Jinghan Ye

Iterative SPECT Reconstruction Using Matched Filtering for Improved Image Quality

SPECT images reconstructed from low count studies suffer either from high noise or poor resolution. We have developed an iterative reconstruction with matched filtering (IRMF) to control image noise while maintaining higher image resolution. IRMF involves filtering the measured projection and re-projection during iterative reconstruction with the same low-pass filter before the two are compared to generate an error projection. Another low-pass filter can be applied to the error projection before it is backprojected to update the current activity distribution estimate. The method is validated with a cardiac phantom filled with a clinical distribution of Tc-99m. A 1-second-per-frame scan was acquired to mimic a single gated segment. The image was reconstructed using ordered-subset expectation-maximization (OSEM) algorithm with depth-dependent resolution recovery. Reconstructions of similar spatial resolution with post-reconstruction Butterworth filtering (OSEM+F) and with matched filtering are compared visually and via standard deviation (SD) and signal-to-noise ratio (SNR) measurements. Results: Images reconstructed with IRMF show strong noise suppression in both the myocardium and background areas as compared to those reconstructed with OSEM+F. The SD in the background is reduced by ~30%, and the SNR is improved by ~100%. IRMF significantly improves image quality by suppressing noise in low count SPECT studies while maintaining higher image resolution.

Investigation of optimization-based reconstruction with an image-total-variation constraint in PET.

In this work, we investigate PET-image reconstruction by using optimization-based algorithms. The work is motivated by the observation that advanced algorithms may be exploited potentially for improving PET-reconstruction quality in current applications and for enabling innovative, advanced PET-scan configurations. may be used for enabling the design of innovative PET systems. Specifically, we investigate image reconstruction from data, collected with PET-scan configuration with sparsely-populated detectors, by formulating it as an image-total-variation (TV)-constrained, Kullback-Leibler (KL)-data-divergence minimization, and then by solving the minimization with a primal-dual optimization algorithm developed by Chambolle and Pock. We carry out IEC-phantom data studies to demonstrate the potential of the reconstruction design and algorithm in enabling PET imaging configurations with reduced number of detectors.

CBCT-subsystem performance of the multi-modality Brightview XCT system (M09-26)

The new Brightview XCT system uses a flat-panel detector to perform CBCT imaging for attenuation correction and localization. Features include a small footprint due to an offset-detector geometry, advanced scatter correction — both software and hardware — isotropic voxels, and GPU-accelerated reconstruction. System performance characteristics of the CBCT system such as spatial resolution, HU linearity, uniformity, noise, low-contrast detectability, and dose measurements are discussed in this paper.

CT-based attenuation correction in PET image reconstruction for the Gemini system

The Gemini system is a combined CT/PET imaging system newly developed by Philips Medical Systems. It has a unique open gantry design that allows for variable separation between the CT and PET gantries. The Gemini system incorporates CT-based attenuation correction (CT-AC) into a three-dimensional row-action maximum likelihood algorithm (RAMLA) for PET image reconstruction. It uses several unique techniques to achieve high accuracy while reducing patient X-ray dose. These new techniques include (1) using low-dose CT protocols to obtain CT images with adequate quality and quantitation for CT-AC while keeping patient X-ray dose low; (2) using a CT truncation compensation technique to improve the accuracy of CT-AC; and (3) using a generalized model for the conversion of CT images to attenuation maps at 511 keV. In this paper, we report the workflow and performance of Gemini CT-AC using phantom and patient studies. For comparison, attenuation maps obtained from PET transmission scans are also used for attenuation correction (TX-AC). Both phantom and patient studies show that PET images with CT-AC have image quality equivalent to or better than those with TX- AC.

Scatter correction with combined single-scatter simulation and Monte Carlo simulation for 3D PET

In positron emission tomography (PET) imaging, scattered gamma photons typically account for more than 30% of total detected coincidence counts. Single scatter simulation (SSS) method is widely used for estimating scatter contribution in PET image reconstruction. Monte Carlo (MC) techniques are more accurate but computationally expensive. When using SSS, the modeled scatter contribution is typically scaled to match the measured data. Typically tail fitting is used for this scaling. However, when the available tail part is either too small or too noisy, the tail fitting technique may lead to artifacts in the reconstructed images. In this study, a hybrid method that combines SSS and MC is presented. The method uses SSS to approximate the shape of scatter contribution, and scales the SSS result by a scaling factor derived from a low-count MC simulation. Effectiveness of the method is evaluated with phantom and patient studies. Images reconstructed using SSS with tail-fitting scaling (SSS-TFS) and SSS with Monte Carlo scaling (SSS-MCS) are compared. Results show that SSS-MCS significantly reduces or eliminates the artifacts that present in SSS-TFS images. For smaller objects, both SSS-MCS and SSS-TFS produce artifact-free images. The MC simulation takes less than a second on a computer with 8 CPU cores to achieve less than 1% of scatter scaling factor variation. A hybrid scatter correction method that combines SSS and Monte Carlo simulation is developed for PET reconstruction. The method demonstrates a significant improvement in scatter correction accuracy. The added computational cost is negligible.

Automated cardiac pose computation from reconstructed myocardial SPECT images

A method to automatically compute cardiac pose from 3D reconstructed cardiac SPECT images is presented. Cardiac region of interest (ROI) is generated from input volume as the first step. Left ventricle is then segmented from the ROI and myocardial pruning algorithm is applied to remove unnecessary myocardial mass at the top and the bottom. Pruned myocardial binary mask is subjected to a myocardial clusterification-based skeletonization process. Skeleton of the myocardium is cleaned using a novel distance transform and binning-based histogram cleaning algorithm. Blood pool enclosed within the myocardial walls is iteratively segmented to compute the center of mass (COM) of the left ventricle. An ellipsoid fitting algorithm along with the COM is then employed on the myocardial skeleton to determine the cardiac angles. The method achieves a success rate of 94.4% when evaluated on 643 reconstructed volumes from 331 patients.

