Method and apparatus for hardware RF receiver channel reduction

Abstract

Method and apparatus for hardware coil compression is disclosed. The coils in an array configured for the same region of interest are grouped into sub-arrays. The coils of each sub-array are pre-combined with a hardware combiner before further processing. The pre-combination converter composed of the pre-combiners is flexible, i.e., applicable to for example non-cylindrical coils; simpler than direct implementation of the software compression algorithm; and commercially feasible.

Claims

1. A method of channel compression for magnetic resonance imaging (MM), the method comprising: pre-combining, using n′ pre-combiners, n outputs (C1, C2, . . . Cn) from n coils of an n coil array configured for imaging the same region of interest (ROI) to obtain fewer n′ pre-combination outputs, where n′ is an integer greater than 1, wherein the n coils are grouped into n′ sub-arrays based on an order of importance of each of the n coils represented by a contribution of the output of each of the respective n-coils to a signal-to-noise ratio (SNR) of a signal obtained by combining the outputs of the n coils and each pre-combiner is configured to pre-combine outputs from a respective one of the n′ sub-arrays to obtain one of the n′ pre-combination outputs; and compressing the n′ pre-combination outputs of the n′ pre-combiners into fewer virtual outputs with highest signal-to-noise ratios (SNRs) for image reconstruction.

2. The method of claim 1, wherein the n coils are grouped into the n′ sub-arrays by: ordering the n coils by the importance from the highest to the lowest; assigning a first n′ coils of the n coils into the n′ sub-arrays, respectively; and repeating the assignment of the remaining n-n′ coils until all the n coils are assigned.

3. The method of claim 1, wherein the sub-arrays and pre-combination coefficients for the sub-arrays are determined by: constructing a channel conversion matrix M of n×n which converts signals s=(s.sub.1, s.sub.2, s.sub.n).sup.T output from the n coils of the n coil array to conversion outputs s′=(s.sub.1′, s.sub.2′, s.sub.n).sup.T, s′=Ms, wherein .sup.T represents matrix transpose operation; selecting rows of the channel conversion matrix M corresponding to a number of conversion outputs with the highest signal-to-noise ratios (SNRs); and optimizing the pre-combination coefficients or both the grouping and the pre-combination coefficients such that the pre-combination coefficients span a space which approaches the space spanned by all or the selected rows of M.

4. The method of claim 3, wherein the channel conversion matrix is a whitening matrix.

5. An apparatus of channel compression for Magnetic Resonance Imaging (MRI) comprising a converter, the apparatus further comprising: n′ pre-combiners configured to pre-combine n outputs (C1, C2, . . . Cn) from n coils of an n coil array configured for imaging the same region of interest (ROI) to obtain fewer n′ pre-combination outputs, where n′ is an integer greater than 1, wherein the n coils are grouped into n′ sub-arrays based on an order of importance of each of the n coils represented by a contribution of the output of each of the respective n-coils to a signal-to-noise ratio (SNR) of a signal obtained by combining the outputs of the n coils and each pre-combiner is configured to pre-combine outputs from one of the n′ sub-arrays to obtain one of the n′ pre-combination outputs, wherein the converter is coupled to the n′ pre-combiners and configured to compress the n′ pre-combination outputs into fewer virtual outputs with highest signal-to-noise ratios (SNRs) for image reconstruction.

6. The apparatus of claim 5, wherein the n coils are grouped into n′ sub-arrays by: ordering the n coils by the importance from the highest to the lowest; assigning a first n′ coils of the n coils into the n′ sub-arrays, respectively; and repeating the assignment of the remaining n-n′ coils until all the n coils are assigned.

7. The apparatus of claim 5, wherein the n′ sub-arrays and pre-combination coefficients for the sub-arrays are determined by: constructing a channel conversion matrix M of n×n which converts signals s=(s.sub.1, s.sub.2, s.sub.n).sup.T output from the n′ coils of the array to conversion outputs s′=(s.sub.1′, s.sub.2′, s.sub.n).sup.T, S′=Ms, wherein .sup.T represents matrix transpose operation; selecting rows of the channel conversion matrix M corresponding to a number of conversion outputs with the highest signal-to-noise ratios (SNRs); and optimizing the pre-combination coefficients or both the grouping and the pre-combination coefficients such that the pre-combination coefficients span a space which approaches the space spanned by all or the selected rows of M.

