JP3560192B2 - Activation area extraction method - Google Patents

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention belongs to a technical field related to a data processing method in brain function measurement MRI (hereinafter, fMRI) using magnetic resonance imaging (hereinafter, MRI). More specifically, the present invention belongs to the technical field related to a method for accurately extracting a cerebral cortex area activated by stimulation even when the SN ratio is low.
[0002]
[Prior art]
A technique called fMRI for analyzing brain functions using MRI has been developed. In this technique, a reaction in the cerebral cortex caused by application of a stimulus such as light or sound is captured as a change in an MR (nuclear magnetic resonance) signal and imaged. That is, a difference image is created from an image captured while applying a stimulus to a subject (hereinafter, a stimulus applied image) and an image captured without applying a stimulus (hereinafter, a rest image). A region having a large signal strength on the difference image is regarded as a region that has responded to the stimulus (hereinafter, an activation region), and identification of a functional area of the brain and evaluation of a signal change rate are performed. Here, in the present specification, image data created using a resting image and a stimulus applied image to extract an activated region, such as a difference image, is referred to as functional image data. The other image data is referred to as morphological image data.
[0003]
In order to accurately identify the functional area, the extraction method of the activated region is particularly important. As a report comparing various extraction methods, see Book of Abstract, 12th Annual Meeting of Society of Magnetic Resonance in Medicine, vol. 1, 449 (1993) are known. In this report, the method of extracting the activation area is based on the signal intensity before the stimulus is applied, and the rate of signal change accompanying the stimulus is evaluated and extracted. The method is classified into two types, that is, a method of performing evaluation and extraction. Although this classification partially overlaps, in practice, the following three types of extraction methods are mainly used. The first method is an extraction method using a signal change rate accompanying application of a stimulus. This method has a shortcoming in that the data processing time for extracting the active region is short, but veins are erroneously extracted in addition to the cerebral cortex. The second method is an extraction method using statistical processing such as a t-test. This method has a feature that the ratio of the signal change rate to noise can be objectively evaluated. The third method is an extraction method using a correlation function. This method is an extraction method for evaluating the degree of synchronization between the application of a stimulus and a signal change. Compared to the first method, the second and third methods have a feature that, although the data processing time for extracting the active area is longer, erroneous extraction of veins is less. Therefore, the second or third method is often used to extract the activated region.
[0004]
In addition to the extraction method, the spatial resolution of the image is also important in identifying functional areas. To perform the identification accurately, it is essential to improve the spatial resolution of the image. However, one problem arises here. That is a decrease in the SN ratio per pixel. The signal amount of one pixel can be considered as the number of hydrogen atoms contained in one pixel. As the spatial resolution improves, the volume of the real space corresponding to one pixel decreases. As a result, the signal amount of one pixel decreases, and the SN ratio decreases. The effect of a decrease in the SN ratio is described in Book of Abstract, 2nd Annual Meeting of Society of Magnetic Resonance, vol. 2, 642 (1994). That is, the area of the extracted active region is reduced for the following reasons. Irrespective of the spatial resolution, the area of the cerebral cortex where the signal changes upon application of the stimulus, and the rate of change are fixed. Here, in the case where the second or third method is used as the extraction method, if the SN ratio is reduced due to the improvement of the spatial resolution, a region having a small signal change rate is not extracted as an active region. Therefore, the area of the extracted active region is reduced.
[0005]
[Problems to be solved by the invention]
As described above, the conventional activation region extraction method has many problems such as erroneous extraction of veins and a change in the area of the activation region due to a change in the SN ratio. An object of the present invention is to provide a method for extracting an activated region, which does not impair the characteristics of the above-described second and third extraction methods, does not cause erroneous extraction of veins and the like, and is hardly affected by a change in SN ratio. It is in.
[0006]
[Means for Solving the Problems]
The activated region extraction method of the present invention includes an extraction method using a t-test or a correlation coefficient that is easily affected by a change in SN ratio, and an extraction method using a signal change rate that is not easily affected by a change in the SN ratio. The feature is that it is used together. In the present invention, by using both the extraction method that is susceptible to the S / N ratio and the extraction method that is less susceptible to the S / N ratio, it is possible to satisfactorily extract the active region even when the S / N ratio is reduced due to improvement in spatial resolution. .
