JP2003116811A - Spectrum estimating method from waveform data, magnetic resonance image-photographing method using the same, magnetic resonance image-photographing system using the same, and program to execute these methods on computer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a spectrum of a high resolution using a autoregressive model from observation data of a small S/N ratio. SOLUTION: The observation data of water and the observation data of a metabolite are obtained, and the phase of the observation data of the metabolite is correlated. Then, the observation data of water is transformed by the fast Fourier transformation to obtain the spectrum. For such a spectrum, the half-amplitude value is measured. A noise-removing filter function which is applied to the observation data is determined from the half-amplitude value. The noise-removing filter function is applied to the observation data of the metabolite, and wave-form data is obtained. Then, the obtained wave-form data is transformed by the fast Fourier transformation, and the S/N ratio of the obtained spectrum is measured. Then, a term number (model degree) for a polynomial of the autoregressive model is determined from the S/N ratio. Regarding the obtained wave-form data, the spectrum wherein a frequency is made a variable can be obtained using the autoregressive model, by using the polynomial comprising the determined term number.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a spectrum estimation method using a so-called autoregressive model, and more particularly to S / S.
A spectrum estimation method for accurately performing spectrum estimation on observation data with a low N ratio, a magnetic resonance imaging method using the spectrum estimation method, a magnetic resonance imaging apparatus using the spectrum estimation method, and these methods executed on a computer Regarding programs to do.

[0002]

2. Description of the Related Art Conventionally, an object to be imaged is placed in a magnetic field atmosphere, and an NMR signal emitted from the object to be imaged due to a nuclear magnetic resonance phenomenon is observed so as to be captured as waveform data. Magnetic resonance imaging that converts the data related to the spectrum, which is the signal intensity distribution with the frequency as a variable, to image the internal structure of the imaging target (hereinafter,
A device called "MRI" is known. In order to improve the resolution of the image captured by the MRI apparatus,
In the process of converting waveform data composed of NMR signals into a spectrum, it is necessary to accurately specify the peak frequency.

The fast Fourier transform is generally used as a method for knowing what spectrum the waveform data has from the waveform data obtained by observation. The fast Fourier transform is the intensity x (at _N times t ₁ , t ₂ , t ₃ , ..., T _i , ..., t _N of the observed waveform data, which are discretized at intervals Δt. This is an algorithm based on performing discrete Fourier transform after picking up t _i ).
When the spectrum is estimated using the Fast Fourier Transform, the frequency distribution of the observed wave can be easily obtained from the observed waveform data, and if the waveform data has periodicity, the peak at a certain frequency in the spectrum Appears, and the frequency of each wave forming the waveform data can be obtained.

However, when spectrum estimation is performed by the fast Fourier transform, the accuracy of estimating the peak frequency is limited, so that there is a problem that an error occurs over a certain range Δω. Specifically, using the time interval Δt between the data picked up from the waveform data and the number of observation points N, Δω = 1 / (NΔt) ... (1) Expression (1) is established based on any algorithm as long as spectrum estimation is performed by the fast Fourier transform. Therefore, the spectrum estimation by the fast Fourier transform has a problem that it has uncertainty in the frequency by Δω.

As a method that does not cause such a frequency error, a method of spectrum estimation using an autoregressive model is known (S. Haykin. Ed, Nonlinear Methods of Sp.
ectral Analysis, Springer-Verlag, 1979). The autoregressive model, the time t _n the data x _n in than t _n before time _{t n-2, t n-} 3, t n-4, that is expressed by linear combination of the data in ... Say. That is, the spectrum estimation using the autoregressive model means that the data x _{n at} time t _n is x _n = a ₁ x _n-1 + a ₂ x _n-2 + a ₃ x _n-3 + ... + a _M x _nM ... On the basis of being expressed in the form of (2), z conversion is performed using an arbitrary number z for x _n expressed by the formula (2),
This is a method for estimating the spectrum by obtaining data on the distribution of signal strength with the frequency as a variable.

FIG. 8 (a) shows a case of a certain phenomenon.
It is time series information indicating the signal strength for a second. And
FIG. 8B is a spectrum obtained by performing fast Fourier transform on the time series information of FIG. On the other hand, FIG. 8C is a spectrum obtained by estimating the time series information of FIG. 8A using an autoregressive model. Comparing FIGS. 8B and 8C, the full width at half maximum of the peak is large in the fast Fourier transform and is 31.5H.
The z peak and the 32.5 Hz peak are not separated. Since the portion originally composed of two peaks is represented by one peak, the frequency obtained from FIG. 8B is different from the actual frequency. On the other hand, FIG.
In (c), the peak half-width is narrow and each peak is well separated, so that an accurate frequency value can be obtained.

[0007]

On the other hand, there are problems with spectrum estimation using an autoregressive model. In the autoregressive model, how to determine the number of terms (hereinafter, referred to as “model order”) M on the right side of the equation (2) is important in spectrum estimation. Generally, a method is known in which M, which is the expected value of FPE, which is the expected value of the square of the prediction error when predicting the current value x _n from the past value x _{ni in the} autoregressive model, is the order of the autoregressive model. ing.
Specifically, FPE is represented by the following formula. FPE = (1+ (M + 1) / N) (1- (M + 1) / N) ⁻¹ (Σ (x _n −Σ a _i x _ni )) (3) where Σa _i x _ni is Sum for 1 ≦ i ≦ M,
Σ (x _n −Σa _i x _ni ) is the sum for 0 ≦ n ≦ N.

However, this method is not sufficient either. This is because FPE is merely a criterion for giving a statistically optimum model order M, and does not necessarily lead to a good spectrum for actual observation data. Therefore, empirically, the model order M for obtaining useful frequency data is 1/3 to N of the data points.
The method of making the value of 1/2 is known, and the S / N of the data is
When the ratio is sufficiently large, the graph regarding the spectrum as shown in FIG. 8C can be obtained by using the order M determined by such a method.

However, there is a problem with this empirical rule. That is, for waveform data for which a sufficient S / N ratio cannot be obtained, an accurate spectrum cannot be obtained even if the above method is applied. In particular, in imaging a magnetic resonance image, noise and the like generated from the human body cannot be ignored, and the measurement time cannot be extended considering the burden on the patient. Therefore, it is difficult to obtain waveform data having a high S / N ratio, and it is not appropriate to use the above-described method of determining the model order M for MRI.

