WO2008030098A2

WO2008030098A2 - Method of reducing artefacts, in particular movement-related artefacts, in mr images

Info

Publication number: WO2008030098A2
Application number: PCT/NL2007/050444
Authority: WO
Inventors: David Gordon Norris; Pieter Buur
Original assignee: Stichting Voor De Technische Wetenschappen
Priority date: 2006-09-08
Filing date: 2007-09-10
Publication date: 2008-03-13
Also published as: NL1032467C2; WO2008030098A3

Abstract

A method of reducing artefacts, in particular movement-related artefacts, in MRI images, comprising: - directing magnetic dipoles of an object to be imaged to a magnetic field; - exciting the magnetic dipoles at at least one moment; - detecting a first echo and at least one second echo, caused by the resonance of the excited dipoles; wherein the first echo is detected so shortly after the said at least one moment that this echo contains substantially no T2* information, that the second echo is only detected at a later time so that the second echo does contain T2* information, and that the first echo is used in a data processing process to reduce or eliminate artefacts possibly present in the (second) echoes, in particular spin history-dependent artefacts.

Description

Title: Method of reducing artefacts, in particular movement-related artefacts, in MRI images

The invention relates to a method of reducing artefacts in MIU images, comprising:

- directing magnetic dipoles of an object to be imaged to a magnetic field; • exciting the magnetic dipoles at at least one moment.

Such a method is, for instance, known from the article "Movement- Related Effects in fMRI Time-Series", Karl F. Friston et al, MRM 35:346- 355 (1999). This article mentions the problem of spin history-dependent artefacts as such (i.e. independent of alignment artefacts). As is generally known, fMRI (Functional Magnetic Resonance

Imaging) typically comprises polarizing (i.e. directing to a magnetic field) of magnetic dipoles of an object to be imaged, and exciting the magnetic dipoles at at least one moment. Echoes (echo signals), caused by the resonance of excited dipoles, are detected and processed to images. As is well-known, particular activity change in the object (for instance brain tissue) can result in a change in T2* values, which is visible on the MRI images. The T2* value depends on the oxygen content in the blood. This mechanism is referred to as BOLD (Blood Oxygen Level Dependent). fMRI movement correction is generally known per se, see for instance the article "Reduction of Signal Fluctuations in Functional MRI Using Navigator Echoes", Xiaoping Hu et al., Magnetic Resonance in Medicine 31 (1994) May, no. 5, and the article "BOLD Contrast Sensitivity Enhancement and Artefact Reduction With Multiecho EPI: Parallel- Acquired Inhomogeneity-Desensitized fMIR", Benedikt A. Poser et al., Magnetic Resonance in Medicine 55: 1227- 1235, 2006. Retrospective movement correction, where brain regions/volumes are aligned with respect to one reference, is now for instance a standard processing step with fMRI data analysis. A problem with fMRI is that movement of an object to be imaged (for instance tissue, a part of the brain or the like) can disturb a measurement. In particular, a movement during prior scans or between scans can influence a next scan, through the spin history effect (see Friston et alS)

Movement-related artefacts are undesired, in the first place with measurements, where an object needs to be motionless but still moves (for instance with patients who are difficult to fix in a position), and in the second place with measurements where movement is an unavoidable consequence of the experimental paradigm (for instance if a subject needs to respond to a test, for instance by pressing a button, or needs to carry out another operation). This retrospective movement correction cannot remove the above-mentioned spin history effects.

It is known to add movement parameters to a regression model, but with this method, the sensitivity to actual activation, occurring simultaneously with the movement, is reduced strongly.

In above-mentioned article of Friston et al., it is, for instance, proposed to remove movement-related artefacts by using a relatively complex autoregressive movement average model for the effects of prior displacements on a (later) signal in question. In particular, it is proposed to estimate movement parameters, by comparing each MRI scan in a time series with a reference scan, then again aligning the time series, and adjusting the data of each voxel. The article mentions various limitations of this method.

