Coherence filtering of magnetic resonance imaging (MRI) signals

Abstract

The device 18 for filtering magnetic resonance (MR) signals includes at least one electronic processor 24. The at least one electronic processor 24 is programmed to receive at least one MR signal 26 excited in a subject placed in the MR imaging device 10, to transform the at least one MR signal with a coherence function representative of the at least one MR signal as a function of time and instantaneous frequency spectrum, to apply spectral filtering to the transformed at least one MR signal, and to reconstruct at least one medical image from the filtered at least one MR signal.

Description

[Technical field]

[0001] The following relates generally to magnetic resonance (MR) imaging, MR image reconstruction, MR signal processing, MR signal acquisition, MR signal filtering, and related techniques. [Background technology]

[0002] Magnetic resonance (MR) imaging involves placing a subject (such as a patient, a veterinary subject, an archaeological mummy, etc.) in a static magnetic field (often referred to as the B0 field), exciting nuclear magnetic resonance in the subject, and detecting the excited magnetic resonance. For imaging, the excited MR is spatially encoded in terms of position, phase, and / or frequency by superimposing a magnetic field gradient on the static B0 field during excitation, during the time interval between MR excitation and MR readout, and / or during MR readout. In a typical design, an MR imaging device (sometimes referred to as an MRI scanner) includes a housing with a central bore in which the MR examination region is located. The static B0 field is generated by solenoidal magnet windings wound around the central bore and contained within the MRI scanner housing. These solenoidal magnet windings are often superconducting windings in modern MRI scanners, and the housing includes a liquid helium (LHe) reservoir that cools the superconducting windings. Magnetic field gradient coils are also located within the housing around the central bore.

[0003]

[0003] For MR excitation in human subjects, a body coil is typically used. The body coil is usually a cylindrical birdcage coil, a TEM coil, or some variation thereof, placed concentrically around a bore. Alternatively, a local coil placed near the body part to be imaged is used for excitation. MR readout is typically performed using a local MR receive coil placed near the body part to be imaged. The local MR receive coil and the local MR excitation coil (if used) may be the same coil or different coils. For various reasons, MR receive coils include MR coil arrays containing multiple coil elements, each of which is usually configured as a loop coil, although other coil element designs are known. Such MR coil arrays may implement several MR receive channels. For example, a single coil element or a group of spatially contiguous coil elements may define an MR channel. The use of such multi-channel MR receive coils has certain advantages. For example, the use of sensitivity encoding (SENSE) or other parallel MR acquisition techniques utilizing multiple receive channels may speed up MR data acquisition. Summary of the Invention

[0004]

[0004] Specific improvements directed to solving these and other problems are disclosed below.

[0005] In some embodiments disclosed herein, a device for filtering magnetic resonance (MR) signals includes at least one electronic processor that is programmed to receive at least one MR signal excited in a subject placed in an MR imaging device, transform the at least one MR signal with a coherence function representing the at least one MR signal as a function of time and instantaneous frequency spectrum, apply spectral filtering to the transformed at least one MR signal, and reconstruct at least one medical image from the filtered at least one MR signal.

[0006]

[0006] In some embodiments disclosed in this specification, the MR receive coil includes at least one MR coil element that receives MR signals excited in a subject placed in the MR imaging device, and electronics that receive MR signals excited in a subject placed in the MR imaging device, transform the MR signals with a coherence function representing the MR signals as a function of time and instantaneous frequency spectrum, apply spectral filtering to the transformed MR signals, and reconstruct at least one medical image from the filtered MR signals.

[0007]

[0007] In some embodiments disclosed herein, the MR receive coil includes at least one MR coil element that receives MR signals excited in a subject placed in the MR imaging device, and electronics that receive a plurality of MR signals excited in a subject placed in the MR imaging device, transform the plurality of MR signals with a coherence function representing each of the plurality of MR signals as a function of time and instantaneous frequency spectrum, apply spectral filtering to the transformed MR signals, and reconstruct at least one medical image from the filtered MR signals.

[0008]

[0008] One advantage is in suppressing incoherent noise from the MRI signal.

[0009] Another advantage resides in improving the signal-to-noise ratio (SNR) of the MR signal.

