WO2021259494A1

WO2021259494A1 - Method and device for acquiring and reconstructing a sequence of diffusion-weighted magnetic resonance images covering a volume

Info

Publication number: WO2021259494A1
Application number: PCT/EP2020/068061
Authority: WO
Inventors: Jens Frahm; Dirk Voit; Oleksandr KALENTEV
Original assignee: Max-Planck-Gesellschaft zur Förderung der Wissenschaften e. V.
Priority date: 2020-06-26
Filing date: 2020-06-26
Publication date: 2021-12-30

Abstract

A method for creating sequences of diffusion-weighted magnetic resonance (MR) images of an object under investigation, wherein each of said sequences of MR images represents the same series of cross-sectional slices covering a volume of the object, comprises the steps of (a) providing multiple sequences of sets of image raw data being collected with the use of at least one radiofrequency receiver coil of a magnetic resonance imaging (MRI) device, wherein each set of image raw data includes a plurality of data samples being generated with a combined diffusion-weighted spin-echo and single-shot stimulated-echo sequence with diffusion-encoding gradients having at least two different strengths and at least three different directions, wherein one of the sequences of sets of raw data is generated with diffusion-encoding gradients of zero strength or lower strength and further sequences of sets of raw data are generated with diffusion-encoding gradients of higher strength, (b) subjecting the sequence of sets of image raw data generated with diffusion-encoding gradients of zero strength or lower strength to a regularized nonlinear inverse reconstruction process to provide a sequence of MR images with no or lower diffusion weighting, and (c) subjecting the sequences of sets of image raw data with diffusion-encoding gradients of higher strengths to a linear inverse reconstruction process to provide sequences of MR images with higher diffusion weighting, each of the images representing one of the cross-sectional slices and being created by using the sensitivity of the at least one receiver coil determined in step (b) for the same cross-sectional slice. Furthermore, an MRI device for creating a sequence of diffusion-weighted MR images of an object under investigation is described.

Description

Method and device for acquiring and reconstructing a sequence of diffusion-weighted magnetic resonance images covering a volume

Field of the invention

The present invention relates to a method for creating, in particular for acquiring and reconstruct ing, a sequence of magnetic resonance (MR) images with diffusion contrast, wherein the MR im ages cover a volume of directly neighbouring cross-sections and provide insensitivity to magnetic field inhomogeneity and improved image quality. Furthermore, the invention relates to a magnet ic resonance imaging (MRI) device configured for implementing the method. Applications of the invention cover the field of MR imaging, in particular medical MR imaging (e.g., brain imaging or prostate imaging) or non-medical investigations in natural sciences (e.g., investigations of a work- piece).

Background of the invention

In the present specification, reference is made to the following prior art illustrating the technical background of the invention, in particular relating to acquisition and reconstruction of MR imag es:

[1] V. Baliyan et al. in "World J Radiol" 8:785-798, 2016;

[2] C. S. Schouten et al. in "Quant Imaging Med Surg" 4:239-250, 2014;;

[3] K. Hirata et al. in "Medicine" 97:19(e0447), 2018;

[4] US 4748409 A (J. Frahm et al.);

[5] J. Frahm et al. in "J Magn Reson" 65:130-135, 1985;

[6] K. D. Merboldt et al. in "Magn Reson Med" 23:179-192, 1992;

[7] U. G. Nolte et al. in "Magn Reson Med" 44:731-736, 2000;

[8] A. A. Khalil et al. in "PLoS ONE" ll:e0161416, 2016;

[9] A. B. Bakushinsky and M. Y. Kokurin "Iterative Methods for Approximate Solution of Inverse Problems" Springer, Dordrecht, The Netherlands, 2005;

[10] A. Merrem et al. in "Invest Radiol" 52:428-433, 2017;

[11] A. Merrem, Dissertation "Undersampled Radial STEAM MRI Methodological Developments and Applications", Georg-August-University of Goettingen, 2018; [12] A. Merrem et al. in "NMR Biomed" 32, doi: 10.1002/nbm.4074, 2019;

[13] US 8384383 B2 (J. Frahm et al.);

[14] M. Uecker et al. in "NMR Biomed" 23:986-994, 2010; and

[15] J. Frahm et al. in "The Open Med Imaging" J 8:1-7, 2014.

Diffusion-weighted (DW) magnetic resonance imaging (MRI) is known as an MRI method provid ing information about dynamic molecular processes such as self-diffusion and/or transport of water molecules by altering the signal strength of MR images. The technique is of high clinical relevance because diffusion characteristics differ between normal and pathologic tissue and thus result in diagnostically relevant contrast, e.g. between normal and ischemic brain tissue (i.e., stroke) or between normal and tumorous tissue (i.e. cancer).

Diffusion encoding of the MRI signal is commonly accomplished by a DW spin-echo sequence which comprises two radiofrequency pulses and a self-compensating set of strong magnetic field gradients. A spin-echo sequence typically comprises an initial excitation radiofrequency pulse, a first interval with a duration of half the spin-echo time TE/2, a refocusing radiofrequency pulse, and another interval TE/2 before formation of the spin echo at time TE (relative to the center of the initial excitation pulse). For diffusion encoding, the spin-echo sequence is typically comple mented by a pair of identical magnetic field gradients with one gradient applied in each of the two spin-echo intervals. These gradients may be of a unipolar or bipolar nature, but also more complex gradient waveforms are possible. In the absence of motion, i.e. without any spatial dis placement of the water molecules, the nuclear spin moments of water protons completely refo cus in the spin-echo signal and therefore provide a maximum spin-echo signal strength which is only attenuated by T2 relaxation. In the presence of molecular diffusion or transport, however, the spin-echo signal is further reduced because a complete refocusing of spin moments is pre cluded when excited water protons move to positions with different magnetic field strengths, or more precisely, when water protons move between the application of the encoding diffusion gra dient in the first spin-echo interval and the application of the identical refocusing diffusion gradi ent in the second spin-echo interval. The relevant "diffusion time" is the duration between (the center of) these each other compensating gradients in the first and second interval of the spin- echo sequence.