Further, when the S / N ratio of the waveform data is small, there are problems other than the above problems. That is, when spectrum estimation is performed using an autoregressive model, not only significant waveform data but also waves forming noise are estimated. Therefore, in the obtained spectrum, the peak of the frequency caused by the waveform data is buried in the peak caused by the noise wave, and it is difficult to utilize it as useful data.

Further, spectrum estimation from waveform data having a small S / N ratio is necessary in various cases such as characteristic analysis of electronic circuits and mechanical systems in addition to the MRI apparatus. Also in these cases, even if spectrum estimation using an autoregressive model is performed by the conventional method,
Unable to get valid data.

The present invention has been made to solve the above-mentioned problems, and in spectrum estimation using an autoregressive model, a method capable of performing spectrum estimation with resolution even with waveform data having a small S / N ratio. An object of the present invention is to provide a magnetic resonance imaging method using the method, a magnetic resonance imaging apparatus, and a program for executing the method on a computer.

[0013]

[Means for Solving the Problems]
In order to achieve the object, the invention according to a first aspect provides a frequency based on a signal strength of a waveform data having time as a variable at a predetermined time and a predetermined number of signal strengths at discrete times until the predetermined time. Is a spectrum estimation method for estimating a signal strength waveform having a variable as a variable, the waveform data deriving step of deriving the waveform data by applying a noise removal filter to the observation data, and the first variable frequency variable from the waveform data. And a determining step of determining the predetermined number from the signal strength and the noise strength of the acquired first signal strength waveform.

According to the invention of the first aspect, since the noise removal filter is applied to the derivation of the waveform data and the predetermined number is determined from the signal strength and the noise strength, the S / N ratio is low. High-resolution spectral estimation can be performed for observation data as well.

The invention according to a second aspect is the invention according to the first aspect, wherein the signal strength of the waveform data having the time as a variable at a predetermined time is a predetermined number at discrete times until the predetermined time. When the signal strength waveform is estimated by the autoregressive model expressed by the linear combination of the signal strengths, the coefficient determination step of determining the coefficient of each term in the linear combination by the least square method is further included. .

According to the invention of the second aspect, since the coefficient of each term in the linear combination is determined by the method of least squares, the observation data having noise is also
The coefficient of each term can be accurately determined.

The invention according to a third aspect is the invention according to the first or second aspect, wherein in the waveform data deriving step, -1 and a frequency obtained by Fourier transforming the observation data are obtained. The observation data is multiplied by a noise removal filter function including an exponential function having an exponential component as a product of a coefficient derived from the half-value width of the maximum peak in the second signal strength waveform as a variable, and It is characterized by deriving waveform data.

According to the invention of the third aspect, the noise removal filter function includes an exponential function, and the exponent component has a coefficient derived from the half width of the peak, and the noise removal filter function is observed. Since the waveform data is derived by multiplying the data, it is possible to accurately remove the noise component of the observation data and use the derived waveform data as a damping oscillation function.

The invention according to a fourth aspect is the invention according to the third aspect, wherein in the waveform data deriving step, when the full width at half maximum of the maximum peak is 0.5 to 3.0 Hz, The value of the full width at half maximum of the maximum peak is the value of the coefficient, and when the full width at half maximum of the maximum peak is 0.5 Hz or less, 0.5 is the value of the coefficient, and the full width at half maximum of the maximum peak is 3 When the value is not less than 0.0 Hz, 3.0 is set as the value of the coefficient.

According to the invention of the fourth aspect, since the exponential component of the noise removal filter function is specifically determined, the noise component can be removed more accurately.

The invention according to a fifth aspect is the first to the first aspects.
In the invention according to any one of the fourth aspect, the determining step is characterized in that the predetermined number is determined from a ratio between the maximum value of the signal strength and the noise strength value in the first signal strength waveform.

According to the invention of the fifth aspect, since the predetermined number is determined from the ratio of the maximum value of the signal intensity and the noise intensity value in the spectrum, the resolution is higher than that when the FPE is used. Spectral estimation can be performed.

The invention according to a sixth aspect is the first to the first aspects.
In the invention according to any one of the fifth aspect, in the determining step, the ratio of the ratio of the maximum value of the signal strength in the first signal strength waveform and the noise strength value to the (-1/2) th power is a constant of proportionality. It is characterized in that the integer part of the value multiplied by is the predetermined number.

According to the sixth aspect of the invention, it is possible to estimate the spectrum with a higher resolution by more specifically determining the predetermined number.

According to a seventh aspect of the invention, the object to be imaged is arranged in a magnetic field atmosphere, and an electromagnetic wave having a center frequency proportional to the magnetic field strength of the magnetic field atmosphere is transmitted to the arranged object to be imaged. Then, the NMR signal obtained from the imaging object is received based on the nuclear magnetic resonance phenomenon, the spectrum is estimated by the autoregressive model based on the waveform data of the received NMR signal, and the estimated spectrum is positioned. In the magnetic resonance imaging method of outputting an image of the internal structure of an imaging target by replacing the waveform data with a luminance removal filter, the spectrum data is derived by applying a noise removal filter to observation data. A step, and a first signal strength waveform having a frequency as a variable is acquired from the waveform data, and the acquired first signal strength waveform is acquired. Characterized in that the signal strength and noise intensity waveform and a determination step of determining the predetermined number.

According to the seventh aspect of the present invention, in the estimation of the spectrum, the noise removal filter is applied and the predetermined number is determined from the signal strength and the noise strength. A high-resolution spectrum is estimated and a high-resolution magnetic resonance image can be formed based on the estimated spectrum.

The invention according to an eighth aspect is the invention according to the seventh aspect, wherein in the step of deriving the waveform data, the waveform data deriving step is performed on the second signal intensity waveform obtained by fast Fourier transforming the observation data. Derivation of the waveform data by multiplying the observation data by a noise removal filter function including an exponential function having a half-value width of a peak caused by hydrogen atoms constituting water molecules existing inside the imaging object as a part of the exponential component It is characterized by doing.

According to the invention of the eighth aspect, with respect to the exponential component in the exponential function included in the noise removal filter function, the half-value width of the peak due to the hydrogen atoms constituting the water molecules present inside the object to be imaged. Since it is included, it is possible to construct a useful noise removal filter function, and it is possible to effectively remove the noise component in the observation data. Therefore, it is possible to capture a magnetic resonance image having a high resolution.

The invention according to a ninth aspect is the invention according to the seventh or eighth aspect, wherein in the determining step, the maximum value of the signal strength in the first signal strength waveform and the noise strength value are calculated. It is characterized in that the integer part of the value obtained by multiplying the value of the power of (-1/2) of the ratio by a proportional constant is set as the predetermined number.

