Stephan Blinder

Accurate Event-Driven Motion Compensation in High-Resolution PET Incorporating Scattered and Random Events

With continuing improvements in spatial resolution of positron emission tomography (PET) scanners, small patient movements during PET imaging become a significant source of resolution degradation. This work develops and investigates a comprehensive formalism for accurate motion-compensated reconstruction which at the same time is very feasible in the context of high-resolution PET. In particular, this paper proposes an effective method to incorporate presence of scattered and random coincidences in the context of motion (which is similarly applicable to various other motion correction schemes). The overall reconstruction framework takes into consideration missing projection data which are not detected due to motion, and additionally, incorporates information from all detected events, including those which fall outside the field-of-view following motion correction. The proposed approach has been extensively validated using phantom experiments as well as realistic simulations of a new mathematical brain phantom developed in this work, and the results for a dynamic patient study are also presented.

NEMA NU 4-2008 Comparison of Preclinical PET Imaging Systems

The National Electrical Manufacturers Association (NEMA) standard NU 4-2008 for performance measurements of small-animal tomographs was recently published. Before this standard, there were no standard testing procedures for preclinical PET systems, and manufacturers could not provide clear specifications similar to those available for clinical systems under NEMA NU 2-1994 and 2-2001. Consequently, performance evaluation papers used methods that were modified ad hoc from the clinical PET NEMA standard, thus making comparisons between systems difficult. Methods: We acquired NEMA NU 4-2008 performance data for a collection of commercial animal PET systems manufactured since 2000: microPET P4, microPET R4, microPET Focus 120, microPET Focus 220, Inveon, ClearPET, Mosaic HP, Argus (formerly eXplore Vista), VrPET, LabPET 8, and LabPET 12. The data included spatial resolution, counting-rate performance, scatter fraction, sensitivity, and image quality and were acquired using settings for routine PET. Results: The data showed a steady improvement in system performance for newer systems as compared with first-generation systems, with notable improvements in spatial resolution and sensitivity. Conclusion: Variation in system design makes direct comparisons between systems from different vendors difficult. When considering the results from NEMA testing, one must also consider the suitability of the PET system for the specific imaging task at hand.

Statistical dynamic image reconstruction in state-of-the-art high-resolution PET.

Modern high-resolution PET is now more than ever in need of scrutiny into the nature and limitations of the imaging modality itself as well as image reconstruction techniques. In this work, we have reviewed, analysed and addressed the following three considerations within the particular context of state-of-the-art dynamic PET imaging: (i) the typical average numbers of events per line-of-response (LOR) are now (much) less than unity, (ii) due to the physical and biological decay of the activity distribution, one requires robust and efficient reconstruction algorithms applicable to a wide range of statistics and (iii) the computational considerations in dynamic imaging are much enhanced (i.e., more frames to be stored and reconstructed). Within the framework of statistical image reconstruction, we have argued theoretically and shown experimentally that the sinogram non-negativity constraint (when using the delayed-coincidence and/or scatter-subtraction techniques) is especially expected to result in an overestimation bias. Subsequently, two schemes are considered: (a) subtraction techniques in which an image non-negativity constraint has been imposed and (b) implementation of random and scatter estimates inside the reconstruction algorithms, thus enabling direct processing of Poisson-distributed prompts. Both techniques are able to remove the aforementioned bias, while the latter, being better conditioned theoretically, is able to exhibit superior noise characteristics. We have also elaborated upon and verified the applicability of the accelerated list-mode image reconstruction method as a powerful solution for accurate, robust and efficient dynamic reconstructions of high-resolution data (as well as a number of additional benefits in the context of state-of-the-art PET).

Implementation of an analytically based scatter correction in SPECT reconstructions

Photon scattering is one of the main effects contributing to the degradation of image quality and to quantitative inaccuracy in nuclear imaging. We have developed a scatter correction based on a simplified version of the analytic photon distribution (APD) method, and have implemented it in an iterative image reconstruction algorithm. The scatter distributions generated using this approach were compared to those obtained using the original APD method. Reconstructions were performed using computer simulations, phantom experiments, and patient data. Images corrected for scatter, attenuation, and collimator blurring were compared to images corrected only for attenuation and collimator blurring. In the simulation studies, results were compared to an ideal situation in which only the primary (unscattered) photon data were reconstructed. Results showed that in all cases, the scatter-corrected images demonstrated substantially improved image contrast relative to no scatter correction. For simulated data, scatter-corrected images had very similar contrast and noise properties to the primary-only reconstructions. Additional work is required to further reduce the computation times to clinically viable amounts.