Because the positional displacement due to a molecular transport process is on a microscopic scale and thus very small, DW MRI usually employs very strong diffusion-encoding gradients in order to ensure sufficient contrast (i.e., signal intensity difference) between a DW image and a non-DW image (i.e., an image acquisition without diffusion-encoding gradients but otherwise identical conditions). Such images with different diffusion weightings serve to calculate quantita tive maps of relevant diffusion characteristics, e.g. of the apparent diffusion coefficient (ADC).

At the same time, however, strong magnetic field gradients render available DW MRI techniques very sensitive to (macroscopic) movements. For example, in medical imaging this problem refers to involuntary body motions as well as movements due to respiration, cardiac pulsation or peri stalsis. It is therefore standard practice to combine a diffusion-encoding spin-echo sequence with a high-speed (preferably single-shot) readout MRI sequence which results in a DW image that effectively "freezes" the actual condition, i.e. an image without motion-induced artefacts. In the past decades, several technical proposals have been made with respect to this strategy.

A preferred solution to overcome the motion sensitivity of DW MRI is the use of an echo-planar imaging (EPI) readout, as described in the recent review [1] Today, DW EPI is the only commer cially available method for DW MRI. It is generally (if not exclusively) used for medical MR imaging and widely accepted for clinical applications - despite some severe problems. The key disad vantage of DW EPI is as follows: As a gradient-echo MRI technique which relies on the application of a single radiofrequency excitation pulse and the acquisition of multiple gradient echoes with necessarily increasing echo times, EPI suffers from a pronounced sensitivity to magnetic field in homogeneity. For example, in the human body, such field distortions are caused by unavoidable differences in the magnetic susceptibilities of different tissues. They lead to often severe geomet ric distortions as well as false positive or negative signal alterations in EPI images and compromise the diagnostic accuracy. For example, inhomogeneity effects occur in lower and frontal parts of the brain (i.e., close to air-filled cavities or dental implants) and even more frequently throughout the body such as for MRI of the prostate (i.e., close to the air-filled rectum). Nevertheless, in a clinical context, DW EPI is currently the only method of choice because the images present with good signal-to-noise ratio (SNR).

An alternative solution for DW MRI which avoids sensitivity to magnetic field inhomogeneity and still allows for high-speed acquisitions is the use of a single-shot spin-echo MRI sequence (see e. g. [2] and [3]). This technique uses multiple radiofrequency-refocused spin echoes (rather than gra dient echoes as for EPI) in order to generate and spatially encode multiple spin-echo signals for image reconstruction. A severe disadvantage of this approach, however, is the need for many high-power radiofrequency excitations (i.e., refocusing pulses with high flip angles) which com monly leads to a violation of the specific absorption rate (SAR) limit for energy deposition within the human body. As a consequence, such single-shot spin-echo methods are not in general use for human applications.

An even further proposal for single-shot DW MRI is a combination of the DW spin-echo sequence with a readout sequence which comprises a series of radiofrequency-refocused stimulated ech oes. A single-shot STEAM (stimulated echo acquisition mode) MRI sequence has been described as early as 1984 (see e. g. [4] and [5]). In this case, multiple stimulated echoes are generated by radiofrequency pulses with low flip angles which avoid any SAR problem. More recently, several versions for DW single-shot STEAM MRI have been developed which solve the susceptibility prob lem of DW EPI, but have the disadvantage of a lower SNR than DW EPI (see [6], [7] and [8]). As a consequence, DW STEAM MRI has not yet found clinical acceptance.

In order to overcome the SNR deficiency of DW STEAM MRI, it has recently been proposed to combine the technique with an undersampled radial acquisition mode and to accomplish the im age reconstruction as a two-step procedure which differs for the non-DW images and the DW images. The joint reconstruction of the non-DW images and their coil sensitivity maps poses a nonlinear inverse problem which is solved by the iteratively regularized Gauss-Newton method, as described e. g. in the textbook [9] In an implementation of this approach, which was specifically designed for applications to the human brain ([10], [11]), the nonlinear inverse reconstruction of the non-DW images was regularized by the coil sensitivity maps of a neighbouring cross-section. However, these maps offer only very low spatial frequency information, so that pertinent recon structions fail to significantly improve the aforementioned problem. Moreover, preliminary clini cal testing in some practical cases revealed numerical instability which most likely relates to insuf ficient convergence because of weak regularization with a coil sensitivity map.

Another preceding version of DW STEAM MRI, which has been developed for specific application to the prostate ([12]), employs a multi-shot acquisition method. Apart from again using weak regularization with neighbouring coil sensitivity maps only, the approach requires a much more complex mathematical treatment in order to accomplish a joint reconstruction of DW images from multiple acquisitions which refer to different states of macroscopic movement.

An extremely accelerated method for acquiring and reconstructing a sequence of dynamic non- DW MR images using e. g. the above regularization described in [9] has been proposed in [13] and [14] The use of a gradient-echo MRI sequence with pronounced undersampling, non-Cartesian trajectories for spatial encoding, and image reconstruction by regularized nonlinear inversion results in acquisition times in the range of tens of milliseconds. Thus, depending on the dynamic process to be studied, temporal changes of the object under investigation can be monitored in real-time with high temporal fidelity ([15]). However, the technique of [13] and [14] is mainly di rected to collect images of a single slice of the object, so that neither coverage of a volume of an object under investigation nor diffusion weighting is obtained. Although collecting images of dif ferent slices of the object is considered in [13] and [14] as well, this is limited to a few slices, like e. g. less than 5 slices. Furthermore, the corresponding applications are realized as interleaved multi-slice data acquisitions, so that the technique of [13] and [14] sacrifices temporal resolution and increases the sensitivity to motion.