Another known method is to monitor the position of the head in an MRI scanner, and to use the thus obtained data to adjust a reference frame of the MRI system thereto, which is a relatively complex and expensive method. Other artefacts which can occur in MRI images may, for instance, be the result of physiological noise, for instance as a result of heartbeat and/or respiration.

The present invention contemplates solving these problems. In particular, the invention contemplates a method where artefacts, in particular spin history-dependent artefacts, in MRI images can be reduced in a relatively simple manner.

According to the invention, to this end, the method is characterized by detecting a first echo and at least one second echo, caused by the resonance of the excited dipoles, where the first echo is detected so shortly after above-mentioned at least one moment that this echo contains substantially no T2* information, that the second echo is only detected at a later time so that the second echo does contain T2* information, and that the first echo is used in a data processing process to reduce or eliminate artefacts, in particular spin history-dependent artefacts, possibly present in the second echoes.

It is found that, in this manner, a relatively large part of artefacts (for instance movement-related artefacts and physiological noise) can be reduced in a simple manner. The present invention is based on the insight that BOLD contrast (T2* contrast) needs time to develop after excitation, in contrast with spin history effects. In an above-mentioned first echo, BOLD contrast is negligible, but since spin history effects are manifest in the signal (the echo) directly after excitation (in the respective intensity Io), these effects can simply be determined on the basis of the intensity (Io) of the first echo, at least be taken into account in the data processing. In other words: the first echo provides spin history information which can be used in data processing with the next (second) echoes to accurately determine real, true, actual activation.

To a skilled person, it will be readily apparent that, in the present application, the term 'echo* (or 'echo signal¹) can be understood to be equivalent to the (complete) MRI image / the MRI scan, i.e.: after transformation from the data domain to image domain via Fourier transformation. In particular, in the present invention, all processes are therefore carried out in the image domain. In addition, it is noted that, for instance, one or more second

(later) echoes can be determined, for instance a series or sequence of second signals, where, for instance, all second echoes do contain T2* information. Further, for instance, successively, different parts of an object can be scanned, while, each time, of each part, both first and second echoes are obtained to image that object part after suitable data processing.

According to a further elaboration of the invention, the data processing process comprises a general linear regression model, utilizing the relation:

Y=β.X+ε (1)

where Y comprises above-mentioned detected second echoes, X is a matrix which comprises above-mentioned detected first echoes, β is a weighting factor and ε corrected second echoes. In particular, Y is the vector with the observed data, and, for each voxel, this is a vector with the length of the number of scans.

In this manner, first echoes are used as a regressor for second echoes, in particular on a voxel-to-voxel basis, for obtaining a corrected echo ε for each voxel. The corrected signals are then used to obtain an MRI image, instead of original second echoes.

An alternative data processing process which can be used in the present invention is, for instance, independent component analysis (ICA). This method known per se from practice can expose spatially separated components with temporal profiles. Here, differential information in the echoes (physiological noise and movement components may be present in both first and second signals, but BOLD activation only in second signals) may be usable for automatic identification and subsequent elimination of irrelevant data parts (for instance as a result of artefacts).

The invention further provides a system for generating MRI images, evidently intended and suitable for carrying out a method according to the invention, while the system is provided with a magnetic resonance system to polarize magnetic dipoles of an object to be imaged and to excite them at at least one moment, and with one or more detectors arranged for detecting a first echo and at least one second echo, caused by the resonance of excited dipoles, while, during use, where the first echo is detected so shortly after the above-mentioned one moment that this echo contains substantially no T2* information, and the second echo is only detected at a later time so that the second echo does contain T2* information, while the system is provided with a data processor arranged for using the first echo in a data processing process to reduce or eliminate artefacts possibly present in the echoes, in particular spin history-related artefacts, more in particular movement-related artefacts.

