The computer facilities of the Department of Radiology is in a constant state of upgrades and additions. With the move to the new building a significant new addition is being made. We now have a large tape archive available for the research needs of the deparment. The automated tape library is a StorageTek 9710 tape library. While MIRL has had a automated library for several years it has been almost exclusively used to perform system backups, this was due to the complexity of the control and management software and the time required to save and restore research data.

The StorageTek has a completely automatic file migration software installed with it. As data is stored on the large RAID disk system the HSM software will monitor the available free disk space. As the disks fill the HSM will automatically move the least recently used files to the tape jukebox. As the file is moved it is deleted from the disk leaving a stub entry in the directory. When the data is accessed again the file will automatically be moved from the tape system back to the disk. With the StorageTek tape jukebox a now 200 gig RAID 5 disk subsystem has been purchased as the cache for the tape archive. As MIRL moves to the new building all of the /v/DATA? disks will be moved to this new RAID disk system.

The Small Bore MR Research System has been undergoing considerable changes in the past few months. In an attempt to make the system more user friendly to the research group at large, in addition to providing a larger suite of resident pulse sequences, we arranged with Picker to attach a Picker Eclipse Console to the small bore magnet. The premise is to provide a research system with an identical development environment to the clinical system in the Bureau of Mines, thus enabling more efficient pre-clinical development of MR techniques. Some of the advantages of the new console include a 4x digital receiver (enabling phased array experiments) and active T/R switching. Available sequences include FSE, 3D FSE, Single, and Multishot EPI, Angio techniques, Spectroscopy etc. The system will have a clinical interface allowing "push-button" NMR for many of the available sequences.

The unique conglomeration of researchers at the Huntsman Cancer Institute has allowed us to initiate a high-resolution NMR study of cell extracts of variety of tumor cell lines. The hypothesis of this study is that tumors of different lineage will have different metabolic profiles, and that we can use high resolution NMR as a fingerprinting tool to identify these tumor specific profiles. Tumors isolated from human patients are grown up in situ and then extracted with perchloric acid. Cell extracts are then analyzed using high field (11.5T) NMR spectroscopy. Spectra are evaluated qualitatively for unique peaks and for different metabolite ratios. Preliminary data have shown distinct variances in glioblastoma's of different lineage and between different grade neuroblastoma's. Specifically, varying levels of alanine, NAA glu/gln and relative levels of phosphoryl-choline and choline could be seen. It is hoped that successful progression of this study will permit differentiation of tumors that are difficult to differentiate by current clinical methods, and that this increased differentiation will provide enhanced therapeutic capabilities.

Accurate interpretation of lung images to diagnose pulmonary embolism (PE) is highly variable between physicians which results in a low sensitivity and specificity. To eliminate diagnostic variability, a pattern recognition system was developed to analyze both ventilation and perfusion lung images. The signal processing module analyzes each image using a smart region growth algorithm to identify the lung shape. The margins of the lung region are then examined by shape analysis to determine areas of diminished radionuclide concentration. Segments are analyzed and correlated between image views. A classification module correlates regional ventilatory and perfusion defects and analyzes them according to the PIOPED criteria that physicians use in visual interpretation.

In a study of 41 patients, the program correctly identified one normal patient and 16 of 17 as low probability. Of twelve intermediate human interpretations, the computer correctly identified nine as low and the algorithm correctly identified all seven patient with a high probability of PE. In 93% of the cases the computer made a decision in a category other than intermediate compared to 36% for human interpreters. The computer reached the correct diagnosis in over 86% of the cases. This algorithm has the potential to assist physicians by improving the accuracy of diagnosis and reducing intermediate results.

Purpose: To evaluate the inter-observer variability and inter-modality variability using conventional angiography and gadolinium-enhanced MRA in the assessment of renal artery stenosis

Methods: 54 patients (31 male, 23 female) with an average of 64 years (range 29-86 years) underwent both conventional angiography and gadolinium-enhanced, 3D gradient echo MRA. MRA was performed on a 1.5T magnet during rapid bolus intravenous injection of 40 ml of gadolinium-based contrast. The studies were performed within 0 to 4 months of each other (average 2.3 weeks). In all patients, no surgical or radiological intervention was performed between the two imaging studies. Three experienced angiographers assessed the angiographic images for renal artery stenosis, blinded to each others interpretations and to the MRA studies. Similarly, three experienced MR radiologists assessed the MRA images.