The effects of center of rotation errors on cardiac SPECT imaging

In SPECT imaging, center of rotation (COR) errors lead to the misalignment of projection data and can potentially degrade the quality of the reconstructed images. In this work, we study the effects of COR errors on cardiac SPECT imaging using simulation, point source, cardiac phantom, and patient studies. For simulation studies, we generate projection data using a uniform MCAT phantom first without modeling any physical effects (NPH), then with the modeling of detector response effect (DR) alone. We then corrupt the projection data with simulated sinusoid and step COR errors. For other studies, we introduce sinusoid COR errors to projection data acquired on SPECT systems. An OSEM algorithm is used for image reconstruction without detector response correction, but with nonuniform attenuation correction when needed. The simulation studies show that, when COR errors increase from 0 to 0.96 cm: 1) sinusoid COR errors in axial direction lead to intensity decrease in the inferoapical region; 2) step COR errors in axial direction lead to intensity decrease in the distal anterior region. The intensity decrease is more severe in images reconstructed from projection data with NPH than with DR; and 3) the effects of COR errors in transaxial direction seem to be insignificant. In other studies, COR errors slightly degrade point source resolution; COR errors of 0.64 cm or above introduce visible but insignificant nonuniformity in the images of uniform cardiac phantom; COR errors up to 0.96 cm in transaxial direction affect the lesion-to-background contrast (LBC) insignificantly in the images of cardiac phantom with defects, and COR errors up to 0.64 cm in axial direction only slightly decrease the LBC. For the patient studies with COR errors up to 0.96 cm, images have the same diagnostic/prognostic values as those without COR errors. This work suggests that COR errors of up to 0.64 cm are not likely to change the clinical applications of cardiac SPECT imaging when using iterative reconstruction algorithm without detector response correction.

Study of the effect of statistical fluctuations on defect detectability at clinical count levels in cardiac SPECT

Cardiac SPECT using Tl-201 suffers from low count statistics. Any statistical studies concerning the evaluation of a reconstruction algorithm, acquisition parameters, diagnostic confidence, etc., for clinical applications are impacted by the difficulty of obtaining data with multiple noise realizations. For this work, we acquired list-mode data of a Tl-201 cardiac phantom with very high counts in three configurations-with an anterior defect, an inferior defect, and no defect. The list-mode data were repartitioned to obtain statistically independent multiple data sets all with the same, clinically relevant noise level. Images were reconstructed from each of the resulting data sets using an iterative algorithm with attenuation correction. Reconstructed images were examined by four human observers, as well as analyzed quantitatively. The ability of observers to differentiate between normal scans and scans with defects varied substantially among datasets. There was correlation between the measured defect detectability and the visual assessment. The fact that the visibility of defects and the uniformity of normal scans varied significantly from one data set to the next, even when both were acquired at the same time, under identical conditions, indicates that the low statistics levels at clinical doses can have a measurable effect on diagnostic confidence.

Sparse-view image reconstruction from gated cardiac data

In recent years, combined SPECT/CBCT systems have been developed and become commercially available for yielding spatially registered SPECT and CT images. The CBCT unit of the Philips SPECT/CBCT system (BrightView XCT) adopts a flat-panel detector for detecting X-rays. Because of heart motion and relatively slow data acquisition speed of the CBCT unit, the reconstruction of a beating heart is challenging for CBCT. Gating is often used in cardiac imaging for acquiring CBCT projection data of the heart. In order to acquire projection data of a certain phase over 360 degree, multiple turns of projection data of the heart have to be acquired, after which the data are sorted to form subsets for different phases. For cardiac data within one source rotation, projection data for each phase are sparsely and non-uniformly distributed over the scanning angular range. In this work, we investigate image reconstruction for different heart phases from both simulation and patient cardiac data sets within one source rotation. The reconstruction task is formulated as a constrained minimization of image total variation (TV) problem, and an ASD-POCS algorithm is used for solving the constrained TV-minimization problem. The approach proposed to image reconstruction can potentially reduce scanning time and imaging dose to the patient. More important for cardiac imaging, this approach may reduce artifacts caused by heart motion. For the cases studied, our results suggest that images of potential practical utility can be reconstructed from both simulation and patient data within a single phase.

Transmission attenuation map with measured downscatter correction

As the attenuation correction becomes clinical reality in SPECT, accurate estimation of emission downscatter to transmission during simultaneous emission-transmission acquisition becomes critical. In this study, we proposed a technique that directly measures the downscatter contamination from 140-keV emission photons to the 100-keV transmission window. The technique combines energy window discrimination and spatial window discrimination to separate emission (EM), transmission (TR) and downscattered (DS) photons. A spatial mask (S) aligned with the opposing scanning line source and a gap (G) at each end of S are first defined. Photons detected inside the S belong to TR if their energy fall inside the 100 keV window; those detected outside the S and G belong to either EM if their energy inside the 140 keV window, or DS if they are inside the 100 keV window. The TR is corrected from the downscatter contamination via subtraction a fraction of the DS. The effectiveness of the technique is evaluated based on the uniformity of the ROI in the reconstructed attenuation map from a series of phantom experiments. The technique is compared with another technique using an adjacent energy window measurement. For typical cardiac studies, the uniformity of the transmission map from both techniques are comparable in the heart area, while for the cardiac study with high concentration present in the gall bladder, the uniformity is significantly improved from the proposed technique.