[0010] Another advantage resides in using the coherence function to suppress short-term noise from the MR signal by allowing dynamically adapting noise filtering as a function of time.

[0011] Another advantage resides in taking advantage of the expected partial coherence between channels of a multi-channel coil to improve MR signal noise filtering.

[0012]

[0012] A given embodiment may provide none of the above advantages, one, more than one, or all of the above advantages, and / or other advantages that will become apparent to a person of ordinary skill in the art upon reading and understanding this disclosure. [Brief description of the drawings]

[0013]

[0013] The present disclosure may take form in various components and arrangements of components, and in various steps and arrangements of steps. The drawings are only for purposes of illustrating preferred embodiments and are not to be construed as limiting the disclosure.

[0014] [Figure 1] FIG. 1 illustrates diagrammatically a magnetic resonance (MR) imaging device including an MR coil element according to the present disclosure. [Diagram 2]

[0015] FIG. 2 shows a module of the device of FIG. [Diagram 3] FIG. 3 shows a module of the device of FIG. [Figure 4]

[0016] FIG. 4 illustrates diagrammatically the suppression of short duration noise events using the disclosed approach. [Diagram 5]

[0017] FIG. 5 shows diagrammatically an MRI imaging method using the device of FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015]

[0018] Magnetic resonance imaging (MRI) signal processing is typically performed by transforming the signal into the frequency domain, for example using a fast Fourier transform (FFT), and applying spectral filtering to the FFT-transformed signal. However, in this approach, the same spectral filtering is applied to the entire time domain. In cases where noise signals appear temporally, such as intermittent noise or other short-time noise, this time-global filter is not adequately adjusted to suppress noise. In embodiments disclosed herein, the MR signal is transformed with a coherence function that represents at least one MR signal as a function of time and instantaneous frequency spectrum, and the spectral filtering is applied to the transformed MR signal. In this way, the spectral filtering can be dynamically adjusted as a function of time, and noise can be more effectively suppressed. When only one MR receive channel is available (or when multiple MR receive channels are filtered individually), the coherence function used is the self-coherence function. On the other hand, in the case of a multi-channel coil, taking into account the expected partial coherence between the channels, the coherence function used is the cross-coherence function applied to pairs of channels.

[0016]

[0019] In current MR designs, MR signals are generated from MR excitations in the subject. Since the Fourier transform (FFT) of the signal does not provide information about the time at which the MR signal was generated, a two-dimensional (2D) representation (e.g., time and frequency) is needed to define the coherence function. Using such a time-frequency representation, the instantaneous frequency of the MR signal is identified (see, for example, Francois Auger, Patrick Flandrin, Paulo Concalves, Olivier Lemoine, "Time-Frequency Toolbox-For Use with MATLAB®", CNRS, France; Rice University, USA, 1995-1996). Other examples of time-frequency representations are the Wigner-Ville distribution (WVD) and / or smoothed pseudo-WVD (SPWVD) of the MRI signal.

[0017]

[0020] In some embodiments disclosed herein, the signals are transformed using a coherence function. The coherence function represents the data as a function of time and instantaneous frequency spectrum. When processing a single MRI signal, the coherence function is suitably an auto-coherence function such as a Cohen class time-frequency transform such as the Wigner-Ville distribution (WVD) or a smoothed pseudo-WVD (i.e., SPWVD). When multiple significantly correlated MRI signal channels are available, as in the case of a multi-channel MRI receive coil, a cross-coherence function such as the mean squared coherence can be used on a channel-pair basis.

[0018]

[0021] To filter MRI signals, an exemplary approach is to calculate the coherence (e.g., self-coherence or cross-coherence) on a scale of 0 to 1 (0 being completely incoherent and 1 being perfect coherence) for a time window around each data point in time, and then multiply the FFT of the signal in that time window by the coherence value. Alternatively, the coherence value can be thresholded before multiplication, allowing the signal to be completely suppressed in time intervals where the coherence is below the threshold.

