Angel R Pineda, Presenter: Nothing to Disclose

Norbert Joseph Pelc ScD, Abstract Co-Author: Nothing to Disclose

High resolution flat panel detectors in computed tomography produce large amounts of data. In order to reduce the time required to extract the data from the panel and to reconstruct, the detector pixels are sometimes binned. We fix the desired resolution and study the noise variance resulting from using binned or unbinned projections to achieve it.

We compute the CT image variance (assuming parallel-beam reconstruction for convenience) where the detector pixels are either not binned or binned in pairs in the limit of a large number of pixels and views. We consider an object which produces, after a bow-tie filter, a fluence of x-rays that is uniform on the detector. In this case, both binned and unbinned data are uncorrelated and have equal variance. We vary the reconstruction kernel of both data sets such that they have the same resolution. First we set the desired resolution to be that of the binned system using a ramp filter up to the Nyquist frequency. Our second resolution goal is to also undo the pixel blur of the binned system up to the Nyquist frequency. Finally we assume quarter offset (or focal spot wobble) was used to double the sampling rate. This system is said to be detector limited since the maximum frequency that can be estimated is the determined by the pixel blur. The third resolution we consider is undoing the pixel blur and reconstructing frequencies up to 0.9 of the Nyquist frequency of the binned system with quarter offset.

Both systems have the same deterministic response but the variance with the unbinned system is always less. If we want the response of the binned system, the ratio of variances is 1.47 which is already significant. If we also undo the pixel blur, the ratio is 1.53. If we introduce quarter offset and require an accurate resolution up 0.9 of Nyquist, the ratio is 10.7. The noise amplification of the filtering process is magnified by binning the projections.

For a fixed resolution, binning the projections before reconstructing results in an increase in variance. As the resolution reaches that of a detector-limited system, the increase in the variance of the reconstructions is dramatic (in a non-Shakespearean way).

A.R.P.,N.J.P.: The authors receive funding from GE Medical Systems

Pineda, A, Pelc, N, To Bin or Not to Bin? A Question Regarding the Noise Properties of CT Reconstructions with or without Binned Projections.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4416050.html