Objective of the invention

The objective of the invention is to provide an improved method for creating, in particular acquir ing image raw data and reconstructing, a sequence of DW MR images, which cover a volume of an object under investigation, said method being capable of overcoming disadvantages of conven tional techniques and/or allowing new applications of MR imaging. In particular, the objective of the invention is to provide an improved method for creating a sequence of cross-sectional DW MR images for gap-free volume coverage with increased acquisition speed, reduced (or without) sen sitivity to motion, improved SNR and/or reduced (or without) sensitivity to magnetic field inho mogeneity. For medical imaging purposes, in particular, the objective is to provide a method for creating a sequence of DW MR images which cover a volume of a human body without gaps, with reduced (or without) sensitivity to motion and with reduced (or without) sensitivity to magnetic field inhomogeneity, thus allowing for a method for medical MR imaging applicable to the entire human body. Furthermore, the objective of the invention is to provide an improved MRI device, in particular being adapted for conducting the method for the acquisition and reconstruction of a spatially contiguous sequence of diffusion-weighted MR images of a volume.

Summary of the invention

The above objectives are solved by an MR image creating method and/or an MRI device compris ing the features of the independent claims. Advantageous embodiments of the invention are de fined in the dependent claims. According to a first general aspect of the invention, the above objective is solved by a method for creating sequences of diffusion-weighted MR images of an object under investigation, in particu lar acquiring image raw data and reconstructing the diffusion-weighted MR images, wherein each of said sequences of MR images represents the same series of contiguous cross-sectional slices covering a volume of the object.

The inventive method comprises a step of providing multiple sequences of sets of image raw data being collected with the use of at least one radiofrequency receiver coil of a magnetic resonance imaging (MRI) device.

Each set of image raw data includes an image content of one of the MR images to be reconstruct ed, and each set of image raw data refers to one of the cross-sectional slices, i. e. each of the MR images to be reconstructed represents one of the cross-sectional slices. The cross-sectional slices provide a series of directly neighbouring cross-sections covering the volume of the object. The position of each cross-sectional slice is shifted by a slice shift in a direction perpendicular to the imaging plane in order to cover the volume of the object under investigation. The slice shift is equal to a predetermined slice thickness of the cross-sectional slices. Providing the image raw data of directly neighbouring cross-sections means that the image raw data are provided without or with negligible overlap of image content of the neighbouring cross-sections and without or with negligible gaps between image content of the neighbouring cross-sections. The spatial orien tation of the imaging plane, e. g. relative to a longitudinal z direction of the main magnetic field of the MRI device, can be selected in dependency on an imaging task, e. g. in dependency on the anatomical orientation of an organ to be imaged in a human body. The spatial orientation of the imaging plane can be set by directions of the spatially encoding magnetic field gradients in the MRI device.

Furthermore, each set of image raw data includes a plurality of data samples being generated with a combined diffusion-weighted spin-echo and single-shot stimulated-echo sequence with diffusion-encoding gradients (indicated in the following as DW single-shot STEAM sequence). The DW single-shot STEAM sequence spatially encodes an MRI signal (i. e. set of image raw data) re ceived with the at least one radiofrequency receiver coil using a non-Cartesian, preferably radial, k-space trajectory.

The diffusion-encoding gradients of said multiple sequences of sets of image raw data have at least two different strengths and at least three different directions. Accordingly, each sequence of sets of image raw data is distinguished by a predetermined diffusion-encoding gradient strength (gradient amplitude, which may be zero) and a predetermined diffusion-encoding gradient direc tion, wherein at least two different sequences of sets of image raw data use different diffusion encoding gradient strengths and at least three different sequences of sets of image raw data use different diffusion-encoding gradient directions.

One of the sequences of sets of raw data is generated with diffusion-encoding gradients of zero strength or lower strength. This sequence, which also can be indicated as "zero or lower strength sequence", is generated without a diffusion-encoding gradient or with a diffusion-encoding gradi ent strength lower than the diffusion-encoding gradient strengths of the remaining sequences of sets of image raw data. The zero or lower strength sequence is collected with the lowest of all applied diffusion-encoding gradient strengths or without diffusion-encoding gradient.

Further sequences of sets of raw data are generated with diffusion-encoding gradients of higher strength. These sequences, which also can be indicated as "higher strength sequences", are gen erated with diffusion-encoding gradient strengths above zero and higher than the diffusion encoding gradient strength of the zero or lower strength sequence.

Optionally, the sequences of sets of image raw data can be repeatedly acquired for advantageous ly obtaining a noise reduction, i. e. the sequences of sets of image raw data may include further sequences of image raw data with equal strength or direction. Accordingly, with this preferred variant, the zero or lower strength sequence and/or at least one of the higher strength sequences can be repeatedly collected.

Each set of image raw data comprises a set of homogeneously distributed lines in k-space with equivalent spatial frequency content, wherein the lines of each set of image raw data cross the center of k-space and cover a continuous range of spatial frequencies and the number of lines of each set of image raw data is selected such that each set of image raw data is undersampled be low a sampling rate limit defined by the Nyquist-Shannon sampling theorem (also known as Whit- taker-Kotelnikow-Shannon sampling theorem). Employing an undersampled radial trajectory for spatial encoding advantageously provides image raw data collected with high acquisition speed. The positions of the lines of each set of image raw data differ in sets of image raw data from neighbouring cross-sectional slices. Furthermore, the inventive method comprises a step (first reconstructing step) of subjecting the sequence of sets of image raw data generated with diffusion-encoding gradients of zero strength or lower strength, i. e. the zero or lower strength sequence, to a regularized nonlinear inverse reconstruction process to provide a sequence of MR images with no or lower diffusion weighting (compared with the remaining MR images). Each of the MR images represents one of the cross- sectional slices and each of the MR images is created by a simultaneous estimation of a sensitivity of the at least one receiver coil and the image content and in dependency on a difference be tween a current estimation of the sensitivity of the at least one receiver coil and the image con tent and an estimation of the sensitivity of the at least one receiver coil and the image content of a neighbouring cross-sectional slice.

