EP1183547A1

EP1183547A1 - Computer for analyzing data from measurements of nuclear magnetic resonance, nuclear magnetic resonance tomograph provided with said computer, and method for analyzing data from measurements of nuclear magnetic resonance

Info

Publication number: EP1183547A1
Application number: EP00936673A
Authority: EP
Inventors: Valeri Kiselev; Stefan Wiese; Stefan Posse
Original assignee: Forschungszentrum Juelich GmbH
Current assignee: Forschungszentrum Juelich GmbH
Priority date: 1999-05-22
Filing date: 2000-05-11
Publication date: 2002-03-06
Also published as: DE19923587B4; DE19923587A1; WO2000072035A1

Abstract

The invention relates to a computer for analyzing data acquired using nuclear magnetic resonance tomography, whereby the data contains at least one relaxation signal of a sample. According to the invention, the computer is designed such that it works with at least one analyzing means which separates the data into at least two parts each having a different dependency on an echo time TE.

Description

description

Computer for evaluating data from measurements of nuclear magnetic resonance, nuclear magnetic resonance tomograph equipped with the computer and method for evaluating data from measurements of nuclear magnetic resonance

The invention relates to a computer for evaluating data from measurements of nuclear magnetic resonance, the data containing at least one relaxation signal of a sample.

The invention further relates to a nuclear magnetic resonance scanner and a method for evaluating data from measurements of nuclear magnetic resonance, at least one relaxation signal of a sample being determined.

Nuclear magnetic resonance (NMR) is used to obtain a contrast image of an object or spectroscopic information about a substance. Magnetic resonance imaging (MRI) and magnetic resonance spectroscopy (MRS) make it possible to investigate regional hemodynamics with changes in blood volumes and blood conditions as well as changes in metabolism in vivo depending on brain activity, see: S. Posse et. al .: Functional Magnetic Resonance Studies of Brain Activation; Seminars in Clinical Neuropsychiatry, Vol. 1, No 1, 1996; p. 76-88.

In medical research in particular, there is a need to obtain information about brain activity by measuring blood flow or deoxyhemoglobin concentration changes. The neuronal activation manifests itself in an increase in blood flow in activated brain areas, with a decrease in the deoxyhemoglobin concentration. Deoxyhemoglobin (DOH) is a paramagnetic substance that reduces the magnetic field homogeneity and thus the

Signal relaxation accelerates. If the DOH concentration drops due to brain activity that triggers blood flow, the signal relaxation is modulated in the active areas of the brain. The protons of hydrogen in water are primarily excited. A localization of

Brain activity is made possible by using an investigation using functional NMR methods that measure the NMR signal with a time delay (echo time). This is also known as susceptibility-sensitive measurement. The biological mechanism of action is known in the literature under the name BOLD effect (Blood Oxygenation Level Dependence - Effect) and leads to susceptibility-sensitive magnetic resonance measurements with a field strength of a static magnetic field of, for example, 1.5 Tesla up to approximately 10% fluctuations in the Image brightness in activated brain regions. Instead of the endogenous contrast agent DOH, other contrast agents can also occur which cause a change in the susceptibility. NMR imaging methods are used to select layers or volumes which, under the appropriate irradiation of high-frequency pulses and the application of magnetic gradient fields, provide a measurement signal which is digitized and stored in a two- or three-dimensional field in the measuring computer.

The desired image information is obtained (reconstructed) from the raw data recorded by means of a two-dimensional or multidimensional Fourier transformation. A reconstructed slice image consists of pixels (= picture element), a volume data set consists of voxels (= volume element). A pixel is a two-dimensional picture element, for example a square. The image is composed of the pixels. A voxel is a three-dimensional volume element, for example a cuboid, which - due to measurement technology - has no sharp boundaries. The dimensions of a pixel are on the order of 1mm ² , those of a voxel of 1mm ³ . The geometries and dimensions can be variable.

Since a strictly two-dimensional plane can never be assumed for layered images for experimental reasons, the term voxel is often used here, which takes into account that the image planes extend into the third dimension.