Results: Inter-observer variability for the degree of renal artery stenosis in the 107 kidneys evaluated in these 54 patients was not significantly different between the two imaging modalities. The mean standard deviation of the percent diameter stenosis as judged by MRA was 6.9% compared to 7.5% by conventional angiography (a<=0.05, p = NS). In 69 of the 107 (64.5%) kidneys evaluated, the average renal artery stenosis estimated by the readers from the two image modalities differed by 10% or less. MRA over-estimated the degree of stenosis as compared to conventional angiography by more than 10% in 22 cases (20.6%). Surprisingly, MRA underestimated the degree of stenosis in 16 cases (15.0%).

Purpose: To determine the clinical utility of estimating renal perfusion using MRI in selecting patients for renal artery interventions.

Materials and Methods: 19 patients suspected of renal artery stenosis (RAS) underwent quantitative evaluation of renal perfusion using a gadolinium-enhanced spoiled gradient echo sequence followed by a gadolinium-enhanced breathhold 3D-TOF renal MRA. Gadolinium (0.05-0.1ml/kg) was infused over two minutes during the perfusion sequence. Signal from the renal cortex and medulla were analyzed separately using a single-compartment pharmacokinetic model. Estimates for the apparent flow/volume ratio (F/V) were calculated for each kidney cortex and medulla.

Results: All patients with RAS>50% suspected by MRA underwent catheter angiography. For normal renal cortex and medulla, F/V= 0.72+0.26 min-1 and 0.41+0.20 min-1, respectively. Eight kidneys with catheter proven RAS >50%, had abnormally low renal cortical and medullary F/V estimates (0.28+0.13 min-1 and 0.15+0.06 min-1, respectively). Three kidneys found not to have significant stenosis by catheter angiography, had low normal cortical and medullary F/V estimates (0.60+0.08 min-1 and 0.43+0.15 min-1, respectively). Conclusions: Renal perfusion estimates using gadolinium-enhanced MRI are a clinically important parameter in selecting patients for renal artery interventions.

A high incidence of seizures and "sub-clinical" epileptiform EEGs has been reported for children with autism spectrum disorders (Autism, Pervasive Developmental Disorder - Not Otherwise Specified, Asperger Syndrome, Retts Syndrome, And Disintegrative Disorder). Incidence of epileptiform discharges is especially high for the sub-population characterized by a noticeable deterioration and regression of language skills and behaviors after 18 months of age, and it is tempting to speculate that electrophysiological perturbation of specific brain subsystems is causally related to the autistic regression. This study used magnetoencephalography to explore patterns of epileptiform activity in 30 children with regressive autism spectrum disorders (ASD). Even though clinical seizures had been reported for only 9 (33%) of these children, epileptiform discharges were observed in 25 (80%). Multiple-dipole, spatio-temporal modeling was used to define patterns of activity. Multifocal activity was found in most cases, although 6 children demonstrated unifocal epileptiform discharges involving the superior and lateral aspects of the superior temporal gyrus. The pattern of activity for these children was reminiscent of that seen for children with Landau-Kleffner Syndrome, an acquired epileptiform aphasia and auditory agnosia that is often amenable to steroid therapy and/or neurosurgical intervention. Overall, 18 of the 25 ASD children with epileptiform activity demonstrated temporal lobe discharges. The data raise the possibility that medical therapies developed for the control of the epilepsy may be useful in the clinical management of children with autism.

Brain activity during sleep has been studied by measuring brain metabolic activity, or alternatively, by recording the changing electromagnetic field patterns associated with the on-going electrical activity of the neurons. Both measurements have shown that the brain manifests different activity during sleep than during wakefulness. In particular, specific rhythmical activities seen in a standard electroencephalogram have been long known for wakefulness and sleep. The rhythms seen during wakefulness are activity dependent, and in fact Pfurtscheller has proposed that each of the awake rhythms is generated by "idling" primary sensory cortex. Thus, alpha is generated by inactive primary visual cortex, mu by inactive sensorimotor cortex, and tau by inactive primary auditory cortex. In sleep, none of these rhythms is seen, but they are replaced by a characteristic rhythm of sleep - spindling. Spindling is similar to the waking rhythms in that it is generated by thalamocortical interaction. The question asked in this abstract is whether there si a characteristic cortex which, when idling, is responsible for generating spindles, just as
there are specific generators of alpha, mu, or tau.


