Yongjie Jin

A Fast Accuracy Crystal Identification Method Based on Fuzzy C-Means (FCM) Clustering Algorithm for MicroPET

A high resolution detector is being developed for our small animal position emission tomography (MicroPET). The detector unit consist of 8x8 crystal blocks, coupled to four photomultiplier tubes (PMTs). Each scintillation event is mapped in a two dimensional (2-D) position through the relative ratio of the output signals of the PMTs. Crystal Look-up table (CLT) used in ThuMicroPET scanner defines the matching relation between signal position of a detected event to a corresponding detector pixel location. It has a direct impact on imaging quality and brings significant influence to the gantry overall performance. However, the currently used method involves a lot of human interaction for CLT corrections, and cannot be implemented as a general process due to its complexity. This paper introduces a fast accuracy method based on Fuzzy C-Means (FCM) Clustering Algorithm for crystal identification. In the FCM, a cluster center and a fuzzy partition matrix of individual events in the 2-D position are defined. By iteratively updating the cluster centers and the membership grades for each event, we can move the cluster center to the right location in a short time, based on minimizing objective function that represents the distance from any given events to a cluster center weighted by its membership grade. The preliminary result shows that FCM can be used effectively in CLT construction, which significantly reduces the time, and brings excellent accuracy than we expected.

A simple smart time-to-digital convertor based on vernier method for a high resolution LYSO MicroPET

We proposed a new and low cost design of time-to-digital converter (TDC) based on Vernier method using only one FPGA EPF10K30ATI144-3 with 1.3 ns timing resolution performance. Neither ECL (emitter-coupled logic) circuit nor high frequency clock was used in this design, which greatly reduced the complication and the power supply. We used two oscillators with slightly different frequencies to measure small time interval. All the time-to-digital converter functions were implemented on only one low-cost Altera FLEX II Family device. Our preliminary results showed: 1) Vernier TDC had a less than 1.3 ns timing resolution that met the demand for coincidence measurement for LYSO MicroPET. 2) No inacceptable degradation was observed in time resolution as of the number of the sliding jaw clock circulations was increased before coincidence. 3) Vernier TDC accomplished on FPGA had good stability with temperature. In a sum, we made a sufficient proof of high resolution and good stability of the proposed Vernier TDC design. Now, we are planning to achieve higher time resolution and higher stability by high-performance FPGA using this smart Vernier TDC method. And such simple design is being applied to our animal PET.

Comparing crystal identification algorithms for PET block detectors

A high resolution detector is being developed for our small animal position emission tomography (MicroPET). Each scintillation event is mapped in a two dimensional (2-D) position through the relative ratio of the output signals of the PMTs. Crystal Look-up table (CLT) used in ThuMicroPET scanner defines the matching relation between signal position of a detected event to a corresponding detector pixel location. It has a direct impact on imaging quality and brings significant influence to the gantry overall performance. This paper compares Fuzzy C-Means (FCM) Clustering Algorithm, Self Organizing Feature Map (SOFM) neural network, and Watersheds algorithm for crystal identification. The primary result shows, the watersheds method has the fast speed and higher accuracy for crystal identification but sensitive with preprocessing; and SOFM has advantage of on-line monitoring and training, with also relatively reasonable accuracy, but unluckily long time cost. FCM have lower stability and accuracy in this experiment. In addition we present a new simple but powerful parameter Interaction Ratio (IR), to quantitatively assess the performance of Crystal Identification of detectors. As proved as experiment, not only it could be quantitatively evaluate the segmentation method generating LUT, but also assess the performance of block detector itself.

A neighborhood standard deviation based algorithm for generating PET crystal position maps

Positron emission tomography (PET) is typically based on 2-D array of scintillation crystals coupled to photon detector and decoded by the Anger-logic. The decoded result is a pseudo-position of the gamma interaction. A crystal position map (CPM) generated from the flood histogram is used as a crystal look-up table (CLT) to assign each pseudo-position to a specific crystal. It is crucial that the accuracy of CPMs affects the detectors spatial resolution. In this paper, we developed a neighborhood standard deviation (NSD) based algorithm for generating CPM. We first calculated the NSD of each pixel in the flood histogram including the x and y directions. NSD maps have strips whose peaks highly correspond to the valley of the flood histogram. The peaks were identified by fitting the profiles of NSD to Gaussian mixture functions using nonlinear least-square method. Using the peaks, the CPM was generated by a scan line algorithm. The proposed algorithm was applied in an animal PET system. 115 of 120 detector blocks can be automatically segmented in ~1000 s. A hot rod phantom experiment was performed, and the reconstruction results showed that the one with CPM generated by NSD based automatic method achieved higher spatial resolution than the one with CPM generated by manual segmentation. We concluded that the proposed method is fast, robust and high accuracy.

High accuracy geometrical calibration for half-mm animal SPECT imaging

In this study we investigate a specific animal SPECT imaging system with a clinical SPECT detector and a multipinhole collimator insert. High accuracy geometrical calibration is a crucial technique in terms of achieving the high SPECT imaging resolution. The calibration steps include: 1) The SPECT system is modeled with a geometrical projection equation; 2) The centroids of the point source projection are calculated from the acquired experimental data; and 3)The geometrical parameters is determined by minimizing a nonlinear least square cost function. With this calibration technique, Experimental studies shows the best achievable reconstruction resolution is 0.5 mm. Furthermore, the uniqueness of the calibration solution is investigated with a SVD based approach. By experimentally calculating centriod variance of point source projections, quantitative calibration accuracy of the proposed calibration method is studied and suggested minimum projection number of the point source experiments to achieve desired calibration accuracy is given.

