JP2939442B2 - Quantitative method and apparatus, cerebral blood flow quantitative method and apparatus - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for quantitatively measuring the amount represented by the density of an image from image data, and a method and apparatus for measuring cerebral blood flow from a medical SPECT image. It concerns the device.

[0002]

2. Description of the Related Art In a diagnosis using a medical image, a method is used in which a part is compared with another part and the relative change is visually observed, or the comparison is performed by calculation using image density. I have been. On the other hand, in the measurement of cerebral blood flow,
It is necessary to measure the absolute value of the amount, for example, an absolute amount such as a flow rate of 40 ml per minute per 100 gr of tissue weight. The classical method of this measurement is to, for example, administer a labeled substance labeled with a radioisotope to a living body, repeat blood collection from arteries and veins several times, and analyze changes over time in the concentration of the labeled substance in the blood. There is something. (Ketty SS & Schmidt
CF, “The nitrous oxiside method for quantitative
determination of cerebral blood flow in men: theor
y, procedure and normal values ", J Clin Invest 27, P
(See pages 476-483, 1948)

[0003] For example, SPECT (Single Photon)
To measure cerebral blood flow from image data such as SPECT images obtained by Emission Computed Tomography (single photon tomography), the density of the image data indicates the relative value of cerebral blood flow. Need to be calibrated and quantified. As a conventional method of such quantification, a method of measuring cerebral blood flow by acquiring and analyzing SPECT images in time series is known. (See Hyugano, " ¹²³ I-IMT Brain SPECT Compartment Model Analysis", Nuclear Medicine 27, PP.51-54, 1990)

[0004]

In the above-described method for measuring cerebral blood flow by Ketty et al., It is necessary to repeatedly collect blood from arteries and veins, which places a heavy burden on the subject. Also,
In any of the above-mentioned methods, analysis of time-series data is used, and for that purpose, data acquisition is required at least three times, generally more times, which makes measurement complicated. There was a problem.

It is an object of the present invention to obtain two measurements obtained by two measurements.
Quantification method and apparatus capable of quantifying the amount represented by the image density from two image data, and a medical S
An object of the present invention is to provide a method and an apparatus for measuring cerebral blood flow from a PECT image.

[0006]

In order to achieve the above-mentioned object, the present invention provides a method for measuring a surface-distributed measurement target amount by an image pickup means and quantifying the measurement target amount from a pixel density of the picked-up image. A quantification method that captures a first captured image capturing the measurement target amount and a second captured image captured when the measurement target amount is changed, and determines a pixel density of each captured image in advance. And the abscissa and ordinate values of the standardized density obtained from the densities of pixels representing the same position on the measurement object on the first and second captured images. The measurement points are plotted on one plane to generate a measurement plane, and the measurement points on the plane are subjected to regression analysis using the information criterion, and the measurement target area representing the measurement target quantity, On the screen where the measurement points of Into the marginal error region representing the measurement point of the margin of the region, the intersection of the boundary between the measurement target region and the marginal error region generated by the separation and the regression line determined in each region. From the coordinates, and the slope of the regression line,
Inputting the measurement target amount, estimating an approximate value of a parameter of a predetermined input / output characteristic with the normalized concentration as an output, further setting a range of parameter values including the approximate value of the parameter, and setting the range. Within the parameters, the output-output characteristics on the measurement plane representing the relationship between the standardized density obtained from the two captured images obtained from the input / output characteristics corresponding to the value of each parameter, A quantitative method characterized by determining a parameter having the best degree of conformity with the distribution of the measurement points, and quantitatively determining a measurement target quantity as an input value from the input / output characteristics having the parameter thus determined. I will provide a.

[0007] The present invention also provides a method for quantifying cerebral blood flow, characterized in that the amount to be measured is cerebral blood flow and the imaging means is single-photon tomographic imaging means.

The present invention also provides a method for increasing the cerebral blood flow at the time of capturing the second captured image and determining the amount of a labeling substance to be administered to a subject to measure the cerebral blood flow. The cerebral blood flow rate is characterized in that the input / output characteristics include a despreading characteristic in which the output is reduced by an amount proportional to the cerebral blood flow rate as the input, while increasing the input / output characteristics as compared with the time of capturing the image. To provide a method for quantification.