The conclusion that each waking frequency is generated by a specific idling is based upon the observations that each displays a different frequency range, a different topographical distribution, and is suppressed by specific sensory input or motor activity. Such characteristics are sough for spindles. Using MEG to study spindles suggests that they have a specific topographical distribution, and a characteristic frequency range. In fact, spindles in one person may have different frequencies in different regions of the head. In addition, we also know that spindles can be suppressed, e.g. by wakefulness, or rapid eye movement sleep. Localization of MEG spindle activity suggests that there are multiple generators, a similar conclusion to that drawn about alpha activity. Spindle activity localizes usually to association cortices, a conclusion in agreement with the multiple frequencies recognized in a single individual.

The conclusion drawn is that spindles are possibly generated from idling association cortices ina manner analogous to alpha, tau and mu. An understanding of spindles may shed light on the connectivity of the brain, and the nature of the sleep state. In addition, disappearances of spindles has been seen Rett's Syndrome and infantile neuronal ceroid lipfuscinosis. It may be a sensitive indicator of a number of brain disorders.

Magnetocardiography (MCG) is a non-invasive method to study the three- dimensional localization of cardiac electrical activity. MCG has been applied in the evaluation of several types of cardiac activation including His-Purkinje activity, pre-excitation via accessory pathway, origin of ventricular extra-systoles, supraventricular arrhythmias, ventricular tachycardia, atrial tachycardia, and activity generated by pacing catheters. Most of the clinical application has focused on Wolff-Parkinson-White syndrome and ventricular arrhythmias. Although it has been suggested that MCG be used for diagnosis of myocardial ischemia and infarct there has been only limited application in this area. Since myocardial ischemia and infarction are the most critical cardiac diseases in terms of mortality and frequency of occurrence it is of interest to investigate whether MCG has potential as a non-invasive diagnostic technique for these important cardiac diseases. One important issue is the need to develop a better understanding of how an ischemic insult causes depolarization and repolarization that precedes cardiac arrhythmias. This presentation will focus on the development of a simulation program based upon a computer model that simulates the spread of activation in the heart and the calculation of the ex vivo magnetocardiogram. Particular emphasis will be placed on simulating the changes of propagation of the excitation wavefront and changes in repolarization due to myocardial ischemia.

The Compton camera has been proposed as an alternative to the Anger camera for SPECT. In its conventional form, the camera consists of two gamma detectors, so that incident gamma photons undergo Compton scattering in the first detector and are absorbed in the second. Coincident registration of these two interaction events defines a "cone of possibility" for the origin of a particular gamma photon. The three vertex electronic collimation requires a Compton camera with three detectors and is achieved by coincident registration of three interaction events: Compton scatterings in the first two detectors and absorption in the third detector. Because of polarization, the plane containing the three vertices defined by the interaction events tends to make a small angle with respect to the plane containing the source point and the first two vertices. This allows significant reduction of the "cone uncertainty" and better approximation of the source location by two semi-lines in the shape of a "V"; the result of the intersection of the plane containing the three vertices and the cone. The data collected by such a camera are integrals over "V" shaped semi-lines called "V"-projections. This paper describes an approach for fully 3D image reconstruction from "V"-projections. The method is based on a special transformation, which maps a certain subset of "V"-projections into a complete set of line-projections. Since the transformation leaves the original image unchanged in the region of interest, the filtered backprojection algorithm can be used for reconstruction.

The qualitative and quantitative accuracy of SPECT images are degraded by a significant number of the measured photons undergoing one or more Compton interactions prior to detection. A promising approach to scatter compensation involves modeling the scatter response function in the operator used during the reconstruction process. This type of method tends to be very computationally intensive, however recent developments in computer hardware have given rise to the possibility of using iterative methods for the reconstruction of SPECT data.

Most of the methods for applying physical corrections use a 2D model, that is to say, photons/matter interactions occur slice by slice. This approximation allows significant reduction of processing times and memory use. However, this model is far from exact for scatter and detector response corrections.

