for More Accurate Radiation Exposure Histories and Quality Assurance in Nuclear Medicine. Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL.
http://archive.rsna.org/2011/11034442.html