Further, according to the present invention, there is provided an image pickup means for picking up an image of an object to be measured which is distributed in a plane, a first image picked up by the image pickup means, and an amount of the object to be measured. Input means for capturing a second captured image captured at the time, and a standardized density for calculating a standardized density by standardizing a pixel density of each captured image captured by the means by a predetermined method. Calculating means, and measuring points at which the standardized density obtained from the density of pixels representing the same position on the measurement target on the first and second captured images is used as its abscissa and ordinate values. A measurement plane is generated by plotting on one plane, and the measurement points on the generated measurement plane are subjected to regression analysis using an information criterion. The area on the screen where you gather Measuring point separating means for separating the measurement target area and the marginal error area representing the measuring point of the marginal area, and a boundary between the measurement target area and the marginal error area generated by the means and the respective areas. Coordinates of the intersection with the regression line
And from the slope of the regression line, a parameter estimating means for estimating an approximate value of a parameter of a predetermined input / output characteristic with the measurement target amount as an input and the standardized concentration as an output, A parameter range including an approximate value of the parameter is set, the parameter is changed within the range, and the value of 2 obtained from the input / output characteristics corresponding to the value of each parameter is set.
Output-output characteristics on the measurement plane representing the relationship between the standardized densities obtained from the two captured images, and parameter correction means for determining a parameter having the best degree of conformity with the distribution of the measurement points. The present invention provides a quantitative device characterized by the following.

Further, the present invention provides a cerebral blood flow rate quantifying device, wherein the measurement target amount is a cerebral blood flow amount, and the imaging means is a single photon tomographic imaging means.

[0011]

Embodiments of the present invention will be described below in detail. FIG. 2 is a block diagram showing a configuration example of a cerebral blood flow measuring device according to the present invention, and FIG. 1 is a flowchart showing a process for measuring cerebral blood flow using this device.
In this apparatus, in order to obtain a SPECT image of the brain, the subject was labeled with H labeled with a radioisotope nuclide Tc-99m.
M-PAO (Hexa Methyl-Propylene Amine Oxime) or ECD (Ethyl Cysteinate Dimer) is administered intravenously. Several minutes after this administration, an image of the head is taken by the SPECT imaging device 201, and a first SPECT image is captured (step 101). Furthermore, administration of a carbonic anhydrase inhibitor (acetazolamide; a diuretic and antiepileptic agent, which is also used to measure cerebral blood flow) to the subject increases cerebral blood flow, which increases the cerebral blood flow. A second SPECT image in the state is captured (step 102). Each of the two images obtained at this time is similar to a head tomographic image or the like by X-ray CT.

Next, in the preprocessing means 202, in order to compare the densities of the pixels at the same position in the two obtained images, the two images are subjected to processes such as shift, rotation, enlargement / reduction, etc. Is corrected to an average form, and at the same time, the two images are aligned. Further, smoothing based on the spatial resolution of the measuring machine and standardization of the pixel density of the image are performed (step 103). "Position" of alignment here
This is a position on coordinates virtually provided on the head, and this position is the same pixel (pixel) on the two images.
By making the positions coincide with each other, the image density at the same position on the head can be compared by comparing the densities at the same pixel position on the image. The smoothing is a process of adding the pixel value of each pixel, for example, by multiplying each of the 3 × 3 pixel values centered on the pixel by an appropriate weighting coefficient. Further, the standardization of the pixel density of the image means the following processing. That is, the first and second SPEs
Each pixel density of CT image

## EQU1 ## p11, p12,... P1n p21, p22,... P2n n: the number of pixels of an image, and the maximum value of these 2n pixel values is pmax, the standardized density q To

[Mathematical formula-see original document] qij = (pij / pmax) * (C-1) i = 1, 2, 1≤j≤n. Here, C is the number of levels representing the density. For example, if an image using eight planes can represent 256 levels, C is set to 256. Also, the right side of the first equation (Equation 2) truncates the decimal part. Note that these preprocessings are necessary because measurement conditions such as position and illuminance at the time of the first and second SPECT image acquisitions do not generally match.

FIG. 3 shows an example of the actual measurement after the pre-processing as described above. The horizontal axis represents the standard obtained by standardizing the data p1 obtained from the first SPECT image by (Equation 2). The normalized density q1, and the vertical axis is the normalized density q2 obtained by normalizing the data p2 obtained from the second SPECT image by (Equation 2). P1j and p2j (j = 1 to n) in (Equation 1) are each SPEC
If they are located at the same position in the T image,
(Q1j, q2j), j = 1 to n obtained by the above q1-
FIG. 3 is a plot on the q2 plane. And FIG.
(A) to (e) show the case where the above-mentioned labeling substance was given to different subjects, and the cerebral blood flow was increased at the second measurement by further giving the carbonic anhydrase inhibitor.
(F) shows the case where measurement was performed twice without giving the carbonic acid dehydrogenation inhibitor.