The method presented here is the extension to 3D of a model developed for 2D scatter and attenuation correction. This model is based on the simultaneous acquisition of transmission and emission projection data and the use of the transmission map to define the inhomogeneous scattering object. The key point is the use of the set of line integrals (ray-tracing method) as the basis of a model of the distribution of scattered events in the emission projection data. The major difference from the original method is that line integrals passing through a given point are now computed in a 3D volume instead of the 2D slice which the point belongs to.

An iterative ML-EM reconstruction algorithm with a projector/backprojector that models photon attenuation and the spatially varying geometric response of the collimator, is used to improve the resolution and the quantitative accuracy in single photon emission computed tomography (SPECT). A key aspect of the implementation of this algorithm is the development of an efficient model for the geometric response by using an incremental slab-by-slab technique instead of a slice-by-slice blurring technique with a small, 2-D depth-dependent, convolution kernel obtained by least square method. In forming the projections from back to front, a collection of coronal slices making-up one slab, are summed for each projection bin. This is done taking into account the effect of attenuation. A depth-dependent 2D kernel is applied to the slab sum and the result is added to the result for the next slab, which is closer to the detector.The implementation maintains the pixel resolution in the axial and trans-axial directions but convolves with a blurring function that is the blurring function at the center of the slab. The application of the incremental blurring model, to the reconstruction of both patient and phantom SPECT data, showed improved resolution and contrast. The slab-by-slab implementation significantly improved computation time over the slice-by-slice implementation without degrading image quality.

An algorithm was developed to obtain reconstructions from truncated projections by utilizing cross-correlation of a "knowledge set" of a priori nontruncated cross-sections with a similar structure. A cross-correlation matrix is constructed for the known set of cross-sectional images. The eigenvectors of this matrix form a set of orthogonal basis vectors for the reconstructed image. The basis set is optimal in the sense that the average of the differences between members of a given set of a priori images, and their truncated linear expansion for any basis set, is minimal for this particular basis set. Therefore, one can represent an image not in the Òknowledge setÓ, but of similar structure by a linear combination of basis vectors corresponding to the larger eigenvalues; thus reducing the number of basis vectors to a number less than the total number of pixels.The projection of an image represented by this linear combination of basis vectors is a linear combination of projected basis vectors. A constrained least squares method is used to evaluate the coefficients of this expansion by minimizing the sum of squares difference between the expansion and the projection measurements taking into account statistical distribution of coefficients over basis vectors. The least squares estimates of the coefficients are used in an expansion of the orthogonal basis to obtain the reconstructed image. Constrained linear inversion stabilized solution removing noise corresponding to computational and systematic error of ill-conditioned problem. It is shown that the reconstruction of truncated projections is significantly improved over that of commonly used iterative reconstruction algorithms.

Scatter correction is an important factor in single photon emission computed tomography (SPECT). Many scatter correction techniques, such as multiple-window subtraction and intrinsic modeling with iterative algorithms, have been under study for many years. Previously, we developed an efficient slice-to-slice blurring technique to model attenuation and system geometric response in a projector/back projector pair, which was used in an ML-EM algorithm to reconstruct SPECT data. This paper proposes a projector/back projector that models the 3D first-order scatter in SPECT, also using an efficient slice-to-slice blurring technique. The scatter response is estimated from a known, non-uniform, attenuation distribution map. It is assumed that the probability of a scatter event at a pixel is proportional to the attenuation coefficient value at that pixel. Monte Carlo simulations were used to verify the accuracy of the proposed projector/back projector model. The non-uniform MCAT phantom was used as the attenuation map. For a 64x64x64 image volume, it took 5 seconds to generate a 64x64 scatter projection array. The main advantage of the proposed method is its ease to implement and efficiency in computer calculation.

Conventional cardiac SPECT suffers from the problem of heart movement during the data acquisition. Thus, the reconstructed image only represents an average estimation of the cardiac activity. Gated SPECT seeks to correct this problem by synchronizing data acquisition with the cardiac cycle. Instead of collecting a single projection data set, in gated SPECT, several "time frames" of projection data sets are acquired during each cardiac cycle. The 3D images in each time frame offer improved spatial resolution as compared to the ungated image since the heart moves less within the time aperture of observation. Unfortunately, since significant increases in total counts are not possible, each time frame of the gated study has a fraction of the total counts of the composite study, and the individual gated images are noisy and are significantly influenced by some physics factors such as the nonuniform attenuation, geometry response and scatter. To improve the quantitative accuracy, we acquired the transmission data separately to reconstruct the transmission map. An EM algorithm was then used to compensate the nonuniform attenuation for each data set. The images between the uncompensated and compensated images will be compared and verify the improvement.