8. The apparatus of claim 7, wherein the channel conversion matrix is a whitening matrix.

9. The apparatus of claim 5, wherein the converter further comprises: a plurality of splitters configured to split each of the n′ pre-combination outputs to a number of output copies; and a plurality of combiners coupled to the splitters and configured to combine the the number of output copies to the virtual outputs, wherein each combiner is characterized by a set of complex combination coefficients whose amplitudes represent amplitude weights and angles represent phase shifts.

10. A controller for channel compression for magnetic resonance imaging (MRI), comprising: a non-transitory computer-readable storage medium that stores machine executable instructions; and a processor that executes the instructions, wherein, when executed by the processor, the machine executable instructions cause a system that includes the controller to implement a process that comprises: pre-combining, using n′ pre-combiners, n outputs (C1, C2, . . . Cn) from n coils of an n coil array configured for imaging the same region of interest (ROI) to obtain fewer n′ pre-combination outputs, where n′ is an integer greater than 1, wherein the n coils are grouped into n′ sub-arrays based on an order of importance of each of the n coils represented by a contribution of the output of each of the respective n-coils to a signal-to-noise ratio (SNR) of a signal obtained by combining the outputs of the n coils and each pre-combiner is configured to pre-combine outputs from a respective one of the n′ sub-arrays to obtain one of the n′ pre-combination outputs; and compressing the n′ pre-combination outputs of the n′ pre-combiners into fewer virtual outputs with highest signal-to-noise ratios (SNRs) for image reconstruction.

11. The controller of claim 10, wherein the process implemented by the system further comprises: grouping the n coils into the n′ sub-arrays by: ordering the n coils by the importance from the highest to the lowest; assigning a first n′ coils of the n coils into the n′ sub-arrays, respectively; and repeating the assignment of the remaining n-n′ coils until all the n coils are assigned.

12. The controller of claim 10, wherein the process implemented by the system further comprises: determining the sub-arrays and pre-combination coefficients for the n′ sub-arrays by: constructing a channel conversion matrix M of n×n which converts signals s=(s.sub.1, s.sub.2, . . . , s.sub.n).sup.T output from the n coils of the array to conversion outputs s′=(s.sub.1′, s.sub.2′, . . . , s.sub.n′).sup.T s′=Ms, wherein .sup.T represents matrix transpose operation; selecting rows of the channel conversion matrix M corresponding to a number of conversion outputs with the highest signal-to-noise ratios (SNRs); and optimizing the pre-combination coefficients or both the grouping and the pre-combination coefficients such that the pre-combination coefficients span a space which approaches the space spanned by all or the selected rows of M.

13. The controller of claim 12, wherein the channel conversion matrix is a whitening matrix.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Features and advantages of the embodiments of the invention will become apparent from the following detailed description in which:

(2) FIG. 1 illustrates a converter with splitters and combiners which converts receiver channels into virtual receiver channels.

(3) FIG. 2 illustrates a converter with pre-combiners, splitters and combiners which converts receiver channels into virtual receiver channels.

(4) FIG. 3 illustrates a degraded converter with pre-combiners which combines receiver channels into combined receiver channels.

(5) FIG. 4 illustrates a graph illustrating an exemplary pre-combination in accordance with aspects of the present invention.

DETAILED DESCRIPTION OF EMBODIMENTS

(6) Various aspects of the novel systems, apparatus and methods are described more fully hereinafter with reference to the figures. The teachings disclosure may, however, be embodied in many different forms and should not be construed as limited to any specific structure or function presented throughout this disclosure. Rather, these aspects are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art. Based on the teachings herein one skilled in the art should appreciate that the scope of the disclosure is intended to cover any aspect of the novel systems, apparatus, and methods disclosed herein, whether implemented independently of or combined with any other aspect of the invention. For example, an apparatus may be implemented or a method may be practiced using any number of the aspects set forth herein. In addition, the scope of the invention is intended to cover such an apparatus or method which is practiced using other structure, functionality, or structure and functionality in addition to or other than the various aspects of the invention set forth herein. It should be understood that any aspect disclosed herein may be embodied by one or more elements of a claim.

(7) The word “exemplary” is used herein to mean “serving as an example, instance, or illustration.” Any aspect described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other aspects.

