Search

The goal of this study was to evaluate the magnitude and origin of scattered radiation in a small-animal PET scanner and to assess the impact of these findings on noise equivalent count rate (NECR) measurements, a metric often used to optimize scanner acquisition parameters and to compare one scanner with another. The scatter fraction (SF) was measured for line sources in air and line sources placed within a mouse-sized phantom (25 mm [phi] x 70 mm) and a rat-sized phantom (60 mm [phi] x 150 mm) on the microPET II small-animal PET scanner. Measurements were performed for lower energy thresholds ranging from 150-450 keV and a fixed upper energy threshold of 750 keV. Four different methods were compared for estimating the SF. Significant scatter fractions were measured with just the line source in the field of view, with the spatial distribution of these events consistent with scatter from the gantry and room environment. For mouse imaging, this component dominates over object scatter, and the measured SF is strongly method dependent. The environmental SF rapidly increases as the lower energy threshold decreases and can be more than 30% for an open energy window of 150-750 keV. The object SF originating from the mouse phantom is about 3-4 % and does not change significantly as the lower energy threshold increases. The object SF for the rat phantom ranges from 10 to 35% for different energy windows and increases as the lower energy threshold decreases. Because the measured SF is highly dependent on the method, and there is as yet no agreed upon standard for animal PET, care must be exercised when comparing NECR for small objects between different scanners. Differences may be methodological rather than reflecting any relevant difference in the performance of the scanner. Furthermore, these results have implications for scatter correction methods when the majority of the detected scatter does not arise from the object itself. Index Terms--MicroPET, positron emission tomography (PET), scatter fraction, small-animal imaging.

Positron emission tomography (PET) with the glucose analog [18F] fluorodeoxyglucose is proving to be useful in cancer diagnosis and treatment. However, as in all nuclear medicine imaging technologies, lesion detection with PET is often hindered by limited spatial resolution and low signal-to-noise ratios. Under such conditions, conventional diagnosis by visual inspection usually becomes difficult and potentially inaccurate. In this paper, we propose use of computer-aided lesion detection methods for PET imaging by applying a maximum likelihood ratio test and a composite hypothesis test, assuming that the mean positron emission rate is deterministic or random, respectively. In our approach, different statistical models characterizing the mean positron emission rate, the raw sinogram data and the filtered backprojection (FBP) reconstructed image are used to derive the test criteria. Three methods to estimate the unknown parameters of the test functions from observations are presented. The performance of one of the proposed methods is evaluated and compared with both simulated and experimental phantom data. In the preliminary trials, the methods detect correctly (with a high probability > 0.9) lesions of diameter [greater than or equal to] 15mm with lesion-to-background contrast 1.1:1. Under the same conditions, the test lesion could not be detected by visual inspection alone in the images reconstructed by either the FBP or the maximum likelihood iterative algorithms. The methods may also be used for the objective assessment of the quality of images reconstructed from different algorithms.

A technique has been devised for independently determining the energy and spatial resolution components of the point-spread functions of electronically collimated (Compton scatter) imaging devices using a Monte Carlo method. Monte Carlo runs are performed for the device being examined, and detailed records of exact energy loss and spatial coordinates of each interaction for each particle in the simulation are made. Taking resolution parameters from the known performance of the detectors, measured data is simulated by sampling from Gaussian distributions (for both energy and position) about the exact interaction energy and position in both detectors, for each Monte Carlo history. By back-projecting cones on an event-by-event basis and calculating the closest approach of the cone to the known source point using the simulated measurement data for one of the energy and position variables and the exact interaction data for the rest of the variables, the effect of the variable in question on the point spread can be isolated. In this paper we present the calculations of the contribution to the point-spread function of the energy and spatial resolution in each detector for various source positions for a representative Compton camera configuration. Index Terms - Compton scatter camera, Monte Carlo, resolution.

Digital x-ray imaging detectors designed for soft x-rays (1 to 50 keV) are significant for medical mammography, dental radiography, microradiography, and microtomography. Detector designs involve either direct absorption of x-rays in solid state devices or thin scintillator screens optically coupled to solid state sensors. Well designed scintillator systems produce 10 or more electrons per detected x-ray and, used with charge coupled devices (CCD), detect 100,000 x-rays per pixel before saturation. However, if the scintillator is directly coupled to the detector, radiation can penetrate to the semiconductor detector with a small number of events producing large charge and noise. We have investigated the degradation of image noise by these direct absorption events using numerical models for a laboratory detector system consisting of a 60 [[micro]meter] CsI scintillator optically coupled to a scientific CCD. Monte Carlo methods were used to estimate the charge deposition signal and noise for both the CsI and the semiconductor. Without a fiber optic coupler, direct absorptions dominate the signal and increase the signal variance by a factor of about 30 at energies above 10 keV. With a 3 mm fiber optic coupler, no significant degradation is observed for input energies below 45 keV.