Here, the meaning of the SPECT image obtained above and the model of a measurement system used for analyzing and quantifying the SPECT image will be described, apart from FIGS. SPEC
The pixel value of the T image is the marker Tc-99m included in the bloodstream.
It is obtained by detecting the radiation dose from the, the greater the cerebral blood flow, the greater the radiation dose per unit time,
The pixel value becomes a higher level value as the cerebral blood flow increases.
Therefore, when the cerebral blood flow is expressed as CBF, the measured pixel value (standardized density) q (hereinafter referred to as output) is calculated for each pixel.

## EQU3 ## There is a functional relationship of q = f (CBF, PARA). However, PARA is a parameter included in the function f, and is generally plural. If the output q is in a proportional relationship with the cerebral blood flow CBF, the relationship between the outputs q1 and q2 obtained by the two measurements is also in a proportional relationship, and if plotting as shown in FIG. 3 is performed, (q1, q2) The points are distributed on a substantially straight line. However, actually, the characteristics of the measurement system are not linear, but have nonlinear characteristics as shown in FIG. Using this characteristic, two cerebral blood flows CBF1, CBF
Assuming BF2,

Q1 = f (CBF1, PARA) q2 = f (CBF2, PARA) CBF2 = CBF1 (1 + △) Outputs q1, q2 are calculated and plotted on the (q1, q2) plane. A characteristic like 5 is obtained.
However, the increase rate in FIG. 5 is the value of Δ in (Equation 4).

Referring to each measurement example of FIGS. 3 (a) to 3 (e),
It can be seen that the distribution of the measurement points shows the same tendency as the curve shown in FIG. FIG. 3F shows the case where △ = 0, which also shows the same tendency as the case of △ = 0 in FIG. Therefore, in the present invention, the relationship of (Equation 3) is assumed as a model of the measurement system, and parameters included in the model are determined so that the assumed model and the measurement data as shown in FIG. Then, the cerebral blood flow rate CBF and the measurement output q can be determined in a one-to-one relationship from the relationship of (Expression 3).
The cerebral blood flow can be quantified using the two SPECT image data.

For this purpose, it is necessary to provide the model of (Equation 3) in a concrete form. This is to provide a function approximating the curve of FIG. 4 as described above. , Commonly used

It is assumed that q = 2.59q0.y / (1 + y) y = CBF / (2.59.CBF0). (For example, Lassen NA, etc. "The retention
of [ ^99m Tc] -d, l-HM-PAOin the human brain after intr
acarotid bolus injection: A kinetic analysis. J Ce
reb Blood Flow Metab 8: s13-s22, 1988) In this model, the parameters PARA in (Equation 3) are q0 and CBF0, but CBF0 is empirically 40 ml / min / 1.
Since this is 00gr, this value is used below. This value is the value of the input CBF near the point where the curve f in FIG. 4 departs from a state substantially along the straight line LM. However, the relationship of (Equation 5) is such that, for a normal value of q0, as shown in FIG. 6, a straight line L (point (CBF
0, q0) and a straight line passing through the origin), this range is a straight line L, the range of CBF> CBF0 is (Equation 5), and the area near the point (CBF0, q0) is smoothed. For this purpose, a curve f connected by spline interpolation is used as a function representing the curve in FIG.

Further, even if the measurement system is the same, the output coefficient q0 included in the above model is actually affected by a great number of factors even in the same measurement system. Generally different values. Therefore, slightly modify (Equation 4)

[Mathematical formula-see original document] q1 = f (CBF1, q01) q2 = f (CBF2, q02) CBF2 = CBF1 (1+ [Delta]), which is hereinafter referred to as an input / output model. Here, as described above, f is a function obtained by smoothly connecting the straight line L of FIG. 6 and the function of (Equation 5).

Returning to FIG. 1 and FIG. 2, in preparation for collating the input / output model with data obtained by actual measurement,
First, statistical processing is performed on two measured values of the standardized density, that is, the output as shown in FIG. The output (q1,
The distribution of q2) is composed of three parts, which are a noise area, a marginal error area, and a measurement target area in order from the lowest level. Since the measurement is performed such that the detection level of the labeled blood flow is a certain level of brightness or more, the level forms a measurement target area, and the vicinity of level 0 is mainly a noise component at the time of measurement, which is a noise component. Form an area. The marginal error region between the two indicates a level portion near a boundary between a portion where the labeling substance exists and a portion where the labeling substance does not exist. Although there is almost no correlation between q1 and q2 in the noise region, there is a correlation between q1 and q2 in the measurement target region and the marginal error region.