[0007]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. First, FIG. 2 shows a measurement example of fMRI, which is one of the fields to which the present invention can be applied. In fMRI, a stimulus application period is provided during measurement, and images are taken at fixed time intervals. In this time-series image, a process of evaluating a time change of a signal for each coordinate and extracting an activation area is an activation area extraction method. When extracting an activated region, one extraction method is usually used, whereas in the present invention, two types of extraction methods are used. That is, the functional image data created by using the extraction method that is easily affected by the SN ratio of the image and the functional image data created by using the extraction method that is hardly affected by the SN ratio are extracted together. First, two types of extraction methods will be described, and then an extraction procedure will be described.
[0008]
The extraction method that is easily affected by the SN ratio is defined as the expression expressing the extraction method that includes 1) a term for inputting the variance or deviation of a signal, and 2) a term whose value always increases as noise increases. Existing extraction method. Here, the noise is a time variation of a signal at the same coordinate of the time-series image, and the variance of the signal is a deviation from a time average value of the signal at the same coordinate of the time-series image. An example of 1) is an extraction method using a t-test. (Equation 1) shows a formula for calculating the t value.
[0009]
(Equation 1)
_{t = (X 1 -X 2)} / {√ {(D 1 + D 2) / (n 1 + n 2 -2)} · √ {1 / n 1 + 1 / n 2}} ... ( Equation 1)
Here, X is the average value of the signals, D ₁ and D ₂ are the sum of squares of the deviation of the signals, and n is the number of images. In (Equation 1), √ ｛｝ takes the square root of ｛｝.
[0010]
The subscripts 1 and 2 represent a rest period and a stimulus application period, respectively. As can be seen from equation (1), the value of D ₁ and D ₂ is increased, the value of t is reduced. Since the activation region is extracted by setting a threshold value to the value of t, it can be said that the extraction method using the t test is easily affected by the SN ratio. As an example of 2), there is an extraction method using a cross-correlation function, and (Equation 2) shows one example.
[0011]
(Equation 2)
f (S, E) = SE · {S‖‖E} (Equation 2) S is a vector having the signal value of each time-series image as an element (FIG. 3). In (Equation 2), {S} and {E} represent the absolute values of the vectors S and E, respectively. E is a user vector that can be arbitrarily determined by the user. The user vector assumes a value of −1 during the rest period and a value of 1 during the stimulus application period. While the signal vector differs for each coordinate, the user vector uses the same vector for each coordinate. (Equation 2) is the inner product (SE) of the signal vector S and the user vector E, where the number of time-series images is the number of dimensions. Therefore, the smaller the angle between the two vectors, the larger the value of (Equation 2). When a correlation function is used to extract an activation region, a threshold value is provided for this value to perform extraction. Now, there is no term in (Equation 2) for inputting the variance of the signal, but as the noise increases, the absolute value of the signal value vector S increases, and as a result, the value of the cross-correlation function decreases. I do. This is shown below. However, for simplification, the description will be made without distinguishing between the rest period and the stimulus application period. Further, the sum of the noise N _i when the number of images is sufficiently large, and becomes 0 as shown in equation (3), the mean value S _m of the signal even SN ratio is changed does not change. In (Equation 3), the subscript i represents the number of the time-series image, and the addition Σ is performed for i = 1, 2,..., N.
[0012]
(Equation 3)
ΣN _i = 0 (Equation 3)
The signal value S _i of each image is considered to be the sum of the average value X _{m of the} signal and the noise N _i . Here, when the coefficients a and b (a <b) representing the magnitude of the noise are used, the signal S _{ai with} a small noise and the signal S _bi with a large noise are expressed as ( _Equation 4) and (Equation 5), respectively. it can.
[0013]
(Equation 4)
S _ai = S _m + aN _i … (Equation 4)
[0014]
(Equation 5)
S _bi = S _m + bN _i … (Equation 5)
Since the same user vector is used regardless of the change in the SN ratio, the magnitude of the denominator (the product of the absolute value of the signal value vector S and the user vector E) in (Equation 2) is equal to the magnitude of the signal value vector S. Depends on. Therefore, the change in the magnitude of the signal value vector with respect to the change in the SN ratio is compared using (Equation 4) and (Equation 5). (Equation 6) is the square of the magnitude of the signal value vector when the noise is small. (Equation 7) is the square of the magnitude of the signal value vector when the noise is large. It can be seen that the difference between (Equation 6) and (Equation 7) is (Equation 8), which is always a positive value. That is, there is a term whose value always increases as the variance of the signal increases. In addition, in (Equation 6), (Equation 7), and (Equation 8), the addition Σ is performed for i = 1, 2,..., N. (Equation 8) shows the difference between the square of the signal value when the noise is large and the square of the signal value when the noise is small.