According to the invention of the ninth aspect, by specifically defining the predetermined number of determinations, the spectrum of high resolution can be obtained from the observation data having a lower S / N ratio as compared with the case of using FPE. It is possible to perform estimation and capture a high-resolution magnetic resonance image.

The invention according to the tenth aspect is the seventh aspect.
~ In the invention according to any one of the ninth aspect, from the distribution of the signal intensity obtained by the spectrum estimating step,
The method may further include a detection step of detecting a temperature change inside the imaging object based on a change in frequency difference between a plurality of peaks.

According to the invention of the tenth aspect,
Since the resonance frequency of the peak due to the atoms constituting the metabolite also fluctuates due to the temperature change, the temperature fluctuation inside the imaging target can be detected by measuring the fluctuation range.

The invention according to an eleventh aspect is the first aspect.
The program according to any one of the invention according to the tenth aspect is executed on a computer,
~ The method according to the invention in any one of the tenth aspects can be executed by a computer.

According to the twelfth aspect of the invention, a static magnetic field generating magnet for generating a uniform magnetic field, a gradient magnetic field generating section for generating a magnetic field that fluctuates depending on position and time, and an electromagnetic wave to an object to be imaged. A transmitting unit for irradiating, and a receiving unit for receiving observation data including an NMR signal from the imaging object,
An image forming processing unit for forming a magnetic resonance image, and a magnetic resonance image capturing unit for forming a magnetic resonance image by transmitting an electromagnetic wave having a frequency proportional to the intensity of the uniform magnetic field and the varying magnetic field and receiving an NMR signal. In the video imaging apparatus, the image forming processing unit obtains waveform data composed of time series information based on the observation data and a noise removal filter function, and obtains the derived waveform data by performing a fast Fourier transform. Magnetic resonance forming a magnetic resonance image by estimating the spectrum using an autoregressive model while determining the number of polynomial terms from the signal intensity and noise intensity in the first signal intensity waveform and estimating the spectrum with the frequency as a variable. It is a video image pickup device.

According to the invention of the twelfth aspect,
The image formation processing unit estimates the spectrum using an autoregressive model, derives the waveform data using a noise removal filter function in the spectrum estimation, and determines the number of polynomial terms based on the signal strength and noise strength. Since it is an image pickup device, it is possible to provide a device capable of picking up a high-resolution magnetic resonance image by performing high-resolution spectral estimation using an autoregressive model even with observation data having a low S / N ratio.

The invention according to a thirteenth aspect is the first aspect.
In the invention according to the second aspect, the noise removal filter function includes water molecules present inside the imaging target in a second signal intensity waveform having a frequency obtained by fast Fourier transform of the observation data as a variable. It is characterized in that it includes an exponential function having the half-width of the peak due to the constituent hydrogen atoms as part of the exponential component.

According to the invention of the thirteenth aspect,
Since the noise removal filter function includes an exponential function that includes the half-value width of the peak due to hydrogen atoms in the exponential component, the image formation processing unit can effectively remove the noise component from the observation data, and high resolution It is possible to provide a magnetic resonance imaging apparatus capable of capturing the magnetic resonance image of

The invention according to a fourteenth aspect is the first aspect.
In the invention according to the second or thirteenth aspect, the number of terms of the polynomial is a value obtained by multiplying a ratio of the signal strength and the noise strength in the first signal strength waveform by (-1/2), and a proportional constant. It is characterized in that it is an integer part of the multiplied value.

According to the invention of the fourteenth aspect,
Provided is a magnetic resonance imaging apparatus capable of more appropriately providing a model order to observation data having a low S / N ratio, as compared with a case where the image forming processing unit determines the model order using FPE. can do.

The invention according to a fifteenth aspect is the first aspect.
In the invention according to any one of the second to fourteenth aspects, the image forming processing unit, in the spectrum obtained by the spectrum estimation, changes the temperature of the imaging target based on the change in the frequency difference between a plurality of peaks. It is characterized by detecting.

According to the invention of the fifteenth aspect,
Since the processing means is the magnetic resonance imaging apparatus which detects the temperature variation inside the object to be imaged by detecting the difference in frequency between the plurality of peaks, the temperature inside the object that cannot be directly measured is accurately measured. Can be detected.

[0042]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiment 1. Hereinafter, the MRI apparatus according to the first embodiment will be described with reference to the drawings. FIG. 1 is a schematic diagram showing an outline of the MRI apparatus according to the first embodiment. The MRI apparatus according to the first embodiment includes a static magnetic field generation unit 1, a gradient coil unit 2 that applies a gradient magnetic field, an RF electromagnetic wave irradiation and an RF signal that receives an NMR signal.
And a coil portion 3. The gradient coil unit 2 is controlled by the gradient driving unit 4, and the RF coil unit 3 is controlled by the transmitting unit 5 and the receiving unit 6. The RF coil unit 3 is connected to the computer 8 via the analog / digital conversion unit 7. Further, the gradient drive unit 4, the transmission unit 5, the reception unit 6, and the analog / digital conversion unit 7 are respectively controlled by the control unit 9, and the computer 8 has an operation unit 10 and a display unit 11, and the control unit 9 To control.

The static magnetic field generator 1 is for applying a magnetic field that is temporally and spatially uniform to the space in which the object to be imaged 12 is placed. The static magnetic field generator 1 is composed of a permanent magnet or a superconducting magnet, and has a function of generating a strong magnetic field of about 1.0T.

The gradient coil section 2 is for generating a gradient magnetic field. As described above, the MRI apparatus uses a nuclear magnetic resonance phenomenon for a specific nuclide (mainly a hydrogen atom) that constitutes the imaging target 12, but in order to examine position information, it is spatially separated from the static magnetic field. This is because it is necessary to apply a gradient magnetic field that fluctuates with time to the imaging target 12.

The RF coil section 3 has a function as an electromagnetic wave transmitting means for irradiating an electromagnetic wave to the image pickup object 12 and receives an NMR signal emitted from atoms constituting the image pickup object 12 by a nuclear magnetic resonance phenomenon. It has a function as an NMR signal receiving means. The frequency of the electromagnetic wave emitted by the RF coil unit 3 is from several MHz when the intensity of the magnetic field applied by the static magnetic field generation unit 1 is 1.0T.
It is preferably several tens Mz.