The invention further provides software provided with code executable by hardware, for instance a computer program, arranged for carrying out a method according to the invention if the software is run by suitable hardware. The software or the computer program may, for instance, be readable and executable by a computer or suitable data processor to carry out the method. A data carrier may, for instance, be provided with the software. Further advantageous elaborations of the invention are described in the subclaims. The invention will now be explained in more detail with reference to an exemplary embodiment and the drawing, in which:

Fig. 1 schematically shows an exemplary embodiment of an MRI system; Fig. 2 shows a diagram with respect to echo intensity and echo time of MRI signals;

Fig. 3 schematically shows a method according to the invention;

Fig. 4 shows a rest pattern A, stimulation pattern B and stimulation/movement pattern C, for use in an activation/movement experiment with subjects;

Fig. 5 shows a number of MRI images with respect to a visual activation under the influence of stimulation pattern B;

Fig. 6 shows non-corrected MRI images with respect to visual activation and movement under the influence of stimulation pattern C; and

Fig. 7 shows corrected MRI images with respect to Fig. 6.

In this patent application, same or corresponding measures are designated by same or corresponding reference symbols.

Fig. 1 schematically shows a system for generating MRI images. The system 1 comprises a magnetic resonance system 2 for directing (at least orienting or polarizing) magnetic dipoles of an object to be imaged (for instance tissue, a part of a patient, brain tissue or the like) and to excite them for at least one moment, and one or more detectors 3 arranged for detecting a first echo (i.e. a first MRI image) and at least one second echo (i.e. a second MRI image), caused by the resonance of excited dipoles. Such a system is known per se from practice. As already known, the system 1 comprises, for instance, a number of suitable magnets 2 (schematically shown) to provide a uniform magnetic field, and means 3 for generating a series of RF and magnetic field gradient pulses, to excite the dipoles, for the purpose of generating MRI images. The latter means (for instance an RF coil) may, for instance, also serve to detect above-mentioned echoes (in particular being or comprising complete MRI images or scans). As already known, the system may be provided with various magnets, in particular a magnet for BO (the strong magnetic field), and the shims. A control 4 may be provided to control the system 1, for the purpose of driving above-mentioned magnetic resonance system 2, above-mentioned detectors, for the purpose of data processing and for the purpose of storing and imaging MRI measurements. The control 4 may, for instance, comprise one or more suitable programmable modules, computers, data processors and editors, microelectronics, hardware, software, memory means, one or more displays or monitors, and/or the like, which will be readily apparent to a skilled person. As is already known, during use, such a system may, for instance, make a scan of a part of an object to be imaged (e.g. a slice having a thickness of a few mm), by suitable control of the means 3 for generating a series of RF and magnetic field gradient pulses, and the detector(s).

Fig. 2 shows a diagram of echo intensity (plotted along the vertical axis) as a function of the echo time, with respect to echoes with different Io and T2* values. Discrete bands Tl and T2 indicate times (echo times TE) within which images (scans) are taken, by measuring respective echoes El, E2. The first echo time Tl is directly after dipole excitation (from t=0), while the second echo time T2 is some time after dipole excitation. In particular, the second echo time T2 is at a moment when the echo signals contain functional (BOLD) contrast.

It follows from Fig.2 that the relation I=I₀.exp(-TE/T2*) applies to the course of the signal curve (however, the signal curve does not need to be exponential for use of the present invention). It further follows from Fig. 2 that one measurement of the second echo time T2 cannot yield sufficient information to determine the relative influence between Io and T2*.

In an advantageous manner, according to the invention, the MRI system 1 is arranged for operatively detecting one or more echoes El so shortly after the above-mentioned moment of excitation that each first echo El contains substantially no T2* information, and a respective second echo E2 only at a later time so that the second echo E2 does contain T2* information. Here, of each voxel, for instance, first and second echo parts can be detected. The first echo El may already comprise various object- related intensity fluctuations, as a result of physiological noise and/or movement.