By comparing the measured signal curve in each pixel with the time curve of a model function, a stimulus-specific neuronal activation can be detected and spatially localized. A stimulus can be, for example, a sensory, acoustic, visual or olfactory stimulus, as well as a mental or motor task. The model function, or the model time series, describes the expected signal change in the magnetic resonance signal as a result of neuronal activation. For example, these can be derived from a paradigm of the respective experiment using empirical rules. It is essential to delay the time

Consider model function versus paradigm (sluggish blood flow response to neuronal activation). It is already known how brain activation can be represented by activation images obtained from magnetic resonance imaging data. The activation images can even be calculated and displayed in real time, which means that a data record can be converted into an image before the next data record is measured. The time interval is about 1 to 3 seconds.

Such calculation and reproduction of the activation images in real time is in the US patent

5,657,758. This method is characterized in that it enables a high temporal and spatial resolution.

Another known method is in the articles.

Jezzard, P. et al., Proc. SMRM 1993, p. 1392; Biswal, B. et al., MRM 34 (1995) p. 537 and Purdon, P. et al., Proc. ISMRM 1998, p. 253. A measurement signal and a measurement paradigm are used in this method. Both signals are subjected to a Fourier transformation.

The known methods evaluate a similarity between the signal of the paradigm and the measurement data.

The invention has for its object to carry out a generic method so that the highest possible contrast-to-noise ratio is achieved.

According to the invention, this object is achieved in that a generic computer is designed so that the

Computer works with at least one evaluation means, the evaluation means separating the data into at least two parts that differ from one echo time T _E Depend way.

In particular, the invention provides to create a computer with which a fast spectroscopic imaging method can be implemented which changes the NMR

Signal relaxation with a time constant T ₂ ^* = on

determined several times after a suggestion.

The spectroscopic imaging method is preferably an echo planar imaging method, in particular a repeated two-dimensional echo imaging method, which consists of a repeated application of a two-dimensional echo planar image coding. Spatial coding takes place in the shortest possible time, which is repeated several times during a signal drop and is preferably 20 to 100 ms. The repetition of the echo planar coding several times during a signal drop shows a course of the signal drop in the sequence of reconstructed individual images.

A convenient conventional echo planar method is called EPI (Echo Planar Imaging). An advantageous implementation of the method according to the invention takes place with TURBO-PEPSI, where PEPSI stands for Proton-Echo-Planar-Spectroscopic-Imaging.

The number of images that are encoded during the signal drop depends on the relaxation time and the

Coding time Δt for a single picture.

The computer is preferably used to evaluate data from nuclear magnetic resonance tomography, wherein the data contain at least one relaxation signal of a sample and the data are separated into components which depend on an echo time T _E and at least one other component which does not depend on the echo time T _E and that the signals are based on the echo time T _E depend, are detected as activation signals.

A noise signal can be determined by the computer working with at least one evaluation means which separates the data into at least a portion which depends on an echo time T _E and at least one other component which does not depend on the echo time T _E and wherein the evaluation means detects the signals which depend on the echo time T _E as activation signals.

A separation of different components of a function to be examined can be determined by ascertaining signals which have a different dependence on the echo time T _E. It is possible, for example, to have an amplitude S ₀ of one

Separate time constants T ₂ ^* and / or from a noise signal g.

The invention also relates to a nuclear magnetic resonance tomograph which contains at least one computer according to the invention.

The invention further provides for a method for evaluating data from nuclear magnetic resonance tomography, wherein at least one relaxation signal of a sample is determined, to be carried out in such a way that the data are separated into at least two parts with a different dependence on an echo time T _E. The method is preferably carried out in such a way that intensity values of the measured data for the same echo times are recorded in at least two different recordings of the relaxation signal and that a dependence of the intensity values on the echo time T _E is subsequently recorded and that the relaxation signal is separated into portions which differ from one another Have dependencies on the echo time T _E.

It is preferred to carry out the method in such a way that the relaxation signal is divided into a component that depends on an echo time T _E and at least a component that does not depend on the echo time T _E and that component that depends on the echo time T _E depends on how an activation signal is detected.