The gradient drive section 4 is for controlling the gradient coil section 2. The magnitude of the gradient magnetic field applied by the gradient coil unit 2 needs to vary spatially and temporally, so the gradient drive unit 4 controls the gradient coil unit 2 to vary the gradient magnetic field.

The transmitter 5 and the receiver 6 are for controlling the RF coil unit 3, respectively. As mentioned above, RF
The coil unit 3 has a function as an electromagnetic wave transmitting unit and an NMR.
It has a function as a signal receiving means. Specifically, an RF pulse is generated based on the current sent from the transmitter 5 to transmit an electromagnetic wave, and the received NMR signal is output to the receiver 6. The receiving unit 6 receives the NMR received from the RF coil unit 3.
The signal is transmitted to the analog / digital converter 7.

The analog / digital converter 7 is for converting analog data relating to the NMR signal received by the RF coil unit 3 into digital data. This is because it is necessary to convert the observation data into a digital data format in order to be emitted from the imaging target 12 and to be spectrally analyzed on the computer 8.

The computer 8 is for estimating the spectrum from the observation data received by the RF coil unit 3 and converted into digital data by the analog / digital conversion unit 7. The computer 8 has a function of performing spectrum estimation, analyzing the internal structure of the imaging target 12 from the result of spectrum estimation, and forming a magnetic resonance image.

The computer 8 is also for controlling the control section 9. Based on the information input to the operation unit 10 connected to the computer 8, the computer 8 has a function of starting imaging of a magnetic resonance image, determining a slice width, and the like, and sending a command to the control unit 9.

The display unit 11 is for displaying a magnetic resonance image formed by the computer 8. Display 1
The structure of 1 may be a CRT display, a device that directly displays an image such as a TFT liquid crystal screen, or a device that outputs an image to another medium such as a printer.

Next, a method for actually capturing a magnetic resonance image using the MRI apparatus according to the first embodiment will be described.
explain. Imaging of magnetic resonance images according to the first embodiment is roughly divided into a process of acquiring observation data by applying a magnetic field, irradiating electromagnetic waves, and receiving, and a process of converting the acquired observation data into a spectrum to form a magnetic resonance image. Consists of.

First, the observation data acquisition process will be described with reference to FIG. First, the imaging target 12 is set to M
The image pickup target 12 is loaded into the RI apparatus and placed in the static magnetic field atmosphere generated by the static magnetic field generator 1 (step S101). When the static magnetic field generating unit 1 is made of a superconducting magnet, it is necessary to cool the superconducting magnet with liquid helium or the like in advance to create a superconducting state.

Then, the gradient coil unit 2 applies a gradient magnetic field in the slice direction, and the RF coil unit 3 irradiates the imaging object 12 with electromagnetic waves (step S10).
2). By applying a gradient magnetic field in the slice direction,
The imaged object 12 is divided into a large number of slices having different resonance frequencies and whose normals coincide with the slice direction. Then, by irradiating the imaging target 12 with an electromagnetic wave composed of a high-frequency pulse having the same frequency as the resonance frequency of an arbitrary slice, it is possible to excite only the spins of the atoms included in the desired slice, and regarding the position of the atom. , The coordinates in the slice direction can be specified.

The gradient coil section 2 is connected to the imaging object 1
A phase gradient magnetic field is applied to 2 (step S10
3). By applying a phase gradient magnetic field, step S
The phases of the electromagnetic waves emitted from the atoms existing in the slice selected in 102 differ according to the distribution of the phase gradient magnetic field. Specifically, the slice is
Electromagnetic waves of different phases are emitted from the atoms that are divided into a plurality of strips having the phase direction as the transverse direction. Therefore, the position of the atom that emits the electromagnetic wave at the end of this step can be specified with respect to the coordinates in the slice direction and the coordinates in the phase direction.

Then, while applying the frequency gradient magnetic field by the gradient coil unit 2, the RF coil unit 3 receives the NMR signal emitted from the atoms constituting the imaging object 12 (step S104). By applying the gradient magnetic field in the frequency direction, the RF coil unit 3 can receive NMR signals of different frequencies along the frequency direction. Therefore, with respect to the atom emitting the NMR signal, the coordinate in the frequency direction can be specified, and the position of the atom emitting the NMR signal can be completely specified together with steps S102 to S103. And step S1
The scanning of 02 to S104 is repeatedly performed by the number of matrices in the phase direction to obtain a two-dimensional data set.

The above steps S101 to S10
By going through the process of 4, it is possible to acquire observation data, which is time-series information indicating at which position and how many atoms constituting the imaging object 12 exist. In addition,
The gradient magnetic field application, electromagnetic wave irradiation, and NMR signal reception in these steps are directly performed by the gradient drive unit 4, the transmission unit 5, and the reception unit 6, respectively, but these operations must be performed in cooperation with each other. Because there is
The control unit 9 issues commands to the gradient drive unit 4, the transmission unit 5, and the reception unit 6, respectively, so that proper synchronization is achieved.

The observation data acquisition step is performed on the metabolites forming the internal organs of the imaged object 12 and the water contained in the imaged object 12. Originally, in order to know the internal structure of the imaging target 12, it is sufficient to have only the information about the internal organs.
This is because the information about water is also needed in the subsequent spectrum estimation. Specifically, the observation data about metabolites and water are acquired as follows. That is, between the hydrogen atoms contained in the molecules of metabolites that compose the internal organs and the hydrogen atoms contained in the water molecules,
The resonance frequency is different due to the difference in molecular structure. Therefore,
By changing the frequency of the electromagnetic waves emitted from the RF coil unit 3, it is possible to acquire observation data regarding internal organs and observation data regarding water.

Next, the step of converting the observation data acquired in the above step into a spectrum and forming a magnetic resonance image will be described with reference to FIGS. In addition, FIG.
The diagram shown in FIG. 6 is a schematic diagram for the purpose of assisting understanding of the contents of this step, and does not necessarily match actual observation data.

First, the phase of the observation data of metabolites is corrected by the observation data of water (step S20).
1). Since the observation data is acquired by applying a gradient magnetic field that changes with time, an induced electromotive force is generated in the imaging target 12 and an eddy current is generated. This eddy current causes a disturbance in the value of the observation data, and this step is performed to correct this disturbance.

Then, the observation data on water is subjected to fast Fourier transform (step S202). The fast Fourier transform in this case can be performed by the conventional method.
As a result of this step, a spectrum for water can be obtained. Then, with respect to the spectrum obtained in step S202, the full width at half maximum of water is measured (step S203).

