Preprocessing for Quantitative Statistical Noise Analysis of MDCT Brain Images Reconstructed Using Hybrid Iterative (iDose) Algorithm

Radiation dose reduction is a very topical problem in medical X-ray CT imaging and plenty of strategies have been introduced recently. Hybrid iterative reconstruction algorithms are one of them enabling dose reduction up to 70 %. The paper describes data preprocessing and feature extraction from iteratively reconstructed images in order to assess their quality in terms of image noise and compare it with quality of images reconstructed from the same data by the conventional filtered back projection. The preprocessing stage consists in correction of a stair-step artifact and in fast, precise bone and soft tissue segmentation. Noise patterns of differently reconstructed images can therefore be examined separately in these tissue types. In order to remove anatomical structures and to obtain the pure noise, subtraction of images reconstructed by the iterative iDose algorithm from images reconstructed by the filtered back projection is performed. The results of these subtractions called here residual noise images and are the used to further extract parameters of the noise. The noise parameters, which are intended to serve as input data for consequent multidimensional statistical analysis, are the standard deviation and power spectrum of the residual noise. This approach enables evaluation of noise properties in the whole volume of real patient data, in contrast to noise analysis performed in small regions of interest or in images of phantoms.