Improved scatter correction using adaptive scatter kernel superposition.

Accurate scatter correction is required to produce high-quality reconstructions of x-ray cone-beam computed tomography (CBCT) scans. This paper describes new scatter kernel superposition (SKS) algorithms for deconvolving scatter from projection data. The algorithms are designed to improve upon the conventional approach whose accuracy is limited by the use of symmetric kernels that characterize the scatter properties of uniform slabs. To model scatter transport in more realistic objects, nonstationary kernels, whose shapes adapt to local thickness variations in the projection data, are proposed. Two methods are introduced: (1) adaptive scatter kernel superposition (ASKS) requiring spatial domain convolutions and (2) fast adaptive scatter kernel superposition (fASKS) where, through a linearity approximation, convolution is efficiently performed in Fourier space. The conventional SKS algorithm, ASKS, and fASKS, were tested with Monte Carlo simulations and with phantom data acquired on a table-top CBCT system matching the Varian On-Board Imager (OBI). All three models accounted for scatter point-spread broadening due to object thickening, object edge effects, detector scatter properties and an anti-scatter grid. Hounsfield unit (HU) errors in reconstructions of a large pelvis phantom with a measured maximum scatter-to-primary ratio over 200% were reduced from -90 ± 58 HU (mean ± standard deviation) with no scatter correction to 53 ± 82 HU with SKS, to 19 ± 25 HU with fASKS and to 13 ± 21 HU with ASKS. HU accuracies and measured contrast were similarly improved in reconstructions of a body-sized elliptical Catphan phantom. The results show that the adaptive SKS methods offer significant advantages over the conventional scatter deconvolution technique.

Zero-quantum filter offering single-shot lipid suppression and simultaneous detection of lactate, choline, and creatine resonances.

A zero-quantum (ZQ) filter offering single-shot lipid suppression and providing for simultaneous detection of the lactate methyl doublet (1.3 ppm) and nonoverlapping singlets including choline (Cho, 3.2 ppm) and creatine (Cr, 3.0 ppm) is described. Filtering is provided by soft mixing and reading pulses (RF(mix), RF(rd)) that are selective for the lactate methine quartet (4.1 ppm), Cho, and Cr resonances but exclude the 1.3 ppm lactate component and overlapping lipids. Surrounding RF(mix) and RF(rd) are magnetic field gradient pulses of equal magnitude but opposite signs to enable the rephasing of the zero-quantum lactate coherence and the creation of a stimulated echo for singlets within the pulse passbands. The sequence is designed to retain half the original lactate and singlet signal intensities. Theoretical predictions were confirmed experimentally at 1.5T using phantom acquisitions. The lipid suppression factor was measured to be over 10(3).

Comparison of relative cerebral blood volume and proton spectroscopy in patients with treated gliomas.

BACKGROUND AND PURPOSE: Elevated relative regional cerebral blood volume (rCBV) reflects the increased microvascularity that is associated with brain tumors. The purpose of this study was to investigate the potential role of rCBV in the determination of recurrent/residual disease in patients with treated gliomas. METHODS: Thirty-one rCBV studies were performed in 19 patients with treated gliomas. All patients also had proton MR spectroscopy and conventional MR imaging. Regions of abnormality were identified on conventional MR images by two neuroradiologists and compared with rCBV and MR spectroscopic data. Metabolites and rCBV were quantified and compared in abnormal regions. RESULTS: In high-grade tumors, rCBV values were proportional to choline in regions of tumor and nonviable tissue. Although the presence of residual/recurrent disease was often ambiguous on conventional MR images, the rCBV maps indicated regions of elevated vascularity in all low-grade tumors and in 12 of 17 grade IV lesions. Regions of elevated and low rCBV corresponded well with spectra, indicating tumor and nonviable tissue, respectively. CONCLUSION: This study suggests that rCBV maps and MR spectroscopy are complementary techniques that may improve the detection of residual/recurrent tumor in patients with treated gliomas. Compared with the spectra, the rCBV maps may better reflect the heterogeneity of the tumor regions because of their higher resolution. The multiple markers of MR spectroscopy enable better discrimination between normal and abnormal tissue than do the rCBV maps.

Motion correction and lipid suppression for 1H magnetic resonance spectroscopy.

Spectral/spatial spin-echo pulses with asymmetric excitation profiles were incorporated into a PRESS-based localization sequence to provide lipid suppression while retaining a sufficient amount of water to allow for correction of motion-induced shot-to-shot phase variations. 1H magnetic resonance spectroscopy data were acquired at 1.5 Tesla from a motion phantom and in vivo from the human liver, kidney, and breast. The results demonstrated that lipids in the chemical shift stopband were completely suppressed and that full metabolite signal intensity was maintained after implementation of a regularization algorithm based on phasing the residual water signal. Liver and kidney spectra contained a large resonance at 3.2 ppm that was ascribed to trimethylammonium moieties (betaine plus choline) and a weaker signal at 3.7 ppm that may result from glycogen. A breast spectrum from a histologically proven invasive ductal carcinoma displayed a highly elevated choline signal (3.2 ppm) relative to that from a normal volunteer.

In vivo 1H MR spectroscopy of human head and neck lymph node metastasis and comparison with oxygen tension measurements.

