K. Erlandsson

Fast accurate iterative reconstruction for low-statistics positron volume imaging

A fast accurate iterative reconstruction (FAIR) method suitable for low-statistics positron volume imaging has been developed. The method, based on the expectation maximization-maximum likelihood (EM-ML) technique, operates on list-mode data rather than histogrammed projection data and can, in just one pass through the data, generate images with the same characteristics as several ML iterations. Use of list-mode data preserves maximum sampling accuracy and implicitly ignores lines of response (LORs) in which no counts were recorded. The method is particularly suited to systems where sampling accuracy can be lost by histogramming events into coarse LOR bins, and also to sparse data situations such as fast whole-body and dynamic imaging where sampling accuracy may be compromised by storage requirements and where reconstruction time can be wasted by including LORs with no counts. The technique can be accelerated by operating on subsets of list-mode data which also allows scope for simultaneous data acquisition and iterative reconstruction. The method is compared with a standard implementation of the EM-ML technique and is shown to offer improved resolution, contrast and noise properties as a direct result of using improved spatial sampling, limited only by hardware specifications.

Intercomparison of four reconstruction techniques for positron volume imaging with rotating planar detectors.

Four reconstruction techniques for positron volume imaging have been evaluated for scanners based on rotating planar detectors using measured and simulated data. The four techniques compared are backproject then filter (BPF), the 3D reprojection (3D RP) method for 3D filtered backprojection (FBP), Fourier rebinning (FORE) in conjunction with 2D FBP (FORE + 2D FBP) and 3D ordered subsets expectation maximization (3D OSEM). The comparison was based on image resolution and on the trade-off between contrast and noise. In general FORE + 2D FBP offered a better contrast-noise trade-off than 3D RP, whilst 3D RP offered a better trade-off than BPF. Unlike 3D RP, FORE + 2D FBP did not suffer any contrast degradation effect at the edges of the axial field of view, but was unable to take as much advantage from high-accuracy data as the other methods. 3D OSEM gave the best contrast at the expense of greater image noise. BPF, which demonstrated generally inferior contrast-noise behaviour due to use of only a subset of the data, gave more consistent spatial resolution over the field of view than the projection-data based methods, and was best at taking full advantage of high-accuracy data.

Preliminary results from the new large-area PETRRA positron camera

The PETRRA positron camera is based on the use of BaF/sub 2/ scintillators interfaced to large area multiwire proportional chambers filled with a photo-sensitive vapour (TMAE). The camera consists of two 60 cm/spl times/40 cm annihilation photon detectors mounted on a rotating gantry. Initial measurements with the camera show that the spatial resolution is /spl sim/6.5/spl plusmn/1 mm FWHM all through the field-of-view, and the timing resolution is between 7 ns and 10 ns FWHM. Detection efficiency for annihilation photons is /spl sim/30% per detector. The count rates obtained, by using a 20 cm diameter by 11 cm long water filled phantom containing /sup 18/F, were /spl sim/1.35/spl times/10/sup 6/ singles and /spl sim/1.2/spl times/10/sup 5/ cps raw coincidences at which point data-rates are limited by the dead-time in the readout system. The randoms rate varies between 5 and 50% with activity in the field of view of 10-100 MBq (0.27-2.7 mCi) with a timing gate of 20 ns. Initial results show that the randoms corrected sensitivity is >3-4 kcps/kBq/ml (120-150 kcps//spl mu/Ci/ml) for activities up to 30 MBq (0.81 mCi) in a 20 cm diameter water-filled phantom. However, these values include a high (60%) scatter fraction due to detector support structures. The camera has not yet been fully optimised and it is expected that the performance will substantially improve by further tuning and a reduction in the scattering structures. With a 40 cm axial FoV the camera is ideally suited to whole-body imaging in oncology.

Attenuation and scatter correction of list-mode data driven iterative and analytic image reconstruction algorithms for rotating 3D PET systems

Image reconstruction directly from list-mode data requires different data correction techniques to standard projection-data based reconstruction, particularly in the case of iterative reconstruction. Attenuation and scatter correction techniques have been developed for two list-mode data driven reconstruction algorithms (FAIR-B (iterative) and Atrax (analytic/iterative)) recently proposed by the authors, and the results compared with two popular projection-data based algorithms (FORE+OSEM (iterative) and FORE+FBP (analytic)). List-mode data driven algorithms require event-by-event correction schemes, or alternatively image space procedures, as no direct operations to the completely sampled projection data set can be practically carried out. The methods developed in this work allow correction of list-mode data driven EM-ML type algorithms, such as FAIR, as well as analytic list-mode algorithms. The correction schemes have been applied to simulated data from various activity and attenuating medium distributions for a rotating 3D PET system. Both list-mode algorithms show improvements over the projection-data based methods in some situations.

Fast accurate iterative three-dimensional Bayesian reconstruction for low-statistics positron volume imaging

Direct use of list-mode data for image reconstruction improves accuracy for some imaging systems, and permits fast reconstructions for low-statistics situations. A list-mode based three-dimensional implementation of an iterative Bayesian reconstruction algorithm has been developed (FAIR-B). The approach starts with an initial 2-D filtered backprojection (FBP) of Fourier rebinned data and employs a Gibbs prior to encourage images with local continuity, using the method of iterative conditional averages to obtain a sequence of estimates. Ten iterations are sufficient to significantly affect the image, incorporating the benefits of list-mode data and the Gibbs prior. The method has been tested with simulated data for rotating planar detector based systems and can offer improved noise contrast behaviour over FBP and list-mode driven expectation maximisation-maximum likelihood (EM-ML). However, for low-contrast regions whilst improved structural accuracy is still obtained, contrast losses are observed.

