JPH0465969B2 - - Google Patents

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention is an analytical method that can be applied to the analysis of blood cell particle size distribution measurements in the field of clinical testing, as well as to the measurement of particle size distribution of fine particles such as latex particles and cell particles. It is related to.

[Conventional technology]

Measurement of the particle size distribution of blood cells has not been widely used in clinical tests due to the difficulty and accuracy of the measurement method, but recently, with the development of fluid control technology such as the sheath flow method, it has become highly accurate. Automatic particle size distribution measurement has become easier, and particle size distribution analysis has become more important.

Conventionally, as a method for analyzing the particle size distribution of red blood cells,
As shown in Japanese Patent No. 47-13299, it is disclosed that the particle size distribution range is expressed by a quartile change coefficient.

On the other hand, regarding the platelet particle size distribution, a curve fitting method has been used in which the platelet distribution is assumed to be a log-normal distribution curve, and those that do not fit the curve are searched for and displayed as abnormal.

[Problem that the invention seeks to solve]

Conventional analysis of red blood cell particle size distribution only displays the spread of the distribution under the assumption that red blood cell particle size is normally distributed and has a single distribution. However, in reality, there are cases where the distribution is not uniform, and in that case, it will give incorrect information.

Furthermore, the curve-fit method for platelet particle size distribution assumes that platelets have a log-normal distribution, and only requires that platelets that do not fit are considered abnormal and be measured using another method. Moreover, some platelets do not have a log-normal distribution, and in such cases they may give erroneous information.

The present invention has been made in view of the above points, and
Assuming that the blood cell particle size distribution is a normal distribution, a lognormal distribution, or multiple overlaps of these distributions, the parameters of the distribution are estimated, the features of the given blood cell particle size distribution are extracted, and the features of the particle size distribution are extracted. The purpose of this paper is to propose a number of parameters and to provide a blood cell particle size distribution analysis method that can classify blood cell particle size distributions that are considered to be abnormal using these parameters.

Furthermore, the present invention aims to provide new medical information by providing new blood parameters that cannot be obtained using conventional hematological blood cell parameters. Furthermore, the present invention makes it possible to significantly improve the accuracy and reproducibility of platelet count measurement, which has been difficult with conventional methods. The present invention can be effectively applied to particle size distribution analysis of particles other than blood cell particle size distribution.

[Means and actions for solving problems]

The blood cell particle size distribution analysis method of the present invention converts a blood cell pulse proportional to volume into a particle size distribution, imports it into a storage device, removes noise components included in the imported particle size distribution data, and smooths it (due to distribution noise, etc.). To analyze the particle size distribution of blood cells, (a) take the natural logarithm of the smoothed particle size distribution, differentiate this natural logarithm, and (b) set an appropriate window. (processing interval on the horizontal axis indicating the granularity), find the straight line part of the differential coefficient within this window, (c) find the mean value and standard deviation of the estimated distribution from this straight line part, (d) find the calculated mean value and standard deviation of the distribution. Substitute the average value and standard deviation into the normal distribution formula to normalize the estimated distribution, (e) find the area difference between the original particle size distribution and the estimated distribution, and flag those with large area differences as abnormal particle size distributions ( memorize it as one of the symbols)
(f) subdistribution is obtained by subtracting the estimated distribution from the flagged particle size distribution; (g) if the area of the subdistribution is small, the operation is terminated; if the area of the subdistribution is large, a new subdistribution is created; The raw particle size distribution is characterized by (a) repeating the following operations.

In other words, the present invention applies and develops Bhattacharya's method, which has been used in distribution analysis, to blood cell particle size distribution, thereby estimating the population of a given blood cell particle size distribution and separating it into each element distribution. By doing so, we aim to improve the accuracy and reliability of each blood cell count, and further devise new characteristic parameters for particle size distribution to provide new medical information.