Experimental verification of 3D detector response compensation using the OSEM reconstruction method

Detector blurring is a major factor in SPECT image degradation and several methods to include detector response compensation (DRC) into iterative reconstruction algorithms have been proposed. Here, we present a study to validate and quantify the performance of such correction using an approach that we have developed and which combines 3D DRC with a non uniform attenuation correction. The objective of this study was to estimate what is the best spatial resolution that could be experimentally achieved in an image reconstructed with both attenuation and resolution recovery corrections. For this purpose we have performed several tests involving physical phantoms and computer simulations. In experiments with capillary tubes in air the resolution came close to the pixel size and was equal to 3.6 mm. DRC convergence was slower when background activity was present but still resulted in an excellent, close to the pixel size, resolution in the image. Similar results were obtained for extended sources scanned in air and with background activity. In this last case, however, the speed of convergence of the reconstruction algorithm was substantially decreased. Additionally, we present the results of a study involving a simulated wrist model which showed that hot and cold defects of the size equal to /spl ap/6 mm can be detected when DRC was used. In summary, our experiments and simulations have demonstrated that including 3D DRC in the iterative reconstruction algorithm resulted in great improvement of image resolution and noise reduction that increased visibility of small details.

A scatter-corrected list-mode reconstruction and a practical scatter/random approximation technique for dynamic PET imaging.

We describe an ordinary Poisson list-mode expectation maximization (OP-LMEM) algorithm with a sinogram-based scatter correction method based on the single scatter simulation (SSS) technique and a random correction method based on the variance-reduced delayed-coincidence technique. We also describe a practical approximate scatter and random-estimation approach for dynamic PET studies based on a time-averaged scatter and random estimate followed by scaling according to the global numbers of true coincidences and randoms for each temporal frame. The quantitative accuracy achieved using OP-LMEM was compared to that obtained using the histogram-mode 3D ordinary Poisson ordered subset expectation maximization (3D-OP) algorithm with similar scatter and random correction methods, and they showed excellent agreement. The accuracy of the approximated scatter and random estimates was tested by comparing time activity curves (TACs) as well as the spatial scatter distribution from dynamic non-human primate studies obtained from the conventional (frame-based) approach and those obtained from the approximate approach. An excellent agreement was found, and the time required for the calculation of scatter and random estimates in the dynamic studies became much less dependent on the number of frames (we achieved a nearly four times faster performance on the scatter and random estimates by applying the proposed method). The precision of the scatter fraction was also demonstrated for the conventional and the approximate approach using phantom studies.

Application of texture analysis to DAT SPECT imaging: Relationship to clinical assessments

Dopamine transporter (DAT) SPECT imaging is increasingly utilized for diagnostic purposes in suspected Parkinsonian syndromes. We performed a cross-sectional study to investigate whether assessment of texture in DAT SPECT radiotracer uptake enables enhanced correlations with severity of motor and cognitive symptoms in Parkinsons disease (PD), with the long-term goal of enabling clinical utility of DAT SPECT imaging, beyond standard diagnostic tasks, to tracking of progression in PD. Quantitative analysis in routine DAT SPECT imaging, if performed at all, has been restricted to assessment of mean regional uptake. We applied a framework wherein textural features were extracted from the images. Notably, the framework did not require registration to a common template, and worked in the subject-native space. Image analysis included registration of SPECT images onto corresponding MRI images, automatic region-of-interest (ROI) extraction on the MRI images, followed by computation of Haralick texture features. We analyzed 141 subjects from the Parkinsons Progressive Marker Initiative (PPMI) database, including 85 PD and 56 healthy controls (HC) (baseline scans with accompanying 3 T MRI images). We performed univariate and multivariate regression analyses between the quantitative metrics and different clinical measures, namely (i) the UPDRS (part III - motor) score, disease duration as measured from (ii) time of diagnosis (DD-diag.) and (iii) time of appearance of symptoms (DD-sympt.), as well as (iv) the Montreal Cognitive Assessment (MoCA) score. For conventional mean uptake analysis in the putamen, we showed significant correlations with clinical measures only when both HC and PD were included (Pearson correlation r = − 0.74, p-value < 0.001). However, this was not significant when applied to PD subjects only (r = − 0.19, p-value = 0.084), and no such correlations were observed in the caudate. By contrast, for the PD subjects, significant correlations were observed in the caudate when including texture metrics, with (i) UPDRS (p-values < 0.01), (ii) DD-diag. (p-values < 0.001), (iii) DD-sympt (p-values < 0.05), and (iv) MoCA (p-values < 0.01), while no correlations were observed for conventional analysis (p-values = 0.94, 0.34, 0.88 and 0.96, respectively). Our results demonstrated the ability to capture valuable information using advanced texture metrics from striatal DAT SPECT, enabling significant correlations of striatal DAT binding with clinical, motor and cognitive outcomes, and suggesting that textural features hold potential as biomarkers of PD severity and progression.