Then, the noise removal filter function is determined from the half width of the peak of water measured in step S203 (step S204). Specifically, it is assumed that the noise removal filter function is an exponential function, and the exponential component is −1, the half value width of the peak, and the time variable t. Here, when the value of the half width is smaller than 0.5 Hz, 0.5 is substituted for the half width of the peak, and when the value of the half width is larger than 3.0 Hz, the half width of the peak is Substitute 3.0 for the price range.

After that, the noise removal filter function determined in step S204 is applied to the observation data on the metabolites (step S205). Specifically, FIG.
As shown in (a), observation data l ₁ with time as a variable
And a noise removal filter function l that also has time as a variable
By multiplying by ₂ , the waveform data l ₃ shown in FIG. 4 (b) is newly obtained. It is clear from FIG. 4B that this waveform data l ₃ has a damped oscillation waveform by multiplying it by a noise removal filter function. By this step, the waveform data from which the noise component is removed can be obtained.

Then, fast Fourier transform is performed on the waveform data obtained in step S205 (step S
206). By performing the fast Fourier transform, the spectrum of the waveform data as shown in FIG. 5 can be obtained.

Then, the spectrum of the waveform data
The maximum value P of the peak intensity shown in FIG. _SAnd the noise component
Size P_NTo derive the S / N ratio, which is the intensity ratio
(Step S207). And based on the S / N ratio
The model order M is determined (step S208). here
If the S / N ratio is less than 250, then the S / N ratio
Is calculated as 250, and the S / N ratio is greater than 400
In that case, the calculation is performed with the S / N ratio set to 400 thereafter.

Specifically, the model order M is a value obtained by multiplying the value of the S / N ratio to the power of −1/2 by the proportionality constant α. The constant of proportionality α is the number of measurement data points,
It is a value determined by the time interval of data acquisition. For example, when the number of data points is 2048, the value of the proportional constant α is 50.

Then, the model determined in step S208 is selected.
Create a polynomial with M number of terms based on Le degree M
Yes (step S209). Specifically, as shown in FIG.
As described above, the discrete time t at which the mutual interval is Δt₀, T₁, T₂,
t₃・ ・ ・, T_M・ ・ ・, T_NAbout, strong
Degree x₀, X₁, X₂, X₃・ ・ ・, X_M, ...
x _NFrom the waveform data and
Create a polynomial.       x_{M + i}= A₁x_{M + i-1}+ A₂x_{M + i-2}+ A₃x_{M + i-3}＋ ・ ・ ・ ・ ・ ＋ a_M-1x_{M + i- (M-1)} + A_Mx_{M + iM}(4) Here, i = 0, 1, 2, ..., N,
Let N = 2M. Therefore, is equation (4) M polynomials?
Consists of Also, x_jIs a concrete example read from the waveform data.
Is unknown, the unknown in equation (4) is a
₁, A₂... a_MIs. Therefore, equation (4)
Are M equations for M unknowns,
Derivation of coefficients in autoregressive model by solving equations
To do.

After that, z conversion is carried out for the expression of i = 0 in the expression (4) (step S210). X (z) obtained by z-transforming the equation of i = 0 in the equation (4) is represented by the following equation. X (z) = (1−Σa _i z ⁻¹ ) ⁻¹ (5) where Σa _i z ⁻¹ is the sum of 1 to M for i.

Then, variable conversion is performed on the equation (5) (step S211). Specifically, the variable z is
Replace with (ifΔt / 2π). Here, i is an imaginary unit, f is a frequency, and Δt is a time interval at which the waveform data is read. Specifically, as a result of the variable conversion, the expression (6) is transformed into the following expression. X (f) = (1−Σa _i (ifΔt / 2π) ⁻¹ ) ⁻¹ (6) where a _i is specifically obtained from the equation (5),
Since Δt, 2π, and i are constants, it can be seen that the expression (7) is a function having the frequency f (Hz) as a variable.

Finally, the absolute value of the equation (6) is calculated (step S212). This is because the expression (6) is an expression including the imaginary number i, and therefore it is necessary to convert the expression (7) into a function composed of a natural number in order to know the spectrum. In particular,
The right side of the equation (6) is multiplied by the conjugate complex number to obtain the square root of the product. This is the end of the process of converting the acquired observation data into a spectrum.

A graph of the spectrum obtained in this step is shown in FIG. 7 (b). In FIG. 7B, spectrum estimation is performed with the peak half-value width, which is the exponential component of the noise removal filter function, being 1.8 Hz and the model order M being 300. Compared with FIG. 7A which is a spectrum obtained by the fast Fourier transform based on the same observation data, the half width of the peak of the spectrum is obviously narrow. Also, by multiplying the observation data with the noise removal filter function as a side effect,
The magnitude of the noise component in the resulting spectrum is also shown in FIG.
It can be seen that the graph in (b) is clearly smaller.

After that, a two-dimensional magnetic resonance image for each slice plane is created on a computer based on the spectrum obtained in the above process, and is output to the display unit 11, whereby the first embodiment is performed. Imaging of the magnetic resonance image using the MRI apparatus is completed.

The noise removal filter function determined in step S204 is an exponential function for the following reason. Since the irradiation of the electromagnetic wave by the RF coil unit 3 in the image data acquisition step is performed in a short time, the intensity of the NMR signal emitted from the atoms forming the imaging target 12 is corresponding to the electromagnetic wave generated by the RF coil unit 3. After the end of irradiation, it should decrease exponentially and form a damped oscillation waveform. However, the observed data does not always decrease exponentially as shown by the straight line l ₁ in FIG. This is because the intensity of the noise component is large, and the reduced amount of the NMR signal emitted based on the nuclear magnetic resonance phenomenon is canceled. Therefore, the noise component is removed from the observation data by multiplying the observation data by the exponential decrease function, and only the NMR signal component based on the nuclear magnetic resonance phenomenon can be extracted as the waveform data.

Further, the reason for determining the exponential component of the noise removal filter function in step S204 is to be proportional to the half-value width of the peak relating to water for the following reason. That is, the half-value width of each spectrum corresponds to the damped oscillation component of the observation data that is time-series data, and the noise component can be effectively removed by using the half-value width of the peak in the spectrum. Further, it is desirable to use a peak having a high peak intensity as a peak for measuring the half width, and a peak due to a hydrogen atom constituting a water molecule is selected.

Thus, the MR according to the first embodiment
The device I can accurately perform spectrum estimation as compared with the conventional imaging of a magnetic resonance image using the fast Fourier transform. As a result, it is possible to capture a magnetic resonance image with high resolution.