Image-space 3D scatter correction following list mode acquisition with a large-area positron camera

Positron volume images will be produced by the new large-area positron camera PETRRA directly from list mode data acquisition. The present method of image reconstruction is based on the backprojection then filter algorithm and no projection data are produced. A 2D method of correcting images in back-projection space has been developed using measurements made with the existing MUP-PET camera. A relationship between the attenuation and scatter fractions has been derived, for a limited axial field-of-view, for a range of cylindrical and elliptical phantoms. Image-space scatter distributions produced using this relationship are converted to backprojection-space by convolution with an experimentally-derived scatter response function (SRF). These scatter distributions are subtracted from the acquired back-projected image using the integrated scatter fraction for normalization. The resulting data are corrected using a multiplicative backprojection space attenuation matrix and deconvolved with an in-air response function. The method has been applied to a Monte Carlo simulated phantom, an experimental phantom and a patient study. The predicted scatter distribution agrees with the simulation to /spl sim//spl plusmn/10% and the correction method improves image contrast for all three data sets studied. However the use of a spatially invariant SRF leads to over-subtraction of scatter in the center of the phantom. In order for the method to be extended to PETRRA using the whole axial extent (40 cm) of the camera, simulations have been used to generate scatter-attenuation relationships for several phantoms over the whole axial FoV. The scatter response function is fitted to a Gaussian with the center displaced from the point of emission. Improved linearity and contrast were obtained with simulated phantoms. The method could be applied to any large-area positron camera acquiring data in list mode.

Singles transmission scanning for rotating positron cameras with large-area detectors

This paper describes the novel geometry and initial results of a singles transmission system for PETRRA, a whole-body PET camera with 400/spl times/600 mm BaF/sub 2/-TMAE detectors. Special frames mounted on to the front face of each detector support a collimated Cs-137 source in various positions. Cone-beam transmission data are acquired during 360/spl deg/ gantry rotation and images are reconstructed using the Feldkamp algorithm. The radial field of view (FOV) can be increased by using two source positions displaced either side of the central position. The truncated data acquired from each source position are combined to form a complete dataset. The axial FOV depends on the amount of collimation in the axial direction and could be increased by horizontal couch motion. Preliminary singles transmission images of both real and simulated thorax phantoms have been obtained. The effects of varying the linear sampling interval, the number of projections and the axial FOV have been studied. The singles transmission scanning system, described here and developed for PETRRA, could be applied to other PET cameras with large planar detectors.

Implementation of spiral PET on cameras with rotating planar detectors

Whole-body PET scanning is achieved by using couch movement to extend the field of view beyond the axial length of the detectors. For PET cameras with rotating planar detectors whole-body scanning can be achieved via spiral PET i.e. by simultaneous and continuous couch translation and gantry rotation. Higher SNRs are obtained when continuous couch motion is used compared with the more conventional step-and-shoot methods of whole-body PET acquisition. The large axial extent of PET cameras with rotating planar detectors and their inherent 3D geometry make these cameras ideally suited to continuous sampling and a spiral mode of acquisition for wholebody imaging. Hardware and software have been developed for list-mode spiral PET acquisition using the PETRRA camera. Image reconstruction is based on backprojection and filtering techniques. Previous optimization of the image reconstruction method and parameters using the MUP-PET camera has been adapted to account for the different geometries of the two cameras. To achieve accurate quantification, the data need to be corrected for non-uniform sampling (by application of both rotational and longitudinal weights), as well as for camera nonuniformity, randoms, deadtime, attenuation and scatter. High-quality whole-body images have been produced of phantom data acquired with short scanning times under ideal conditions of minimal scatter and randoms. Image quantification is still to be achieved on PETRRA via implementation of corrections for scatter and randoms. Spiral PET, as implemented on PETRRA, could be applied to gamma-camera PET systems, provided they offer continuous gantry rotation, couch translation and list-mode data acquisition.

Optimisation of source geometry for flood images to be used for nonuniformity corrections on large area detectors

Prior to reconstruction, emission data must be normalised to account for crystal nonuniformities and for geometric effects. The method adopted for normalisation of data acquired with the novel positron camera PETRRA is a direct method based on the inversion of a pair of flood images. In this work sources of several geometries for the flood acquisitions were tested on tomographic data. The different source geometries were: a /sup 68/Ge point source in static and tomographic mode, a 2D plane source filled with /sup 18/F, and a 3D uniform cylinder filled with /sup 18/F.

Backprojection-space 3D scatter correction for the PETRRA positron camera

Positron volume images have been produced by the new large-area positron camera PETRRA directly from list-mode data acquisition then reconstruction using backprojection then filter methods (BPF). These images have been corrected for scatter and attenuation using backprojection space methods. Two methods have been developed to provide scatter distributions. The first method uses a spatially-invariant scatter response function derived from measured point spread functions in a water-filled phantom. The second uses the scatter distributions measured at various points in the same phantom to allow for the spatial variance of the scatter. The correction methods have been tested using a uniformly-filled radioactive elliptical phantom (major and minor axes of 29.5 cm and 19.6 cm). Preliminary results show that both methods can improve the quantitative nature of the reconstructed images but that the use of spatially variant scatter distribution is superior. The methods could be applied to any large-area, rotating positron camera system which acquires data in list mode.