BACKGROUND AND PURPOSE: Current diagnostic methods for head and neck metastasis are limited for monitoring recurrence and assessing oxygenation. 1H MR spectroscopy (1H MRS) provides a noninvasive means of determining the chemical composition of tissue and thus has a unique potential as a method for localizing and characterizing cancer. The purposes of this investigation were to measure 1H spectral intensities of total choline (Cho), creatine (Cr), and lactate (Lac) in vivo in human lymph node metastases of head and neck cancer for comparison with normal muscle tissue and to examine relationships between metabolite signal intensities and tissue oxygenation status. METHODS: Volume-localized Lac-edited MRS at 1.5 T was performed in vivo on the lymph node metastases of 14 patients whose conditions were untreated and who had primary occurrences of squamous cell carcinoma. MRS measurements were acquired also from the neck muscle tissue of six healthy volunteers and a subset of the patients. Peak areas of Cho, Cr, and Lac were calculated. Tissue oxygenation (pO2) within the abnormal lymph nodes was measured independently using an Eppendorf polarographic oxygen electrode. RESULTS: Cho:Cr ratios were significantly higher in the nodes than in muscle tissue (node Cho:Cr = 2.9 +/- 1.6, muscle Cho:Cr = 0.55 +/- 0.21, P = .0006). Lac was significantly higher in cancer tissue than in muscle (P = .01) and, in the nodes, showed a moderately negative correlation with median pO2 (r = -.76) over a range of approximately 0 to 30 mm Hg. Nodes with oxygenation values less than 10 mm Hg had approximately twice the Lac signal intensity as did nodes with oxygenation values greater than 10 mm Hg (P = .01). Cho signal intensity was not well correlated with pO2 (r = -.46) but seemed to decrease at higher oxygenation levels (>20 mm Hg). CONCLUSION: 1H MRS may be useful for differentiating metastatic head and neck cancer from normal muscular tissue and may allow for the possibility of assessing oxygenation. Potential clinical applications include the staging and monitoring of treatment.

Lactate detection at 3T: compensating J coupling effects with BASING.

Detection of lactate by in vivo 1H magnetic resonance spectroscopy may provide a means of identifying regions of metabolic stress in brain and other human tissue, potentially identifying regional ischemia in stroke or necrosis in tumors. At higher field strengths (3 and 4 T), which have recently become available for whole-body human studies, the chemical shift difference between the doublet from the methyl protons and the quartet from the methine proton becomes comparable to the available radiofrequency (RF) pulse bandwidth. In this case "anomalous" J modulation occurs in PRESS and STEAM because the coupling partner of the observed resonance may or may not be refocused by the RF pulses depending on the position of the molecule within the voxel and the size of the chemical shift misregistration artifact. These anomalies lead to signal cancellation for echo times near odd multiples of 1/J (often used to highlight the inverted lactate doublet against nearby lipid peaks) in single voxel studies, and spatial variation of the doublet lineshape in chemical shift imaging studies, producing erroneous determination of relative lactate concentrations. While increasing the band-width of the RF pulses can reduce this effect by reducing the signal cancellation, some cancellation will always remain. A means of eliminating this effect using BASING/ MEGA (Mescher M et al. Solvent suppression using selective echo dephasing J Magn Reson A 1996;123:226-229; Star-Lack J et al. Improved water and lipid suppression for 3D PRESS CSI using RF band selective inversion with gradient dephasing (BASING). Magn Reson Med 1997;38: 311-321) water suppression pulses will be described, along with some of its limitations.

Optimal gradient waveform design for projection imaging and projection reconstruction echoplanar spectroscopic imaging.

The use of modulated B0 projection gradient waveforms is proposed for magnetic resonance imaging (MRI) and magnetic resonance spectroscopic imaging (MRSI) to shape the k-space sampling density function to match the profile of an applied reconstruction window function. This allows for a time-efficient means of maximizing the signal-to-noise ratio when, for example, a low-pass spatial filter, designed to reduce k-space truncation artifacts and corresponding point-spread function sidelobe energies, is implemented. Both the two-dimensional (2D) and 3D cases are investigated. To create the projection gradient waveforms, a design method is developed that uses nonlinear constrained optimization (NLCO) to minimize the variance of the reconstruction noise. The design is subject to both equality and inequality constraints, which include the maximum gradient magnitude and slew rate. It is shown that NLCO can also be applied to twisting the projection trajectories for purposes of reducing the data acquisition time while still maintaining the desired sampling density. Applications to 1H MRSI are investigated via simulations. Advantages and limitations of the new sampling schemes are discussed.

Four-dimensional 1H and 23Na imaging using continuously oscillating gradients.

A class of fast magnetic spectroscopic imaging methods using continuously oscillating gradients for four-dimensional (three spatial and one spectral) localization is introduced. Sampling may start immediately following the application of an RF excitation pulse, thus enabling measurement of spin density, chemical shift, and relaxation rates of short-T2 species. For spatial localization, steady-state sinusoidal gradient waveforms are used to sample a ball in k space. The two types of trajectories presented include: (1) continuously oscillating gradients with continuously rotating direction used for steady-state free-precession imaging and (2) continuously oscillating gradients followed by a spoiler directed along discrete projections. Design criteria are given and spatial-spectral and spatial-temporal reconstruction methods are developed. Theoretical point-spread functions and signal-to-noise ratios are derived while considering T2*, off-resonance effects, and RF excitation options. Experimental phantom, in vivo, and in vitro 1H and 23Na images collected at 2.35 T are presented. The 1H images were acquired with isotropic spatial resolution ranging from 0.03 to 0.27 cm3 and gradient-oscillation frequencies ranging from 600 to 700 Hz, thus allowing for the separation of water and lipid signals within a voxel. The 23Na images, acquired with 500 and 800 Hz gradient waveforms and 0.70 cm3 isotropic resolution, were resolved in the time domain, yielding spatially localized FIDs.