(8) Although particular aspects are described herein, many variations and permutations of these aspects fall within the scope of the disclosure. Although some benefits and advantages of the preferred aspects are mentioned, the scope of the disclosure is not intended to be limited to particular benefits, uses, or objectives. Rather, aspects of the disclosure are intended to be broadly applicable to different technologies, system configurations, some of which are illustrated by way of example in the figures and in the following description of the preferred aspects. The detailed description and drawings are merely illustrative of the disclosure rather than limiting, the scope of the disclosure being defined by the appended claims and equivalents thereof.

(9) In a MRI system, many coils may be placed around a Region of Interest (ROI) to improve imaging quality. The image of the inside of the ROI may be obtained by combining a portion or all of the outputs of the coils for the optimum Signal to Noise Ratio (SNR). For example, an array of coils may be placed around the head of a human body, another array of coils may be placed around the arm of a human body, and a further array of coils may be placed around the leg of a human body. The arm, the head, and the leg are three ROIs, and the three arrays are placed for the reconstruction of the images of the three regions of the human body. The three arrays may provide parallel imaging acceleration with a factor of three. The outputs of the coils of each array may be input to an imaging computer and combined in the imaging computer to increase the SNR of the image. However, the number of inputs of a commercial imaging computer is limited to, for example, 4 channels. The number of coil elements in an array for a ROI may be large, for example, up to 64. It may have to be reduced or compressed to no more than 4 channels and coil compression may be necessary.

(10) Coil compression, or receiver channel reduction, MRI data compression may be used to compress data from many receiver channels for a ROI into fewer virtual receiver channels. FIG. 1 illustrates a converter with splitters and combiners which converts receiver channels into virtual receiver channels.

(11) As illustrated in FIG. 1, n coils c.sub.1, c.sub.2, . . . , c.sub.n are deployed for a ROI. Each of the n coils outputs a signal in the process of imaging and the n coils constitute n signal channels or receiver channels. The n signals may be amplified, digitized and input into a computer if, for example, the computer has n inputs. Imaging of the ROI may be completed in the computer. However, the growing number n may be large, such as up to 32, 64, 128, or even larger, while the number of the inputs of a commercial imaging computer may be limited to only 4 for cost reasons. Thus, signal combination, signal compression, channel compression, or coil compression may be inevitable, which selects the signals after combination with the highest qualities, such as highest SNRs.

(12) As a general model for channel compression, FIG. 1 illustrates n splitters followed by n combiners. Each of the splitters provides n copies of the signal output from one of the n coils. The outputs from the n splitters may then be combined with a combiner to produce one combination output. There may be totally n combiners and n combination outputs. There may be up to n linearly independent combination outputs for n coils.

(13) Each of the combiners may be characterized by a set of complex combination coefficients. The combination is actually the linear combination using the complex combination coefficients as weights. The amplitudes of the combination coefficients represent the amplitude weights, and the angles or the arguments of the combination coefficients represent the phase shifts added to the signal channels. Optimized linear combinations may produce combinations with highest signal qualities or highest signal SNRs. The splitters, the combiners, and the phase shifters and attenuators between the splitters and the combiners constitute a converter.

(14) The converter illustrated in FIG. 1 may be implemented with software, hardware components, or a combination of both of them. The converter implemented with hardware components provides fast compression. The converter implemented with software components provides flexibility. Hardware components and software components may be selected or adequately combined for feasibility, imaging quality, cost effectiveness, simplicity, etc.

(15) For hardware implementation, the output of each of the n coils illustrated in FIG. 1 and configured for the same ROI is pre-amplified with a Low Noise Amplifier (LNA) first, and the output of the LNA may be adjusted with a phase shifter and an attenuator and then connected to a hardware splitter. The hardware splitter splits the pre-amplified, phase shifted and amplitude attenuated signal into a number of copies. In the case that the splitters are hardware splitters, there may be splitting loss in addition to the simple copying. The split signals are then adjusted with a set of phase shifters and attenuators and combined using a hardware combiner into a virtual receiver channel. Different virtual receiver channels may be obtained with different sets of phase shifters and attenuators. The splitters, the combiners, and the phase shifters and attenuators between the splitters and the combiners constitute a converter. There may be up to n splitters, n×n phase shifters and attenuators and n combiners for linearly independent n combination outputs, in addition to the phase shifters and attenuators between the coils and the splitters.

(16) For software implementation, the combiners illustrated in FIG. 1 represent weighted additions or linear combinations. The splitters illustrated in FIG. 1 represent the simple copy operation. The phase shifters and attenuators represent complex combination coefficients. A/D conversions are necessary before digital processing.

