Abstract: A magnetic resonance imaging apparatus includes one or more digital transmitters (B), one or more digital receivers (C), and digital data processing circuitry (D) which are all clocked and controlled by a single clock (F). Each digital transmitter includes a numerically controlled modulated oscillator (20) which processes digital phase and frequency signals to produce an output which addresses a wave-form map stored in a PROM (22). Each wave-form output of the PROM is multiplied (24) by a digital amplitude profile signal to generate a phase, frequency, and amplitude modulated digital RF signals. A clock gate (30) controls clocking of the digital modulation to create RF pulses. A digital-to-analog converter (28) converts the digital information to an analog RF pulse which is applied to a subject in an image region. The receivers each include an analog-to-digital converter (60) which digitizes the magnetic resonance signal emanating from the subject in the image region with four fold oversampling.

Abstract: A gradient magnetic field control (20) and a transmitter (30) are operated under the control of a timing and control computer (40) to generate magnetic resonance excitation pulse sequences. Each sequence provides phase encoding with one of a plurality of phase angles to the resultant resonance signals. A receiver (34) receives the phase encoded magnetic resonance signals which are digitized by an analog to digital converter (50) to form a plurality of views which are stored in a view memory (52). A larger plurality of views are generated adjacent a central or zero phase angle, e.g. views -63 to +64 of FIG. 2, and only one or a smaller plurality of views are generated adjacent peripheral phase angles, e.g. views -127 to -64 and +65 to +128 of FIG. 2. The slower, low frequency motion artifacts, such as respiratory motion artifacts, manifest themselves in the low frequency phase encoded views adjacent the zero phase encode angle.

Abstract: Magnetic resonance imaging data lines or views are generated and stored in a magnetic resonance data memory (56). The number of views or phase encode gradient steps N along each of one or more phase encode gradient directions is selected (70) to match the dimensions of the region of interest. A discrete Fourier transform algorithm (94) operates on the data in the magnetic resonance data memory to generate an image representation for storage in an image memory (96). Unlike a fast Fourier transform algorithm which requires a.sup.N views or data lines, where a and N are integers, the discrete Fourier transform has a flexible number of data lines and data values which can be accommodated. More specifically to the preferred embodiment, the discrete Fourier transform operation is performed by a CHIRP-Z transform or a Goertzel's second order Z-transform which can accommodate any number of data lines or values.

Abstract: A multi-echo magnetic resonance imaging sequence is implemented such that a radio frequency receiver (34) receives magnetic resonance signals during each of a plurality of magnetic resonance echoes. The resonance data received during each echo are digitized and the resultant echo data are stored in a corresponding echo memory (40, 42). The locations of the data within the memories are brought into registration (52) such that corresponding data in each memory is disposed at the same memory address. Because data from later echoes tends to be weaker or at a lower magnitude, the magnitude of the data stored in each memory is normalized (60). The phase of the data in each memory is brought into coordination by a zero order phase correction (70). A high pass filter (84) and a complementary low pass filter (86) separate complementary portions of the data from the memories. The separated portions are combined into a single synthesizied data set for storage in memory (82).

Abstract: In the absence of signals supplied to gradient field coils (14), radio frequency signals, or magnetic resonance signals received by coil (18), analog-to-digital converters (44, 46) indicate background noise and DC offset. An attenuator (26) attenuates signals from the receiving coil in the absence of magnetic resonance signals. A phase sensitive detector (38) produces real and imaginary signal components both in the absence of magnetic resonance signals during calibration and subsequently during the processing of magnetic resonance signals. The analog-to-digital converters, low pass filters (40, 42), and the phase sensitive detectors inject an undesirable DC offset into the signals. In the absence of a magnetic resonance signal, the digitized output of the analog-to-digital converters is substantially the DC offset plus noise. A statistical analysis routine (64) analyzes the sampled analog-to-digital data in the absence of a magnetic resonance signal.

Abstract: A spin echo (52) and a gradient echo (60) are generated in each magnetic resonance sequence repetition. The spin echo is phase encoded by a phase encode gradient (44) in regular steps spanning about a quarter of k-space. More particularly, steps from -n to G.sub.max /2, where n is a small integer and G.sub.max is the maximum phase encode gradient. An off-set phase encode gradient (58) shifts the phase encoding of the gradient echo by G.sub.max /2 relative to the first phase encoding gradient. Data to fill the empty portions of k-space (142, 167) between -n and -G.sub.max are generated from the complex conjugate (140, 160), of the first echo data (74) and the second echo data (76). The first and second echo data and the complex conjugate data are transformed (122, 132, 146, 166) to generate parted image representations (124, 134, 148, 168).

Abstract: An MRI or other scanner (A) generates medical diagnostic data d(x,y) which has a Gaussian noise distribution for reconstruction by an imager (B) into an electronic image representation P(i,j) which may have a Gaussian or Rayleigh noise distribution. An image improving circuit (C) replaces each image pixel value P(i,j) from an image reconstruction means (32) with an improved pixel value P*(i,j) defined as follows:P*(i,j)=G(i,j)[P(i,j)-P(i,j)]+P(i,j)-n,where G(i,j) is a weighting function uniquely defined for each pixel (i,j), P is the mean of pixel values of neighboring pixels and n is the mean image noise. The weighting function is based on a diagnostic data noise variance and a pixel value variance V(i,j) corresponding to the same pixel. The data noise variance is derived by comparing a data value difference between each data value d(x,y) and its neighboring data values in a data memory (30). The smallest data value difference is indicative of the image noise variance.