Therefore, the statistical processing device 203 shown in FIG.
First, the division positions of the noise region and the marginal error region are determined. For this purpose, first, the entire measurement points distributed on the (q1, q2) plane are approximated by one straight line. This straight line is, for example, a straight line connecting the point (0, 0) and the point (q1m, q2m) by taking the maximum values q1m and q2m of q1 and q2 for the entire measurement point. Next, when orthogonal lines are drawn at a sufficiently small interval so as to be orthogonal to this straight line, a band-like region is formed between these orthogonal lines, and this band-like region is sequentially formed from the origin (0, 0) side. Number j
= 1, 2,..., And the number of measurement points included in the band-shaped area of number j is Nj. Then, a histogram of the measurement points along the straight line direction is obtained. In this histogram, due to the nature of noise, Nj decreases as the distance from the origin increases and reaches a minimum point. From here, Nj increases again, and this minimum point approximately gives a branch point between the noise area and the marginal error area. Then, measurement points smaller than the branch point are removed and subsequent analysis is performed. This is the process of step 104 in FIG.

Subsequently, the statistical processing unit 203 separates the marginal error region from the measurement target region. This is the process of step 105 in FIG. Since the marginal error area and the measurement target area each have a linear correlation, each area can be represented by a regression line. Therefore, the information criterion (AIC) developed by Akaike et al. For separating the two regions is used. (See, for example, Suzuki, "Read-Ahead Statistics-What is the" Information Criteria "?", Bluebacks B-855, 1994, 4th print.) In general, the regression line is a sample point distributed with the obtained regression line. Is calculated based on the sum of squares of the differences of the dependent variables of, that is, the amount called the residual sum of squares. However, when a model with an increased number of independent variables is created, the residual sum of squares decreases as the number of independent variables increases. It is difficult to judge the quality of the model using only the residual sum of squares, so the number of parameters included in the model is also taken into consideration.

It has been proposed to use AIC = Nlog (QN / N) + 2 · (number of parameters) as an evaluation criterion.
In Equation (7), N is the number of samples, and QN is the residual sum of squares.

The separation of the marginal error region and the measurement target region using the AIC is performed as follows. The points on the (q1, q2) plane to be processed here are a set of points (q1j, q2j) created from the normalized density qij obtained by normalizing the pixel density of (Equation 1) by (Equation 2). Yes, step 10 in FIG.
4 except for the points included in the noise area obtained in step 4. If the point set to be processed is written as G, the set G is divided into two sets G1 and G2 as follows;

G1 = {(q1j, q2j)} G | q2j ≧ qs G2 = {(q1j, q2j)} G | q2j <qs} That is, the horizontal plane of q2 = qs on the (q1, q2) plane A straight line is drawn, and a set of G points above the straight line is G1, and a set of G points below the straight line is G2. The qs is set as a search level, and the AIC is calculated for each of the two sets of data G1 and G2 while moving the search level qs little by little, and the search level at which the sum of the two AICs is minimum is calculated. qsm2 is used as the boundary between the marginal error region and the measurement target region. This is the process of step 105 in FIG.

FIG. 7 schematically shows regression lines L1 and L2 of each region obtained by the AIC analysis.
The curve inferred from the model between the two outputs described in FIG. 5 corresponds to the one obtained by smoothly connecting the two straight lines L1 and L2 in FIG. The connection between the two straight lines substantially corresponds to the point where the curvature of the (q1, q2) characteristic in FIG. 5 is almost maximum, and therefore also substantially corresponds to the point Q0 in the input / output characteristic (model) of the measurement system shown in FIG. doing. Therefore, the above qsm2
Is set as q02 of (Equation 6), and qsm1 in FIG.
The coordinate value of q1 when the two coordinates become qsm2) is (Equation 6)
Q01.

Further, 2 obtained in the marginal error region
The value of the data q2 of the second SPECT image is obtained in the peripheral portion of the region where the labeling substance exists, and the blood flow does not increase in this region as in the measurement target region.
For this reason, the data in this area approximately corresponds to the increase rate の of the cerebral blood flow in (Equation 6) being 0. Further, in the measurement target region, q2> q1
Becomes Therefore, the difference between the slopes of the two regression lines in FIG. 7 obtained by the AIC analysis can be regarded as 1 + △ as a first-order approximation if the function f of (Equation 6) is regarded as a straight line.