Using Monte Carlo simulation, we investigate algorithms to identify and correct for detector Compton scatter in hypothetical PET modules with 3x3x30 mm BGO crystals coupled to individual photosensors. Rather than assume a particular design, we study three classes of detectors: (1) with energy resolution limited by counting statistics, (2) with energy resolution limited by electronic noise, and (3) with depth of interaction (DOI) measurement capability. For the first two classes, selecting the channel with the highest signal as the crystal of interaction yields a 22-25% misidentification fraction (MIF) for all reasonable noise fwhm to signal (N/S) ratios (i.e. < 0.5 at 511 keV). Algorithms that attempt to correctly position events that undergo forward Compton scatter using only energy information can reduce the MIF to 12%, and can be easily realized with counting statistics limited detectors but can only be achieved with very low noise values for noise limited detectors. When using position of interaction to identify forward scatter, a MIF of 12% can be obtained if the detector has good energy and position resolution.

Performance measurements of the General Electric ADVANCE Positron Emission Tomograph operating with the septa retracted (3D mode) were made. All reconstructions were performed with the GE ADVANCE 3D package. Performance tests were carried out with: the NEMA phantoms; a 3D Hoffman phantom; a Data Spectrum torso phantom with lung and cardiac inserts; and the 'Utah' 3D evaluation phantom. Data collected included: transaxial and axial resolution, uniformity, recovery coefficients, count rate performance, dead time accuracy, and effect of scatter correction.

The contribution of the scattered events originating from Radioactivity Outside the tomograph Field of View (ROFOV) is treated differently by different scatter correction algorithms. Using measured data we have evaluated the amount of scatter originating from ROFOV, its axial and transaxial distribution in two energy windows for direct and a subset of oblique planes in 3D PET. Nonuniform distributions were observed with the total contribution found to be non negligible when the amount of ROFOV was comparable to the radioactivity in the Field of View (FOV).

Since many different configurations of brain SPECT systems are available for clinical imaging, it is informative to investigate whether there are significant differences in their performance potentials. However, the difference in configurations and performance characteristics make direct comparison between systems difficult. To clarify the issue, we have developed a method to analyze the relative merits of different basic system geometries and configurations. This method is based on the calculated system geometric efficiency (SGE) which is obtained when the system spatial resolution is normalized to a reference level under a specified imaging condition. During the normalization process, the design parameters and performance of the collimators are iteratively adjusted to match the system resolution of an appropriately-defined reference specification. After the collimator is normalized, the geometric efficiency (GE) of the collimator is calculated by summing the point source geometric efficiency (PSGE) over a brain source model while imposing narrow-beam attenuation effect. The SGE is then derived for each system configuration based on the calculated GE for each collimator. This analysis has been applied to many existing geometries and detector-collimator configurations of current brain SPECT systems.

We have shown in previous works that distinct non-stationary analytical scatter kernels can be extracted from line source measurements and used to independently subtract object scatter and subtract or restore detector scatter in high resolution PET. In this work, the dependence of the scatter components on energy threshold and source position was investigated. Line source measurements were acquired in multispectral mode using the Sherbrooke PET simulator. Scatter parameters were extracted from data summed in energy windows with a lower threshold varying from 129 keV to 516 keV in steps of 42 keV, and a fixed upper threshold of 644 keV. Decreasing the lower threshold from 344 keV to 129 keV increases the trues by only 25%, but increases object scatter by 136% and almost triples detector scatter. A gain in efficiency by a factor of 2 or more would result from recovering the latter by restoration in the broad window. The intensity and shape of the scatter functions for both object and detector are shown to have a significant dependence on energy and position. This dependence needs to be taken into account in the design of kernels for accurate scatter correction over a broad energy range.

We investigated the shift of local energy spectra with SPECT rotation for the GE 400 AT and the Picker Prism 3000 tomographs. A Co-57 flood source was taped to the parallel-beam collimator of the GE 400 AT; a Tc-99m line source was placed at the focus of the fan-beam collimator of one head of the Picker Prism. The count-based method, which employs a narrow window (about 4 keV) on the maximum slope of the photopeak, was used with both systems. Non-linear, polynomial spectral fitting was applied to x-y-E data acquisitions with the GE camera. The fitting yielded either shifts or shifts and width changes. Results show 1) the shifts are pseudo-sinusoidal with angle and similar for different spatial locations, 2) the average of their absolute value is 0.71 keV and 0.13 keV for the GE and Picker cameras, respectively, 3) width changes for the GE camera are small and appear random, 4) the calculated shifts from the count-based method for the central part of the GE camera are correlated with those from the spectral fitting method. They are 12% smaller. The conclusion is that energy shifts with angle may be present with many rotating cameras although they may be smaller with newer cameras. It might be necessary to account for them in schemes designed for high-accuracy compensation of Compton-scattered gamma rays although they possibly could be ignored for newer cameras.