It is particularly advantageous here that at least one signal is determined which is proportional to T _E exp (-T _E / T ₂ ^* ), the value of T ₂ ^{* being determined} in particular by a preferably separate fit procedure on the same data.

It is particularly expedient here that T ₂ ^{* is} calculated using the formula S = S ₀ exp (-T _E / T ₂ ^* ) + g.

It is also advantageous to carry out the method in such a way that statistical fluctuations of ΔT ₂ ^{* are} determined.

It is particularly expedient here that a standard deviation σ (ΔT ₂ ^* ) is calculated. It is also advantageous that a quotient σ (ΔT ₂ ^* ) / T ₂ ^{* is} formed and recorded as a measure of an activity.

It is particularly expedient that a statistical deviation of a starting intensity S _{0 is} determined.

It is advantageous here that a standard deviation σ (S ₀ ) is calculated.

It is preferred that a quotient σ (S ₀ ) / S _{0 is} calculated.

It is particularly preferred to carry out the method in such a way that a statistical fluctuation from a noise signal g is determined.

It is particularly advantageous that a standard deviation σ (g) of g is formed.

Furthermore, the method is preferably carried out in such a way that the recorded data are recorded in an at least two-dimensional field, one field axis (DTE) recording echo times T _E and another field axis (DTR) repeating excitations at a time interval of T _R reproduces.

It is particularly expedient here that σ (ΔT ₂ and σ (g) are determined by the following steps:

(i) adapting signals averaged over the other field axis (DTR) to an exponential decay depending on the one field axis (DTE) and determining So and T ₂ ^* ; (ii) Calculation of σ (ΔS ₀ ), σ (ΔT ₂ ^* 'and σ (g) for several Voxel and various T _E followed by averaging these values over at least a region under investigation (ROI); (iii) Customize

and determination of σ (ΔS) / So as a function of T _E.

It is particularly advantageous here that the expression <ΔS ₀ ΔT ₂ ^* > = 0 is set when adapting σ (ΔS) / S ₀ .

Further advantages, special features and expedient developments of the invention result from the subclaims and the following illustration of preferred exemplary embodiments of the invention on the basis of example calculations, drawings and a table.

From the drawings shows:

1 multi-echo sequence with several measurement sequences, each of which follows a spin excitation (*) and with detection of several echo times T _E ,

Fig. 2 is a schematic diagram illustrating a

Method of processing data separately for each of the echo times,

3 shows an experimental difference signal of a functional relaxation time change in one selected picture element depending on the measuring time after a signal excitation,

4 ΔS from various voxels averaged over a few ROIs as a function of T _E for 2 representative persons,

5 shows a detection of in an upper partial image

Brain activation in four steps by means of a conventional imaging method and in a lower partial image a detection of brain activation using a method according to the invention.

The table shows a summary of experimental sample data.

1 shows a multi-echo sequence with a plurality of measurement sequences, each of which follows a spin excitation (*) and with detection of several echo times T _E.

The measurement sequences of the multi-echo sequence were determined using the turbo? Ξ? SI method. Each of the measurement sequences contains twelve echo signals with echo times that are between 12 and 213 ms. The echo times were each recorded as an 18.3 ms time interval ΔT _E.

The specified values for the echo times and the time intervals are each adapted to the speed of the data processing. In particular, with a further improvement in scanner technology, the number of echo signals can be increased and the time intervals ΔT _{E can be} shortened. FIG. 2 shows a schematic diagram which shows how a signal is acquired from different measuring sequences at a first echo time or at a second or subsequent echo time.

In the curve shown in FIG. 3, a measurement signal σ (S) is recorded as a function of the echo time. Here is a principle with a fit procedure for dividing the measurement signal σ (S) into contributions that depend on T ₂ * and noise independent of T _E. The measurement signal σ (S) is composed of a component that depends on an amplitude S ₀ , a component that depends on a relaxation time T ₂ * and a constant noise signal g.

In particular, the invention provides for a distinction to be drawn between activation signals and noise by analyzing a time profile of the measurement data and / or its statistical distribution.

