Timothy Peter Szczykutowicz PhD, Presenter: Equipment support, General Electric Company Research Grant, Siemens AG

Frank N. Ranallo PhD, Abstract Co-Author: Grant, General Electric Company

Develop a metric for quantifying noise non-uniformity in clinical CT images and evaluate a current metric for calculating noise magnitude in clinical images.

The general noise index (GNI) method developed by Christianson et al. (Med. Phys. 40:6 2013) has been modified to provide information related to noise uniformity. The GNI is computed by taking the difference between adjacent axial slices, dividing the difference image into small ROIs, and then computing the pixel standard deviation for each ROI (neglecting those ROIs containing bone or air). The GNI is taken as the maximum value of the histogram of all ROI noise values. This method can be modified by calculating the standard deviation of ROI noise levels used to calculate the original GNI, or the stdGNI. Simulated elliptical phantoms with varying major to minor axis ratios (1 to 2) were simulated at off-centering amounts (0 to 10 cm). A bowtie filter was also simulated. The GNI and stdGNI were calculated for all combinations. In addition, an anthropomorphic phantom and clinical CT images were used to assess the ability of the GNI and stdGNI metrics to identify cases of image noise non uniformity due to patient positioning.

The GNI broke down in cases of large image noise non-uniformity. Large variations in image noise made the noise histogram used to calculate the GNI bimodal which made the GNI results have a large variation and not correlate with overall noise level or noise non-uniformity. The stdGNI increased with the eccentricity of the elliptical phantom images and with off-centering amount as expected.

We propose stdGNI is capable able of capturing the degree of noise non-uniformity in clinical images. In addition, for images with large image noise non-uniformity, the GNI metric may not provide accurate results and simply taking the mean of the noise across the image provides more clinically relevant results.

Noise uniformity in CT is a surrogate for a poorly positioned patient; quantifying the degree of noise uniformity can aide in protocol optimization and in developing better CT acquisition methods.

Szczykutowicz, T, Ranallo, F, A Metric for Measuring Noise Non-uniformity in Clinical CT Images.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14045462.html