[0019]

[0022] With reference to FIG. 1, an exemplary magnetic resonance (MR) imaging system or device 10 for imaging a subject S (e.g., an exemplary human subject S, such as a patient, a veterinary subject, or an archaeological mummy, etc.) includes a magnetic resonance (MR) imaging scanner (also referred to herein as an MRI scanner). The MRI scanner includes a housing or gantry 2 that includes, in an exemplary embodiment, various components shown in FIG. 1, including, by way of non-limiting example, a superconducting or resistive magnet 4 for generating a static (B0) magnetic field, magnetic field gradient coils 6 for superimposing magnetic field gradients on the B0 field, a whole-body RF coil 8 for applying radio frequency (RF) pulses to excite and / or spatially encode magnetic resonances in a patient to be imaged disposed in an MR bore 12 or other MR examination region, etc. The magnet 4 and gradient coils 6 are arranged concentrically around the bore 12. Using a robotic patient couch 14 or other patient support, a patient, a patient undergoing medical screening, or other patient to be imaged can be loaded into the MR bore 12 for imaging.

[0020]

[0023] Magnetic resonances excited in a subject S to be imaged are read out by an MR receive coil 18, which in the illustrated embodiment includes a number of MR coil elements 22 (in the extreme case the number of coil elements is one, i.e. the coil may include only a single coil element). Each coil element 22 is a radio frequency antenna for receiving MR signals excited in a subject disposed in the MR imaging device 10. Each coil element 22 typically forms an MR receive channel.

[0021]

[0024] FIG. 1 illustrates an exemplary MR coil 18 (or coil array 18) having a plurality of exemplary coil elements 22. It will be understood that the coil 18 may generally include any number of coil elements 22, e.g., 16 coil elements, 20 coil elements, 32 coil elements, etc. Each coil element 22 is typically part of an MR receive channel. The MR receive channel includes the MR coil elements 22 for receiving MR signals in the MR frequency range, as well as a preamplifier and often other signal processing electronics. Each exemplary coil element 22 is a single loop of copper, copper alloy, or another conductive material, for example, formed as a copper layer deposited on a circuit board, plastic sheet, plastic former, or other electrically insulating substrate, or alternatively formed as a separate metal loop. More generally, however, the coil elements 22 may be any suitable antenna (such as a multi-loop coil or other shaped antenna) capable of coupling with MR signals in the MR frequency range. In some embodiments, the MR coil 18 including the MR coil elements 22 is disposed in the examination region (i.e., the MR bore 12) as shown in FIG. 1.

[0022]

[0025] 2 and 3, and with continued reference to FIG. 1, electronics 24 (such as an electronic processor, more specifically a microprocessor) receives MR signals from the coil elements 22. Typically, at least one MR signal 26 excited in a subject S in the MR imaging device 10 is received by the electronics 24 from the MR 18 coil. In an exemplary embodiment using a multi-channel coil 18, the MR signal 26 typically includes an MR signal for each channel. The one or more MR signals 26 are then processed by an analog-to-digital converter (ADC, or ADC bank in the case of multi-channel signals) 28 to convert the one or more MR signals 26 into corresponding digital signals.

[0023]

[0026] The electronics 24 is programmed to transform the at least one MR signal 26 using a coherence function module 32 (see Figs. 2 and 3) that represents the at least one MR signal as a function of time and instantaneous frequency spectrum. In the embodiment shown in Fig. 2, the at least one MR signal comprises a single MR signal 26. In these embodiments, the coherence function 32 comprises a self-coherence function, such as a Cohen class time-frequency transform. Examples include the Wigner-Ville distribution (WVD) and / or the smoothed pseudo-WVD (SPWVD). In such an embodiment, when the at least one MR signal comprises a single MR signal 26, the coherence function comprises a SPWVD that includes a time window in the time domain and an attenuation filter window in the frequency domain.

[0024]

[0027] In the embodiment shown in Fig. 3, the at least one MR signal 26 includes a plurality of MR signals 26 (Fig. 2 shows this with a second MR signal 26' and a second ADC 28, but any number of ADCs 28 corresponding to the number of MR signals 26 may be included), and the coherence function 32 includes a cross-coherence function. For example, each MR signal of the plurality of MR signals is received from a corresponding channel of the illustrated multi-channel MR coil 18, and the cross-coherence function includes a mean squared coherence used on a channel pair basis. In such an embodiment, when the at least one MR signal 26 includes a plurality of MR signals, the coherence function includes a mean squared coherence function, and each MR signal 26 is transformed by setting the MR signal 26 to zero when the mean squared coherence function outputs a coherence value below a zero threshold.

