Automated analysis tools vary in pulmonary nodule measurements

Different automated analysis tools vary in determining the size of pulmonary nodules, according to research presented at the 2007 Society for Imaging Informatics in Medicine (SIIM) annual meeting in Providence, RI.

"There was a significant difference in maximum dimension that different vendor packages measured," said Dr. Woojin Kim, a musculoskeletal fellow at the University of Pennsylvania School of Medicine in Philadelphia. "And this was noted throughout all nodule categories."

Kim presented the research from the Imaging Informatics Lab at the VA Maryland Health Care System in Baltimore during a scientific session at SIIM 2007.

Lung nodules need to be accurately measured over time to help with diagnosis, clinical management, and monitoring treatment response, Kim said. While interobserver variations in nodule diameter and volume have been shown for manual measurements in previous studies, automated tools have yielded more consistency.

To quantify the differences in pulmonary nodule maximum diameter and volume among different automated analysis software employed by multiple users, the researchers examined a dataset from the Baltimore VA Medical Center (BVAMC) lung nodule study evaluating the impact of ultralow-dose CT and dual-energy subtraction digital radiography (DR) in detecting lung nodules.

The dataset of consecutive chest CT scans included low-dose and conventional-dose studies acquired during a single scanning session on either a Sensation 64 or 16-slice scanner (Siemens Medical Solutions, Malvern, PA). Low-dose scans were acquired at 11 mAs, while conventional dose studies were at 180 mAs. All scans were performed using 120 kVp, 0.75-mm slice thickness, 0.75-mm collimation, and B40 reconstruction kernel.

Overall, 29 nodules for eight patients in five nodule size categories (the four categories used in the Fleischner Society nodule guidelines and a fifth category with nodules larger than 10 mm) were included in the study. Four investigators from the VA Maryland Healthcare, University of Maryland, and the University of Pennsylvania Hospital then analyzed the nodules using commercially available lung nodule analysis packages from TeraRecon (San Mateo, CA), Vital Images (Minnetonka, MN), and Siemens.

Investigators one and three used TeraRecon and Vital Images software, while the other two used Siemens' Leonardo software. JPEG snapshots of each nodule were provided to the investigators, who recorded the nodule dimension and volume as provided by the software. Researchers then performed analysis of variance (ANOVA) and t-test analysis to compare the tools.

Significant differences were found in maximum dimension across workstations (p < 0.01), as well as significant differences in volume measurements for nodule categories 1, 2, and 4, Kim said. There was no significant difference in maximum dimension or volume between investigators, however, and no significant difference in overall determination of nodule volume between workstations (p = 0.329).

"Even though there was a difference in maximum dimension, when it came to volume measurements, pretty much all the vendors demonstrated the same kind of measurements," he said. "However, differences did exist in volume measurements when we decided to look at the small subcategories -- grades 1, 2, and 4."

The significant difference in nodule maximum dimension between workstations, across all nodules, and within individual nodule grades is due to the difference in calculation approaches used by the various workstation packages, Kim said.

Unless standardized, maximum dimension is an unreliable marker of tumor size, even when using automated analysis tools, according to Kim.

"One (finding) that is potentially good news is that when you're using the same software package it really didn't matter who used it," he said. "It tended to produce consistent results among different investigators."

By Erik L. Ridley staff writer
June 25, 2007

Related Reading

Big Japanese study finds benefit in CT lung screening, May 24, 2007

CAD performs well in lung nodule detection, February 5, 2007

Lung CT CAD boosts performance of less experienced radiologists, November 27, 2006

Pulmonary nodule decisions must be individualized, August 16, 2006

CAD improves lung nodule detection, June 1, 2006

Copyright © 2007

Page 1 of 156
Next Page