(17) In general, the converter including n splitters, n×n phase shifters and attenuators, and n combiners may be expressed as a channel conversion matrix M of n×n which converts signals s=(s.sub.1, s.sub.2, . . . , s.sub.n).sup.T output from the n coils of the array to conversion outputs s′=(s.sub.1′, s.sub.2′, . . . , s.sub.n′).sup.T. The conversion may be expressed as the following equation:

(18) $\begin{array}{cc}{s}^{\prime }=\mathrm{Ms}& \left(1\right)\\ \mathrm{or}& \\ \left(\begin{array}{c}{s}_1^{\prime }\\ s_2^{\prime }\\ \mathrm{.Math.}\\ s_n^{\prime }\end{array}\right)=\left(\begin{array}{cccc}{m}_{11}& m_{12}& \mathrm{.Math.}& m_{1n}\\ m_{21}& m_{22}& \mathrm{.Math.}& m_{2n}\\ & & \ddots & \\ m_{n1}& m_{n2}& \mathrm{.Math.}& m_{nn}\end{array}\right)\left(\begin{array}{c}{s}_1\\ s_2\\ \mathrm{.Math.}\\ s_n\end{array}\right)& \left(2\right)\end{array}$

(19) where s.sub.i is the signal output from the i.sup.th coil c.sub.i; i=1, 2, . . . , n; n is the number of coils for the same ROI; and .sup.T represents transpose operation.

(20) Different converters or different conversion matrix M may result in different sets of conversion output s′. An optimum conversion matrix M may result in an optimum conversion output s′. For example, the conversion matrix M may be chosen or optimized to produce converted or conversion outputs s.sub.1′, s′.sub.2, . . . , s.sub.n′ with the highest signal qualities. More specifically, the conversion matrix M may be chosen or optimized to produce conversion outputs s.sub.1′, s′.sub.2, . . . , s.sub.n′ with the highest SNRs. With the conversion matrix M and the conversion outputs s.sub.1′, s′.sub.2, . . . , s.sub.n′ a portion of the n conversion outputs with the greatest qualities or greatest SNRs may be chosen for imaging in an imaging computer. For example, 4 of the n conversion outputs with the greatest SNRs may be chosen and input into a 4 channel imaging computer. For example, if the conversion outputs s.sub.1′, s′.sub.2, . . . , s.sub.n′ are ordered by quality from the highest to the lowest, s.sub.1′, s.sub.2′, s.sub.3′, s.sub.4′ with the highest qualities may be chosen. Coil compression, receiver channel reduction or MRI data compression, signal compression, channel compression may be achieved by choosing only a portion of the n conversion outputs, for example m of the n conversion outputs. In practice, only the chosen conversion outputs are necessary to be input into the imaging computer for further processing. In that case, the number of combinations is m, and the number of phase shifters and attenuators immediately before the combination may be reduced from n×n to m×n, where m<n.

(21) For example, a typical 16 coil element system providing 16 receiver channels or signal channels may be linearly combined using 16 sets of combination coefficients to produce 16 virtual channels. The 16 sets of combination coefficients constitute a conversion matrix M. If the conversion matrix is well chosen, the 16 virtual channels contain all information contained in the original 16 receiver channels, and the virtual channel with the highest SNR may be the virtual channel with theoretically obtainable highest SNR. The 4 virtual channels with the highest SNRs may be input into a 4 channel imaging computer for image reconstruction. In that case, the number of combinations is 4, and the number of phase shifters and attenuators immediately before the combination may be reduced from 16×16 to 4×16. However, the number of combinations and the number of phase shifters and attenuators are still too large for commercial implementation.

(22) In the case that the conversion is implemented with software, the computation for combinations is a heavy load to the computer when n is large, and n Analog-to-Digital (A/D) converters have to be used after the pre-amplifications of the n channels and before the n data channels are processed digitally. In the case that the conversion is implemented with hardware, the computation for combinations is eliminated. Only m A/D converters have to be used after the channel compression and before the digital processing or before the compressed m data channels are input into the imaging computer. However, at least m×n phase shifters and attenuators are required, which is too large a number and difficult to be commercially implemented when n is large.