[0025]

[0028] To perform the transformation, the digital MR signal 26 is processed by an FFT module 30 (or, if there are multiple MR signals 26, at least a second FFT module 30′ is included) to transform the digital MR signal 26 into the frequency domain. In some embodiments, the transformed MR signal 26 can be input to a signal-to-noise ratio (SNR) module 34, which suppresses noise in the transformed MR signal 26.

[0026]

[0029] The electronics 24 then applies spectral filtering to the transformed at least one MR signal 26. To do this, the coherence function module 32 is programmed to calculate a coherence function at a predefined scale for a selected time window around each data point in time of the at least one MR signal 26 to generate a coherence value. The FFT of the MR signal 26 is multiplied by the coherence value in the selected time window. In some embodiments, prior to the multiplication, the suppression module 36 thresholds the coherence value to completely suppress the signal in time intervals where the coherence is below a predefined threshold, while allowing the signal to pass unattenuated in time intervals where the coherence is above the predefined threshold.

[0027]

[0030] The filtered MR signals 26 are then further processed by an inverse FFT (IFFT) module 38 (FIG. 2), or by a first IFFT module 38 and a second IFFT module 38' (FIG. 3), to further suppress noise in the filtered MR signals 26. An image reconstruction module 40 is then programmed to process the filtered MR signals 26 into one or more reconstructed medical images of the subject S.

[0028]

[0031] With reference to Fig. 4, a simulation calculation shows diagrammatically how the disclosed approach using the coherence function module 32 suppresses short duration signal events. In Fig. 4, a signal 50 is plotted as a function of time. In this signal 50, a second signal event 51 is observed. On the left side of Fig. 4, the frequency domain of the main signal 50 is shown as a corresponding energy spectral density (ESD) 52. The signal component 50 is represented in the energy spectral density 52 by a main peak 50. ESD and the second signal event 51 is seen as a main signal peak 50 in terms of energy spectral density 53. ESD Low amplitude high frequency peaks well separated from ESDHowever, the energy spectral density 52 is not resolved as a function of time. This is also true for other typical frequency domain representations of the MR signal 50, such as FFT (not shown). Because the frequency domain representation 52 is not resolved as a function of time, a noise filter acting on the frequency domain representation 52 must be applied globally over time to remove noise. This can result in suboptimal noise removal, since filtering is not applied in some time areas of the short-duration signal events 51.

[0029]

[0032] With continued reference to FIG. 4, the coherence function transform of the MR signal 50 is shown as a smoothed pseudo-Wigner-Ville distribution (SPWVD) 54. The coherence transform 56 represents at least one signal as a function of time (plotted on the abscissa or horizontal x-axis) and instantaneous frequency spectrum (plotted on the ordinate or vertical y-axis). The signal components of the signal 50 are represented in the SPWVD 54 as structures 50 that extend throughout the plotted time interval. SPWVD This is because, as can be seen in the plot above signal 50, the signal extends over the entire plotted time interval. Meanwhile, the second signal event 51 is seen in SPWVD 54 as being part of the main signal structure 50. SPWVD and within the time interval t signal2 High frequency features that are temporally restricted to SPWVD Since the second signal event 51 is resolved as a function of time in the SPWVD 54, a noise filter acting on the SPWVD 54 to remove noise may be applied in a time-limited manner, e.g., over a time interval t signal2 or signal2 This allows for dynamically adapting noise filtering as a function of time, avoiding undesirable signal loss and suboptimal noise rejection that can occur in the presence of short duration signal events when filtering using FFT, ESD, or other frequency domain representations that are not resolved as a function of time.

[0030]

[0033] With reference to Fig. 5 and again to Figs. 1 and 2, an exemplary imaging method 100 using the MR device 10 is shown diagrammatically as a flow chart. To start the method 100, a patient is loaded onto the couch 1 and into the bore 12. An MR coil 18 is placed on the patient. In step 102, MR signals 26 are received by the electronics 24. In step 104, the MR signals 26 are transformed with a coherence function that represents the MR signals 26 as a function of time and instantaneous frequency spectrum. In step 106, the transformed MR signals 26 are spectrally filtered. In step 108, the filtered MR signals 26 are reconstructed into at least one medical image.