In this way, the approximate values of the parameters q01, q02, and Δ of the input / output model of the measurement system represented by (Equation 6) were obtained from the AIC analysis in step 105 in FIG. Therefore, next, a more accurate model of the measurement system is generated by the model generation device 204 of FIG. Therefore, if q01, q02, and Δ are parameters in (Equation 6), the variable CB
Via F1 the output q2 is represented as a function 出力 of the output q1;

## EQU9 ## q2 = ψ (q1, q01, q02, Δ) On the contrary, the output q1 is also represented as a function of q2 as the inverse function;

## EQU10 ## q1 = １０ (q2, q01, q02, Δ) The two functions （and の of (Expression 9) and (Expression 10) are derived from the input / output model of the measurement system of (Expression 6). Where
A model showing the functional relationship between two measurement outputs q1 and q2,
Hereinafter, this is called an output model. The model generation device 204 performs the following processing using this output model.

As described in the description of the AIC analysis, among the measurement points, the points belonging to G1 in (Equation 8)
05, parameters q01, q02, measured as initial values,
The point corresponding to the output model of (Expression 9) or (Expression 10) using Δ and belonging to G2 of (Expression 8)
This corresponds to an output model in which only 0 is set. Therefore, as an evaluation amount for checking the correspondence, a square sum ERS of a difference between the output model and the measurement point in the horizontal direction (q1 direction) and the vertical direction (q2 direction) is used.
Consider UM.

Equation 11] ERSUM = E1 + E2 E1 = Σ {ψ (q1j, q01, q02, Δ) -q2j} 2 + Σ {ξ
(Q2j, q01, q02, Δ) -q1j｝ ² , (q1j, q2j) ∈G1 E2 = Σ ｛ψ (q1j, q01, q02,0) -q2j｝ ² + Σ ｛ξ
(Q2j, q01, q02,0) -q1j｝ ² , (q1j, q2j) ∈G2 And the parameters q01, q02, and Δ are slightly within a range of about ± several% around the initial values obtained from the AIC analysis. The parameter that minimizes the evaluation amount ERSUM in (Equation 11) is obtained. This is step 10 in FIG.
As a result, in the processing of No. 6, parameters that match the measured data as a model more than the above initial values are obtained. Since q01 and q02 among the parameters thus obtained are parameters of the input / output model, the measured data q1
By solving (Expression 6) from the above, the cerebral blood flow rate CBF1 can be quantified. However, (Equation 6) of the input / output model and (Equation 9) and (Equation 10) of the output model derived therefrom are not one simple function. An up table is created, and a function value or its inverse function is obtained using the up table.

As described above, by performing each processing shown in FIG. 1, the cerebral blood flow can be quantitatively measured from only two image data, but this processing method can be modified. One of them is that in the process of making the output model more consistent with the measurement data in step 106, the measurement point is set to the point set G2 of the marginal error region as shown in (Expression 11). The target area is divided into a set of points G1, and each is set to Δ
= 0 and Δ ≠ 0, but all points in the marginal error region and the measurement target region are set to Δ ≠.
An evaluation method that matches the output model of 0 may be used. As the evaluation amount in this case, the following ERSUM1 is used instead of (Equation 11).

[Equation 12] ERSUM1 = 1 (q1j, q01, q02,
Δ) −q2j｝ ² + {(q2j, q01, q02, Δ) −q1j}
² , (q1j, q2j) G. Further, in (Equation 11) or (Equation 12), the amount of calculation can be reduced by using the square error of only the q1 direction or only the q2 direction as an evaluation amount.

In the AIC analysis in step 105, the separation into the marginal error region and the measurement target region is performed by q2 =
It is also possible to perform the division by dividing q1 = the left and right sides of a certain straight line, instead of dividing the top and bottom of a certain straight line. For these deformations, a more suitable one may be used according to the characteristics of the measurement object and the measurement system.

Next, in the embodiment described above, (Equation 6) is assumed as an input / output model of the measurement system. This function approximates the input / output characteristics as shown in FIG. 4 by smoothly connecting a part of the hyperbolic function of (Equation 5) and a straight line. In particular, the input, that is, the cerebral blood flow CBF is large. However, they use a hyperbolic function. However, considering in more detail, as the cerebral blood flow rate increases, the blood flow velocity also increases.Therefore, the factor that the amount of the labeling substance labeled by the nuclide in the brain decreases also increases. The output q tends to decrease as the input CBF is larger than the input / output model shown in FIG. This phenomenon is called back diffusion. In the following, a method for quantifying cerebral blood flow based on a more precise model taking into account this back diffusion will be described.