The evaluation method according to the invention is experimentally checked, for example, using magnetic resonance imaging examinations of the brains of test subjects. A light source, in particular a matrix of luminescent diodes (Light Emitting Diode LED), was positioned in the immediate vicinity of the test person's face and stimulated to produce signal flashes. The excitation frequency is 8 Hz. The signal flashes act on a time interval of several seconds, for example 5 seconds, which is synchronized with a carrier signal of a scanner, which is followed by an approximately equally long rest interval. The scanner is a Vision 1.5 Tesla full body scanner from Siemens Medical Systems, Erlangen, Germany Magnetic field gradients of 25 mT / m. Such a scanner is able to switch gradient fields within approximately 600 μs.

TURBO-PEPSI (Proton-Echo-Planar-Spetroscopic-Imaging) was used as the spectroscopic imaging method.

Data is adjusted according to the exponential function:

-7V7;

S = S _Q e ^X E '2

using a non-linear least-square fit.

A distinction between activation and noise using multi-echo fMRI is shown below.

The detection of physiological noise (for example caused by a heartbeat) requires a stationary frequency spectrum, sufficient temporal resolution and prior knowledge of the spatial and temporal

Noise characteristics. According to the invention a new method for the differentiation between BOLD-related variations and other fluctuations of the MR signal (such as caused by thermal noise), it is proposed that does not need any prior knowledge of a stimulation paradigm ^'. The method is based on a single-shot ^ ltiecho sequence, such as that described in the article by Posse, S. et al. PROC. ISMRM 1998, p. 299 Turbo PEPSI technology shown. Full reference is made to this publication.

After signal excitation, its relaxation behavior is shown in equidistant time intervals T _E recorded. This is repeated several times with a time interval of T _R seconds. In such an experiment, the signal of each voxel forms a 2-dimensional field with the echo times T _E in one direction (DTE) and the repetitions at a distance T _R in the other direction (DTR). The relaxation is assumed to be monoexponential, S = S ₀ exp (-T _E / T ₂ ^*, + g, with a hardware-dependent noise g, which we can see as white in both domains, DTE and DTR. The values S ₀ and T ₂ ^* are constant in DTE, but vary in DTR: S ₀ due to hardware instabilities or blood flow effects and T _R due to subject stimulation, variations in T ₂ ^* show changes in local blood flow, for relatively small changes ΔS ₀ and ΔT ₂ ^* the signal changes can be formulated as follows:

where <A> and σ (A) correspond to the mean and standard deviation of a quantity A in DTR. A further analysis depends on the current size of the terms used in [1]. It is expedient for the experimental conditions ΔS _{0 to} be negligible both in the resting and in the activation phases (except in the sagital sinus). The quantities σ (ΔT ₂ ^* 'and σ (g) are determined as follows: (i) adapting the signal averaged over the DTR to the -nor-oexponential decay as a function of DTE to determine S ₀ and T ₂ ^* ; ( ii) Calculation of σ (ΔT ₂ ^* ) and σ (g) for each voxel and each T _E and averaging these values about the region of interest (ROI); (iii) fitting [1] with ΔSo = 0 to these values as a function of T _E. This is possible because local brain activation shows an increase in T ₂ ^* , which has a characteristic T _E dependence, proportional to T _E e ^{~ τ} _E ^τ ₂ ^* , whereas the contribution of white noise does not depend on T _E ( see pictures). The T _E dependence of the signal outside the brain is approximated by a constant. To validate this method, the contribution of white noise is compared to the noise outside the brain, taking into account that σ (g) outside the brain is reduced. For a Gaussian distribution, this reduction factor is 0.6028.

Visual stimulation experiments on 4 healthy people were carried out on a Siemens Vision-1, 5-Tesla-

Scanner. With a multi-layer turbo PEPSI sequence, 12

EPI images (matrix size: 64 x 32 pixels, pixel size: 3 x 6 mm2) of a single FID, 90 ° flip angle acquired at echo times from 12 to 228 ms. A conventional correlation analysis using the software package Stimulate was carried out using a box car

Reference vector.