[0031] Working Example

[0034] Continuing with reference to FIG. 2, based on M observations of MRI signals in a single channel, each MRI signal has a signal plus noise ratio of y i (m)=x i (m)+n i (m), and i (m) is the signal to be estimated. Index i is the signal component of channel i and m is the sample value. It is assumed that the signal is highly correlated with itself and accompanied by uncorrelated noise sources (signals that carry no information). The value of the coherence function is the primary metric used to determine if a signal is present. Furthermore, no learning of noise statistics is required, only the received current observation of the signal is needed. If the amplitude of the coherence function between noisy signals (same channel or different channels) is equal to or close to 1, then the signal dominates and should pass without distortion.

[0032]

[0035] The MRI signals are spatially separated and therefore each signal is expressed according to Equation 1: y i [m]=x i [m]+n i [m], [i=1, 2, 3,...] (1) This allows n i (m) represents noise, and xi (m) represents the signal of interest without noise, and y i (m) represents the signal with noise, and there are i channels and m samples. The FFT can be performed on blocks of M samples, or it can be performed on M subsets (Q < M) using the overlap-add algorithm of sliding window FFT filtering. Generally, the FFT resolution can be changed from Q to M using standard zero-padding techniques. After obtaining the FFT of the signal, the signal is represented according to Equation 2: Y i [f,k]=X i [f,k]+N i [f,k], [i = 1, 2, 3, ···] (2) Here, f is the frequency bin and k is the frame index. The signal and the noise are not correlated. In the example of a two-channel coil as shown in Figure 3, the cross-power spectral density of Y i [f,k] is represented according to Equation 3:

Equation

Equation

[0033]

[0036] Then, the mean squared coherence (MSC) is

Equation

[0034]

[0037] The above-mentioned two-channel coherence process can be applied to all other channel pairs, and the determination of the frequency bins to be set to zero is performed based on the OR algorithm of the suppression function output.

[0035]

[0038] To cancel the incoherent noise of the signal, several suppression functions can be designed (see, for example, Nima Yousefian, Kostas Kokkinakis, Philipos C. Loizou, "A Coherence-Based Algorithm for Noise Reduction in Dual-Microphone Applications" (the 18th European Signal Processing Conference (EUSIPCO-2010), August 2010). Here, the suppression function is represented according to Equation 5:

Equation

[0036]

[0039] Here,

Equation

[0037]

[0040] Next, the final signal value (in the frequency space) is calculated as Y1[f,k]*G[f,k] (in the case of the signal Y1 which is the FFT of y1[m]), and similarly as Y2[f,k]*G[f,k] (in the case of the signal Y2 which is the FFT of y2[m]).

[0038]

[0041] Furthermore, in single-channel coherence-based noise reduction, the SPWVD algorithm is used. According to Equation 6, instead of using the coherence function, the auto-coherence can be directly calculated from the sampled signal:

Equation

number

[0039]

[0042] The final signal value (in frequency space) is then calculated as Y[f,k]*G[f,k] (for signal Y which is the FFT of y[m]).

[0040]

[0043] Note that the signal notation here is continuous rather than discrete.

[0041]

[0044] For an N-channel array of coil elements, the method is repeated for each pair of coupled coils. Knowledge of the coupling (noise resistance coupling and mutual inductance signal coupling) can be incorporated into the method.

[0042]

[0045] This description is for 1D SPWVD using a sliding window FFT in the frequency dimension since this is the fast data acquisition direction typically associated with fast A / Ds, however, the technique can also be applied in the k-space line dimension (k) after all k-space lines have been collected, and therefore 2D noise reduction algorithms exist.

[0043]

[0046] The present disclosure has been described with reference to the preferred embodiments. Modifications and alterations will occur to those upon reading and understanding the preceding detailed description. It is intended that the exemplary embodiments be construed as including all such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.