In order to make the above-mentioned reverse diffusion remarkable in the measurement result, the amount of the labeling substance in the second measurement for increasing the cerebral blood flow is administered more than in the first measurement. Further, as the input / output model, a function shown in (Expression 13) is used as a base. (Y.Yonekura et al, "Bra
in Perfusion SPECT with ^99m Tc-Bicisate: Comparsionw
ith PET Measurement and Linearization Based on Per
meability-Surface Area Product Model ", J. Cerebral B
lood Flow and Metabolism 14 (Suppl.1): S58-S65, 1994
reference)

[Mathematical formula-see original document] q = fe (CBF, PARA) = q0 * (CBF / CBF0) * H (CBF) / H (CBF0) H (x) = 1-exp (-PS / x) where q0 and CBF0 are The parameters are the same as in the case of (Equation 6), and PS is an experimentally given constant.

Therefore, the input / output characteristic q1 at the time of the first measurement
(Normalized pixel density) -CBF1 (First cerebral blood flow)
(Expression 13) is used as it is.

Q1 = fe (CBF1, PARA) On the other hand, in the second measurement, the cerebral blood flow is increased by Δ × 100% as described above, and at the same time, the labeling substance is administered more than in the first measurement. From the above, assuming that the dose is a (> 1) times the first time, the contribution q22 of the count value of the labeling substance administered at the second time to the image density is

## EQU15 ## q22 = a.fe (CBF2, PARA), and CBF2 = CBF1. (1 + .DELTA.). In addition, at the time of the second measurement, the labeling substance administered at the time of the first measurement is also counted, but the labeling substance decreases with time. So the first
Assuming that b (<1) times the second dose remains at the time of the second measurement, the contribution q21 of the count value of this labeling substance to the pixel density is

## EQU16 ## q21 = b.q1. What is actually measured at the time of the second measurement is the sum q22 + q21 of q22 and q21 of (Equation 15) and (Equation 16), but this sum is not output as it is, and as described above. The value reduced by despreading is measured. The amount of decrease by the inverse diffusion is assumed to be proportional to the output at the time of the second measurement. If the proportional coefficient is γ / a (a is a coefficient of (Equation 15)), the second measurement value (pixel Concentration) q2

[Mathematical formula-see original document] q2 = q22 + q21-([gamma] / a) * q22 = (a- [gamma]) * fe (CBF2, PARA) + b * fe (CBF1, PARA)

As described above, based on (Equation 13), the first and second calculations based on the model considering despreading are performed.
As can be seen from (Equation 13) to (Equation 17), the output q1 and q2 of the second time are the ratio of the dose of the second labeling substance to that of the first time a, and the second measurement of the labeling substance administered at the first time. Decay rate b, parameters PS and CB included in (Equation 13)
F0 is included as a parameter. However, since these quantities are determined by preliminary experiments, etc., the actual output q1,
The only parameters remaining in q2 are q0, Δ and γ. FIG.
5 shows an example of the output-output characteristic thus obtained, that is, an example of an output model.
Of the output model. However, in this example,

## EQU18 ## This is the characteristic when PS = 66.2, CBF0 = 55.0, a = 1.5, b = 0.95, q0 = 55.0, and γ ≒ 0.2. As shown in FIG. 8, the output q2 shows a tendency of saturation where the output q1 is lower than that of FIG. 5 when the output q1 is large by considering the despreading, and this tendency also appears when Δ = 0. You can see that

Thus, an output model as shown in FIG. 8 was obtained from the new input / output model represented by (Equation 14) and (Equation 17). Then, as in the case of the model of (Equation 6), after the processing shown in FIG. 1, that is, the acquisition of two image data, the preprocessing thereof, and the separation of the noise region, the approximate values of the parameters are measured. Estimated by AIC analysis from data,
Further, by performing an analysis that the model parameters are determined by changing the obtained approximate parameters so that the output-output curve indicated by the model most closely approximates the measured data, the following equations (Equation 14) and (Equation 17) are obtained. The parameter q0 (and Δ,
γ) can be estimated, and the cerebral blood flow can be quantitatively determined. However, it is assumed that the initial value of γ takes a value of about 0 or 2.