[0015]
(Equation 6)
Σ ｛S _ai ｝ ² = Σ (S _m + aN _i ) ² = ΣS _m ² + a ² ΣN _i ² + 2aS _m ΣN _i
= ΣS _m ² + a ² ΣN _i ² … (Equation 6)
[0016]
(Equation 7)
Σ ｛S _bi ｝ ² = Σ (S _m + bN _i ) ² = ΣS _m ² + b ² ΣN _i ² + 2bS _m ΣN _i
= ΣS _m ² + b ² ΣN _i ² … (Equation 7)
[0017]
(Equation 8)
Σ ｛S _bi ｝ ^2- S _ai ｝ ² = (b ² -a ² ) ΣN _i ² ... (Equation 8)
As described above, it can be determined that the extraction method using the t-test and the extraction method using the correlation function are easily affected by the change in the SN ratio. Note that an autocorrelation function may be used for extracting an activation region.
[0018]
On the other hand, the extraction method that is not easily affected by the SN ratio includes, for example, an extraction method using a signal change rate. This is because the averaging process is performed when calculating the signal change rate, so that the influence caused by the change in the SN ratio is reduced.
[0019]
Next, a procedure for extracting an activated region using these extraction methods will be described.
[0020]
The method for extracting an activated region using the present invention is divided into two stages. In the first stage, the active region is extracted without erroneous extraction, and in the second stage, the active region is extended with the region extracted in the first stage as a starting point. In the field of image processing, various area expansion methods have been proposed, and these methods can be used in the present invention. However, these methods differ in that the user determines and designates the start point of the area expansion processing from the structure of the target object, whereas the present invention extracts the start point in the first processing. Further, the information of the functional image data created by the extraction method used in the first stage is different from the information of the functional image data created by the extraction method used in the second stage. Is also different from the conventional area expansion processing in that the area is expanded by using.
[0021]
FIG. 1 is a flowchart showing an example of a data processing procedure in an activated area extraction method using the present invention. In the first process (process 1), functional image data (functional image data susceptible to the SN ratio) is created using an extraction method that is susceptible to the SN ratio, and the active region is extracted. These extraction methods can extract an activated region without erroneously extracting a vein. In the second processing (second processing), the area not extracted as the activated area in the first processing is reevaluated. For this evaluation, signal change rate image data, which is functional image data created using an extraction method that is not easily affected by the SN ratio, is used. First, the activation area extracted in the first processing is set as a start point, and the signal change rate of a pixel adjacent to the start point is evaluated. That is, if the signal change rate is equal to or more than the predetermined threshold, the area is expanded as the activation area. Next, the signal change rate of the pixel in the vicinity of the activated area after the expansion is evaluated. If the signal change rate is equal to or more than a predetermined threshold, the pixel is similarly expanded as the activated area. The expansion of the activation region is performed until there is no pixel that satisfies the threshold value of the signal change rate among the adjacent pixels.
[0022]
FIG. 4 shows this state. When expanding the activation region, not only the signal change rate but also a spatial gradient of the signal change rate may be used. By performing a series of processes related to the area expansion, it is possible to prevent the active area from being reduced due to a change in the SN ratio. On the other hand, in order to prevent the activated region from overflowing outside the cerebral cortex, a threshold value may be provided for the signal value of the time-series image to extract the cerebral cortex, and a region to be searched for the activated region may be determined. In this case, the information of the two types of functional images and the information of the morphological images are used together.
[0023]
In the first process and the second process, it is necessary to set a threshold value. In the present invention, the setting method of each threshold value is not restricted at all. That is, the threshold value may be changed each time the activated region is extracted, or the extraction may be performed by repeatedly using a preset threshold value. Further, in the above description of the embodiment, only the measurement using the MRI has been described, but the application target of the present invention is not limited to the MRI.