Claims

1. A device for filtering magnetic resonance (MR) signals, comprising at least one electronic processor, wherein the at least one electronic processor: Receives at least one MR signal excited within a subject disposed in an MR imaging device; Converts the at least one MR signal with a coherence function representing the at least one MR signal as a function of time and instantaneous frequency spectrum; Applies spectral filtering to the converted at least one MR signal; Is programmed to reconstruct at least one medical image from the filtered at least one MR signal; The conversion includes converting the at least one MR signal into a frequency domain using a fast Fourier transform (FFT), and the filtering: Calculates the coherence function at a predetermined scale for a selected time window around each data point in time of the at least one MR signal to generate a coherence value; Multiplies the coherence value by the FFT of the at least one MR signal in the selected time window; And includes The at least one MR signal includes a plurality of MR signals, the coherence function includes a cross-coherence function, Each MR signal of the plurality of MR signals is received from a corresponding channel in an MR coil, and the cross-coherence function includes an average squared coherence used on a channel pair basis; A device.

2. The device according to claim 1, wherein the at least one MR signal includes one MR signal, and the coherence function includes an auto-coherence function.

3. The device according to claim 2, wherein the auto-coherence function includes a Cohen's class time-frequency transform.

4. The device of claim 3, wherein the Cohen's class time-frequency transform includes a Wigner-Ville distribution (WVD).

5. The device of claim 3, wherein the Cohen's class time-frequency transform includes a smoothed pseudo-Wigner-Ville distribution (SPWVD).

6. The filtering further Includes thresholding the coherence value before the multiplication to completely suppress the signal at time intervals where the coherence is below a predetermined threshold, the device according to claim 1.

7. The at least one MR signal includes a plurality of MR signals, and the coherence function includes a mean squared coherence function, the device according to claim 1.

8. The at least one MR signal includes a plurality of MR signals, and the coherence function includes, where includes the cross-power spectral density of the first MR signal among the plurality of MR signals based on the frequency bin and the frame index of the first MR signal, includes the cross-power spectral density of the second MR signal among the plurality of MR signals based on the frequency bin and the frame index of the second MR signal, includes the cross-power spectral density of the multiplication of the first MR signal and the second MR signal, the device according to any one of claims 1 to 6.

9. The at least one MR signal includes a single MR signal, and the coherence function includes a smoothed pseudo-Wigner-Ville distribution including a time window in the time domain and an attenuation filter window in the frequency domain, the device according to claim 1.

10. The at least one MR signal includes a single MR signal, and the coherence function includes, where SPW includes the coherence function, t represents time, v represents frequency, h(t) represents the time window in the time domain, and g(t) represents the attenuation window filter in the frequency domain, the device according to claim 1.

11. An MR receiving coil including at least one MR coil element that receives an MR signal excited in a subject placed in a magnetic resonance (MR) imaging device and an electronic device, wherein the electronic device receives an MR signal excited in a subject placed in the MR imaging device, converts the MR signal with a coherence function representing the MR signal as a function of time and instantaneous frequency spectrum, applies spectral filtering to the converted MR signal, reconstructs at least one medical image from the filtered MR signal, the conversion includes converting the at least one MR signal into the frequency domain using a fast Fourier transform (FFT), and the filtering is For each selected time window around each data point in time of the at least one MR signal, calculating the coherence function at a predetermined scale to generate a coherence value; Multiplying the coherence value by the FFT of the at least one MR signal in the selected time window; comprising; the at least one MR signal includes a plurality of MR signals, and the coherence function comprises; wherein, includes the cross-power spectral density of the first MR signal among the plurality of MR signals based on the frequency bin and the frame index of the first MR signal; includes the cross-power spectral density of the second MR signal among the plurality of MR signals based on the frequency bin and the frame index of the second MR signal; includes the cross-power spectral density of the multiplication of the first MR signal and the second MR signal; MR receiving coil. **Claim 12**: The MR receiving coil according to claim 11, wherein the coherence function includes a Cohen class time-frequency transform. **Claim 13**: The MR receiving coil according to claim 11, wherein the filtering further includes thresholding the coherence value before the multiplication to completely suppress the signal in a time interval where the coherence is below a predetermined threshold. **Claim 14**: The at least one MR signal includes a single MR signal, and the coherence function comprises; wherein SPW includes the coherence function, t represents time, v represents frequency, h(t) represents a time window in the time domain, and g(t) represents an attenuation window filter in the frequency domain, the MR receiving coil according to claim 11.

Images