Furthermore, the inventive method comprises a step (second reconstructing step) of subjecting the sequences of sets of image raw data with diffusion-encoding gradients of higher strengths, i. e. the higher strength sequences, to a linear inverse reconstruction process to provide sequences of MR images with higher diffusion weighting. Each of these MR images represent one of the cross-sectional slices and being created by using the sensitivity of the at least one receiver coil determined in the previous step of subjecting the zero or lower strength sequence to the regular ized nonlinear inverse reconstruction process for the same cross-sectional slice.

Advantageously, in contrast to [10] and [11], the present invention extends the regularization of the numerical minimization problem to both a directly neighbouring image (with low and high spatial frequency components) and all its associated coil sensitivity maps. The resulting recon structions allow for a higher degree of data undersampling and therefore acquisition speed. They further achieve computational robustness as well as high spatial fidelity and increased quality of each cross-sectional image.

In fact, in the first reconstruction step, the numerical solution formally adopts the same regulari zation as originally developed for dynamic real-time MRI ([13] and [14]), but now in space rather than in time. In other words, the invention exploits the similarity of successive, i.e. directly neigh bouring, images and coil sensitivity maps, and therefore achieves the same degree of spatial fidel ity as previously demonstrated for temporal fidelity when using temporal regularization by a pre ceding image and its coil sensitivity maps ([15]).

In other words, the nonlinear inverse reconstruction process applied in the first reconstructing step is an iterative process which in each iterative step solves a regularized linearization of a non- linear MRI signal equation which maps the unknown spin density to be measured and its coil sen sitivities to the data acquired from the at least one receiver coil. The inventors have found that the nonlinear inverse reconstruction process employing the similarity of temporally successive images of a given image plane as described in [13] and [14] can be used for reconstructing spatial ly successive images of the contiguous cross-sectional slices, i. e. images of different imaging planes. This is a surprising result as it was not expected before the invention that adjacent cross- sectional slices have sufficient similarity for successfully applying the nonlinear inverse recon struction process, even with objects having step-wise changes of the spin density therein. Contrary to [13] and [14], the first reconstructing step of the inventive method primarily does not provide a sequence of temporally changing (dynamic) MR images, but a sequence of spatially distributed (static) MR images of the object which furthermore have diffusion contrast.

In the second reconstruction step, the DW images of the same cross-section are computed with use of the coil sensitivity maps from the corresponding non-DW images. The procedure reduces this part of the reconstruction to a linear inverse problem which is solved by a conjugate gradient method.

In contrast to [11] and [12], the present invention relies on a DW single-shot STEAM MRI acquisi tion in conjunction with a nonlinear inverse reconstruction of the non-DW images with regulariza tion to a directly neighbouring image and its associated coil sensitivities and a subsequent linear inverse reconstruction of the DW images. The method according to the invention is therefore adapted for improving the spatial resolution as, compared to [11] and [12], higher spatial fre quencies are introduced to the nonlinear inverse reconstruction. Advantageously, the invention provides particular advantages for medical imaging, and it is directly applicable to different organ systems, in particular to both the human brain and prostate.

In summary, the invention provides to a technique for DW single-shot STEAM MRI of a volume without susceptibility artefacts and with improved image quality compared to preceding meth ods. The invention acquires non-DW and DW datasets for a series of directly neighbouring cross- sections which cover a volume of an object under investigation without gaps. The joint recon struction of the non-DW images and associated coil sensitivity maps is accomplished as the solu tion to a nonlinear inverse problem with regularization to a neighbouring image and its coil sensi tivities. The determined coil sensitivities are then used to compute the corresponding DW images of the same cross-section as solution to a linear inverse problem. The invention offers (i) DW MRI of a volume without any sensitivity to magnetic field inhomogeneity and therefore avoids geo- metric distortions, signal void and false positive and negative signal alterations in contrast to DW EPI. In comparison to preceding DW STEAM MRI methods the invention further offers (ii) a higher degree of data undersampling and therefore reduced measuring times, (iii) improved computa tional stability, speed and spatial fidelity, (iv) improved spatial resolution and SNR, (v) general applicability to different organ systems, and (vi) in case of medical imaging, clinically feasible scan times.

According to a second general aspect of the invention, the above objective is solved by an MRI device being configured for creating a sequence of diffusion-weighted MR images of an object under investigation and comprising an MRI scanner and a control device. The MRI scanner in cludes a main magnetic field device, at least one radiofrequency excitation coil, three magnetic field gradient coils and at least one radiofrequency receiver coil. According to the invention, the control device is adapted for controlling the MRI scanner for collecting the multiple sequences of sets of image raw data and reconstructing the sequences of MR images with the method accord ing to the first aspect of the invention or one of the embodiments thereof.

According to a preferred embodiment of the invention, the method for creating sequences of DW MR images comprises a further step of subjecting said sequences of MR images to a calculation of quantitative maps of the object under investigation, comprising a pixelwise calculation of at least one of a mean diffusion-weighted MRI signal averaged across gradient directions, a representa tion of diffusion properties, in particular the diffusion tensor, a trace of the diffusion tensor, an apparent diffusion coefficient, a fractional anisotropy, and a representation of diffusion direc tions, in particular a main diffusion direction. The calculation of the quantitative maps has particu lar advantages for medical imaging as the quantitative maps provide a basis for separate succes sive diagnostics.

According to a further preferred embodiment of the invention, the reconstruction processes of at least one of the first and second reconstructing steps includes a filtering process suppressing im age artefacts. Advantageously, filtering improves the image quality. With a particularly preferred variant, the filtering process includes applying a spatial filter, in particular a non-local means filter for each MR image.

