Judson Jones

Optimal gating compared to 3D and 4D PET reconstruction for characterization of lung tumours

PurposeWe investigated the added value of a new respiratory amplitude-based PET reconstruction method called optimal gating (OG) with the aim of providing accurate image quantification in lung cancer.MethodsFDG-PET imaging was performed in 26 lung cancer patients during free breathing using a 24-min list-mode acquisition on a PET/CT scanner. The data were reconstructed using three methods: standard 3D PET, respiratory-correlated 4D PET using a phase-binning algorithm, and OG. These datasets were compared in terms of the maximum SUV (SUVmax) in the primary tumour (main endpoint), noise characteristics, and volumes using thresholded regions of SUV 2.5 and 40% of the SUVmax.ResultsSUVmax values from the 4D method (13.7 ± 5.6) and the OG method (14.1 ± 6.5) were higher (4.9 ± 4.8%, p < 0.001 and 6.9 ± 8.8%, p < 0.001, respectively) than that from the 3D method (13.1 ± 5.4). SUVmax did not differ between the 4D and OG methods (2.0 ± 8.4%, p = NS). Absolute and relative threshold volumes did not differ between methods, except for the 40% SUVmax volume in which the value from the 3D method was lower than that from the 4D method (−5.3 ± 7.1%, p = 0.007). The OG method exhibited less noise than the 4D method. Variations in volumes and SUVmax of up to 40% and 27%, respectively, of the individual gates of the 4D method were also observed.ConclusionThe maximum SUVs from the OG and 4D methods were comparable and significantly higher than that from the 3D method, yet the OG method was visibly less noisy than the 4D method. Based on the better quantification of the maximum and the less noisy appearance, we conclude that OG PET is a better alternative to both 3D PET, which suffers from breathing averaging, and the noisy images of a 4D PET.

Variance reduction on randoms from coincidence histograms for the HRRT

A new algorithm for variance reduction on random coincidences (VRR) has been validated for the HRRT. VRR is crucial to achieve quantitation for low statistics dynamic studies reconstructed with iterative methods based on ordinary Poisson model. On HRRT, VRR cannot be performed in projection space since individual LORs are mixed after histogramming in parallel projection space using nearest neighbor approximation and axial compression. The proposed algorithm uses the classical random rate equation on the 4.5 109LORs. However, crystal singles are registered at block level and have lower deadtime than coincidences. Variations in layer identification with countrate were reported biasing random estimation from block singles. Our method overcomes these problems by estimating the singles per crystal from delayed coincidences. A singles map is created histogramming every delayed event into 2 singles. Each element represents the number of coincidences between that crystal and the ones in the 5 opposite coincident heads. The algorithm finds iteratively the crystal singles rates compatible with the delayed coincidence events. The method has been validated on decaying phantoms. We compared estimated and measured block singles to identify deadtime difference between singles and coincidences

SPMD cluster-based parallel 3D OSEM

This study empirically compares two approaches to parallel 3D OSEM that differ as to whether calculations are assigned to nodes by projection number or by transaxial plane number. For projection space decomposition (PSD), the forward projection is completely parallel, but backprojection requires a slow image synchronization. For image space decomposition (ISD), the communication associated with forward projection can be overlapped with calculation, and the communication associated with backprojection is more efficient. To compare these methods, an implementation of 3D OSEM for three PET scanners is developed that runs on an experimental, 9-node, 18-processor cluster computer. For selected benchmarks, both methods exhibit speedups in excess of 8 for 9 nodes, and comparable performance for the tested range of cluster sizes.

Impact of a high-performance communication network on cluster-based parallel iterative reconstruction

Assuming balanced computation, the scalability of an iterative reconstruction algorithm on a cluster computer will be determined by the tradeoff between shorter computation times on larger clusters versus longer communication times. To investigate the impact of improving communication performance, we assembled a cluster equipped with dual communication networks: Gigabit Ethernet and PCI-X Infiniband. The cluster consisted of 8 compute nodes and one fileserver node; each node had dual 3.6 GHz Intel Xeon processors and the two dual-ported communication networks. For image synchronization alone, PCI-X Infiniband was 3.8 times faster than Gigabit Ethernet. We benchmarked two parallel OSEM-3D reconstruction algorithms representing a range of image and sinogram sizes, for both a brain-only scanner (HRRT) and a whole-body scanner (HIREZ). In every case PCI-X Infiniband provided significantly superior reconstruction performance at all cluster sizes, and in all cases but one extended the range of cluster sizes over which we observed performance improvements.