(23) A lot of efforts have been made for the further compression of the receiver channels. For example, as mentioned above, a hardware coil compression method is based on the Butler matrix. However, it is applicable to only cylindrical coil arrays, making it inefficient for linear arrays or normal surface coil arrays. The TIM method is also developed, but it is applicable to only linear arrays such as spine and torso coil arrays, and its compression ratio is limited to no more than 3:1. Another method providing a 32:8 conversion matrix directly implements the software coil compression algorithm with hardware, and is not subject to the above mentioned 2 shortcomings. However, it requires a lot of compression hardware, making it inefficient for commercial implementation. A novel converter with pre-combination for channel compression may be used to solve the above problem which is illustrated in FIG. 2 and FIG. 3.

(24) FIG. 2 illustrates a novel converter with pre-combination prior to channel compression of the converter in FIG. 1. FIG. 3 illustrates a degraded one. As illustrated in FIG. 2, a set of n coils c.sub.1, c.sub.2, . . . , c.sub.n are configured for the same ROI. Though they are shown in FIG. 2 as a linear array, they may be deployed in any other way. The only restriction is that they are deployed for the same ROI. The n coils in the array are grouped into n′ sub-arrays. For each sub-array, the outputs of all the coils in the sub-array are pre-combined with a pre-combiner and one pre-combination output is produced. In that way, the n channels are reduced or compressed into n′ channels. The pre-combiners may be software combiners or hardware combiners. The converter constructed with hardware pre-combiners is simple and cost effective for commercial implementation.

(25) Mathematically, a pre-combination matrix V of n′×n may be constructed which converts signals s=(s.sub.1, s.sub.2, . . . , s.sub.n).sup.T output from the coils c.sub.1, c.sub.2, . . . , c.sub.n of the array to pre-combination outputs s″=(s.sub.1″, s.sub.2″, . . . , s.sub.n′″).sup.T:
s″=Vs  (3)

(26) In each column of the pre-combination matrix V, only one element is nonzero.

(27) $\begin{array}{cc}V=\left(\begin{array}{cccccccccc}{v}_{11}& \mathrm{.Math.}& v_{1d_1}& 0& \mathrm{.Math.}& 0& \mathrm{.Math.}& 0& \mathrm{.Math.}& 0\\ 0& \mathrm{.Math.}& 0& v_{21}& \mathrm{.Math.}& v_{2d_2}& \mathrm{.Math.}& 0& \mathrm{.Math.}& 0\\ \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}\\ 0& \mathrm{.Math.}& 0& 0& \mathrm{.Math.}& 0& \mathrm{.Math.}& v_{n^{\prime }1}& \mathrm{.Math.}& v_{n^{\prime }{d}_{n^{\prime }}}\end{array}\right)& \left(4\right)\end{array}$

(28) The pre-combination matrix V shown above is only an example which indicates that sub-arrays are composed respectively of d.sub.1, d.sub.1, . . . , d.sub.n′ coils, where d.sub.1+d.sub.1+ . . . +d.sub.n′=n, and the coils of each sub-array are coils adjacent in the array. In the case that the coils of a sub-array are not coils adjacent in the array, the pre-combination matrix V looks different from shown above, but may look similar to the pre-combination matrix V shown above after a linear transformation. The transformation is equivalent to renumbering the coils without changing the physical deployment.

(29) For the purpose of producing the best image of the ROI, the grouping and pre-combination may be based on importance of each of the coils. The importance of each of the coils may be represented by the contribution of the output of the coil to the Signal-to-Noise Ratio (SNR) of the signal obtained by combining the coils.

(30) The array of coils may be grouped into subgroups in many ways. For example, the coils may be grouped into sub-arrays by assigning coils with the highest importance into the sub-arrays, one coil for each sub-array, and repeating the assignment of the remaining coils until all coils are assigned.

(31) More specifically, without loosing generality, coils c.sub.1, c.sub.2, . . . , c.sub.n are assumed to be ordered from the most important to the least important. For example, coil c.sub.1 is the most important, and c.sub.n is the least important. In that case, c.sub.1 is assigned to sub-array 1, c.sub.2 is assigned to sub-array 2, . . . , c.sub.n′ is assigned to sub-array n′. The remaining coils c.sub.n′+1, c.sub.n′+2, . . . , c.sub.2n′ are assigned respectively to sub-arrays 1, 2, . . . , n′ in the same way. That is, c.sub.n′+1 is assigned to sub-array 1, c.sub.n′+2 is assigned to sub-array 2, . . . , c.sub.2n′ is assigned to sub-array n′. The remaining coils c.sub.2n′+1, c.sub.2n′+2, . . . , c.sub.n are assigned to sub-arrays 1, 2, . . . , n′ in the same way, again. The process is repeated, until all the coils are assigned to sub-arrays. There may be many other ways of grouping the coils in the array for the same ROI. For example, the most important k coils c.sub.1, c.sub.2, . . . , c.sub.k are assigned to sub-array 1, the most important k coils c.sub.k+1, c.sub.k+2, . . . , c.sub.k+k of the remaining n−k coils are assigned to sub-array 2, and so on. As an example, k=n\n′+1. “n\n′” represents the largest integer which is less than n/n′. The coils may also be grouped by assigning c.sub.1 and c.sub.n to sub-array 1, assigning c.sub.2 and c.sub.n−1 to sub-array 2, and so on.