As a further advantage of the invention, the image raw data can be selected with a high degree of undersampling, i.e. relative to a fully sampled reference which - e.g. for radial encoding with ro tated straight lines and according to the sampling theorem - is given by p/2 times the number of data samples per line. The degree of undersampling can be at least a factor of 5, in particular at least a factor of 10, thus accelerating the data acquisition in the same manner as described for real-time MRI (e.g., see [14]). Accordingly, the number of lines of each set of image raw data can be reduced. In particular for medical imaging, it has been found that a number of lines equal or below 30, in particular equal or below 20 is sufficient for obtaining high quality MR image se quences.

According to a further preferred embodiment of the invention, the lines of each set of image raw data can be selected such that the lines of successive sets of image raw data are rotated relative to each other by a predetermined angular displacement. As an advantage, this rotation improves the effect of the regularization within the image reconstruction.

According to another preferred embodiment of the invention, the strength of the diffusion encod ing gradients of zero or lower strength is equal to or below a b-value of 100 s mm ², in particular zero. With a diffusion encoding gradient strength below this limit, the coil sensitivity maps calcu lated with the first reconstructing step are provided for the second reconstructing step in an ad vantageous manner with improved contrast and signal quality. As a further advantage, the result ing (sequences of) images with no or lower diffusion-weighting provide an improved reference for the combination with corresponding (sequences of) images with higher diffusion-weighting to calculate (sequences of) quantitative maps of a variety of diffusion characteristics such as the apparent diffusion coefficient which, e. g. in medical MR imaging, are of importance for diagnostic purposes.

Contrary to the diffusion encoding gradients of lower strength, the strengths of the diffusion en coding gradients of higher strength are preferably equal to or above a b-value of b = 200 s mm ², in particular equal to or above a b-value of b = 800 s mm ², and equal to or lower than b = 3000 s mm ². For example, b-values of up to 1000 s mm ² are typically used for standard neu roimaging applications [1]

With a further preferred variant of the invention, the collection of each set of image raw data is preceded by a radiofrequency pulse and magnetic field gradient module for frequency-selective saturation. Advantageously, the frequency-selective saturation allows for achieving specific con trasts depending on the imaging task. As mentioned above, imaging quality can be improved by repeatedly collecting zero or lower strength sequences. Preferably, at least one further sequence of sets of raw data is generated with diffusion-encoding gradients of zero strength or lower strength and the at least one further sequence of sets of raw data generated with diffusion-encoding gradients of zero strength or lower strength is subjected to a linear inverse reconstruction process to provide at least one fur ther sequence of MR images with no or lower diffusion weighting. Each of the images represents one of the cross-sectional slices and is created by using the sensitivity of the at least one receiver coil determined in the first reconstructing step for the same cross-sectional slice.

Advantageously, the inventive method for reconstructing a sequence of MR images can be con ducted during and/or immediately after collecting the image raw data with the at least one ra diofrequency receiver coil of the MRI device. In this case, providing the series of sets of image raw data comprises the steps of arranging the object in the MRI device including the at least one re ceiver coil, subjecting the object to the DW single-shot STEAM sequence, and collecting the series of sets of image raw data using the at least one receiver coil. Reconstructing the sequence of MR images can be completed within a few seconds after the end of the image raw data collection.

According to an alternative embodiment, the inventive method for reconstructing the sequence of MR images can be conducted independently of collecting the image raw data with predeter mined measurement conditions. In this case, the sets of image raw data can be received, e.g. from a data storage, like in a data cloud storage, and/or a data transmission from a distant MRI device.

Brief description of the drawings

Further details and advantages of the invention are described with reference to the attached drawings, which schematically show in

Figure 1: a combined diffusion-weighted spin-echo and single-shot stimulated-echo sequence with diffusion-encoding gradients used for raw data acquisition with preferred em bodiments of the MR imaging method according to the invention;

Figure 2: a preferred embodiment of an MRI device according to the invention; Figure 3: a flowchart illustrating features of preferred embodiments of the MR imaging meth od according to the invention;

Figure 4: a flowchart illustrating the first reconstructing step of the MR imaging method ac cording to the invention;

Figure 5: a flowchart illustrating the second reconstructing step of the MR imaging method according to the invention; and

Figures 6 and 7: examples of clinically relevant applications of the inventive DW MRI.

Description of the preferred embodiments

Preferred embodiments of the invention are described in the following with particular reference to the data flow of the inventive raw data acquisition and reconstruction process, the basic com ponents of an inventive MRI device and practical application examples.

The details of the design of k-space trajectories and the mathematical formulation and implemen tation of the regularized nonlinear inverse reconstruction are preferably provided as disclosed in [13] or [14]. In particular, the regularized nonlinear inverse reconstruction process is preferably implemented as disclosed in [13] for the reconstruction of time series of MR images of an object under investigation. Thus, [13] is incorporated to the present specification by reference in its en tirety, in particular with regard to all details of image reconstruction of the sequence of cross- sectional gradient-echo MR images of the object under investigation in the first reconstructing step. All procedural steps applied to time series of sets of raw data and sequences of MR images in [13] can be applied in the same manner to slice sequences of sets of raw data generated with the zero or lower strength sequence.