FIG. 9 shows an example in which this analysis is performed on the actually measured data, and the values of the assumed parameters are the same as (Expression 18). A curve L3 is an output-output characteristic (output model) obtained from (Equation 14) and (Equation 17) using the estimated parameter values, and a straight line L4 is a regression line (Δ = 0).

In the above description, the case of quantifying the cerebral blood flow from the SPECT image of the head has been described as an example. However, the present invention is not limited to this, and the saturation characteristics as shown in FIG. It is clear that the present invention can be applied to any measurement system having the following as its input / output characteristics.

[0035]

According to the present invention, it is possible to quantitatively measure the amount represented by only two image data, and the efficiency of the quantitative measurement is greatly improved. In particular, when used for measurement of cerebral blood flow, the time for restraining the subject to be measured can be reduced, and blood sampling is not required, so that the burden on the subject can be greatly reduced.

[Brief description of the drawings]

FIG. 1 is a flowchart showing a cerebral blood flow measurement method according to the present invention.

FIG. 2 is a block diagram showing a configuration example of a cerebral blood flow measurement device of the present invention.

FIG. 3 is a diagram illustrating an example of distribution of measurement points when the pixel densities of two images are represented by vertical and horizontal coordinates.

FIG. 4 is an explanatory diagram of an input / output model of a measurement system.

FIG. 5 is a diagram showing an output model created by vertically and horizontally outputting two images from the input / output model of FIG. 4;

FIG. 6 is an explanatory diagram of an input / output model creation method.

FIG. 7 is a diagram showing an example of a regression line obtained by AIC analysis.

FIG. 8 is a diagram illustrating an example of an output model when despreading is considered.

FIG. 9 is a diagram illustrating an example of a parameter estimation result when despreading is considered.

[Explanation of symbols]

 Reference Signs List 201 SPECT imaging device 202 Preprocessing device 203 Statistical processing device 204 Model generation device

Claims (15)

