The use of multivariate regression analysis in contrast detail studies of CT scanners

Manuel Trajtenberg*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

Previous studies of the imaging performance of computed tomography (CT) scanners, and other imaging modalities, have failed to apply appropriate statistical methods to data analysis, thus impairing the accuracy and significance of results. Given that imaging performance involves a number of interrelated variables and an element of randomness, its empirical assessment requires multivariate regression analysis. This method is used here to analyze anew a set of contrast detail data from a previous study on CT scanners. The main issues considered are the specification of the proper functional form linking perceptibility, dose and contrast, the estimation of the contrast and dose coefficients, and of scanner specific constants to be used in computing indices of imaging quality. One of the main empirical findings is that the dose coefficient of the CT scanners studied is significantly less than that predicted by the theoretical model: 1/5 instead of 1/3. This result suggests that actual dose used in routine clinical studies could be reduced substantially without impairing much the quality of the images. On the other hand, the coefficient of contrast does correspond to its predicted value, i.e., 2/3. The methodology used here is not limited to the contrast detail framework, but is applicable to, and indeed essential in, empirical studies of the performance of any imaging modality.

Original languageEnglish
Pages (from-to)456-464
Number of pages9
JournalMedical Physics
Volume11
Issue number4
DOIs
StatePublished - Jul 1984
Externally publishedYes

Keywords

  • CAT SCANNING
  • IMAGE FORMING
  • OPTICAL MODULATION
  • REGRESSION ANALYSIS
  • STATISTICS

Fingerprint

Dive into the research topics of 'The use of multivariate regression analysis in contrast detail studies of CT scanners'. Together they form a unique fingerprint.

Cite this