Further details of the MRI device, the numerical implementation of the mathematical formulation using available software tools and optional further image processing steps or calculating steps for providing quantitative maps are not described as far as they are known from conventional MRI techniques. Furthermore, exemplary reference is made in the following to parallel MR imaging wherein the image raw data comprise MRI signals received with a plurality of radiofrequency re ceiver coils. It is emphasized that the application of the invention is not restricted to parallel MR imaging, but rather possible even with the use of one single receiver coil. Figure 1 schematically shows an embodiment of a combined diffusion-weighted spin-echo and single-shot stimulated-echo sequence with diffusion-encoding gradients as applied and collected with an embodiment of an MRI device 100 as illustrated in Figure 2. The MRI device 100 includes an MRI scanner 10 with a main magnetic field device 11, at least one radiofrequency excitation coil 12, three magnetic field gradient coils 13 and radiofrequency receiver coils 14. The main magnetic field device 11 creates a static main magnetic parallel to the longitudinal axis of the MR scanner bore, e. g. z-axis. The object 1 to be investigated is accommodated in the MRI device 100. Furthermore, the MRI device 100 includes a control device 20 being adapted for controlling the MRI scanner 10 for collecting the sequences of sets of image raw data and reconstructing the sequence of MR images with step SI of the illustrated embodiment as shown in Figures 3 to 5.

The control device 20 includes at least one GPU 21, which is preferably used for implementing the regularized nonlinear inversion of Figure 4 and the linear inversion of Figure 5.

According to Figure 3, the preferred embodiment of the inventive method includes the raw data acquisition step SI of providing multiple sequences of sets of image raw data collected with the DW single-shot STEAM sequence, the first reconstructing step S2 of regularized nonlinear inverse reconstruction of zero or lower strength sequence, preferably including a repeated reconstruc tion with a reversed order of slices, the second reconstructing step S3 of linear inverse recon struction of higher strength sequences, optionally combined with a further linear inverse recon struction of at least one further zero or lower strength sequence, and a calculating step S4 of cal culating quantitative maps.

Thus, one part of the inventive method includes the reconstruction of a sequence of cross- sectional STEAM MR images for a number of slices without or with lower diffusion-encoding gra dients, wherein the STEAM MR images are obtained by a nonlinear inverse reconstruction process with regularization to a neighbouring image and its coil sensitivity maps (first reconstructing step S2). Another part of the inventive method includes the reconstruction of multiple cross-sectional STEAM MR images for the same slices (second reconstructing step S3) with as many different dif fusion-encoding gradients as required for a particular application (e.g., with gradients of different strength or different orientation) which are obtained by a linear inverse reconstruction process. Raw data acquisition

For acquiring the raw data, a number of directly neighbouring slices 2 with constant slice thick ness is defined in the volume under investigation, like an organ of a patient, as schematically shown in the insert of Figure 2. The number and thickness of slices 2 is selected in dependency on the size of the organ, the slice thickness employed, the measuring time and/or the imaging reso lution to be obtained. As a practical example of imaging a human brain about 50 slices with a slice thickness of 3 mm are defined. For each slice, varying DW single-shot STEAM sequences are ap plied as described in the following.

In the raw data acquisition step SI of the inventive method, multiple sequences of sets of image raw data are provided which are collected with the DW single-shot STEAM sequence as shown in Figure 1. Each sequence of sets of image raw data represents one of the slices. The object 1, e. g. a tissue or organ of a patient, is subjected to the DW single-shot STEAM sequence of Figure 1 en coding the MRI signal received with the radiofrequency receiver coils 14. The section in brackets is repeated for each line in k-space. The DW single -shot STEAM sequence is constructed such that data samples are collected along non-Cartesian k-space trajectories. A slice shift D being equal to the slice thickness is accomplished by changing the radiofrequency excitation pulse.

Each DW single-shot STEAM MRI sequence of Figure 1 combines a leading spin-echo sequence (with or without diffusion-encoding gradients) and a single-shot STEAM MRI readout sequence. This version refers to a preferred example with a pair of diffusion-encoding gradients (hatched) in the spin-echo sequence (SE = spin echo) and radial encoding gradients (G_x, G_y) in the single-shot STEAM MRI sequence (STE = stimulated echoes, G_s= slice-selection gradient, RF = radiofrequency pulses, dotted = spoiler gradients).

In the practical implementation of the invention, the DW single-shot STEAM sequence of Figure 1 is applied a number n of times. As an example, one of the sequences of sets of raw data (the zero or lower strength sequence) is generated without diffusion-encoding gradients in the leading spin-echo sequence, while further sequences of sets of raw data are generated with diffusion encoding gradients of higher strength. Implementations with a minimum number of n = 4 repeti tions represent applications with only one diffusion-encoding gradient of zero or lower strength and three diffusion-encoding gradients with higher strength and different directions. Flowever, the number n may also correspond to several hundred repetitions, e.g. when using 128 gradient directions and multiple averages to allow for advanced post-processing applications such as fiber tractography.

To avoid ambiguities with directional differences of diffusion properties, clinical applications of DW MRI techniques usually measure DW images in at least 3, but preferably 6, directions with use of 3 or 6 differently oriented diffusion-encoding gradients, respectively. Such protocols are used for the calculation of a mean DW image averaged across diffusion-encoding directions in order to obtain an "isotropic" (i.e., orientation-independent) diffusion weighting.

Moreover, clinical applications of DW MRI usually rely on two or more different gradient strengths (e.g., characterized by b values of 0 and 1000 s mm ²). Together with DW images with multiple gradient directions, such information are used for the calculation of an orientation- independent mean DW image, a diffusion tensor, and a corresponding map of the ADC as well as of other diffusion characteristics (e.g., a fractional anisotropy map). With particular applications of DW MRI, more than two, e. g. up to 10, gradient strengths can be applied.

For increasing the SNR, the zero or lower strength sequence of sets of raw data and/or the higher strength sequences of sets of raw data can be repeatedly collected. The number of repetitions can be selected in dependency on the application conditions of the MRI method.

Image reconstruction

First, the sequence of MR images with no or lower diffusion weighting is obtained by a regularized nonlinear inverse reconstruction process (first reconstructing step S2) which jointly estimates each image and its associated coil sensitivity maps while exploiting the spatial similarity to a di rectly neighbouring image and its coil sensitivities (e.g., using the iteratively regularized Gauss- Newton method).

