Eiji Yoshida

First Human Brain Imaging by the jPET-D4 Prototype With a Pre-Computed System Matrix

The jPET-D4 is a novel brain PET scanner which aims to achieve not only high spatial resolution but also high scanner sensitivity by using 4-layer depth-of-interaction (DOI) information. The dimensions of a system matrix for the jPET-D4 are 3.3 billion (lines-of-response) times 5 million (image elements) when a standard field-of-view (FOV) of 25 cm diameter is sampled with a (1.5 mm)3 voxel . The size of the system matrix is estimated as 117 petabytes (PB) with the accuracy of 8 bytes per element. An on-the-fly calculation is usually used to deal with such a huge system matrix. However we cannot avoid extension of the calculation time when we improve the accuracy of system modeling. In this work, we implemented an alternative approach based on pre-calculation of the system matrix. A histogram-based 3D OS-EM algorithm was implemented on a desktop workstation with 32 GB memory installed. The 117 PB system matrix was compressed under the limited amount of computer memory by (1) eliminating zero elements, (2) applying the DOI compression (DOIC) method and (3) applying rotational symmetry and an axial shift property of the crystal arrangement. Spanning, which degrades axial resolution, was not applied. The system modeling and the DOIC method, which had been validated in 2D image reconstruction, were expanded into 3D implementation. In particular, a new system model including the DOIC transformation was introduced to suppress resolution loss caused by the DOIC method. Experimental results showed that the jPET-D4 has almost uniform spatial resolution of better than 3 mm over the FOV. Finally the first human brain images were obtained with the jPET-D4.

Annihilation photon acollinearity in PET: volunteer and phantom FDG studies

Annihilation photon acollinearity is a fundamental but little investigated problem in positron emission tomography (PET). In this paper, the cause of the angular deviation from 180.00 degrees is described as well as how to evaluate it under conditions of a spatially distributed radiation source and a limited acquisition time for the human body. A relationship between the shape of the photopeak spectrum and the angular distribution is formulated using conservation laws of momentum and energy over the pair annihilation. Then the formula is used to evaluate the acollinearity for a pool phantom and the human body with FDG injected. The angular distribution for the pool phantom agrees well with that for pure water which had been directly measured by Colombino et al in 1965 (Nuovo Cimento 38 707-23), and also with that for the human body determined in this study. Pure water can be considered as a good approximation of the human body regarding the angular deviation. The blurring coefficient to be multiplied by the ring diameter in calculations of the PET spatial resolution is experimentally determined for the first time as 0.00243 +/- 0.00014; this is 10% larger than the value widely used by investigators.

Evaluation of static physics performance of the jPET-D4 by Monte Carlo simulations

The jPET-D4 is the first PET scanner to introduce a unique four-layer depth-of-interaction (DOI) detector scheme in order to achieve high sensitivity and uniform high spatial resolution. This paper compares measurement and Monte Carlo simulation results of the static physics performance of this prototype research PET scanner. Measurement results include single and coincidence energy spectra, point and line source sensitivities, axial sensitivity profile (slice profile) and scatter fraction. We use GATE (Geant4 application for tomographic emission) as a Monte Carlo radiation transport model. Experimental results are reproduced well by the simulation model with reasonable assumptions on characteristic responses of the DOI detectors. In a previous study, the jPET-D4 was shown to provide a uniform spatial resolution as good as 3 mm (FHWM). In the present study, we demonstrate that a high sensitivity, 11.3 +/- 0.5%, is provided at the FOV centre. However, about three-fourths of this sensitivity is related to multiple-crystal events, for which some misidentification of the crystal cannot be avoided. Therefore, it is crucial to develop a more efficient way to identify the crystal of interaction and to reduce misidentification in order to make use of these high performance values simultaneously. We expect that effective sensitivity can be improved by replacing the GSO crystals with more absorptive crystals such as BGO and LSO. The results we describe here are essential to take full advantage of the next generation PET systems that have DOI recognition capability.

A multiplex “OpenPET” geometry to extend axial FOV without increasing the number of detectors