(57) [Claims] 1. A quantification method for imaging a measurement target quantity distributed in a plane by an imaging means and quantifying the measurement target quantity from a pixel density of the picked-up image, wherein: The captured image and the second captured image captured when the measurement target amount is changed are captured, and the pixel density of each captured image is set as a standardized density standardized by a predetermined method, The measurement points where the standardized densities obtained from the densities of the pixels representing the same position on the measurement target on the captured image are taken as their abscissa and ordinate values are plotted on one plane, and the measurement plane is plotted. By generating the measurement points on the plane and performing regression analysis using the information criterion, the measurement target area representing the measurement target quantity and the measurement points at the edge of the area on the screen where the measurement points of the area are collected are obtained. Separated into marginal error regions From the coordinates of the intersection of the boundary between the measurement target region and the marginal error region generated by the separation and the regression line determined in each region, and the slope of the regression line, the measurement target amount is input, and the reference Approximate values of the parameters of the predetermined input / output characteristics are estimated using the chemical concentration as an output, and further, a parameter value range including the approximate values of the parameters is set, and the parameters are changed within the range. The output-output characteristic on the measurement plane representing the relationship between the standardized densities obtained from the two captured images obtained from the input / output characteristics corresponding to the values of A parameter to be best determined, and the input / output characteristic having the parameter thus determined is quantitatively determined as a measurement target quantity as an input value. Method. 2. Prior to separation of the measurement target area and the marginal error area, a regression line approximating all measurement points is determined, and a histogram of the number of the measurement points in a direction along the straight line is determined. The method according to claim 1, wherein a minimum point of the histogram that appears first from the origin along the straight line is obtained, and a measurement point on the side closer to the origin along the regression line from the minimum point is removed. Quantitation method. 3. The regression analysis using the information amount criterion is such that when one of the ordinate values of the measurement plane is set as a search level, the information amount criterion for the set of the measurement points above the search level is determined. A search level at which the sum of the set of measurement points below the search level and the information amount criterion is minimized, and a search target area above the search level and a margin error area below the search level are separated. The quantification method according to claim 1, wherein: 4. The regression analysis using the information amount criterion is such that, when one of the abscissa values of the measurement plane is a search level, the information amount criterion for the set of measurement points to the left of the search level is A search level that minimizes the sum of the measurement point set to the right of the search level and the information criterion, and separates the right from the obtained search level as a measurement target area and the left as a marginal error area. The quantification method according to claim 1, wherein: 5. The degree of conformity between the output-output characteristic on the measurement plane and the distribution of the measurement points is determined by calculating the sum of squares of the horizontal distance between each of the measurement points and the output-output characteristic and the vertical direction. The method according to claim 1, wherein one or both of the sums of squares of the distances are determined as an evaluation value. 6. An output-output characteristic when calculating the evaluation value, wherein, for a measurement point belonging to the marginal error region, the same measurement target amount is used for the second captured image and the first captured image. The quantification method according to claim 5, wherein an output-output characteristic when is used. 7. The quantification method according to claim 1, wherein the amount to be measured is cerebral blood flow, and the imaging means is a single photon tomographic imaging means. A method for quantifying cerebral blood flow. 8. The method according to claim 8, wherein the cerebral blood flow is increased at the time of capturing the second captured image, and the amount of the labeling substance administered to the subject for measuring the cerebral blood flow at the time of capturing the first captured image. 8. The brain according to claim 7, wherein the input / output characteristics include a despreading characteristic in which the output decreases by an amount proportional to the cerebral blood flow as the input. A method for quantifying blood flow. 9. An image pickup means for picking up an image of a measurement target amount distributed in a plane, and a first picked-up image obtained by picking up the measurement target amount by the image pickup means and an image pickup when the measurement target amount is changed. Second
Input means for capturing a captured image of the reference image; reference density calculating means for calculating a standardized density by standardizing a pixel density of each captured image captured by the means by a predetermined method; and The measurement points where the standardized densities obtained from the densities of the pixels representing the same position on the measurement target on the first and second captured images are used as their abscissa and ordinate values are plotted on one plane. To generate a measurement plane, and the measurement points on the generated measurement plane are subjected to regression analysis using information criterion,
A measurement point separation unit for separating the measurement target region representing the measurement target amount and a marginal error region representing a measurement point at an edge of a region on the screen where measurement points of the region are gathered; From the coordinates of the intersection of the boundary between the measurement target area and the marginal error area generated separately and the regression line determined in each area thereof, and the slope of the regression line, the measurement target amount is input, Parameter estimating means for estimating an approximate value of a parameter of a predetermined input / output characteristic using the standardized density as an output; and setting a parameter value range including the approximate value of the parameter estimated by the means. The parameters are changed within the range, and the relationship between the standardized densities obtained from the two captured images obtained from the input / output characteristics corresponding to the values of the parameters is on the measurement plane. And force over output characteristics, quantification apparatus being characterized in that and a parameter correcting means for goodness of fit and distribution of the measurement points defines the best become parameters. 10. A regression line approximating all measurement points is obtained before separation of the measurement target area and the marginal error area by the measurement point separation means, and the number of the measurement points in a direction along the straight line is determined. A noise removal means for obtaining a histogram, obtaining a minimum point of the histogram which appears first from the origin along the regression line, and removing a measurement point closer to the origin along the regression line from the minimum point; The quantification device according to claim 9, further comprising: 11. The method according to claim 11, wherein, when one of the ordinate values of the measurement plane is set as a search level, the measurement point separating unit sets an information amount criterion for the set of the measurement points above the search level and the search level. Finding a search level that minimizes the sum of the information point criterion for the set of measurement points below it, measuring above the found search level, and separating the lower part as a marginal error area. The quantification device according to claim 9, which performs the measurement. 12. The measurement point separation means, wherein one of the abscissa values of the measurement plane is set as a search level, an information amount criterion for the set of measurement points to the left of the search level, and the search level Finding a search level that minimizes the sum of the information point criterion for the set of the measurement points on the right, and separating the right from the found search level as a measurement target area and the left as a marginal error area. The quantification device according to claim 9, which performs the measurement. 13. The parameter correction means according to claim 1, wherein the degree of conformity between the output-output characteristics on the measurement plane and the distribution of the measurement points is determined by determining the horizontal distance between each of the measurement points and the output-output characteristics. The quantification device according to claim 9, wherein one or both of the sum of squares and the sum of squares of the vertical distance are determined as an evaluation value. 14. The apparatus according to claim 1, wherein the parameter correction unit determines the output-output characteristic when calculating the evaluation value for the measurement point belonging to the marginal error region. 14. An output-output characteristic when the images have the same measurement target amount is used.
The quantification device according to 1. 15. The quantification device according to claim 9, wherein the amount to be measured is cerebral blood flow,
An apparatus for quantifying cerebral blood flow, wherein the imaging means is a single-photon tomographic imaging means.