Deviating from [13], each set of the image raw data represents another one of contiguous cross- sectional slices 2 as shown in the schematic insert of Figure 2. Furthermore deviating from [13], this step preferably is performed twice, the second time with a reverse order of slices. This is be cause the first few images of the sequence of images from neighbouring cross-sectional slices do not fully benefit from spatial regularization. In particular, for the nonlinear inverse reconstruction of the first image of the sequence no prior information exists, i.e. no image from a neighbouring slice. The repeated reconstruction in reverse order avoids this problem by initializing the recon- struction with the last image and its coil sensitivities from the initial reconstruction. As an alterna tive, it is possible to solve this problem by slice oversampling, wherein sets of image raw data are additionally collected for cross-sectional slices outside the target volume to be covered.

Figure 4 summarizes a data flow including the first reconstructing step S2, as described in [13] after the raw data acquisition step SI of collecting measured raw data and an optional prepro cessing step Sl.l of preprocessing the measured data.

With preprocessing step Sl.l, the image raw data are subjected to a gradient delay correction S1.1A, wherein possible unintended shifts of the k-space lines from the center of k-space are cor rected, and to an interpolation step SI. IB, wherein an interpolation of the non-Cartesian data onto a Cartesian grid is conducted. Steps S1.1A and SI. IB can be implemented as disclosed in [13].

Subsequently, with the first reconstructing step S2, the sequence of MR images of the object 1 is reconstructed by the regularized nonlinear inverse reconstruction process, which is described in [13] Starting from an initial guess S21 for the MR image of a first cross-sectional slice and the coil sensitivities, each of the MR images is created by an iterative simultaneous estimation S22 of sen sitivities of the receiver coils and the image content. Step S22 comprises the nonlinear inverse reconstruction using an iteratively regularized Gauss-Newton method including a convolution- based conjugate gradient algorithm S23. The number of iterations (Newton steps) is selected in dependency on the image quality requirements of a particular imaging task. Steps S21 to S23 are preferably repeated in reverse order of slices for improving the reconstruction of the images of the first few slices. Finally, the reconstructed sequence of MR images is output (S25). Further steps of conventional processing, storing, displaying, or recording of image data can follow.

As the result of step S2, a sequence of cross-sectional STEAM MR images, each representing one of slices 2 is obtained, wherein the STEAM MR images and associated coil sensitivity maps are generated without or with lower diffusion-encoding gradients.

Subsequently, the second reconstructing step S3 of Figure 3 is conducted as shown in Figure 5. The complex coil sensitivity maps of the individual cross-sectional images are provided with step S31 as a result of the first reconstructing step S2. Due to the provision of the coil sensitivity maps, the number of unknowns is reduced and the further MR images with higher diffusion weighting can be calculated with a linear inverse reconstruction using the known complex coil sensitivity maps. As illustrated with step S32 in Figure 5, the coil sensitivity maps are taken to calculate the DW MR images (MR images with higher diffusion weighting) of the same cross-sections as the solution to a linear inverse problem. As an example, the generally known conjugate gradient method is applied on the higher strength sequences of sets of raw data for obtaining the DW MR images. If the higher strength sequences of sets of raw data have been repeatedly collected, the conjugate gradient method is applied individually on all sequences of sets of raw data collected with equal diffusion-encoding gradients.

As the result, the DW MR images to be obtained are provided with step S33 for further pro cessing. Multiple MR images with identical high diffusion weighting and associated coil sensitivity maps can be averaged for further processing. Again, further steps of conventional processing, storing, displaying, or recording of image data can follow.

If the zero or lower strength sequence of sets of raw data is repeatedly collected in step SI for increasing the SNR, only a first sequence of MR images is calculated with the above nonlinear inverse reconstruction of step S2. Further sequences of MR images can be calculated in step S3 using the complex coil sensitivity maps of the individual cross-sectional images known from the first sequence of MR images in step S2. Again, the number of unknowns is reduced and the fur ther MR images with no or lower diffusion weighting are calculated with a linear inverse recon struction (e.g. with the conjugate gradient method) using the known complex coil sensitivity maps. Calculating the further MR images with no or lower diffusion weighting has advantages for calculating the maps in step S4 (see below). The MR images with no or lower diffusion weighting obtained with steps S2 and S3 and associated coil sensitivity maps can be averaged for further processing.

For medical imaging, the sequences of MR images obtained with the reconstructing steps S2 and S3 are used for a calculation of quantitative maps of the object under investigation (step S4 in Figure 3). The calculation of the quantitative maps relates to a pixelwise calculation of a variety of diffusion properties such as e.g. a mean diffusion-weighted MRI signal averaged across gradient directions and the apparent diffusion coefficient. Step S4 is implemented as it is known as such from conventional techniques. Experimental examples

Experimental examples of the invention are described in the following with particular reference to applications in medical imaging. All examples refer to studies of healthy human subjects.

An example representing the clinically relevant application of the inventive DW MRI to the brain is shown in Figure 6. It summarizes three selected cross-sections (top to bottom) of a DW single shot STEAM MRI study of the normal human brain where images without and with diffusion weighting (b = 1000 s mm ², 6 diffusion directions) were acquired for a total of 51 directly neigh bouring cross-sections in a measurement time of 2 minutes and 30 seconds. The results represent the non-DW image (left), the mean DW image (middle) and the calculated ADC map (right). Re gardless of whether the images represent transverse brain sections with little (top row) or strong (bottom row) susceptibility differences, all images are free from inhomogeneity artefacts (i.e., geometric distortions and signal alterations) which are typical for DW EPI measurements.

The same finding holds true for abdominal scans. Figure 7 presents selected results of a DW sin gle-shot STEAM MRI study of the normal human prostate which corresponds to another field of high clinical importance. In this case the non-DW image, the mean DW image (b = 600 s mm ²,

6 gradient directions), and the ADC map of the three sections (same organization as in Fig. 6) were selected from a total of 31 directly neighbouring cross-sections acquired in a measurement time of 4 minutes and 30 seconds.