At the last MIC conference, we proposed an “OpenPET” geometry, which consists of two axially separated detector rings of axial length W. A long and continuous field-of-view (FOV) including a 360-degree open gap G between two detector rings can be imaged though iterative image reconstruction. In addition to providing stress-less PET scanning and simultaneous PET/CT, the OpenPET is expected to lead to realization of in-beam PET. The OpenPET also extends the axial FOV with a limited number of detectors. The range of G to obtain the axially continuous FOV is represented as aW (0≪a≪=1). The extension gain, which is defined as [extended axial FOV size] / [original axial FOV size], is limited to (2+a)/2. In order to increase the gap and the axial FOV, therefore, we have to extend W itself. However the extension of W results in cost enhancement of the scanner. In this paper, we propose a new “multiplex OpenPET” geometry, which can extend axial FOV without increasing the number of detectors. In the first step, two detector rings of axial length W are separated by the gap aW, and we have a pseudo scanner, unit [1] , of axial length W [1] =(2+a)W. In the second step, two unit [1] s can be separated by the gap aW [1] , and we have unit [2] of axial FOV W [2] =(2+a)2W. After repeating these steps N times, we finally have unit [N] , which has a long axial FOV of W [N] =(2+a)NW. Here the extension gain is represented as {(2+a)/2}N. Theoretically it can be increased to an unlimited extent, although there is no gain in sensitivity. Simulation results show that fine reconstructed images are obtained by the multiplex OpenPET geometry. In addition to the extended gap, the proposed geometry is expected to help realize an affordable entire body PET scanner that enables whole body dynamic imaging.

A Multiplex “OpenPET” Geometry to Extend Axial FOV Without Increasing the Number of Detectors

We have proposed an ldquoOpenPETrdquo geometry, which consists of two axially separated detector rings, each of axial length W. A long and continuous field-of-view (FOV) including a 360-degree open gap G between two detector rings can be imaged through iterative image reconstruction. In addition to providing stress-less PET scanning and simultaneous PET/CT, the OpenPET is expected to lead to realization of in-beam PET. The OpenPET also extends the axial FOV with a limited number of detectors. However, the axial FOV is limited to 3 W because the maximum limit of G to obtain the axially continuous FOV is W. In this paper, therefore, we propose an alternative geometry to extend axial FOV even more without increasing the number of detectors. The proposed geometry consists of multiple detector rings separated by multiple gaps. By optimizing each width of the gaps based on a new concept of multiplex geometry of the OpenPET, the axial FOV can be theoretically increased to an unlimited extent without increasing the number of detectors. The multiplex OpenPET geometry was compared with the standard OpenPET and the conventional PET using numerical simulation data and experimental data. The results show that similar reconstructed images are obtained by three geometries. The proposed geometry is expected to help realize an affordable entire body PET scanner that enables whole body dynamic imaging.

A DOI-Dependent Extended Energy Window Method to Control Balance of Scatter and True Events

In a conventional PET scanner, coincidence events are limited by the energy window for detection of photoelectric events. In contrast, Compton scatter events occur both in a patient and in detector crystals. Scatter events within the patient cause scatter coincidences, but scatter events within crystals have useful information for an activity distribution. The PET scanner with an extended energy window has higher sensitivity but higher scatter fraction than the PET scanner with a default energy window. In this work, we develop a scatter reduction method using a depth-of-interaction (DOI) detector. The DOI detector can make an upper layer act as an absorber of patient scatter, thus shielding a lower layer. Therefore, Compton scatter events with an interaction at the lower layer mainly have more crystal scatter events than patient scatter events. Our method proposes a different energy window for each layer. The energy window of the upper layer is limited to the region of photopeak events, the same as done for the conventional PET scanner. The energy window of the lower layer is extended to the region of crystal scatter events and photopeak events. We tested our proposed technique for a whole-body PET scanner using GATE simulations. Simulation results show the DOI-PET scanner can provide more true events while keeping a low scatter fraction using our scatter reduction method. We conclude this method is promising for obtaining high image quality using DOI information and energy information.

Preliminary study on potential of the jPET-D4 human brain scanner for small animal imaging

ObjectiveOne trend in positron emission tomography (PET) instrumentation over the last decade has been the development of scanners dedicated to small animals such as rats and mice. Thicker crystals, which are necessary to obtain higher sensitivity, result in degraded spatial resolution in the peripheral field-of-view (FOV) owing to the parallax error. On the other hand, we are developing the jPET-D4, which is a dedicated human brain PET scanner that has a capability for depth-of-interaction (DOI) measurement. Although its crystal width is about twice that of commercially available small animal PET scanners, we expect the jPET-D4 to have a potential for small animal imaging by making full use of the DOI information. In this article, we investigate the jPET-D4’s potential for small animal imaging by comparing it with the microPET Focus220, a state-of-the-art PET scanner dedicated to small animals.MethodsThe jPET-D4 uses four-layered GSO crystals measuring 2.9 mm × 2.9 mm × 7.5 mm, whereas the microPET Focus220 uses a single layer of LSO crystals measuring 1.5 mm × 1.5 mm × 10.0 mm. First, the absolute sensitivity, counting rate performance and spatial resolution of both scanners were measured. Next a small hot-rod phantom was used to compare their imaging performance. Finally, a rat model with breast tumors was imaged using the jPET-D4.ResultsThanks to the thicker crystals and the longer axial FOV, the jPET-D4 had more than four times higher sensitivity than the microPET Focus220. The noise equivalent counting-rate performance of the jPETD4 reached 1,024 kcps for a rat-size phantom, whereas that of the microPET Focus220 reached only 165 kcps. At the center of the FOV, the resolution was 1.7 mm for the microPET Focus220, whereas it was 3.2 mm for the jPET-D4. On the other hand, the difference of resolution became smaller at the off-center position because the radial resolution degraded faster for the microPET Focus220. The results of phantom imaging showed that the jPET-D4 was comparable to the microPET Focus220 at the off-center position even as the microPET Focus220 outperformed the jPET-D4 except for the peripheral FOV.ConclusionsThe jPET-D4 human brain PET scanner, which was designed to achieve not only high resolution but also high sensitivity by measuring DOI information, was proven to have a potential for small animal imaging.