PET imaging of ¹⁸FDG is the gold standard for determination of myocardial viability. Static SPECT imaging of the positron emitting ¹⁸FDG has recently been realized clinically, by using 511 keV collimators and standard gamma cameras. A possible improvement to imaging ¹⁸FDG is to acquire a time series of images and fit the changing distribution of the radionuclide to a three-compartment model. This type of dynamic PET imaging of ¹⁸FDG has been reported to be more accurate and have better contrast than static PET. Dynamic SPECT imaging, however, may not be feasible due to low photon detection efficiency and poor resolution. It is thus of interest to determine if dynamic imaging of myocardial ¹⁸FDG with a three detector SPECT system and a compartmental modeling approach can result in accurate kinetic parameter estimates.

An anatomically realistic simulation was used to investigate imaging ¹⁸FDG with dynamic SPECT. Time-activity curves for blood and the myocardium were modeled from published dynamic PET data. The simulated projections were reconstructed and useful parameter estimates were obtained from large regions of interest. In addition to the simulation, dynamic ¹⁸FDG imaging of a 22kg canine was performed using a high (17mCi) dose. Although the blood pool could not be reliably defined on this small heart, fits using plasma samples for the input function resulted in reasonable parameter estimates.

Rotating Slant Hole (RSH) collimators have been designed for accurate lesion detection in SPECT imaging by improving photon sensitivity and thereby improving the resolution of the gamma camera images. Photon sensitivity is enhanced by increasing solid angular beam coverage using converging collimators.

An RSH collimator rotates about its own axis while orbiting a patient. Two types of RSH collimators have been designed - a 2 segment and a 4 segment collimator. Both types requires the smallest field of view for the best accuracy. Since the location of the heart is different in each patient, due to different sizes and gender, the smallest common volume of heart is evaluated by grouping the heart location of male and female patients falling within particular weight ranges. Therefore, all patients are first classified by gender and then grouped by weight into categories of extra-large, large, and small/medium by frequencies 0.15, 0.35 and 0.5. Each group is imaged using separate collimator with a separate slant angle covering the common volume of the field of view.

In designing this collimator, the geometric transfer function response, edge effects and center effect (in the case of a 4 segment collimator) have been considered. A database of 300 patients was sampled for this research.

This presentation describes a proposed research project, whose funding is anticipated to begin in January 1997. My current project on RSH SPECT will be presented by Swapna Devi.

Reconstruction algorithms for cone-beam tomography usually consider the projection data to be sampled along a well defined vertex path in space. In order to provide a tomographically complete dataset, the path should satisfy the celebrated Tuy's condition, and none of the projections should be truncated. There has been some progress in relaxing each of these constraints separately.

This "limited data" project addresses the general cone-beam image reconstruction problem, where the vertices do not necessarily lie on a Tuy path, and the projections may be truncated. The objective is to provide tomographic images of the internal structure of high contrast objects (such as metal containers). Some preliminary images, showing reconstructions from truncated projections and arbitrarily placed vertices will be presented. A model predicting achievable local resolution in terms of the vertex distribution will also be presented.

The velocity of blood in the anterior cerebral arteries of 12 patients was measured as a function of vessel diameter using digital subtraction angiography. Contrast bolus time density curves were measured at multiple points along artery segments. The time of arrival for each point along the vessel was determined by a linear least squares fit. Velocities were computed from the slopes of distance vs. Time of arrival curves. Vessel diameters were measured from the angiogram and corrected for magnification.

A general trend of decreasing velocity with decreasing vessel size was observed. Mean (diameters: velocities) calculated for seven diameter ranges were:(0.36mm:9.6cm/sec), (0.45mm:12.1cm/sec), (0.55mm:16.8cm/sec), (0.65mm:18.6cm/sec), (0.74mm:16.4cm/sec), (0.85mm:21.3cm/sec), (21.3mm:29.0cm/sec), (1.13mm:33.1cm/sec).