Further, with the spectrum estimation according to the first embodiment, it is possible to perform spectrum estimation with high resolution using the autoregressive model even for observation data with a small S / N ratio. Therefore, in addition to the magnetic resonance imaging by the MRI apparatus, for example, in the analysis of meteorological phenomena, the characteristic analysis of electronic circuits and mechanical systems,
A spectrum with high resolution can be obtained from time series information with a small S / N ratio.

In the observation data acquisition process, steps S102, S103, and S104 are steps for specifying the positions of atoms in the image-capturing object 12, as described above. On the other hand, the observed data on water is only used for spectral conversion of the observed data on metabolites,
The application of the gradient magnetic field may be omitted in the step of acquiring the observation data regarding water. For the same reason, it is not necessary to obtain a two-dimensional data set when acquiring the observation data regarding water, and the repetition of steps S102 to S104 may be omitted.

Further, although the waveform data has the noise reduced by the noise removal filter function, the noise is not completely removed. Therefore, step S2
In 09, strictly, the equal sign between the left side and the right side of the equation (4) may not hold. Therefore, more accurately, a ₁ , a ₂ ,.
It is desirable to derive _M. Specifically, first, i =
Equation (4) regarding 0, 1, 2, ..., NM is expressed by a determinant. Here, (x _M , x _{M + 1} , ...
x _n ) is defined as a column vector, and (a ₁ , a ₂ ,
Let a be a column vector consisting of a ₃ , ..., A _M ).
Also, let H be an M × N matrix having M _kl = x _{M-1 + kl} as an element. From these d, a, and H, the equation (4) can be expressed in the form of d = H · a (7). Therefore, by the method of least squares, a can be expressed as a = H ^t H ^-1 H ^t d (8), and a can be specifically obtained by taking the formula (8). . By using the least squares method,
More accurately, a ₁ , a ₂ , ..., A _M can be obtained. In addition to the above, a ₁ , a ₂ , a ₃ , ..., A _M by the Yule-Walker method or the Burg method
It is also effective to make a decision.

Further, in FIG. 1, the RF coil unit 3 has a structure that covers only the head of the patient, which is the imaging target 12, but the MRI apparatus according to the first embodiment is not limited to this structure. For example, the shape of the RF coil unit 3 may be changed in order to image the abdomen of the patient, or the shape may be such that the whole body of the patient can be imaged. further,
The MRI apparatus shown in FIG. 1 is a so-called horizontal magnetic field type MRI apparatus that applies a uniform magnetic field in the horizontal direction, but other than this, for example, a so-called vertical magnetic field type structure that applies a static magnetic field in the vertical direction. The MRI apparatus may be composed of

Further, in the first embodiment, FIG.
As shown in FIG. 7, the MRI structure is such that the gradient drive unit 4, the transmission unit 5, and the like that control the apparatus are independent of each other. this is,
This is for the purpose of facilitating the description of the structure of the MRI apparatus according to the first embodiment, and does not necessarily show that each part has a separated structure. Therefore, for example, the control unit 9 may be built in the computer 8, or the analog / digital conversion unit 7 may be built in the computer 8 to directly output analog data to the computer 8.
The structure may be input to

Further, with respect to the observation data acquisition step in the first embodiment, the slice direction is usually the direction from the head of the patient who is the imaging target 12 to the toes, but is not limited to this. . The slice direction can be changed according to the region of the magnetic resonance image to be imaged. Further, the frequency direction, the phase direction, and the slice direction usually form an orthogonal coordinate system, but each direction in the first embodiment is not necessarily limited to the orthogonal coordinate system.

Regarding the spectrum estimation according to the first embodiment, the operator may manually perform the spectrum estimation through the operation unit 10, but as a more preferable mode, the above-mentioned spectrum estimation program is executed on the computer 8. It is desirable to do. By making the program execute on the computer 8,
The spectrum can be estimated automatically. Although only the spectrum estimation may be executed by the program, the control of the MRI apparatus and the step of forming the magnetic resonance image from the estimated spectrum may be executed by the program.

Embodiment 2. Next, the MRI apparatus according to the second embodiment will be described. The MRI apparatus according to the second embodiment has a function of capturing a magnetic resonance image,
It has a function of detecting a temperature change inside the object to be imaged.
The structure of the MRI apparatus according to the second embodiment is similar to the structure of the MRI apparatus according to the first embodiment shown in FIG.

Further, it is assumed that the observation data acquisition step and the spectrum estimation in the second embodiment are performed in substantially the same manner as the steps in the first embodiment. Then, from the spectra obtained by these steps, the temperature variation inside the imaging target 12 is observed.

Specifically, after the spectrum is estimated by the same method as in the first embodiment, the frequency of the peak due to the hydrogen atoms constituting water and the hydrogen atoms constituting the metabolite are displayed on the computer 8. The difference from the frequency of the peak due to is read from the estimated spectrum.

Then, by comparing the read frequency difference with the reference data, it is detected how the internal temperature of the object 12 to be imaged has changed. For example, by previously acquiring data regarding the frequency difference at 36 degrees, and comparing the frequency difference read in the second embodiment with known data, it is examined how much the internal temperature has changed from 36 degrees. .

It is generally known that the resonance frequency of metabolites fluctuates depending on temperature. On the other hand, since the resonance frequency of hydrogen atoms that compose water molecules does not depend on temperature, by determining the difference between the resonance frequency of hydrogen atoms that compose metabolites and the resonance frequency of hydrogen atoms that compose water molecules,
It is possible to detect temperature fluctuations. Since the variation of the resonance frequency of the hydrogen atom which constitutes the metabolite was a small value, it was difficult to detect it by the conventional spectrum estimation by the fast Fourier transform. However, in the spectrum estimation according to the first embodiment, it is possible to perform the spectrum estimation with high accuracy, and thus it is possible to sufficiently detect the fluctuation of the resonance frequency with respect to the temperature change. That is, the MRI apparatus according to the second embodiment can detect the temperature variation inside the object that cannot directly measure the temperature.

By measuring the temperature fluctuation inside the image pickup object 12 in this way, the following advantages occur. For example,
The MRI apparatus according to the second embodiment can be used as a diagnostic apparatus for the human brain.

In brain diseases such as brain contusion, it is extremely important to know the temperature of the brain. That is, when the temperature of the brain rises, cerebral edema is accompanied and the volume of the brain increases,
Patients may die. In such cases,
If the MRI apparatus according to the second embodiment can detect the rise in the brain temperature, the patient can be treated by performing appropriate treatment, for example, hypothermia therapy.