The control (or data processor) 4 is arranged for using the first echoes El in a data processing process to reduce or eliminate object-related artefacts possibly present in the (second) echo, in particular spin history-dependent artefacts, for instance object movement-related artefacts and artefacts as a result of heartbeat and/or respiration. It is found that, in this manner, such artefacts can be reduced in a very simple manner, so that reliable MRI images can be obtained, and, for instance, be imaged on a monitor and stored in a memory. Such a method may, for instance, be carried out or controlled by suitable software provided with code executable by control hardware, for instance a computer program, at least when the software is run by suitable hardware.

According to a further elaboration, the system is arranged for detecting first echoes El at a first measurement time Tl within a period of 0-20 ms after above-mentioned of excitation, in particular at a first measurement time Tl which is within a period of approximately 0-10 ms. In a further elaboration, the system 1 can detect an above-mentioned second echo E2 for instance later than 20 ms after above-mentioned moment of excitation, in particular at least approximately 30 ms after above-mentioned excitation moment, and more in particular at least approximately 40 ms after this. To a skilled person, it will be readily apparent that various other first and second measurement times are possible, which, for instance, depends on the magnetic field strength at which measurements are done, on the configuration of the receiver or receivers measurements are done with, and on the manner in which measurements are done. Thus, a method according to the invention, for instance utilizing above-mentioned system 1, comprises:

- directing magnetic dipoles of an object to be imaged to a magnetic field; - exciting the magnetic dipoles at at least one moment;

- detecting a first echo El and at least one respective second echo E2, as a result of the resonance of the excited dipoles (from respective voxels); wherein the first echo El is detected so shortly after the moment of excitation that this echo contains substantially no T≥* information, that the second echo E2 is only detected at a later time so that the second echo does contain T2* information, and that the first echo is used in a data processing process to reduce or eliminate artefacts possibly present in the (second) echoes, in particular spin history-dependent artefacts. As already mentioned, for instance both first and second echoes El, E2 can be determined for a large number (i.e. a plurality of) voxels, to provide respective MBI images of an object.

In a non-limitative example, above-mentioned first echo El may, for instance, be detected at a first measurement time Tl within 0-20 ms after above-mentioned moment of excitation, in particular within a period of approximately 0-10 ms. Above-mentioned second echo E2 may, for instance, be detected at or from a moment later than 20 ms after above-mentioned moment of excitation, in particular at a second measurement time T2 at least approximately 30 ms after above-mentioned excitation moment and more in particular at least approximately 40 ms after this (a respective second measurement time is designated by T2 in the Figures).

In particular, the data processing process may comprise a general linear regression model, where the first echo/echoes El is/are used as a regressor for the second echo/echoes E2, utilizing the relation:

Y=βX+ε (1) where Y comprises above-mentioned observed second echoes E2, X is a matrix comprising above-mentioned first echoes El, β is a weighting factor and ε corrected second echoes E2'. In particular, β is a coefficient vector, representing the relative amount of variance of the regressors (in this case the first echoes El).

Preferably, further, movement correction (for instance via MCFLIRT) and/or baseline drift removal is applied to the (obtained) echoes (in particular prior to the above-mentioned data processing). Here, the various processes are carried out in the image domain, which will be readily apparent to a skilled person. The brain regions (volumes, or voxels) of the different scans El, E2 are, for instance, preferably aligned at one reference, or the scans El, E2 are aligned with one another, which will also be readily apparent to a skilled person.

An example of the method is shown in Fig. 3. Fig. 3 schematically shows the result of a first measurement Ml, which comprises an MRI scan (with various voxels) which does not yet comprise T2* information. The first measurement Ml is, for instance, taken at (at least coming from) a first measurement time Tl directly or shortly after excitation (see above-mentioned times). The time curve of a respective first echo measured with respect to one voxel v from the measurement Ml is designated by El(V).

In addition, Fig. 3 shows the result of a second measurement M2, which comprises an MRI scan which does comprise T2* information. The second measurement M2 is, for instance, taken at (at least coming from) a second measurement time Tl. The time curve of a respective detected second echo part E2(v) from the second measurement M2, which second echo part E2(v) relates to the same voxel v as above-mentioned voxel of which the first echo part El(v) is given, is designated by E2(v).