The strategy of elastic Motion corrections

The Motion in PET studies degrades image quality and introduces bias and partial volume artifacts, which are critical considerations for high resolution scanners. There are two kinds of motion, such as rigid (e.g. brain) and nonrigid (e.g. respiratory and cardiac). Elastic motion correction is needed for nonrigid-motion artifacts. There are three basic steps in this approach, acquisition of a gating signal, extraction of elastic motion, and reconstruction. First, the gating signal is acquired by hardware, such as EKG, Belt, RPM, and MR, or from the analysis PET list-mode data. This is the most important step because, if the data are not properly gated, it is not possible to extract accurate motion vectors. Second, motion information for each gated signal can be extracted from CT for PET/CT or MRI for MR/PET. The motion information can also be extracted from the PET data themselves, and optical flow methods have been shown to be very robust in this approach. Third, image reconstruction with motion correction is commonly performed through summing gated images in a common reference frame. However, the combination of processed data with poor statistics generally results in high image noise and bias in the final image. A better approach is to incorporate motion information into the reconstruction process itself. Motion-correction reconstruction has been shown to produce less noise and bias in the image domain than conventional summing methods. The ideal method for motion correction in emission data should produce quantitatively accurate images which retain noise properties of conventional images, all while introducing no additional subject dose or inconvenience.

Data Processing Methods for a High Throughput Brain Imaging PET Research Center

We describe a computer system designed to meet the data processing needs of a high-volume brain PET research center based on the High Resolution Research Tomograph (HRRT). Listmode data are collected by an acquisition computer and stored on a high-speed disk. A workflow management program transfers the data through a gigabit network, rebins events into sinograms, and calculates correction factors. Reconstruction jobs are performed on a 64 processor cluster. We developed methods for dynamically allocating subclusters from the pool of available nodes, and reconstructing multiple images on multiple subclusters simultaneously. We also studied overall workflow. In our initial plan, scatter and randoms calculation unexpectedly became a bottleneck. We therefore adjusted our plan so that scatter estimation was performed initially in low resolution, and later expanded to high resolution.

FORE(J)+OSEM2D versus OSEM3D reconstruction for large aperture rotating LSO panel detector PET prototype

3D reconstruction on large aperture PET systems based on rotating LSO panel detectors is a challenge due to the large amount of data. When reconstruction time is critical, fast reconstruction methods are attractive despite the potential image quality loss due to rebinning approximation and the large polar aperture angle. This work validates a new data processing suite aimed at processing 3D data for two new PET scanner prototypes using two and five rotating LSO panel detectors, respectively. The processing includes /spl mu/-map reconstruction, generation of a 3D attenuation correction, scatter correction and image reconstruction with FORE(J)+OSEM2D and weighted OSEM3D schemes. The validation is based on reconstructing images for simulated wholebody FDG data with realistic random and scatter fractions and given statistics. The same processing suite was applied on experimental data acquired at intermediate statistics on a contrast sphere phantom and on a low statistics FDG wholebody study. Despite the large polar aperture angle, the image quality obtained with FORE(J)+OSEM2D was in excellent agreement with those obtained with time consuming methods such as OSEM3D. The loss in axial resolution from FOREJ to FORE was assessed from noiseless data. We also observed that, on low statistics wholebody data, FOREJ as implemented, required sinogram rebinning and smoothing to control the noise which offsets the benefits of the exact rebinning. The best images in terms of resolution and contrast were obtained with OSEM3D but they required precise normalization and deadtime corrections to avoid ring and banding artifacts.