4 shows ΔS from various voxels averaged over a few ROIs as a function of T _E for 2 representative persons. The variability of all values across ROIs was small (10-20%). The ROIs were in the visual cortex (vc), in the motor cortex (mc), in the white matter (w) and outside the brain, bypassing areas outside the brain known as ghosting (out). The

Filter results from [1] are summarized in the table, where the abbreviated ROI designations indicate the number of voxels in brackets follows, the mean correlation coefficient over an ROI, σ (g) the ROI outside the brain is normalized to the mean SO of the inner ROIs and the errors in all values are defined as a standard deviation.

Table 1

ROI ξ σ (ΔT ₂ ^* ) / T ₂ (%) σ (g) / S ₀ (%) vc (20) - 0.62 ± 0.21 4.3 ± 0.1 0.75 ± 0.05 mc (20) -0.11 ± 0.14 0.26 ± 0.16 0.79 ± 0.05 wm (21) -0.009 ± 0.19 -0.001 ± 51 0.93 ± 0.07 out (21) -0.19 ± 0.11 not fitted 0.66 ± 0.01 vc (28) 0.67 ± 0.12 3.6 ± 0.1 0.42 ± 0.07 mc (32) -0.22 ± 0.14 0.6 ± 0.8 0.72 ± 0.06 wm (32) -0.29 ± 0.06 0.4 ± 1.2 0.64 ± 0.06 j out (38) -0.12 ± 0.25 not fitted 0.45 ± 0.01

The value of σ (ΔT ₂ ^* ) / T ₂ ^{* was} significantly increased in the activated voxels in all persons, whereas there was no significant deviation from 0 in the non-activated voxels. Therefore, this value is deterministic with a negligible stochastic component.

Consequently, σ (ΔT ₂ ^* ) / T ₂ ^* can be used as an indicator of regional brain activity as well as the correlation coefficients of a conventional correlation analysis. In contrast to the latter, σ (ΔT ₂ ^* ) / T ₂ ^* indicates brain activity for any stimulation course, so that knowledge of a paradigm is not necessary. The small variability of this value over the ROIs suggests that the results for individual voxels are similar to those presented here. This allows the creation of σ (ΔT ₂ ^* ) / T ₂ ^* maps. The level of the T _E -independent white noise is very low, which suggests that it stems from the hardware. The So noise is so small that a closer examination of the So noise is difficult due to the presence of white noise.

The invention provides a method for distinguishing between activation, in particular brain activation, and noise, with no correlation analysis being required. Of course, the invention can also be used in combination with a correlation analysis, such as for example a calculation of correlation coefficients, Z-scores, or an application of the t-test, in order to check the results found in this way. However, a correlation analysis with two different measurements, one with stimulation and the other without stimulation, is not necessary. The inclusion of a correlation analysis, in which correlation coefficients are determined between the time course of the stimulation (“reference vector”) and signal changes in pixels of the image, can, however, be used for comparison purposes.

High values of the correlation coefficient found in this way can be viewed as an activity indicator and, for example, can be reproduced as additional information in a graphical representation of the measurement data in slice images or volume images.

The invention is particularly suitable for use in areas in which complicated activations take place. Therefore, the method and the computer according to the invention are particularly suitable for analyzing higher cognitive brain functions such as emotions, memory and imagination. The invention has a number of advantages. This includes optimizing the measurement sensitivity for a quantitative measurement of the relaxation time and the qualitative change in relaxation time. This makes it possible to use imaging with the highest possible bandwidth (shortest coding time) for the least possible spatial distortion and to achieve maximum measurement sensitivity by measuring an optimal number of codings after signal excitation.

The evaluation method can be used in real-time measurements in order to analyze the relaxation changes immediately.

The evaluation methods according to the invention are also particularly versatile. It has proven to be expedient to use a summation or, what is even more advantageous, a .weighted summation, which can be done at a higher speed and without loss of sensitivity compared to curve fitting. A summation, or a weighted summation, has the advantage that it represents a particularly robust evaluation method.

All subjects showed strong activation in the primary visual cortex (Vx) and in neighboring areas. The observed changes in the functional signal measured with TURBO-PEPSI are up to 10%, depending on the relaxation time T ₂ ^* , the position and the respective test subject.

The excitation has a maximum near TE = T ₂ ^* . at A comparison of EPI and TURBO-PEPSI images with TE = 72.5 ms revealed very similar activation images.