In this manner, of all (or of a large number of) voxels of the measurements Ml and M2, first and corresponding second echoes El, E2 can be obtained. As Fig. 3 shows, these measurements El, E2 are then inputted in the general linear regression model (see equation 1), (preferably after above-mentioned signal processing) and the correct second echo ε (at least E2¹) can be determined. It is found that, in the thus corrected second echo (Le. the corrected image), the number of movement-related artefacts is reduced significantly.

As already mentioned, in an alternative method, the first and second echoes El, E2 may, for instance, be used in an independent component analysis to reduce such artefacts.

In an experiment, MRI brain scans were made of 7 subjects, in a 1.5 T

Siemens Sonata. Here, the following parameters were used: first echo measurement time Tl=IO ms, second measurement time T2=40 ms, repetition time=3050 ms. Here, each time, 31 slices were measured. The voxel dimension was 3.5x3.5x3.5 mm. The subjects were visually stimulated at particular times during the measurement, utilizing flashing checkerboard patterns and optional arrows, and were to move their heads upon perception of an arrow. Fig. 4 shows the three different stimulation patterns, namely a first pattern A (rest, baseline image), a second pattern B (checkerboard for visual stimulation) and a third pattern C (checkerboard with arrow, for stimulation and movement) which were applied to the subjects, during the making of the respective MRI scans.

For each MRI scan, a first and a second echo El, E2 were recorded, namely at 10 ms and 40 ms. Then the corrected second echo images E2* were determined on the basis of the above-described linear regression method, to obtain corrected MRI images.

Statistical analysis was carried out of the non-corrected second echoes E2, and of the corrected data set E2' (via the so-called, generally known two-tailed T test, with p<5 -10*). Here, the number of activated voxels was determined, and max/average t values were determined in activation ROIs and movement ROIs (ROIs, regions of Interest), defined by significant voxels resulting from t tests of the second stimulation patterns B compared to the rest patterns A and the third stimulation patterns C compared to the rest patterns A, respectively. In addition, voxels overlapping with the activation ROIs were removed from the movement ROI. The results are shown in Tables 1 and 2, and in Figβ. 5-7.

TABLE 2. Results for 7 persons, with respect to the activation

It clearly follows from Tables 1 and 2 that the number of movement artefacts is reduced significantly (by 1/3) by the present method. As Table 2 shows, observed activation as such is, on the whole, not substantially influenced, and can even increase. In other words: in the corrected echo E2\ movement artefacts are considerably reduced while real activation is maintained and even increases.

Fig. 5 shows 12 digitally processed MRI slices of the activation images of one of the subjects, after visual stimulation (by showing the second pattern B to the person) but without movement. Arrows q indicate the observed activated regions.

Fig. 6 shows activation images of non-corrected E2 values at which movement took place as well (to this end, the third pattern C was shown to the person), and Fig. 7 shows respective images obtained with the corrected echo values E2\ Arrows r indicate regions where many artefacts are present. It is clearly visible that false activation has decreased considerably after the correction.

It follows from the above that the present method can prevent object-related artefacts as a result of object movement well, even if object movement is a result of the experiment. An idea behind the invention is to use a very early echo (El) which contains no or very little BOLD information but can contain spin history-related Io fluctuations, which fluctuations have, for instance with an above-mentioned regression, no or hardly any influence on the desired information from a later echo (E2) which does contain BOLD contrast.

In the case of the movement correction, the movement may particularly be correlated with the experimental paradigm and therefore the expected activation, where movement of the head is, for instance, not desired, but is unavoidable. The present method can at least partly remove such movement which is correlated with the experimental paradigm, while maintaining the activation (see also the above-described experiment and results).

It goes without saying that the invention is not limited to the exemplary embodiments described. Various modifications are possible within the framework of the invention as set forth in the following claims.