The application of the invention is not restricted to medical imaging, like in the above examples, but correspondingly possible for imaging other objects, like workpieces or other technical objects.

The features of the invention disclosed in the above description, the drawings and the claims can be of significance individually, in combination or sub-combination for the implementation of the invention in its different embodiments.

Claims

1. Method for creating multiple sequences of diffusion-weighted magnetic resonance

(MR) images of an object under investigation, wherein each of said sequences of MR images rep resents the same series of contiguous cross-sectional slices covering a volume of the object under investigation, comprising the steps of:

(a) providing multiple sequences of sets of image raw data being collected with the use of at least one radiofrequency receiver coil of a magnetic resonance imaging (MRI) device, wherein

- each set of image raw data includes an image content of one of the MR images to be recon structed,

- each set of image raw data refers to one of the cross-sectional slices,

- each set of image raw data includes a plurality of data samples being generated with a combined diffusion-weighted spin-echo and single-shot stimulated-echo sequence with diffusion-encoding gradients, said combined sequence spatially encoding an MRI signal received with the at least one radiofrequency receiver coil using a non-Cartesian k-space trajectory,

- said diffusion-encoding gradients of said multiple sequences of sets of image raw data having at least two different strengths and at least three different directions, wherein one of the sequences of sets of raw data is generated with diffusion-encoding gradients of zero strength or lower strength and further sequences of sets of raw data are generated with diffusion-encoding gradi ents of higher strength,

- each set of image raw data comprises a set of homogeneously distributed lines in k-space with equivalent spatial frequency content,

- the lines of each set of image raw data cross the center of k-space and cover a continuous range of spatial frequencies,

- the number of lines of each set of image raw data is selected such that each set of image raw data is undersampled below a sampling rate limit defined by the Nyquist-Shannon sampling theo rem, and

- the positions of the lines of each set of image raw data differ in sets of image raw data from neighbouring cross-sectional slices,

(b) subjecting the sequence of sets of image raw data generated with diffusion-encoding gradi ents of zero strength or lower strength to a regularized nonlinear inverse reconstruction process to provide a sequence of MR images with no or lower diffusion weighting, each of the images representing one of the cross-sectional slices and being created by a simultaneous estimation of a sensitivity of the at least one receiver coil and the image content and in dependency on a differ ence between a current estimation of the sensitivity of the at least one receiver coil and the im age content and an estimation of the sensitivity of the at least one receiver coil and the image content of a neighboring cross-sectional slice, and

(c) subjecting the sequences of sets of image raw data with diffusion-encoding gradients of higher strengths to a linear inverse reconstruction process to provide sequences of MR images with higher diffusion weighting, each of the images representing one of the cross-sectional slices and being created by using the sensitivity of the at least one receiver coil determined in step (b) for the same cross-sectional slice.

2. Method according to claim 1, comprising a further step of

(d) subjecting said sequences of M R images to a calculation of quantitative maps of the object under investigation, comprising a pixelwise calculation of at least one of

- a mean diffusion-weighted MRI signal averaged across gradient directions,

- a representation of diffusion properties, in particular the diffusion tensor,

- a trace of the diffusion tensor,

- an apparent diffusion coefficient,

- a fractional anisotropy, and

- a representation of diffusion directions, in particular a main diffusion direction.

3. Method according to one of the foregoing claims, wherein

- the reconstruction process of at least one of steps (b) and (c) includes a filtering process sup pressing image artefacts.

4. Method according to claim 3, wherein

- the filtering process includes applying a spatial filter, in particular a non-local means filter for each MR image.

5. Method according to one of the foregoing claims, wherein

- the number of lines of each set of image raw data is selected such that a resulting degree of undersampling is at least a factor of 5, in particular at least a factor of 10.

6. Method according to one of the foregoing claims, wherein

- the number of lines of each set of image raw data is at most 30, in particular at most 20.

7. Method according to one of the foregoing claims, wherein

- the lines of each set of image raw data are selected such that the lines of the sets of image raw data of neighbouring cross-sectional slices are rotated relative to each other by a predetermined angular displacement.

8. Method according to one of the foregoing claims, wherein

- the strength of the diffusion encoding gradients of zero or lower strength is equal to or below a b-value of 100 s mm ², in particular zero.

9. Method according to one of the foregoing claims, wherein

- the collection of each set of image raw data is preceded by a radiofrequency pulse and magnetic field gradient module for frequency-selective saturation.

10. Method according to one of the foregoing claims, wherein

- at least one further sequence of sets of raw data is generated with diffusion-encoding gradients of zero strength or lower strength, and

- the at least one further sequence of sets of raw data generated with diffusion-encoding gradi ents of zero strength or lower strength is subjected to a linear inverse reconstruction process to provide at least one further sequence of MR images with no or lower diffusion weighting, each of the images representing one of the cross-sectional slices and being created by using the sensitivi ty of the at least one receiver coil determined in step (b) for the same cross-sectional slice.

11. Method according to one of the foregoing claims, wherein the sequences of sets of image raw data are provided by at least one of

- arranging the object in the MRI device including the at least one receiver coil, subjecting the object to the combined diffusion-weighted spin-echo and single-shot stimulated-echo sequence, and collecting the sequences of sets of image raw data using the at least one receiver coil, and

- receiving the sequences of sets of image raw data by a data transmission collected from a dis tant MRI device.

12. MRI device being configured for creating a sequence of diffusion-weighted MR images of an object under investigation, comprising

- an MRI scanner including a main magnetic field device, at least one radiofrequency excitation coil, three magnetic field gradient coils and at least one radiofrequency receiver coil, and - a control device being adapted for controlling the MRI scanner for collecting the multiple se quences of sets of image raw data and reconstructing the sequences of MR images with the method according to one of the foregoing claims.