A dedicated 3D brain PET scanner has several advantages, most notably increased sensitivity, over a whole body scanner for neurological studies. However brain scanners have higher scatter fractions, random count-rates and deadtime for the same activity concentration. We have used noise effective count-rate (NECR) analysis to compare brain scanners of 53, 60, and 66 cm diameter with the GE ADVANCE whole body scanner (93 cm diameter). Monte Carlo simulations of a brain-sized phantom (16 cm diameter, 13 length) in the ADVANCE geometry were used to develop a model for NECR performance, which was reconciled to results from a decay series measurement. The model was then used to predict the performance of the brain scanner designs. The brain scanners have noise effective sensitivities (the slope of the NECR curve at zero activity) as much as 40% higher than the body scanner. However, their NECR, advantage diminishes quickly as t;he activity concentration increases. The brain scanners' NECR equals the body scanner with about 0.7-0.8 mCi in the phantom; the body scanner has superior NECR performance at higher activity levels. An imaging center concentrating on only very low activity imaging tasks would find the efficiency advantage of a smaller detector diameter valuable, while a center performing higher activity studies such as bolus water injections or 5mCi FDG injections might prefer the count rate performance of a whole body scanner.

A series of 2D and 3D count rate measurements have been carried out on the Advance PET system. The measurements include: NEMA standard protocols; simultaneous imaging of two small vials containing [O-15]water and [F-18]; a cardiac insert placed in a water filled elliptical tank; two different head sized cylindrical tanks; and a Hoffman 3D phantom. The data were processed with and without dead time and scatter corrections. Plots were made of the total count rate, randoms, trues + scatter, scatter corrected trues, and noise equivalent count rate as a function of total activity in the machine and as the [Mu]Ci/cc in the phantoms. The accuracy of deadtime corrections was also measured. In 2D mode, the NEMA standard protocol resulted in maximum 'trues' count rates of 540,000 cps in high sensitivity mode. In 3D mode the head sized cylindrical phantoms yielded a maximum trues rate of approximately 950,000 cps. Recovery of the water half-life was accurate with up to 50 mCi for the cardiac insert in 2D mode.

Precision of medical image registration is very important in clinical diagnosis and treatment, which is usually assessed by visual inspection or by referring to other methods that require special expertise and extensive experience. In this study, we proposed a novel automatic approach based on statistical theory to estimate confidence intervals of the registration parameters and allow the precision of registration results to be objectively assessed. Under the assumption of local linearity, statistical confidence intervals of model fitting (regression) can be used to evaluate registration precision. Monte Carlo simulations using the Hoffman brain phantom with various amounts of displacement, noise and spatial filtering were conducted to evaluate the formula for estimating the confidence intervals in 2D image registrations. Monte Carlo simulation results are consistent with the calculated confidence intervals, and the agreement is applicable to different amounts of translation, angular rotation and spatial smoothing. The estimated parameter values fall within the predicted 90%, 95% and 99% confidence intervals with less than [+ or -] 1% of errors. The present results indicate that the use of statistical confidence intervals can provide an objective assessment of individual image registration results.

We compared 12 pairs of cerebral [sup.18F]-fluoro-deoxyglucose (FDG) 2D/3D image sets from a GE/Advance PET scanner, incorporating the actual corrections used on human subjects. Differences in resolution consistent with other published values [1-5] were found. There is a significant difference in axial resolution between 2D and 3D, and we focused on this as it is a scanner feature that cannot be readily changed. Previously published values for spatial axial resolution in 2D [1] and 3D [2] modes were used to model the differential axial smoothing at each image voxel. This model was applied to the 2D FDG images, and the resulting smoothed data indicate the published differences in axial resolution between 2D and 3D modes can account for 30-40% of the differences between these image sets. We then investigated the effect this difference might have on analysis typically performed on human FDG data. A phantom containing spherical hot- and cool-spots in a warm background to mimic a typical human cerebral FDG PET scan was scanned for a variety of time durations (30, 15, 5, 1 min). Only for the 1-minute frame (total counts 2D:6M, 3D:30M) is there an advantage to using 3D mode; for the longer frames which are more typical of a human FDG protocol, the reliability for extracting regions-of-interest is the same for either mode while 2D mode shows better quantitative accuracy.

Positioning algorithms for scintillation cameras are investigated. Maximum Likelihood (ML) estimation techniques and centroid-based methods (e.g., weighted and local centroid) are evaluated. Three implementation methods to reach the two-dimensional ML solution are proposed. First, we investigate a 1D based recursive ML positioning algorithm. The technique overcomes the dimensional separable violation for hexagonally packed photomultiplier tube (PMT) arrays. Second, we examine a 2D based ML method that uses local PMT clusters to reduce the computational demands of full 2D ML implementation. Third, we develop a correlation method that maps event characterization vectors to the associated position. The methods are tested with data sets generated using Monte Carlo simulation and verified with experimental study. Simulation (experimental) results show that the full 2D ML algorithm is superior to the others. It has a 7(10)%, 67(47)% and 69(30)% improvement over the weighted centroid with 3% bias subtraction method in terms of spatial resolution, linearity and MSE, respectively.