Picker International, Inc.

Case Western Reserve University

E Mark Haacke

John L Patrick, Iii

Francis H Bearden

Carolyn A Kershaw

Zhi-pei Liang

Janice E Hahn

Nagarjuna R Linga

An imaging apparatus and method which provides for reconstruction of images from sampled data with improved resolution, improved signal-to-noise, optimal edge detection, and pattern recognition. The apparatus and method are characterized by the use of a priori information (constraints) and a high pass filter on the data to obtain superresolution reconstruction of image data. An object function is approximated by a series of known model functions such as box car and polynomial functions. Solution of the object function depends on a finite number of unknowns whereupon the model parameters can be perfectly resolved to obtain infinite or superresolution. The apparatus and method have particular application in nuclear magnetic resonance imaging as well as other imaging modalities.

Parametric image reconstruction using a high-resolution, high

A portion of a subject (22) which is undergoing respiratory or other motion is disposed in an image region (20) to be examined. A respiratory or other motion monitor (50) monitors the cyclic respiratory motion and provides output signals indicative of chest expansion. A phase encoding gradient selector (60) selects the phase encoding gradient that is to be applied by a gradient magnetic field controller (40) and coil (42). A central phase encoding gradient is selected corresponding to a chest relaxation extreme and minimum and maximum phase encoding gradients are selected corresponding to a chest expansion extreme (FIG. 2). Intermediate degrees of monitored physical movement cause the selection of corresponding intermediate phase encoding gradients. Resonance signals collected during each phase encoding gradient are Fourier or otherwise transformed (80) into a corresponding view. A filter (92) weights each view such that views closest to the central phase encoding gradient are weighted most heavily and views adjacent the minimum and maximum phase encoding gradients are weighted least heavily. The physical position of pixels within each view are scaled (94) to adjust each view in accordance with the degree of physical expansion. The weighted and scaled views are transformed into an image memory (120) for display on a video display (122) or the like.

Centrally ordered phase encoding

A shimming magnetic field control (22) causes shimming magnetic fields for improving uniformity of a main magnetic field generated by main magnets (10). A resonance excitation control (32) selectively applies a resonance excitation pulse (34) and inversion pulse (36) for inverting the spin magnetization of water and lipid dipoles. A phase sensitive detector (30) selectively receives resonance signal components which are transformed by a transform algorithm (70) into a real image (72) and an imaginary image (74). The inversion pulse (36) is shifted by a time such that the water and lipid spin magnetizations are out of phase by a predetermined amount. With a 90.degree. phase difference, the real image represents water and the imaginary represents lipid. A phase image (80) is derived from a reference image pair (76, 78). A phase unwrap circuit (82) removes ambiguities attributable to the spin magnetizations becoming dephased by multiples of 2.pi. to create a phase map (84). A shim adjustment circuit (94) measures uniformity of the phase map and causes the shim field control circuit (22) to adjust the shimming magnetic fields to optimize the measured uniformity. The phase map is inversely transformed (100) to generate a field map (102) indicative of geometric distortion. A geometric distortion correction circuit (106) selectively interpolates values in the real and imaginary images in accordance with the field map to correct the image for geometric distortion.

Magnetic resonance imaging with phase encoded chemical shift correction

A reference object (24) is disposed in an image region (20) with a subject (22) to be examined. The reference object has known parameters such as relaxation time, spin density, dimensions, and position. Magnetic resonance signals in which the spatial position of resonating nuclei is encoded in the relative phase and frequency thereof are sampled and temporarily stored in a view memory (56). A Fourier transform (60) is performed to convert the stored signals view into a representation of at least the positions and spin density of the resonating magnetic dipoles of the subject and reference object. The parameters of the reference object measured from the image representation are compared or inverse transformed back to data space for comparison with actual parameters of the reference object or thresholds. Based on the comparison, the resonance signals or the image representation are adjusted. For example, the magnetic resonance signals are rotated (78) through a phase correction to correct for offsets in the phase encoding. As other examples, the spin density may be adjusted, low magnitude data discarded, spatially offset data shifted, or the like. Alternately, the reference object may be interconnected with the subject to undergo movement therewith. The position of the reference object is then utilized to select views in which the movement of the subject is within preselected ranges or to approximately correct for the motion. The image can be shifted to a selected location or interpolated.

Magnetic resonance reconstruction and scanning techniques using known

The resonance signals from a magnetic resonance imaging apparatus are discretely sampled. The sampled data are non-uniform in data space due to motion, eddy currents in the magnetic field, inaccurate timing in the application of the gradient magnetic fields, non-uniform sampling of the data, or the like. The discretely sampled data, s(k.sub.x, k.sub.y), is mapped from data space to image space by a generalized transform. The generalized transform, unlike a Fourier transform, concurrently corrects for the non-uniformity as it transforms the non-uniform data space data into image space data with a preselected distribution, e.g. uniform. For a given y position of non-uniformly sampled data s(k,y), the estimated spin density .rho.(x,y) is related to the actual spin density .rho.(x,y) by the equation: ##EQU1## where the matrix A describes the non-uniformity and can be calculated or determined experimentally. When the matrix B=A.sup.-1, the estimated and actual data match. In one embodiment, the generalized transform algorithm is determined by inverting the matrix A and operating on a data matrix therewith.

Generalized reconstruction technique

Hybrid fast scan magnetic resonance imaging is performed by using, for example, both a two dimensional Fourier transform (2DFT) method using phase encoding prior to data collection and an echo planar technique which phase encodes by using an oscillating gradient during data collection. In this hybrid imaging, the amplitude of the oscillating gradient determines the time savings achieved in imaging. The hybrid scan has particular application for medical diagnostic imaging since it is advantageous that such imaging be conducted as rapidly as possible.

Growing community of inventors

University Heights, OH, United States of America

E Mark Haacke