RSNA 2004 > How Accurate is Automatic Nodule Volume Measurement? ...
Scientific Papers
  CODE: SSM05-01
  SESSION: Chest (Lung Nodules: Growth)
  How Accurate is Automatic Nodule Volume Measurement? A Comparison of Two Automatic Nodule Analysis Software Products Using a Lung Nodule Phantom

Add to Briefcase
  Email Event
Mary Roddie MD
Abstract Co-Author
Stephen Ellis MBBCh
Peter Armstrong
Manlio Valdiviesco-Casique
Jamshid Dehmeshki
- Author stated no financial disclosure

- Disclosure information unavailable
  DATE: Wednesday, December 01 2004
  START TIME: 03:00 PM
  END TIME: 03:10 PM

To compare the accuracy of automated measurement of lung nodules in a lung phantom using two different computer-assisted nodule analysis programs.
A lung phantom containing spherical nodules was scanned using a GE Lightspeed 16 slice MDCT (1.25mm, 300mA, 120kV, soft tissue reconstruction algorithm). The phantom nodules used ranged in size from 3 to 10 mm in diameter, 15 adjacent to a simulated blood vessel and 7 isolated. Commercially available (X) and in-house (Y) nodule analysis software extracted (created a 3D model) each nodule and calculated 3 orthogonal diameters and the volume. A single reader observed the accuracy of extraction by viewing a 3D graphic produced by the software and calculated a sphericity index (SI), the ratio of the longest and shortest diameters (for an exact sphere SI=1)
For isolated nodules, software X consistently under-estimated nodule volume (mean 47.1%; range-36 to -64%). Volume measurement by software Y was significantly more accurate (mean 9.3%; range –17% to +6%)[p<0.001]. The ranges of SI for software X (1.02 to 1.13) and software Y (1.0 to 1.02) reflect accurate extraction of the isolated nodules from the adjacent simulated lung parenchyma.
However, for the nodules adjacent to vessels the SI values for software X reflected the inaccuracy of nodule extraction (SI mean 1.38; range 1.07 to 1.98), software Y performing significantly better (SI mean 1.12; range 1.07 to 1.24)(p<0.002). The high SI values for software X resulted from the incorporation of segments of adjacent simulated vessels.
The value of nodule follow up by volume calculation is questionable if the volume being calculated is not accurate. Two errors in the automatic analysis of lung nodules have been identified in a commercially available nodule analysis program. 1: under-estimation of nodule size despite apparently accurate extraction (sphericity value close to 1), which is of concern because the underestimation was not readily apparent from viewing the 3D graphic presented. 2: inclusion of adjacent vessel causing an increase in SI, which was readily appreciated from 3D graphic. The in-house software demonstrated that these errors can be reduced.
S.M.E.: Author works for the company funding the study on a consultancy basis
M.V.,J.D.: Authors are full time employees of the company funding the study
M.R.: Employed on a part-time basis by Medic Sys.
P.A.: Member of the clinical advisory board for the company funding the study; paid advisor to Medisight (company developing software)
Copyright © 2004 Radiological Society of North America, Inc.
820 Jorie Boulevard, Oak Brook, IL 60523-2251 || (630) 571-2670 || fax (630) 571-7837
U.S. and Canada: Main (800) 381-6660, Membership (877) RSNA-MEM (776-2636)