For instance, successively, different second echoes (i.e. images, scans) in a series may be detected and corrected for errors, utilizing at least one or more of above-mentioned first echoes. Further, the present invention is, for instance, usable in combination with other correction methods, for instance with an above-mentioned regression model with the movement parameters, and/or with a method comprising monitoring object movement with movement sensors and adjusting the settings of the MRI system thereto.

Claims

1. A method of reducing artefacts, in particular movement-related artefacts, in MRI images, comprising:

- detecting a first echo and at least one second echo, caused by the resonance of the excited dipoles; wherein the first echo is detected so shortly after the said at least one moment that this echo contains substantially no T2* information, that the second echo is only detected at a later time so that the second echo does contain T2* information, and that the first echo is used in a data processing process to reduce or eliminate artefacts possibly present in the second echoes, in particular spin history-dependent artefacts.

2. A method according to claim 1, wherein the artefacts possibly present in the second echoes are spin history-dependent artefacts.

3. A method according to claim 1 or 2, wherein the first echo provides spin history information, which information is used in the data processing process to reduce spin history-dependent artefacts.

4. A method according to any one of the preceding claims, wherein said first and second echo each comprise a respective MRI image or MRI scan.

5. A method according to any one of the preceding claims, wherein, further, movement correction, for instance alignment, is applied to the echoes.

6. A method according to any one of the preceding claims, wherein the first echo (El) contains no or very little BOLD information, but can contain spin history-related effects, in particular intensity fluctuations.

7. A method according to any one of the preceding claims, wherein said first echo is detected within a period of 0-20 me after said moment of excitation, in particular within a period of approximately 0-10 ms.

8. A method according to any one of the preceding claims, wherein said second echo is detected at least at a moment later than 20 ms after said moment of excitation, in particular at least approximately 30 ms after said excitation moment and more in particular at least approximately 40 ms after this.

9. A method according to any one of the preceding claims, wherein movement of an object is or has been an unavoidable result of a test or experiment, wherein said data processing process at least partly removes artefacts related to such movement.

10. A method according to any one of the preceding claims, wherein the data processing process comprises a general linear regression model, utilizing the relation:

Y=βX+ε

wherein Y comprises said detected second echoes, X is a matrix comprising respective said detected first echoes, β is a weighting factor and ε corrected second echoes.

11. A method according to any one of claims 1-9, wherein said data processing process comprises an independent component analysis process.

12. A method according to any one of the preceding claims, wherein said artefacts comprise object movement-related artefacts and/or artefacts related to physiological noise.

13. A method according to any one of the preceding claims, wherein the first and second echoes (El, E2) are each determined for a plurality of voxels.

14. A system for generating MRI images, evidently intended and suitable for carrying out a method according to any one of the preceding claims, wherein the system is provided with a magnetic resonance system for polarizing magnetic dipoles of an object to be imaged and exciting them at at least one moment, and with one or more detectors arranged for detecting a first echo and at least one second echo, caused by the resonance of excited dipoles, wherein, during use, the first echo is detected so shortly after the moment of excitation that this echo contains substantially no T2* information, and the second echo is only detected at a later time so that the second echo does contain T2* information, wherein the system is provided with a data processor arranged for using the first echo in a data processing process to reduce or eliminate artefacts possibly present in the echoes, in particular spin history-dependent artefacts.

15. Software provided with code executable by hardware, for instance a computer program, arranged for carrying out a method according to any one of claims 1-13 if the software is run by suitable hardware.

16. Use of a first MRI echo to reduce or eliminate spin-related artefacts in at least one second MRI echo, wherein:

• the first echo (El) is detected so shortly after a particular excitation moment that this echo contains substantially no T2* information;

- the second echo (E2) is only detected at a later time so that the second echo does contain T2* information; and

- wherein the first echo (El) contains spin history-related information (Lo), which information is used in a data processing process to reduce the said spin history-dependent artefacts in the at least one second echo (E2).