Typical values statistical analysis for adult chest and abdomen-pelvis CT examinations

Abstract

The use of DRLs has been extensively documented in the literature as a tool for protocol optimization across different x-ray imaging modalities in different countries. It recognizes the importance of developing and validating methods capable of correlating DRL quantities (CTDIvol, DLP and SSDE) with the technical parameters employed in CT studies. Such correlations must be supported by robust statistical methodologies in order to ensure the adoption of adequate optimization decisions. The aim of this work was to apply the Generalized Additive Model (GAM) statistical analysis in adult non-contrast chest and abdomen-pelvis CT typical values. These patient cohorts were statistically evaluated to identify correlations with key-parameters associated to the demographic patient information and machine dependent data, taking into account patients’ effective diameters, d, and body mass indexes (BMI). GAM was implemented considering each anatomical region in order to correlate the log-transformed DRL quantities (DRLq's) as outcomes given different key predictors related to image acquisition and patient characteristics. A total of 956 CT patient data were collected in this retrospective single-center study. Demographic variables demonstrate that age is not or it is just weakly-correlated to the DRLq's resulting from chest procedures, but it is strongly correlated when considering abdomen-pelvis examinations. Gender is correlated to the DRLq's for chest examinations adopting d as a key predictor but it is only correlated with DLP adopting the BMI as a key predictor. The level of accuracy provided by the GAM was adequate for interpreting the large fluctuations of the DRLq's, technical parameters and demographic data observed for the studied patient cohorts. Our results reflect the importance of a comprehensive statistical evaluation of typical values. The domain of this technique is important to different CT imaging chain stakeholders and its application can be a key tool for decision-making process to effective optimization strategies.