In the second embodiment, since the purpose of imaging the magnetic resonance image itself is not intended, the step of applying a gradient magnetic field for imaging is omitted in the observation data acquisition step, and only the static magnetic field is applied. Then, the RF coil unit 3
The electromagnetic wave may be transmitted by. Imaging target 1
This is because it is sufficient to detect the resonance frequency of the metabolite and the resonance frequency of water in order to detect the temperature fluctuation inside the No. 2. On the contrary, the observation data acquisition process is performed in the same manner as in the case of the first embodiment to measure the temperature variation and to capture the image of the object
The magnetic resonance image 2 may be captured.

Although the spectrum estimation method using the autoregressive model has been described in the first and second embodiments, the spectrum estimation method using the autoregressive model is the same as the so-called MEM (Maximum Entropy Method). Proven to be a spectrum estimation method
(SL Marple, Digital Spectral Analysis, Prentice
-Hall, 1987). Therefore, it goes without saying that the present invention can also be applied to this MEM.

[0092]

As described above, according to the invention of the first aspect, the noise removal filter is applied to the derivation of the waveform data, and the predetermined number of decisions are obtained from the signal strength and the noise strength. , It is possible to perform spectrum estimation with high resolution even for observation data having a low S / N ratio.

According to the invention of the second aspect,
Since the coefficient of each term in the linear combination is determined by the method of least squares, there is an effect that the coefficient of each term can be accurately determined even for observation data having noise.

According to the invention of the third aspect,
The noise removal filter function includes an exponential function, and the exponential component has a coefficient derived from the half-width of the peak, and the observation data was derived by multiplying the observation data with the noise removal filter function. There is an effect that the noise component of the data can be accurately removed and the derived waveform data can be used as the damping oscillation function.

According to the invention of the fourth aspect,
Since the exponential component of the noise removal filter function is specifically determined, the noise component can be removed more accurately.

According to the invention of the fifth aspect,
Since the predetermined number is determined from the ratio of the maximum value of the signal intensity in the spectrum to the noise intensity value, there is an effect that it is possible to perform spectrum estimation with higher resolution than in the case of using FPE.

According to the invention of the sixth aspect,
With the configuration in which the predetermined number of terms is more specifically determined, there is an effect that it is possible to perform spectrum estimation with higher resolution.

According to the invention of the seventh aspect,
In the estimation of the spectrum, a noise removal filter is applied and the predetermined number is determined from the signal strength and the noise strength, so a high-resolution spectrum is estimated from the observation data with a low S / N ratio, and a high resolution is based on that spectrum. The effect of being able to form a magnetic resonance image of is obtained.

According to the invention of the eighth aspect,
Concerning the exponential component in the exponential function included in the noise removal filter function, it was decided to include the half-value width of the peak due to the hydrogen atoms that compose the water molecules inside the imaged object, thus forming a useful noise removal filter function. Therefore, the noise component in the observation data can be effectively removed. Therefore, there is an effect that a magnetic resonance image having high resolution can be captured.

According to the invention of the ninth aspect,
By specifically defining the predetermined number of determinations, it is possible to perform high-resolution spectrum estimation from observation data having a lower S / N ratio than that when FPE is used, and to capture high-resolution magnetic resonance images. It has the effect of being able to.

Further, according to the invention of the tenth aspect, the peak due to the atoms constituting the metabolite also changes the resonance frequency due to the temperature change. Therefore, by measuring the fluctuation range, the inside of the object to be imaged can be measured. It is possible to detect the temperature fluctuation of

According to the invention of the eleventh aspect, the method according to any one of the inventions of the first to tenth aspects is used as a program to be executed on a computer. There is an effect that the method according to the invention in any of the aspects can be executed by a computer.

According to the twelfth aspect of the invention, the image forming processing section estimates the spectrum using the autoregressive model, and in the spectrum estimation, the waveform data is derived using the noise removal filter function. Since the magnetic resonance imaging apparatus determines the number of polynomial terms based on the signal strength and the noise strength, high-resolution spectral estimation is performed using an auto-regression model even for observation data with a low S / N ratio. It is possible to provide an apparatus capable of capturing the magnetic resonance image of 1.

Further, according to the invention of the thirteenth aspect, since the noise removal filter function includes an exponential function whose exponent component includes the full width at half maximum of the peak due to hydrogen atoms, the image forming processing unit observes It is possible to effectively remove noise components from data and provide a magnetic resonance imaging apparatus capable of capturing a high-resolution magnetic resonance image.

Further, according to the invention of the fourteenth aspect, it is more appropriate for the observation data having a low S / N ratio as compared with the case where the image forming processing unit determines the model order using FPE. It is possible to provide the magnetic resonance imaging apparatus capable of providing the model order.

According to the invention of the fifteenth aspect, there is provided a magnetic resonance imaging apparatus, wherein the processing means detects a temperature variation inside the object to be imaged by detecting a frequency difference between a plurality of peaks. Therefore, there is an effect that it is possible to accurately detect the temperature inside the object that cannot be directly measured.

[Brief description of drawings]

FIG. 1 is a schematic diagram showing a structure of an MRI apparatus according to a first embodiment.

FIG. 2 is a flowchart showing a process of acquiring observation data according to the first embodiment.

FIG. 3 is a flowchart showing steps of spectrum estimation in the first embodiment.

FIG. 4A is a schematic diagram for multiplying observation data by a noise removal filter function in spectrum estimation in the first embodiment. (B) is a schematic diagram showing the waveform data derived as a result of multiplication.

FIG. 5 is a schematic diagram showing peak intensity and noise intensity when determining a model order M in spectrum estimation in the first embodiment.

FIG. 6 is a schematic diagram when deriving a polynomial by an autoregressive model from observed data in spectrum estimation in the first embodiment.

FIG. 7A is a spectrum diagram obtained by performing fast Fourier transform on observation data obtained by the MRI apparatus according to the first embodiment. (B) is a spectrum diagram in which the same observation data is derived using the spectrum estimation method according to the first embodiment.

FIG. 8A is a graph showing observed data that is time-series information, and FIG. 8B is a spectrum diagram obtained by performing a fast Fourier transform on the data shown in FIG. 8A.
Further, (c) is a spectrum diagram obtained by performing spectrum estimation from the data of (a) using an autoregressive model.