Scanning rats on the high resolution research tomograph (HRRT): A comparison study with a dedicated micro‐PET

PURPOSE The Siemens ECAT high resolution research tomograph (HRRT) is a dedicated human brain PET camera with a 6% absolute sensitivity and a (2.3 mm)(3) spatial resolution, improving to (1.8 mm)(3) when point spread function (PSF) modeling algorithms are used. These values are very close to those of dedicated small animal PET cameras such as the Siemens microPET FOCUS 120 (F120). The larger axial and transaxial field of view of the HRRT compared to the F120 allows, in principle, for simultaneous imaging of several rodents thus potentially reducing scanning costs and time. This study investigates the feasibility of using the HRRT for quantitative small animal brain studies. METHODS We compare, in terms of magnitude, reproducibility, and asymmetry, the nondisplaceable tissue input binding potentials (BP(ND)) in the striata obtained from [(11)C]methylphenidate scans of the same rats imaged on both the F120 and the HRRT. The animal studies are complemented by a phantom study aimed at investigating noise properties relevant to the size of typical regions of interest used in rat brain image analysis. RESULTS (i) The BP(ND) values obtained from HRRT data are lower than those obtained on the F120 by 38% when PSF modeling is not used, while they are 7% higher with PSF modeling. (ii) The within animal reproducibility on the HRRT is 18% without PSF modeling, worse than the 6% reproducibility on the F120, and is even further degraded to a value of 27% with the use of PSF modeling. (iii) The asymmetry between the left and right striatum in healthy rats worsens from 4.7% in the F120 images to 7.8% in the HRRT images reconstructed without PSF modeling, and is even worse with a value of 14.8% when PSF modeling is used. (iv) Overshooting artifacts and clumpiness in the noise structure of the HRRT images reconstructed with PSF modeling are clearly visible. CONCLUSIONS The spatial resolution achieved on the HRRT without the use of resolution recovery techniques is not sufficient to allow for reliable quantitative small animal brain imaging. While PSF modeling in the reconstruction of the HRRT images in principle improves the resolution close to the level of the F120, it also introduces small scale nonuniformity artifacts and overshooting artifacts which preclude reliable quantitative small animal brain imaging on the HRRT.

A Scatter Calibration Technique for Dynamic Brain Imaging in High Resolution PET

We describe a scatter calibration technique which improves the quantitative accuracy of the positron emission tomography data in specific scanning conditions: i.e., scans with high random fraction (RF) and/or low number of counts. Such a situation is often encountered in dynamic imaging on scanners with a large number of lines-of-response (LOR) such as the high resolution research tomograph (HRRT). In this paper, we first describe how high RFs and low number of counts affect the scatter scaling process. We then demonstrate experimentally, with phantom studies, the bias in the scatter estimate introduced by the commonly used tail-fitting technique employed in the single scatter simulation (SSS) method. A significant bias in scatter fraction (SF) was found for frames which contain a RF higher than 50% and/or with a number of counts less than 20 M. Finally, we present a new scatter scaling technique which compensates this bias. The scatter calibration technique is based on using the scatter estimate obtained from a reference frame, in which the bias due to high RFs and low number of counts is minimized, to calibrate the scatter in each dynamic frame. The calibration also separately accounts for the change in SF due to the pulse pile-up effect. A much more consistent and accurate SF value was assigned to each segment of the scatter sinogram thus leading to a more quantitative reconstructed image with a better axial uniformity after the scatter calibration. The new calibration technique was tested with phantom, monkey, and human data and was found to significantly improve the quantitative aspect of the early frames: such improvement is expected to positively affect the feasibility of rather novel image analysis methods, such as determination of image derived input function.

Investigation of Subject Motion Encountered During a Typical Positron Emission Tomography Scan

Subject motion has a known detrimental effect on brain Positron Emission Tomography image resolution. Numerous motion compensation techniques exist to address this issue, however prior to their application every effort should be made to limit subject motion. Using a Polaris motion tracking system subject motion was observed under typical scanning conditions for both healthy and Parkinsons disease (PD) volunteers. Motions in the range of 0 to 5 mm were observed for healthy subjects, and 0 to 20 mm for PD subjects. The most common source of motion was due to interaction between the subject and the attending nurse/scanning staff, especially during examination of the subjects symptoms (motions up to 8 mm). Less common activities resulting in significant motions were the use of a bedpan (20 mm), the removal of a cushion from under the subjects legs (5 mm) and leg readjustments (3 mm). Awareness of the effect each of these activities had on head motion can be used to motivate further limitations on these motions. Measured motions were also extrapolated to various regions in the brain, specifically the cerebellum, occipital cortex, and striatum. Subject head rotation about the vertical and horizontal axes resulted in the greatest displacement of regions in the cerebellum, while rotations about the subjects long axis primarily impacted the displacement of the occipital cortex region. This measurement provides motion related information about the expected accuracy of time activity curves for different brain regions.