Steven Kohlmyer

Implementation of ML based positioning algorithms for scintillation cameras

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, the authors investigate a 1D based recursive ML positioning algorithm. The technique overcomes the dimensional separable violation for hexagonally packed photomultiplier tube (PMT) arrays. Second, they examine a 2D based ML method that uses local PMT clusters to reduce the computational demands of full 2D ML implementation. Third, they 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.

Coincidence imaging using a standard dual head gamma camera

Coincidence electronics and a data acquisition system were developed to explore coincidence detection using a conventional dual head gamma camera. A high impedance pick-off circuit provides position and energy signals without interfering with normal camera operation. The signals are pulse-clipped to reduce pileup effects. Thin lead-tin-copper filters are used to reduce the flux of low energy photons to the detectors. The data are stored in list mode format. The measured coincidence timing resolution for the system is 9 nsec FWHM (450 kcps/detector) and the energy resolution is 11% (650 kcps/detector). The system sensitivity is 46 kcps//spl mu/Ci/cc for a 20 cm diameter (18 cm length) cylindrical phantom centered in the field of view. A scatter fraction of 31% was measured using the 20 cm cylindrical phantom. The sensitivity and scatter fraction measurements were made using a 450-575 keV energy window, 63.0 cm detector spacing, and 1 mm thick lead filters. The maximum recommended singles rate (full spectrum) for coincidence imaging is /spl sim/800 kcps per detector. The 3D reprojection algorithm has been implemented. Example images of the 3D Hoffman brain phantom and patient tumor images are shown.

Performance of a dual headed SPECT system modified for coincidence detection

A NIM/CAMAC/Macintosh data acquisition system was developed to collect/sup /spl plusmn//X, /sup /spl plusmn//Y, and E information from a GE Maxxus SPECT system. A high impedance signal pick-off circuit was used so the data could be collected without disturbing normal operation of the camera. The measured coincidence timing resolution for the system is 8 msec FWHM. The energy resolutions for the detector heads are /spl sim/7.5% (using pulse clipping), at 511 keV. The system sensitivity is 70 cps//spl mu/Ci for a point source centered in the field of view (450-575 keV energy window, 62.6 cm detector spacing). The intrinsic spatial resolution of the detector heads are /spl sim/3.7 mm FWHM in the focal plane using linear tomography. Linear tomography image reconstruction and a volume image reconstruction based on the Kinahan-Rogers approach have been implemented. Images of a line source phantom, the Data Spectrum torso phantom, and the 3D Hoffman brain phantom.

Scatter and attenuation correction for 111In based on energy spectrum fitting

A combined scatter and attenuation correction that does not require a transmission scan is proposed for 111In imaging. Estimates of the unscattered intensity at both 171 and 245 keV are obtained by fitting the observed energy spectrum at each pixel or region of interest using the measured scatter-free spectrum and a simple model for scatter. The scatter model for the 171 keV peak accounts for scatter contributed by both the 171 and 245 keV emissions. After correcting for scatter, the attenuation is estimated from the observed ratio of photopeak intensities using the known difference in attenuation at the two emission energies and a model based on a point source in water. Accurate scatter correction is a prerequisite for the success of this method because scatter from the higher energy emission will otherwise contaminate the lower photopeak. This differential attenuation method (DAM) of estimating attenuation is demonstrated and calibrated using a series of point source measurements with a wedge-shaped attenuator. The observed absolute and differential attenuation are in good agreement with the narrow-beam linear attenuation coefficients for water. Estimates of precision suggest a depth resolution of 1.0-2.5 cm for realistic count densities over the clinically relevant depth range (0-25 cm). The accuracy of DAM in a more realistic attenuation environment is assessed using a hot sphere inside the anthropomorphic data spectrum torso phantom viewed from several angles (with differing attenuation). Finally, the potential of DAM for SPECT attenuation correction was investigated by computer simulation using the SIMSET Monte Carlo software. Preliminary results based on measured planar data and simulated SPECT data indicate that DAM can improve the quality and quantitative accuracy of 111In images. In one SPECT simulation study, the average error in tumor to soft-tissue ratios was reduced from 32% for uncorrected data to 8% for data corrected with DAM. However, the technique is susceptible to significant noise amplification and can cause substantial streak artifacts in low-count SPECT studies if sufficient smoothing of the depth estimates is not performed.

Effect of lower energy threshold on single and multiple scatter distributions in positron volume imaging

Simulations of the scatter distributions in the Advance tomograph were performed using the SimSET software package. The Zubal phantom was used, centered at the heart with activity only in the heart or in all soft tissues except the lungs. Data were binned axially for: (1) -0.884 to 0.844 cm; (2) 2.53 to 4.22 cm; and (3) -7.6 to 7.6 cm (the full axial field of view). Simulations were performed with both a 45 cm long (Full object) and a 15.5 cm long (Short object) section of the phantom with 250 million decays. The total scatter to trues ratio for the full object vary from 1.20 (300 keV, all soft tissues) to 0.35 (425 keV, heart only). The ratio of single scatters from the short object simulation compared to the total scatters from the full object varied from 40% at 300 keV to 60% at 425 keV. While the multiple scatters were well estimated with a Gaussian convolution of the single scatter events (for the full object), significant errors will occur if the attenuation and activity outside the FOV is not included in the estimation of the single scatters.

The effect of scatter on quantitation in positron volume imaging of the thorax

In positron volume imaging (PVI) a high percentage of detected coincidences are scatter events. The authors performed a series of phantom and simulation studies of scatter contamination in thorax imaging. Scatter fractions of 50 to 70% were measured for a torso phantom in a General Electric Advance positron emission tomograph with the septa retracted. The authors also measured scatter contribution from outside the field-of-view (FOV) using a small cylindrical source (1 cm long, 0.5 cm diameter) at various axial and transaxial positions in a solid Lucite cylinder (43 cm long, 20 cm diameter). A source 9 cm outside the FOV contributed approximately 30% the scatter that the same source would contribute if placed near the center of the FOV. To isolate the effect of scatter from outside the FOV in realistic imaging situations the authors performed Monte Carlo simulations of PVI for heart studies. Organs outside the FOV can contribute significantly to detected scatter. For instance, with equal activity concentrations the liver and heart contributions to scatter were virtually identical even though 90% of the liver was outside the FOV.<<ETX>>

Optimization of input parameters for a segmented attenuation algorithm for the GE advance/sup TM/ scanner

Segmented attenuation correction (SAG) techniques provide the possibility for increasing patient throughput by decreasing transmission scan times from those required for standard measured attenuation correction (MAC) methods. A SAC algorithm using an unsupervised least biased fuzzy clustering method has been implemented on the GE Advance/sup TM/ system. This implementation has the advantage of requiring no a priori region definitions. There are, however, several parameters which affect both the bias and variance produced by the SAC algorithm. The authors used statistical design of experiment (DOE) techniques to determine the subset of critical input parameters that have the largest impact on both bias and variance. These critical parameters were then analyzed to determine their influence on bias and variance. A second DOE was then carried to optimize the remaining parameters. Settings are identified which can be used to reduce both bias and variance in the segmentation algorithm.