Influence of TOF information in OpenPET image reconstruction

We have proposed an “OpenPET” geometry, which consists of two axially separated detector rings. The central point of the field of view, where the highest sensitivity is obtained, is opened. The OpenPET mainly has three applications, namely, simultaneous PET/CT, extension of the axial FOV, and in-beam PET. In our previous report, we showed that axial spatial resolution, which is degraded with the extended gap due to the parallax error, can be recovered by use of depth-of-interaction (DOI) detectors. On the other hand, image reconstruction of the OpenPET is an incomplete problem because projection data do not satisfy Orlovs condition. Low-frequency components are missing in oblique LORs, i.e., LORs with large ring differences. The gap would suffer from loss of low-frequency components because the gap, where direct LORs (i.e., LORs in the same ring) do not exist, is imaged only from the oblique LORs. In this paper, we focused on time-of-flight (TOF) information, which is expected to compensate for the loss of low-frequency components in the gap. We investigated influence of TOF information in the OpenPET image reconstruction through numerical simulations. Simulated OpenPET scanner had dual detector rings (827.0 mm in diameter and axial length of 153.6 mm each) separated by a gap of 153.6 mm. We supposed that the detectors had a DOI capability of 8 layers and had a TOF capability of 400 ps FWHM resolution. For the non-TOF case, hot-spot objects, which are commonly seen in cancer diagnosis with use of FDG, were imaged without any artifacts, but objects containing more low-frequency components suffered from strong distortion. However, these artifacts were effectively reduced by using TOF information. These results showed that TOF information can compensate for low-frequency components missing in the gap.

A DOI-dependent extended energy window method to control balance of scatter and true events

In a conventional PET scanner, coincidence events are limited by the energy window for detection of photoelectric events. In contrast, Compton scatter events occur both in a patient and detector crystals. Scatter events within the patient cause scatter coincidence, but scatter events within crystal have useful information for an activity distribution. The PET scanner with an extended energy window has higher sensitivity but higher scatter fraction than the PET scanner with a default energy window. In this work, we develop a scatter reduction method using a depth-of-interaction (DOI) detector. The DOI detector can make an upper layer act as an absorber of patient scatter for a lower layer. Therefore, crystal scatter with an interacting lower layer has a potential to discriminate patient scatter. Our method has a different energy window for each layer. The energy window of the upper layer is limited to the region of photoelectric events. The energy window of the lower layer is extended to the region of crystal scatter and photoelectric events. We applied our proposal technique for a whole-body PET scanner using GATE simulation. The DOI- PET scanner can provide more true events while keeping a low scatter fraction using this scatter reduction method. Based on simulation results, this method is judged to be promising for obtaining high image quality using DOI information and energy information.

Parallel Implementation of 3-D Iterative Reconstruction With Intra-Thread Update for the jPET-D4

One way to speed-up iterative image reconstruction is by parallel computing with a computer cluster. However, as the number of computing threads increases, parallel efficiency decreases due to network transfer delay. In this paper, we proposed a method to reduce data transfer between computing threads by introducing an intra-thread update. The update factor is collected from each slave thread and a global image is updated as usual in the first K sub-iteration. In the rest of the sub-iterations, the global image is only updated at an interval which is controlled by a parameter L. In between that interval, the intra-thread update is carried out whereby an image update is performed in each slave thread locally. We investigated combinations of K and L parameters based on parallel implementation of RAMLA for the jPET-D4 scanner. Our evaluation used four workstations with a total of 16 slave threads. Each slave thread calculated a different set of LORs which are divided according to ring difference numbers. We assessed image quality of the proposed method with a hotspot simulation phantom. The figure of merit was the full-width-half-maximum of hotspots and the background normalized standard deviation. At an optimum K and L setting, we did not find significant change in the output images. We also applied the proposed method to a Hoffman phantom experiment and found the difference due to intra-thread update was negligible. With the intra-thread update, computation time could be reduced by about 23%.