The gain in sensitivity is particularly advantageous for real-time measurements, because even with a few

Measured values a change in relaxation can be determined effectively. In summary, it can be said that an optimal sensitivity at different magnetic field strengths is achieved by multi-echo detection of the difference signal.

In addition, the invention can be used both in echo planar imaging (echo planar imaging EPI), in phase-coded imaging methods and in spectroscopic imaging methods.

The examples shown explain the computer and the evaluation method using NMR measurements on the human brain. Of course, both the computer and the magnetic resonance tomograph can also be used

Evaluation methods can also be used to examine other samples of living or non-living material.

Claims

Patent claims

1. Computer for evaluating data from nuclear magnetic resonance, the data being at least one

Contain relaxation signal of a sample, characterized in that the computer works with at least one evaluation means that separates the data into at least two parts with a mutually different dependence on an echo time T _E.

2. Computer according to claim 1, characterized in that the evaluation means separates the data into at least one component which depends on an echo time T _E and into at least one other component which does not depend on the echo time T _E and wherein the evaluation means the signals depend on the echo time T _E , detected as activation signals.

3 Nuclear resonance tomograph, so that it contains at least one computer according to one of claims 1 or 2.

4. Method for evaluating data from nuclear magnetic resonance, whereby at least one relaxation signal of a sample is determined, characterized in that the data are separated into at least two parts with a mutually different dependence on an echo time T _E.

5. The method according to claim 4, characterized in that intensity values of the measured data are recorded and separated into at least two mutually different dependencies on the echo time T _E.

6. The method according to claim 5, characterized in that an extent of a statistical variation in the intensities is recorded.

7. The method according to claim 6, characterized in that a standard deviation of the intensities is determined.

8. The method according to one of claims 4 to 7, characterized in that the relaxation signal is divided into at least one portion which depends on the echo time T _E and into at least one portion which does not depend on the echo time T _E.

9. The method according to any one of claims 4 to 8, characterized in that at least one signal is determined which is proportional to T _E exp (-T _E /T ₂ ^* ).

10. The method according to claim 9, characterized in that T ₂ ^* is determined using the formula S = S ₀ exp (-T _E /T ₂ ^* ) +g.

11. The method according to one or more of claims 4 to 10, characterized - shows that statistical fluctuations of ΔT ₂ ^* are determined.

12. The method according to claim 11, d a d u r c h g e k e n n z e i c h n e t that a

Standard deviation σ(ΔT ₂ ^* ) is determined.

13. The method according to claim 12, characterized in that a quotient σ(ΔT ₂ ^* )/T ₂ ^* is formed and recorded as a benchmark for an activity.

1 . Method according to one or more of claims 4 to 13, characterized in that a statistical deviation of a starting intensity S ₀ is determined.

15. The method according to claim 14, characterized in that a standard deviation σ (ΔS ₀ ) is determined.

16. The method according to claim 15, characterized in that a quotient σ(ΔS ₀ )/S ₀ is determined.

17. The method according to one or more of claims 4 to 16, so that a statistical fluctuation of a noise signal g is determined.

18. The method according to claim 17, characterized in that a Standard deviation σ(g) of g is formed,

19. The method according to one or more of claims 4 to 18, characterized in that the recorded data are recorded in an at least two-dimensional field, with one field axis (DTE) recording echo times T _E and another field axis (DTR) recording repetitions of excitations in one time distance from T _R represents.

20. The method according to claim 19, characterized in that σ (ΔT ^* ' and σ (g) are determined by the following steps: (i) adapting signals averaged over DTR to an exponential decay as a function of DTE and

Determination of S ₀ and T ₂ ^* ; (ii) Calculation of σ(ΔT ₂ ^* ' and σ(g) for several voxels and different T _E with subsequent averaging of these values over at least one to be examined

region (ROI); (iii) Customize

σ(AS)

and determining σ(ΔS/So) as a function of T _E .

21. The method according to claim 20, characterized in that when adjusting σ (ΔS) / S ₀ , the expression <ΔS ₀ ΔT ₂ ^* > = 0 is set.