(32) The sub-arrays and pre-combination coefficients for the sub-arrays may alternatively be determined by constructing a channel conversion matrix M of n×n which converts signals s=(s.sub.1, s.sub.2, . . . , s.sub.n).sup.T output from the coils of the array to conversion outputs s′=(s.sub.1′, s.sub.2′, . . . , s.sub.n′).sup.T, s′=Ms; selecting rows of the channel conversion matrix M corresponding to a number of conversion outputs with the highest SNRs; and optimizing the pre-combination coefficients or both the grouping and the pre-combination coefficients such that the pre-combination coefficients span a space which approaches the space spanned by the selected rows of M or all the rows of M. The space spanned by a matrix refers to a space obtained by linear combinations of row vectors or column vectors of the matrix. The conversion outputs s.sub.1′, s.sub.2′, . . . , s.sub.m′ with the greatest importance, for example, with the greatest SNRs, may be selected.

(33) Another channel conversion matrix M′ of n′×n′ may be used to convert the pre-combination outputs s″=(s.sub.1″, s.sub.2″, . . . , s.sub.n′″).sup.T to conversion outputs s′″=(s.sub.1″′; s.sub.2′″, . . . , s.sub.n′′″).sup.T as shown in FIG. 2:
s′″=M′s″=M′Vs  (5)

(34) The pre-combination matrix V may be such that the linear combination of the n′ conversion outputs: (a.sub.1 a.sub.2 . . . a.sub.n′)s″=(a.sub.1 a.sub.2 . . . a.sub.n′) Vs approaches the linear combination of the first n′ conversion outputs (b.sub.1 b.sub.2 . . . B.sub.n′ 0 . . . 0)Ms, or the space of 16-element row matrix (a.sub.1 a.sub.2 . . . a.sub.n′) V approaches the space of 16-element row matrix (b.sub.1 b.sub.2 . . . b.sub.n′ 0 . . . 0)M or the space of 16-element row matrix (b.sub.1 b.sub.2 . . . b.sub.n′ b.sub.n′+1 . . . b.sub.n)M. Where a.sub.i, i=1, . . . , n′ and b.sub.i, i=1, n are arbitrary complex numbers.

(35) The optimization may be expressed in more detail as follows:

(36) $\left(\begin{array}{cccc}{a}_1& a_2& \mathrm{.Math.}& a_{n^{\prime }}\end{array}\right)\left(\begin{array}{cccccccccc}{v}_{11}& \mathrm{.Math.}& v_{1d_1}& 0& \mathrm{.Math.}& 0& \mathrm{.Math.}& 0& \mathrm{.Math.}& 0\\ 0& \mathrm{.Math.}& 0& v_{21}& \mathrm{.Math.}& v_{2d_2}& \mathrm{.Math.}& 0& \mathrm{.Math.}& 0\\ \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}\\ 0& \mathrm{.Math.}& 0& 0& \mathrm{.Math.}& 0& \mathrm{.Math.}& v_{n^{\prime }1}& \mathrm{.Math.}& v_{n^{\prime }{d}_{n^{\prime }}}\end{array}\right)=\left(\begin{array}{cccccccccc}{a}_1{v}_{11}& \mathrm{.Math.}& a_1{v}_{\mathrm{dl}1}& a_2{v}_{21}& \mathrm{.Math.}& a_2{v}_{2d_2}& \mathrm{.Math.}& a_{n^{\prime }}{v}_{n^{\prime }1}& \mathrm{.Math.}& a_{n^{\prime }}{v}_{n^{\prime }{d}_{n^{\prime }}}\end{array}\right)$$\mathrm{approaches}\left(\begin{array}{ccccccc}{b}_1& b_2& \mathrm{.Math.}& b_{n^{\prime }}& 0& \mathrm{.Math.}& 0\end{array}\right)\left(\begin{array}{cccc}{m}_{11}& m_{12}& \mathrm{.Math.}& m_{1n}\\ m_{21}& m_{22}& \mathrm{.Math.}& m_{2n}\\ \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}\\ m_{n1}& m_{n2}& \mathrm{.Math.}& m_{nn}\end{array}\right)=\left(\underset{i=1,n^{\prime }}{.Math.}{b}_i{m}_{i1}\underset{i=1,n^{\prime }}{.Math.}{b}_i{m}_{i2}\mathrm{.Math.}\underset{i=1,n^{\prime }}{.Math.}{b}_i{m}_{in}\right),\mathrm{or}\left(\begin{array}{ccccccc}{b}_1& b_2& \mathrm{.Math.}& b_{n^{\prime }}& b_{n^{\prime }+1}& \mathrm{.Math.}& B_n\end{array}\right)\left(\begin{array}{cccc}{m}_{11}& m_{12}& \mathrm{.Math.}& m_{1n}\\ m_{21}& m_{22}& \mathrm{.Math.}& m_{2n}\\ \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}& \mathrm{.Math.}\\ m_{n1}& m_{n2}& \mathrm{.Math.}& m_{nn}\end{array}\right)=\left(\underset{i=1,n}{.Math.}{b}_i{m}_{i1}\underset{i=1,n}{.Math.}{b}_i{m}_{i2}\mathrm{.Math.}\underset{i=1,n}{.Math.}{b}_i{m}_{in}\right)$

