Evaluation of reconstruction algorithms in SPECT neuroimaging. I. Comparison of statistical noise in SPECT neuroimages with 'naive' and 'realistic' predictions

Least-squares dual characterization for ROI assessment in emission tomography

Our aim is to describe an original method for estimating the statistical properties of regions of interest (ROIs) in emission tomography. Drawn upon the works of Louis on the approximate inverse, we propose a dual formulation of the ROI estimation problem to derive the ROI activity and variance directly from the measured data without any image reconstruction. The method requires the definition of an ROI characteristic function that can be extracted from a co-registered morphological image. This characteristic function can be smoothed to optimize the resolution-variance tradeoff. An iterative procedure is detailed for the solution of the dual problem in the least-squares sense (least-squares dual (LSD) characterization), and a linear extrapolation scheme is described to compensate for sampling partial volume effect and reduce the estimation bias (LSD-ex). LSD and LSD-ex are compared with classical ROI estimation using pixel summation after image reconstruction and with Huesman's method. For this comparison, we used Monte Carlo simulations (GATE simulation tool) of 2D PET data of a Hoffman brain phantom containing three small uniform high-contrast ROIs and a large non-uniform low-contrast ROI. Our results show that the performances of LSD characterization are at least as good as those of the classical methods in terms of root mean square (RMS) error. For the three small tumor regions, LSD-ex allows a reduction in the estimation bias by up to 14%, resulting in a reduction in the RMS error of up to 8.5%, compared with the optimal classical estimation. For the large non-specific region, LSD using appropriate smoothing could intuitively and efficiently handle the resolution-variance tradeoff.

A priori selection of the regularization parameters in emission tomography by Fourier synthesis

This paper describes how the stability of the inverse problem underlying emission tomography can be measured and controlled in clinical settings. We show how the Lanczos approximation provides a way to regularize a certain class of iterative reconstruction algorithms through a given level of noise or resolution in the slices and for a given acquisition protocol. Moreover, we show how the same Lanczos approximation can be used to decide when the iterative reconstruction algorithm actually converges for a given machine precision. These ideas are illustrated by means of reconstructions of simulated and actual emission datasets.

Experimental study of stochastic noise propagation in SPECT images reconstructed using the conjugate gradient algorithm

Thanks to an experimental study based on simulated and physical phantoms, the propagation of the stochastic noise in slices reconstructed using the conjugate gradient algorithm has been analysed versus iterations. After a first increase corresponding to the reconstruction of the signal, the noise stabilises before increasing linearly with iterations. The level of the plateau as well as the slope of the subsequent linear increase depends on the noise in the projection data.

Experimental performance assessment of SPM for SPECT neuroactivation studies using a subresolution sandwich phantom design

The validity domain of voxel-based statistical analysis of SPECT neuroactivation studies with statistical parametric mapping (SPM) has been investigated by a limited number of theoretical and simulation studies. In this work, an experimental setup is described with an assessment of its activation detection performance together with the influence of acquisition and processing parameters. A subresolution sandwich phantom was constructed using a printed high-resolution digital Hoffman phantom with a (99m)TcO(4)-ink mixture. Activations of 8, 16, and 24 mm diameter, with an intensity ranging from 5 to 60%, were constructed in the right frontal cortex, anterior and posterior cingulate, and left striatum. Triple-headed SPECT acquisitions were carried out using both fan-beam and parallel beam geometry. The impact of activation characteristics (size, intensity and location), study size, physical degradation factors, and reconstruction technique were studied using SPM99 in a group comparison design with correction for multiple comparisons. For a 15 x 15 design, all 24-mm activations of 5% intensity were detected for the fan-beam data. Decreased focus or study size, lower SPECT resolution, absence of scatter, and attenuation correction resulted in an increase in minimally detectable activation. For a single study referred to 15 control studies, only 24-mm activation foci with a minimal intensity of 10% were detected in the optimal configuration. This approach allows experimental parameter optimization of SPM-based group or single-subject SPECT activation studies compared to normal data, as used in clinical applications. In principle, these findings can be extended to SPECT receptor studies or PET data.

Assessing the performance of SPM analyses of spect neuroactivation studies. Statistical Parametric Mapping

Several simulations of SPECT neuroactivation studies have been performed in order to determine the influence of both study size and activation focus characteristics on the detection of brain activation foci following a pixel-based statistical analysis. This was achieved by developing a methodology based on the Hoffman software brain phantom, SPECT acquisition simulation software, standard reconstruction software, and the Statistical Parametric Mapping (SPM96) package. We present results on the minimal activation levels required for focus detection. Furthermore, the improved sensitivity of the analysis resulting from the use of an iterative reconstruction technique (OSEM) with regard to the classical filtered backprojection (FBP) is assessed quantitatively, and the various physical, processing, and physiological parameters that potentially influence the detection of foci are discussed. Finally, the influence is investigated of the height threshold as implemented in SPM96 upon the size of the detected foci. Practical guidelines are proposed with regard to the number of subjects per group for SPECT activation studies following the split-dose design.