Elastic motion correction for cardiac PET studies

Motion in PET studies degrades image quality and introduces bias, which is critical for high resolution scanners such as mCT, mMR, and HRRT. In this work we describe and compare motion corrected cardiac studies of cardiac gating and dual (cardiac and respiratory) gating. Elastic motion information is extracted using Mass Preservation Optical Flow (MPOF), and motion correction is performed during the reconstruction process. Elastic motion corrected PET cardiac studies show less noise than individual gated images, and higher resolution than images without motion correction. Studies utilizing three different tracers (FDG, Ammonia, and Rubidium) are shown in the results.

The effect of time-of-flight and point spread function modeling on 82 Rb myocardial perfusion imaging of obese patients

BackgroundThe effect of time-of-flight (TOF) and point spread function (PSF) modeling in image reconstruction has not been well studied for cardiac PET. This study assesses their separate and combined influence on 82Rb myocardial perfusion imaging in obese patients.MethodsThirty-six obese patients underwent rest-stress 82Rb cardiac PET. Images were reconstructed with and without TOF and PSF modeling. Perfusion was quantitatively compared using the AHA 17-segment model for patients grouped by BMI, cross-sectional body area in the scanner field of view, gender, and left ventricular myocardial volume. Summed rest scores (SRS), summed stress scores (SSS), and summed difference scores (SDS) were compared.ResultsTOF improved polar map visual uniformity and increased septal wall perfusion by up to 10%. This increase was greater for larger patients, more evident for patients grouped by cross-sectional area than by BMI, and more prominent for females. PSF modeling increased perfusion by about 1.5% in all cardiac segments. TOF modeling generally decreased SRS and SSS with significant decreases between 2.4 and 3.0 (P < .05), which could affect risk stratification; SDS remained about the same. With PSF modeling, SRS, SSS, and SDS were largely unchanged.ConclusionTOF and PSF modeling affect regional and global perfusion, SRS, and SSS. Clinicians should consider these effects and gender-dependent differences when interpreting 82Rb perfusion studies.Spanish AbstractAntecedentesEl efecto de los algoritmos de reconstrucción “time of flight” (TOF) y “point spread function” (PSF) en la reconstrucción de imágenes no ha sido bien estudiado para el PET cardiaco. Este estudio evalúa su influencia en por separado y combinado en los estudios de imagen de perfusión miocárdica con 82Rb en pacientes obesos.MétodosTreinta y seis pacientes obesos fueron sometidos a un PET cardiaco 82Rb en estrés y en reposo. Las imágenes fueron reconstruidas con y sin TOF y PSF. La perfusión fue comparada cuantitativamente utilizando el modelo segmentario AHA17 para pacientes agrupados por IMC, área corporal transversal in el campo de vista del escáner, sexo y volumen ventricular izquierdo miocárdico. Los puntajes sumados de reposo (SRS), los puntajes sumados de estrés (SSS) y el puntaje diferencial sumado (SDS) fueron comparados.ResultadosEl TOF mejoró la uniformidad visual del mapa polar e incrementó la perfusión de la pared septal hasta un 10%. Este incremento fue mayor para pacientes más grandes, más evidentemente en pacientes agrupados por área transversal que por IMC, y siendo más prominente en mujeres. El PSF aumentó la perfusión por cerca de 1.5% en todos los segmentos cardiacos. El TOF generalmente disminuyó el SRS y el SSS con disminuciones significativas entre 2.4 y 3 (P < .05), lo cual podría afectar la estratificación por riesgo; el SDS permanece igual. Con el modelamiento PSF, el SRS, el SSS y el SDS no presentaron cambios.ConclusiónEl TOF y el PSF afecta a la perfusión regional y global, el SRS y el SSS. Los clínicos deberían considerar estos efectos y las diferencias dependientes de sexo cuando se interpretan los estudios de perfusión con 82Rb.Chinese Abstract背景飞行时间(TOF)和点扩散函数(PSF)建模对于心脏 PET 成像重建的影响尚未完善建立。