This technique works well when the velocity ranges between 5 and 30 cm/s. Blood velocities below this level tend to be in vessels below 0.3 mm diameter and are difficult to measure with precision due to the decreased signal to noise ratio. Velocities above 30 cm/s are difficult to measure in the short vessel segments of the anterior cerebral vessels at a frame rate of 7.5 frames per second.

It is difficult to measure the blood flow by using contrast bolusinjection method in a short and moving vessel segment. The mass conservation method has the capability to overcome some of these problems. However, this approach usually overestimates flow at the stage of bolus leading edge because of the assumption of plug flow and uniform mixing of the contrast agent. With the knowledge of the transport velocity, we found that the accuracy of the velocity measurement could be improved dramatically by carefully choosing the initial bolus leading edge. This process actually removes the high transport velocity part of the leading edge from later frames and results in a more accurate measurement. Simulation and phantom studies validated this modified technique. For a straight tube of 4.76mm diameter with a flow rate ranging from 2cc to 12cc, the accuracy for the measurement in a segment of 4.0 to 8.0 cm length is within 15% by using about 40% front leading edge as the initial frame.

The linescan pulse technique is a method for selecting a line of interest. This line selection technique is based on two sets of field gradients which are applied to two directions. An echo is produced from a line by a 90^o RF pulse in the presence of a gradient followed by a 180^o RF pulse with a gradient perpendicular to the first. The linescan pulse technique is developed using the pulse gradient Hahn spin echo sequence (PGSE) on a GE 1.5T MR scanner to precisely measure the diffusion coefficient (Dw) of water.

The error factor for Dw measurements with linescan PGSE caused from inhomogeniety of field gradient and signal to noise ratio(SNR) is examined using a water phantom. The distribution of field gradient inhomogeniety along the line, assuming the water Dw as 2.1x10-5 cm2/sec, is visualized and changes gradually according to sample position. The effect of SNR on this calculation method is also estimated by adding white noise to simulation data. The standard deviation (SD) of our diffusion estimates for SNR=90 of a water signal is calculated as 3.4x10-7.

A 2D image is made by selecting several more lines across the object adjacent to each other. Using a diffusion weighted linescan image, it is possible to measure the distribution of Dw values. This is currently being developed. The linescan method is a simple and old technique. However, it is useful to measure the Dw value of water precisely.

Recently, I implemented a combination of linear prediction, singular value decomposition and non-linear variable projection least-square fitting techniques to noisy data obtained from an MRS fatigue study. The goal was to obtain the relative concentrations of P31 metabolites, chemical shifts and transverse relaxation times from the time-domain data obtained from the arm muscle (flexur carpi radialis). This avoids the different time-consuming and manually intensive techniques applied in the frequency domain which require a trained-user interaction (i.e peak picking, phase correction, etc).

The problem is based on modeling the NMR FID and establishing complex data matrices by the linear prediction method which gives a first guess to the non-linear least square optimizer. There are many different variables that need to be optimized for the problem -- the noise, the coil tuning sensitivity, the model assumptions, the short acquisition time and the dynamics of the fatigue exercise (i.e peak variation with pH) -- makes the problem challenging.

Quality assessment is an important but under utilized field in medical imaging. With increased demands for justifying spending sparse resources on new technologies, medical imaging technology development needs to be directed towards optimizing image quality. In radiology image quality is usually defined authoritatively in terms of some signal detection task. Here typically an observer is asked to perform a binary classification task: is a signal (e.g., a lesion) present or absent. The observer's classification is compared to some truth standard to compute sensitivities and specificities for the task. However, signal detection experiments are difficult and expensive to conduct, primarily because of the difficulty in obtaining and using a truth standard. Therefore, establishing the relationship between signal detection measures and either physical measurements

or observer tasks not requiring a truth standard is an important problem. The focus of our work is to develop such measurements for assessing image quality in magnetic resonance angiography (MRA). We have previously developed MRA signal detection experiments (both ROC and 2AFC) using cerebral x-ray angiograms as the truth standard. In this talk we will present our preliminary work on relating signal detection results with three alternative measures: (1) a rank evaluation of observer's perceived image quality, without reference to a truth standard; (2) detailed contrast-to- noise ratio measurements; and (3) linear discriminate functions based on small sub-images. We will apply these measures to evaluate the impact of asymmetric echo correction and zero-filled interpolation in MRA image quality. The limitations of these measures will be discussed and work on second generation measures will be presented.

