Affordable Access

A new robust statistical method for treatment planning systems validation using experimental designs

Authors
  • Dufreneix, S.
  • Briand, K.
  • Di Bartolo, C.
  • Legrand, C.
  • Bremaud, M.
  • Mesgouez, J.
  • Tiplica, Téodor
  • Autret, Damien
Publication Date
Jan 01, 2016
Source
HAL-UPMC
Keywords
Language
English
License
Unknown
External links

Abstract

<p>Introduction</p><p>Dose computation verification is an important part of acceptance testing. The IAEA Tecdoc 1540 and 1583 suggest comparing computed dose to measurements for several beam configurations. However, this process is time-consuming and results out of tolerance are often left unexplained.</p><p>Purpose</p><p>To validate a treatment planning system using experimental designs which allow evaluating several parameters in a few tests selected by a robust statistical method.</p><p>Materials and methods</p><p>The Taguchi table L36 (211 × 312) was used to determine the 72 beams needed to test the 7 parameters chosen: energy, MLC, depth, jaw field size in X, Y1 and Y2 directions and wedge. Measurements were conducted in water using a CC04 (IBA) on a TrueBeam STx, a TrueBeam Tx, a Trilogy and a C-serie clinac (Varian). Dose was computed using the AAA algorithm (Eclipse, version 11). The same raw data was used for all accelerators during the algorithm configuration.</p><p>Results</p><p>The mean difference between computed and measured doses was 0.1 ±± 0.5% for all tested beams and all linacs with a maximum difference of 2.4% (under the 3% tolerance level). For all beams, the measured doses were within 0.6% for all linacs. No studied parameter led to statistically significant deviation between computed and measured doses.</p><p>Conclusion</p><p>Experimental design is a robust statistical method to validate an algorithm. Only 2 h of measurements were needed to evaluate 7 parameters. Furthermore, the commissioned accelerators were found dosimetrically equivalent even though the linac characteristics differ.</p>

Report this publication

Statistics

Seen <100 times