Design and feasibility studies of a high-resolution and low-cost small animal SPECT system

In this study we investigate the feasibility of performing small animal SPECT imaging on a clinical SPECT scanner with a dedicate pinhole collimator. Analytical formulae are used to calculate spatial resolution and absolute sensitivity given a pinhole collimator design scheme. Two optimal pinhole designs are proposed in terms of achieving the best trade-off between spatial resolution and absolute sensitivity. One consists of one single pinhole aperture and the other one has 7 pinholes. In the 7-pinhole design scheme, the pinholes are arranged in the way that each pinhole only covers part of the region of the object to maximize the sensitivity in the central slice of field of view (FOV) and to make use of the entire detection area without much overlapping. Monte Carlo simulation studies confirm the predicted spatial resolution values from analytical formulae. Imaging reconstruction results of a simulated ultra-micro hotrod phantom demonstrate that the 0.5 mm hot rods are clearly identified with the 7-pinhole collimator. Despite that the intrinsic resolution of the SPECT detector is only 3.55 mm, its large detection area allows large magnification and good spatial resolution. We conclude that high-resolution animal SPECT imaging can be performed through the proposed approach with relatively low cost for clinical SPECT users comparing to purchasing a dedicated small animal SPECT.

SVD reconstruction algorithm in 3D SPECT imaging

Singular value decomposition (SVD) method was used for image reconstruction in single photon emission computed tomography (SPECT). The 3D system transition matrix and the projection data were produced by Monte-Carlo simulation based on NCAT human torso phantom. Generalized matrix inverse of system transition matrix was computed. NMSE and Contrast parameters were chosen to evaluate the image quality. The relationship between reserve singular value number and reconstructed image quality is discussed. Reconstructed image in best quality was obtained when the optimized number of preserved singular value was chosen, and compared with routine OSEM reconstruction methods. Results show that SVD reconstruction algorithm, which can reduce noise influence effectively and improve the reconstruction result greatly, is a valuable image reconstruction algorithm. It can be improved to solve the coded mask SPECT imaging problem.

Front-end Electronics Design based on Vernier Method for a High Resolution MicroPET

The design of a high resolution small animal positron emission tomography (MiroPET) camera was presented. This MicroPET has 10 photomultiplier tube (PMT)-Sharing-modular detectors. Each detector has 21 PMTs decoded with 180 BGO(Bi4Ge3Ol2 ) crystal elements. A new and low cost design of Time-to- Digital Converter (TDC) based on Vernier method was developed. We use only one FPGA EPF10K30A and achieve 1.3ns timing resolution performance. Such TDC design gives us chance to realize TDC and ADC all together on one Printed Circuit Board (PCB), which would also significantly decrease complication of coincidence processing compared with our previous design. Our preliminary results showed: Vernier TDC had a less than 1.3ns timing resolution that met the demand for coincidence measurement for MicroPET. No inacceptable degradation was observed in time resolution when the temperature changes from 20degC to 40degC We obtain the 16.6% energy resolution, using this new easily compact modular electronic architecture.

A high resolution and high sensitivity small animal SPECT system based on H8500

A compact multi-detector SPECT system for small animal (MicroSPECT) has been developed. The detector of MicroSPECT system is composed of a NaI scintillation crystal array and four H8500 flat panel position sensitive photon multiplier tubes(PS-PMT). The pixel size of the scintillation crystal array is 1.45 mmtimes1.45 mm and the total detection area is 103.75 mmtimes103.75 mm with 61times61 pixels. Four H8500 PS-PMTs are jointed together, coupling with the scintillation crystal array. Single pinhole and multi-pinhole collimators are used and special designed to satisfy the need of high spacial resolution or high sensitivity respectively. The DETECT2000 & MCNP Monte Carlo simulation programs were used for verifying the MicroSPECT system detector and pinhole collimator design. A high speed readout circuit was developed for the detectors. Each detector can achieve the counting rate at 300 k/s. The results show that the scintillation crystal array has an effective area of 55times55 pixels and that the joint parts of the PS-PMTs are still available for detection. We aim that the MicroSPECT system can achieve 1 mm spatial resolution or better.

Efficient analytical scatter modeling in fully 3-D iterative single photon emission computed tomography reconstruction

Quantitative single photon emission computed tomography (SPECT) images are degraded by several physical factors, among which Compton scatter is the most difficult to compensate. An analytical scatter modeling (ASM) method is proposed to model scatter in SPECT. The reconstruction architecture includes following steps: 1) a look-up table describing patient independent factors was precalculated, and a numerical integration method based on number theory resulted in 1 to 2 order of magnitude speed up in precalculation. 2) The transition matrix was generated based on the patient-specific attenuation map. 3) OSEM reconstruction was performed with an unmatched projector/backprojector. ASM was validated by Monte-Carlo simulation, considering the case of the point source in a homogeneous and a non-homogeneous medium. Experiment was performed using a HAMAMATSU BHP6601 SPECT and a SPECT performance phantom. Projection data was reconstructed by OSEM method with/without ASM, filtered back projection(FBP) and FBP with dual energy window (DEW) scatter correction. Results show that OSEM/ASM is more accurate and gain higher contrast than others. For a 64 /spl times/64 /spl times/ 64 image array. The computation time of the transition matrix is 80 min, and the reconstruction takes 4 min per iteration on a 1.54 GHz processor PC. This work proposed a computationally efficient method to model scatter in SPECT reconstruction within clinically acceptable time.