(37) The channel conversion matrix M may be selected to be a theoretically optimal channel conversion matrix. The software compression method may be used to obtain the theoretically optimal channel conversion matrix M. For example, the theoretically optimal channel conversion matrix M is a whitening matrix. Then array elements in the coil array may be grouped into sub-arrays based on the optimal channel conversion matrix M.

(38) Taking a 16-element coil array as an example, a channel conversion matrix M of 16×16 may be constructed which converts signals s=(s.sub.1, s.sub.2, . . . , s.sub.16).sup.T output from the 16 coils c.sub.1, c.sub.2, . . . , c.sub.16 of the array to conversion outputs s′=(s.sub.1′, s.sub.2′, . . . s.sub.16′).sup.T. Its rows correspond to the new virtual channels. We have s′=Ms, and M=(M.sup.1.sub.1×16, M.sup.2.sub.1×16, . . . , M.sup.16.sub.1×16).sup.T, M.sup.i is the i.sup.th row of M, and the corresponding covariance matrix is C.sub.16×16 of the signals s′=(s.sub.1′, s.sub.2′, . . . s.sub.16′).sup.T with its diagonal components as C.sub.1, C.sub.2, . . . C.sub.16, representing respectively the SNRs for the virtual channels. The matrix M is arranged such that the diagonal components of the covariance matrix are in a descending order. In that way, C.sub.1 corresponds to the virtual channel which has the greatest SNR, while C.sub.16 corresponds to the virtual channel which has the least SNR.

(39) Using hardware to implement M directly, the four virtual channels with the greatest SNRs (say, M.sup.1 through M.sup.4, in the case of 16:4 compression) will be directly implemented with hardware, resulting in sixteen 1-to-4 power splitters, 4 16-to-1 power combiners, and 64 phase shifters and attenuators. The hardware converter is too complicated for commercial implementation.

(40) Using the method of pre-combination, the 16 elements of the array are grouped into 4 sub-arrays, each having 4 elements. There are 16 input channels and 4 output channels for the pre-combination. For the grouping, the first 4 rows of M (M.sup.1 through M.sup.4), for example, are used to determine how the grouping should be implemented. An analysis is performed to figure out the most and least important coil array elements, and then the 16 coil elements are rated and ordered by their importance. The 16 elements are grouped into 4 sub-arrays based on the order of importance.

(41) After the 16 coil elements are grouped into 4 sub-arrays, one 4-input-1-output RF hardware pre-combiner is used for each 4-element sub-array, so we have four pre-combination matrixes, say v.sup.1.sub.1×4, v.sup.2.sub.1×4, v.sup.3.sub.1×4, v.sup.4.sub.1×4, to be determined.