There are a variety of applications where 3D multislab imaging is advantageous. However, imperfect excitation profiles can cause intra-slab variations in intensity producing an inter-slab discontinuity, the so-called "slab boundary artifact" or SBA. One can reduce SBA by overlapping slabs, but this increases the total acquisition time, reducing efficiency. We are currently exploring novel k-space sampling techniques which seek to maintain the resolution and SNR efficiency of 3D acquisitions while reducing the artifacts which degrade their performance. Preliminary work with the Sliding Interleaved ky (SlInky) technique developed for MRA has produced some excellent results. Yet, before the technique can be considered robust, a number of challenges remain: the determination of the optimal sampling order, design of the RF pulse, and proper measurement of the slab excitation profile. We will discuss our work in addressing these issues and our progress in applying these techniques to other 3D modes.

Magnetic resonance images acquired with phased array coils have severe intensity modulation due to the inhomogeneous fields of the coils. Several approaches have been devised to correct the intensity modulation. They generally consist of approximating the field pattern of the phased array coils, and dividing the acquired image by that pattern to correct the received intensities. A simple analysis shows that this approach does correct image intensities, but greatly amplifies image noise in regions of low image intensity. The amplified noise rivals the intensity of small blood vessels, and clouds their visiblity in maximum intensity projections. An alternative approach is to subtract the field pattern from the acquired image. Subtraction preserves a constant noise amplitude across the entire image, so that small vessels viewed in Maximum Intensity images can be seen with much more clarity.

The goal of this research has been to develop an end-cap birdcage radio frequency (RF) head resonator for the purpose of increasing the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) in Magnetic Resonance Angiography (MRA) images. This improves visualization of the smaller vessels in the brain. The birdcage resonator has become a standard coil for magnetic resonance imaging because of its large homogeneous imaging volume, its high characteristic Q-factor, and relatively high SNR. Its simple symmetric design also lends itself easily to quadrature excitation and detection of the spins of interest. However, because of the characteristic RF profile of the standard birdcage coil, less signal is obtained towards its open ends. In this work, special consideration was given to increase the field homogeneity of the coil towards the top of the head, improve the coil Q- factor and achieve better SNR than available with a standard (commercial) birdcage coil. The prototype coil measures 24 cm x 20.5 cm, resulting in a better filling factor than the standard birdcage head coil. The conductive end-cap increases field homogeneity and helps reduce resistive losses. The desired characteristics of this coil have been demonstrated through phantom experiments and the use of vessel Contrast-to-Noise (CNR) measurements in MRA images obtained with this prototype coil.

The objective of most conventional magnetic resonance angiography (MRA) is to obtain a strong signal from flowing blood while suppressing the signal from the surrounding soft tissues. An alternative approach is to suppress signal from blood and image the other tissues. Recently, we have developed such "black-blood" MRA techniques for imaging cerebral vessels and the carotid arteries using fast spin echo (FSE) pulse sequences. Spatial saturation bands and signal loss due to gradient moments cause the blood signal to be dark. For imaging vessels in the brain, we have optimized the imaging parameters for a 3D FSE sequence to maximize the contrast between the brain and vessels. We have similarly developed 2D FSE and localized 3D FSE sequences for imaging the carotid arteries. Various strategies, such as inversion nulling, have also been examined for improving the blood signal suppression. Initial studies have demonstrated that black-blood techniques are capable of showing similar and possibly improved vessel detail relative to that of conventional "white-blood" MRA techniques. Example cerebral and carotid black blood images will be shown and the strengths and weaknesses of black-blood MRA will be discussed.

Magnetic Resonance Angiography (MRA) techniques have greatly improved in recent years but some challenges remain to be addressed. This presentation addresses the problem of small vessel visualization in intracranial MRA. The hypotheses that small vessel detail is lost due to motion and due to signal saturation in slowly moving blood are both examined. An ECG correlated multiple phase 3D time-of-flight (TOF) MRA pulse sequence was developed and utilized to test the hypothesis of blurring due to vessel motion. Contrast enhancement is proposed to remove the signal saturation effects. Solutions to the Bloch equations are used to provide imaging parameters for maximized blood/tissue contrast as a function of Gd-DTPA concentration. These solutions provide an understanding that can be used to improve vessel detail in all types of contrast enhanced MRA.