Using comparative histogram analysis to guide the acquisition and reconstruction choices that minimize quantitative disparities in PET images obtained with different scanners.

Objectives Quantitative values from PET images often show significant differences when acquired with different scanner models or when acuired and processed differently. The objective of this study was to examine whether features and metrics from phantom-derived, image histograms offered guidance regarding processing and acquisition modifications that may minimize quantitative image disparities (such as SUV values) between any pair of devices or reconstruction schemes.

Methods Images from a Ge68 solid cylinder with uniform activity (UA) or an F18-filled ACR-type phantom with structured activity (SA) distributions were acquired for different durations and with different acquisition parameters. They were also reconstructed using different algorithms, matrix sizes, filter bandwidth and zoom factors. ROI’s and VOI’s of identical area (volume) were applied to identical slices and structures within the cylinders. Histograms of SUV were generated from the voxels contained in these ROIs (VOIs) and then binned using a uniform approach. The differences in the acquisition and processing parameters were then associated with their effects on the overall appearance of the histogram and on the value of parameters extracted from the histograms such as the median, mode,FWHM, mean, max and min, kurtosis and skew.

Results Histogram analysis provided quantitative measures and visually revealed quantitative-image mismatch. Post-filtering showed a predictable effect on the spread of the SUV values. Deficient calibration and attenuation correction resulted in a shift in the mode and in a visible skewness in the histogram of UA slices. Histogram analysis of subsamples of the images with contrasting structures of various sizes, facilitated the matching of the SUV spread. Unexpectedly, this analysis showed which matrix or zoom choice would result in LOR rebinning as opposed to simple interpolation. This information is essential to establish if one could meaningfully manipulate matrix size and zoom to qualitatively match images (i.e. whether it is appropriate to assume that a constant matrix-zoom product will produce quantitatively matching voxels volumes)

Conclusions Comparative histogram analysis on uniform and nonuniform input revealed the effects of acquisition choices. By characterizing voxel volume, accuracy of calibration and attenuation, and degree of filtering, this approach is advantageous in assessing the quantitative equivalency of images produced by different scanners . Detailed equivalency for SUV or SUVmax, required a non uniform phantom with contrasting and structurally diverse objects. More importantly, the effects of known interventions (acquisition or processing settings) on the histogram parameters of phantom images suggested acquisition and processing modifications that reduced disparities between the quantitative values derived from images obtained in different scanner models or reconstruction schemes.