[0024]
FIG. 5 is a schematic configuration diagram of a magnetic resonance apparatus to which the method of the present invention is applied. 5, the magnet for generating a static magnetic field _{H 0} 621, 622 is measured, 623 a coil for detecting the magnetic resonance signals resulting from the generation and the measurement target 622 of the high frequency magnetic field. Reference numerals 6624, 625, and 626 denote gradient magnetic field generating coils that generate gradient magnetic fields in the x-axis direction, the y-axis direction, and the z-axis direction, respectively. 627, 628, and 629 are gradient magnetic field driving devices that supply current to the respective gradient magnetic field generating coils 624, 625, and 626. The gradient magnetic field generating coil and the gradient magnetic field driving device are collectively referred to as a gradient magnetic field system. 630 is a computer for calculating the measured data, and 631 is a display for displaying the calculation result of the computer 630.
[0025]
Next, an outline of the operation of the inspection apparatus will be described. High frequency magnetic field H ₁ to excite the nuclear spins of the measuring object 622, a waveform shaping the high frequency generated by a synthesizer 632 with modulator 633, and a power amplifier, it generates and supplies a current to the coil 623. The gradient magnetic field generating coils 624, 625, and 626 supplied with currents from the coil driving devices 627, 628, and 629 generate gradient magnetic fields and modulate magnetic resonance signals from the measurement target 622. The modulated signal is received by the coil 623, amplified by the amplifier 634, detected by the detector 635, then AD-converted, and input to the computer 630. After the calculation, the calculator 630 displays the calculation result on the display 631. The computer 630 performs control so that each device operates at a timing and intensity programmed in advance. Among these programs, a program that particularly describes a high-frequency magnetic field, a gradient magnetic field, and the timing and intensity of signal reception is called a pulse sequence.
[0026]
Finally, to summarize the present invention,
A stimulus is applied to the subject for a predetermined period to capture a time-series image, and the signal intensity at the same coordinates of the time-series image is input to a predetermined determination formula, and functional image data having the result of the determination formula as a value is created. In the active area extraction method, which extracts active areas in the brain using functional image data (hereinafter referred to as “active areas”), fluctuations (hereinafter referred to as “noise”) that are superimposed on the signal and have no correlation with the stimulus A process of creating a functional image data reflecting the change in the S / N ratio and extracting an active area using a judgment formula including a term that always increases when the noise increases, and including a term that always increases when the noise increases An active area extraction method is characterized in that it includes a process of generating functional image data that does not reflect a change in the SN ratio and extracting an active area by using a determination formula that does not have a determination formula. The method is the SN ratio A process of setting the region extracted using the functional image data reflecting the change as the start region of the activation region, and a process of extracting the region using the functional image data adjacent to the activation region and not reflecting the change in the SN ratio. Processing that merges the activated area with the activated area, and area expansion processing that repeats the acquisition processing until there is no more area to be newly captured in the activated area, and the area obtained by the area expansion processing is used as the activated area. The feature is to extract.
[0027]
【The invention's effect】
ADVANTAGE OF THE INVENTION According to this invention, even if the SN ratio falls with the improvement of a spatial resolution, an active area can be extracted correctly.
[Brief description of the drawings]
FIG. 1 is a flowchart showing an activation area extraction procedure using the present invention.
FIG. 2 is a diagram showing a measurement example in FMRI.
FIG. 3 is a diagram showing signal vectors.
FIG. 4 is a view for explaining a procedure of an activation region according to the present invention.
FIG. 5 is a diagram showing a configuration example of an MRI apparatus as an example to which the present invention is applied.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... 1st processing in the active area extraction method using this invention 2 ... 2nd processing in the activated area extraction method using this invention 621 ... Magnet for generating a static magnetic field 622 ... Measurement object 623: High frequency magnetic field generation and signal detection coil, 624, 625, 626: Gradient magnetic field generation coil, 627, 628, 629: Coil driving device, 630: Computer, 631: Display, 632: Synthesizer, 633: Modulation device , 634: an amplifier, 635: a detector.

Claims (1)

In an activation region extraction method of extracting a brain function activation region from time-series image data of a magnetic resonance diagnostic apparatus including information on a response to an applied stimulus and creating a functional image,
  Image SN A first functional image created by a first active area extraction method that is susceptible to the ratio, SN A second functional image created by a second active area extraction method that is less affected by the ratio is created, and an activity on the first functional image is created based on an active area on the second functional image. An activated area extraction method characterized by expanding an activated area.

Images