Separating components of variation in measurement series using maximum likelihood estimation: Application to patient position data in radiotherapy

J P Sage, W P M Mayles, H M Mayles, I Syndikus

Research output: Contribution to journalArticle

Abstract

Maximum likelihood estimation (MLE) is presented as a statistical tool to evaluate the contribution of measurement error to any measurement series where the same quantity is measured using different independent methods. The technique was tested against artificial data sets; generated for values of underlying variation in the quantity and measurement error between 0.5 mm and 3 mm. In each case the simulation parameters were determined within 0.1 mm. The technique was applied to analyzing external random positioning errors from positional audit data for 112 pelvic radiotherapy patients. Patient position offsets were measured using portal imaging analysis and external body surface measures. Using MLE to analyze all methods in parallel it was possible to ascertain the measurement error for each method and the underlying positional variation. In the (AP / Lat / SI) directions the standard deviations of the measured patient position errors from portal imaging were (3.3 mm / 2.3 mm / 1.9 mm), arising from underlying variations of (2.7 mm / 1.5 mm / 1.4 mm) and measurement uncertainties of (1.8 mm / 1.8 mm / 1.3 mm), respectively. The measurement errors agree well with published studies. MLE used in this manner could be applied to any study in which the same quantity is measured using independent methods.

Original languageEnglish
Pages (from-to)6019-6030
Number of pages12
JournalPhysics in Medicine and Biology
Volume59
Issue number20
DOIs
Publication statusPublished - 21 Oct 2014
Externally publishedYes

Keywords

  • Algorithms
  • Humans
  • Likelihood Functions
  • Patient Positioning
  • Radiotherapy
  • Journal Article

Fingerprint Dive into the research topics of 'Separating components of variation in measurement series using maximum likelihood estimation: Application to patient position data in radiotherapy'. Together they form a unique fingerprint.

  • Cite this