本研究评估其单独以及联合使用对肥胖病人行铷 82 心肌灌注成像的影响。方法36 个肥胖病人接受静息-负荷的铷 82 心脏 PET 成像扫描。图像分别在有无 TOF 和 PSF 建模的情况下被重建。病人按照 BMI、扫描仪视野下横断面的体表面积、性别和左室容积进行分组,采用 AHA 17节段模型量化对比灌注情况,比较静息灌注总积分(SRS),负荷灌注总积分(SSS)和灌注总积分差值(SDS)。结果TOF 改进了靶心图的视觉一致性, 间隔壁的灌注增加了10%。这种增加表现为: 体型越大的病人增加越大, 以横断面体表面积分组的病人比用 BMI 分组的病人增加更明显,女性比男性增加更突出。在所有的心脏节段中, PSF 建模增加了约 1.5% 的灌注。TOF 建模总体上显著降低了SRS和 SSS(在 2.4 和 3.0 之间, P < .05), 这会影响风险分层; SDS 保持不变。利用 PSF 建模, SRS, SSS 和 SDS 在很大程度上保持不变。结论TOF 和 PSF 建模影响局部和整体灌注、SRS 以及 SSS。当阅读铷 82 灌注图像时, 临床医生应该考虑这些因素的影响以及性别导致的不同。French AbstractContexteL’effet de la modélisation du temps de vol (TOF) et de la fonction d’étalement ponctuel (PSF) pour la reconstruction d’images n’a pas été bien étudiée pour la TEP en cardiologie. Cette étude évalue l’influence séparée et combinée des ces deux facteurs sur la perfusion myocardique par imagerie au 82Rb chez les patients obèses.MéthodesTrente-six patients obèses ont été soumis à une étude TEP repos-effort au 82Rb au repos. Les images ont été reconstruites avec et sans modélisation TOF et PSF. Les résultats de la perfusion myocardique a été comparée quantitativement en utilisant le modèle de 17 segments de l’American Heart Association (AHA). Les patients ont été groupés selon l’index de leur masse corporelle (IMC), et selon leur dimension corporelle transversale dans le champ de vision du scanner, sexe et volume myocardique ventriculaire gauche. Les score de perfusion myocardique au repos (SRS), après effort (SSS) et les scores différentiels (SDS) ont été comparés.RésultatsTOF améliore l’uniformité visuelle de la carte polaire et augmente la perfusion de la paroi septale de 10%. Cette augmentation est plus importante chez les patients de grande taille et plus apparente chez les patients groupés selon leur dimension corporelle transversale zone plutôt que par l’IMC, et plus élevée chez les femmes. La modélisation PSF augmente la perfusion d’environ 1,5% dans tous les segments cardiaques. La modélisation TOF diminue significativement les scores SRS et le SSS de 2,4 et 3,0 points (P < 0,05), ce qui peut changer la stratification; le score SDS est dans l’ensemble inchangé. Avec la modélisation PSF, SRS, SSS et SDS sont largement inchangés.ConclusionLa modélisation TOF et PSF affectent la perfusion régionale et globale, SRS et SSS. Les cliniciens devraient tenir compte de ces effets et des différences entre les sexes lors de l’interprétation 82Rb études de perfusion.

Ultrafast Elastic Motion Correction via Motion Deblurring

Patient motion during PET studies degrades image quality. Some types of motion (e.g. brain) can be modeled as rigid-body transformations, whereas others (e.g. respiratory and cardiac), are more complex, involve deformations of the imaged organs, and require Elastic Motion Correction (EMC). The conventional way (cEMC) to handle the dense information needed for EMC is to divide the acquired data into multiple respiratory, cardiac, or dual “gates”, where motion is minimal within each gate. Motion fields can then be calculated between a reference gate and all other gates via optical flow. These motion fields can then be used in a cEMC iterative reconstruction process by warping the reference image to each gated image before forward projection and transposing the gated correction factors back to the reference image after backward projection. In this algorithm, the number of forward and backprojections, processing time, and memory requirements are proportional to the number of gates. In this paper, we introduce a faster algorithm, Elastic Motion Deblurring (EMDB), which does not depend on the number of gates. Instead, a Mass Preservation Optical Flow (MPOF) algorithm is used to calculate a blurring kernel from the reference gate to the static (motion blurred) image only. This novel approach reduces the processing time and hardware requirements for iterative EMC reconstruction.