(42) $s_1^{\prime }={v^1\left(s_1,s_2,s_3,s_4\right)}^T=\left(v_1^1,v_2^1,v_3^1,v_4^1\right)\left(\begin{array}{c}{s}_1\\ s_2\\ s_3\\ s_4\end{array}\right),s_2^{\prime }={v^2\left(s_5,s_6,s_7,s_8\right)}^T=\left(v_1^2,v_2^2,v_3^2,v_4^2\right)\left(\begin{array}{c}{s}_5\\ s_6\\ s_7\\ s_8\end{array}\right),s_3^{\prime }={v^3\left(s_9,s_{10},s_{11},s_{12}\right)}^T=\left(v_1^3,v_2^3,v_3^3,v_4^3\right)\left(\begin{array}{c}{s}_9\\ s_{10}\\ s_{11}\\ s_{12}\end{array}\right),s_4^{\prime }={v^4\left(s_{13},s_{14},s_{15},s_{16}\right)}^T=\left(v_1^4,v_2^4,v_3^4,v_4^4\right)\left(\begin{array}{c}{s}_{13}\\ s_{14}\\ s_{15}\\ s_{16}\end{array}\right).$

(43) The four pre-combination matrixes v.sup.1, v.sup.2, v.sup.3 and v.sup.4 span a sub-space: W.sub.result=(a.sub.1v.sup.1, a.sub.2v.sup.2, a.sub.3v.sup.3, a.sub.4v.sup.4), in which a.sub.1, a.sub.2, a.sub.3 and a.sub.4 are arbitrary complex numbers. W.sub.result is a space of 16-element row matrix.

(44) The target of choosing sub-array groups, as well as choosing the RF pre-combination matrixes v.sup.1, v.sup.2, v.sup.3 and v.sup.4, is to construct another space W.sub.target=b.sub.1M.sup.1+b.sub.2M.sup.2+b.sub.3M.sup.3+b.sub.4M.sup.4, or W.sub.target=b.sub.1M.sup.1+b.sub.2M.sup.2+b.sub.3M.sup.3+b.sub.4M.sup.4+b.sub.5M.sup.5+ . . . +b.sub.16M.sup.16, and W.sub.result should be as close to the W.sub.target as possible, in which b.sub.1, b.sub.2, b.sub.3, . . . , b.sub.16 are also arbitrary complex numbers, and M.sup.i is the i.sup.th row of M as indicated above.

(45) The pre-combination may be shown as illustrated in FIG. 4.

(46) One simple example for the proposed method may be as follows. For the 16-element coil array, the optimal channel conversion matrix M may be constructed. The first row M.sup.1=(M.sup.1.sub.1, M.sup.1.sub.2, . . . , M.sup.1.sub.16)=m.sub.1,2, m.sub.1,2, . . . , m.sub.1,16) may be picked out, which represents the most important virtual channel.

(47) The importance of the coils may be determined and the order is adjusted. It is assumed that the 16 coils c.sub.1, c.sub.2, c.sub.16 after the adjustment are ordered by importance. That is, it is assumed that c.sub.1 is the most important coil, and c.sub.16 is the least important coil, and M represents the optimal conversion matrix for the ordered coils.

(48) The 16 ordered coils are then grouped into 4 sub-arrays as (c.sub.1, c.sub.5, c.sub.9, c.sub.13), (c.sub.2, c.sub.6, c.sub.10, c.sub.14), (c.sub.3, c.sub.7, c.sub.11, c.sub.15), and (c.sub.4, c.sub.8, c.sub.12, c.sub.16) in the following way: c.sub.1, c.sub.2, c.sub.3, c.sub.4 are first assigned to sub-arrays 1, 2, 3 and 4, respectively; c.sub.5, c.sub.6, c.sub.7, c.sub.8 are then the most important coils of the remaining coils and assigned to sub-arrays 1, 2, 3 and 4, respectively; and the process is repeated until all the coils are assigned into the sub-arrays.

(49) The RF pre-combination matrixes v.sup.1, v.sup.2, v.sup.3 and v.sup.4 may be determined for the sub space W.sub.result to be as close to the W.sub.target as possible.

(50) In brief, for the pre-combination method, a conversion matrix M is constructed first for the coils in the array configured for the same ROI. The conversion matrix M is optimized for the conversion outputs to have highest qualities. The coils are ordered and grouped into sub-arrays based on their importance. The pre-combination coefficients are determined for the sub-space spanned by the pre-combination coefficients to be as close as possible to the space spanned by all rows or the most important rows of the optimized conversion matrix M for the ordered coils. The converter constructed with hardware pre-combiners is simple and cost effective for commercial implementation and is of high performance.