[Explanation of symbols]

1 Static magnetic field generator 2 gradient coil 3 RF coil section 4 Gradient drive 5 transmitter 6 Receiver 7 Analog-to-digital converter 8 computers 9 control unit 10 Operation part 11 Display

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Go Matsuda             127, 4-7 Asahigaoka, Hino City, Tokyo             GE Yokogawa Medical System Co., Ltd.             Within F-term (reference) 4C096 AA13 AA14 AA20 AB07 AB50                       AD13 AD14 AD25 DA04 DA08                       DB07 DB10 DB12 DC05    (54) [Title of Invention] Spectrum estimation method from waveform data, magnetic resonance imaging method applying the spectrum estimation method,                     Magnetic resonance imaging device applying spectrum estimation method and these methods on computer                     The program to execute

Claims (15)

[Claims] 1. A signal strength waveform having a frequency as a variable is estimated based on a signal strength of a waveform data having time as a variable at a predetermined time and a predetermined number of signal strengths at discrete times up to the predetermined time. A spectrum estimation method, comprising: a waveform data deriving step of deriving the waveform data by applying a noise removal filter to observation data; and acquiring a first signal strength waveform having a frequency as a variable from the waveform data, and acquiring the waveform. And a determining step of determining the predetermined number from the signal strength of the first signal strength waveform and the noise strength, the spectrum estimating method comprising: 2. The signal strength waveform is obtained by an autoregressive model in which the signal strength of the waveform data having the time as a variable is expressed by a linear combination of a predetermined number of signal strengths at discrete times up to the predetermined time. The spectrum estimating method according to claim 1, further comprising a coefficient determining step of determining a coefficient of each term in the linear combination by a least square method when estimating. 3. The waveform data deriving step comprises: -1 and a coefficient derived from a half-value width of a maximum peak in a second signal intensity waveform having a variable obtained by Fourier transforming the observation data. 3. The spectrum estimation method according to claim 1 or 2, wherein the waveform data is derived by multiplying the observation data by a noise removal filter function including an exponential function having a product with a variable relating to time as an exponential component. . 4. The waveform data deriving step uses the value of the full width at half maximum of the maximum peak as the value of the coefficient when the full width at half maximum of the maximum peak is 0.5 to 3.0 Hz. When the full width at half maximum is 0.5 Hz or less, 0.
5 as the value of the coefficient, and the full width at half maximum of the maximum peak is 3.
The spectrum estimation method according to claim 3, wherein when the value is 0 Hz or more, 3.0 is set as the value of the coefficient. 5. The determination step determines the predetermined number from the ratio of the maximum value of the signal strength and the noise strength value in the first signal strength waveform. The spectrum estimation method described in one. 6. The determining step comprises (-) a ratio of the maximum value of the signal strength in the first signal strength waveform to the noise strength value.
6. The spectrum estimating method according to claim 1, wherein an integer part of a value obtained by multiplying a value of 1/2 power by a proportional constant is set as the predetermined number. 7. An object to be imaged is arranged in a magnetic field atmosphere, and an electromagnetic wave having a center frequency proportional to the magnetic field intensity of the magnetic field atmosphere is transmitted to the arranged object to be imaged, based on a nuclear magnetic resonance phenomenon. NMR obtained from the imaged object
A signal is received, a spectrum is estimated by an autoregressive model based on the received waveform data of the NMR signal, and the estimated spectrum is replaced with luminance information for a position to output an image of the internal structure of the imaging target. In the magnetic resonance imaging method, the estimation of the spectrum includes a waveform data deriving step of deriving the waveform data by applying a noise removal filter to the observation data, and a first signal strength having a frequency as a variable from the waveform data. And a determining step of determining the predetermined number from the signal strength and the noise strength of the acquired first signal strength waveform, the magnetic resonance imaging method. 8. The peak derived from hydrogen atoms constituting water molecules existing inside the imaging target in a second signal intensity waveform obtained by fast Fourier transforming the observation data in the waveform data deriving step. 9. The magnetic resonance imaging method according to claim 7, wherein the waveform data is derived by multiplying the observation data by a noise removal filter function including an exponential function having a half-value width as a part of the exponential component. 9. The determining step comprises (-) a ratio of the maximum value of the signal strength in the first signal strength waveform to the noise strength value.
9. The magnetic resonance imaging method according to claim 7, wherein an integer part of a value obtained by multiplying a value of the power of 1/2) by a proportional constant is set as the predetermined number. 10. The method further includes a detection step of detecting a temperature change inside the imaged object based on a change in frequency difference between a plurality of peaks, from a distribution of signal intensity obtained by the spectrum estimation step. 7. The method according to claim 7,
8. The magnetic resonance imaging method according to 8 or 9. 11. A program for executing the method according to any one of claims 1 to 10 on a computer. 12. A static magnetic field generating magnet that generates a uniform magnetic field, a gradient magnetic field generating unit that generates a magnetic field that fluctuates depending on position and time, a transmitting unit that irradiates an image pickup object with electromagnetic waves, and the image pickup target. An electromagnetic wave having a frequency proportional to the uniform magnetic field and the varying magnetic field intensity, which has a receiving unit for receiving observation data including an NMR signal from an object and an image forming processing unit for forming a magnetic resonance image. Is a magnetic resonance image pickup device that picks up a magnetic resonance image by transmitting a waveform data including time series information based on the observation data and a noise removal filter function. The autoregressive model is determined while determining the number of polynomial terms from the signal strength and the noise strength in the first signal strength waveform obtained by performing fast Fourier transform on the derived waveform data. Performs a spectrum estimation by Le, magnetic resonance image pickup apparatus characterized by forming a magnetic resonance image by estimating the spectrum of the frequency as a variable. 13. The noise removal filter function is a hydrogen forming a water molecule existing inside the imaging target in a second signal intensity waveform having a frequency obtained by fast Fourier transform of the observation data as a variable. 13. The magnetic resonance imaging apparatus according to claim 12, further comprising an exponential function having a half-value width of a peak caused by atoms as a part of an exponential component. 14. The number of terms in the polynomial is an integer part of a value obtained by multiplying a value obtained by multiplying a ratio of the signal strength and the noise strength in the first signal strength waveform by (−1/2) by a proportional constant. The magnetic resonance imaging apparatus according to claim 12 or 13, wherein 15. The image forming processing unit detects temperature fluctuation of an image pickup object based on fluctuation of frequency difference between a plurality of peaks in the spectrum obtained by the spectrum estimation. The magnetic resonance imaging apparatus according to any one of 12 to 14.