Assuming that a given particle size distribution is a normal distribution, a lognormal distribution, or an overlap of multiple distributions thereof, the Bhattakarya method can be applied, which utilizes the fact that the logarithm of the frequency of a normal population is a parabola. The normal distribution is given by the following equation.

f(x)=1/√2πσexp{−(x−μ) ² /2σ
² } In the above formula, μ is the average value and σ is the standard deviation. Taking the natural logarithm of f(x), lnf(x) = ln1/√2πσ−(x−μ) ² /2σ ² = ln1/√2πσ−x ² /2σ ² +μx/σ ² −μ ² /2σ ² If we set g(x)=lnf(x), g(x) becomes a quadratic expression regarding x.

g(x)=-1/2σ ² x ² +μ/σ ² x+ln1/√2πσ
−μ ² /2σ ² When g(x) is differentiated, g'(x) becomes as shown in the following equation.

g'(x)=-1/σ ² x+μ/σ ² =-1/σ ² (x-
μ) g′(x) is a linear equation with μ at the intersection with the x-axis and slope −(1/σ ² ), so for the given distribution
By determining g′(x), the population parameters μ, σ
can be estimated. Furthermore, by defining the matching distance D between the estimated normal distribution and the original distribution using the following equation, it can be used as a characteristic parameter of a given distribution.

D=∫{|f(x)−h(x)|}dx In the above equation, f(x) is the given distribution, h
(x) is the estimated distribution.

Hereinafter, the method of the present invention will be explained in detail based on the flowchart shown in FIG. First, a blood cell pulse proportional to volume is converted into a volume distribution using a comparator circuit, etc., and then stored in a storage device. The captured particle size distribution data is then smoothed to facilitate particle size distribution analysis. That is, noise components included in the particle size distribution are removed. Thereafter, g'(x) is calculated by taking the natural logarithm of the particle size distribution and calculating the difference between adjacent particles, as shown by the Bhattakarya method. Next, as a means of searching for the main distribution of the given distribution, we set an appropriate window and search for the point within this window where g'(x) is closest to a straight line. That is, the part of the particle size distribution curve that is closest to a normal distribution is searched for. A regression linear equation for g'(x) that is closest to a straight line is found, and the mean value and standard deviation of the estimated distribution are calculated from the slope and intercept of this equation.

Next, the obtained mean value and standard deviation are substituted into the normal distribution formula, and the maximum value is normalized using the original particle size distribution. This makes it possible to compare the original particle size distribution and the estimated distribution. Thereafter, by determining the area difference between the original particle size distribution and the estimated distribution, it is possible to quantify how closely the original particle size distribution fits the normal distribution. This value is defined as matching distance D. This D value can be used as a characteristic parameter of particle size distribution. If the area difference is large, that is, the matching distance D is large, it is assumed that the distribution is far from a normal distribution, and is flagged as an abnormal particle size distribution. However, this does not apply to particles whose particle size distribution is known from the beginning to be an overlap of multiple distributions.

In order to check whether the subdistribution of the flagged particle size distribution fits the normal distribution, the estimated distribution is subtracted from the original particle size distribution to obtain the subdistribution. Next, it is determined whether this sub-distribution has a sufficient area to perform particle size distribution estimation again. If it has a sufficient area, g'(x) is calculated using the sub-distribution as a new raw particle size distribution. If the area of the sub-distribution is small, the operation is terminated.

By repeating the above process, it is possible to flag particle size distributions that do not fit the normal distribution, and if multiple distributions overlap,
The distribution of each element can be estimated.

〔Example〕

Hereinafter, the results of implementing the method of the present invention on commercially available E-2000 to 5000 series manufactured by Toa Medical Electronics Co., Ltd. (the applicant of the present application) will be explained.

When applying the multiple distribution Bhattakarya method assuming that the red blood cell (RBC) particle size follows a normal distribution,
The type shown in Figure 2 (hereinafter referred to as type), the type shown in Figure 3 (hereinafter referred to as type), the type shown in Figure 4 (hereinafter referred to as type),
Three types were observed.

The type shown in FIG. 2 shows an almost perfect normal distribution. It can also be perfectly estimated using the single distribution Bhattakarya method. The solid line in the figure shows the given particle size distribution, the broken line shows the estimated normal distribution, and the diagonal plot points show the difference in the natural logarithm of the frequency. The dashed line is not visible because the estimated normal distribution overlaps the given distribution. Q in the figure indicates the parallel distance q between the plot point and the regression line. DIST 1 shows the red blood cell count equivalent value of the estimated normal distribution, and DIST 2 shows the red blood cell equivalent value of the remaining portion after subtracting the estimated normal distribution from the given distribution. As can be seen from FIG. 2, the Q and matching distance D (DEFF) values are becoming smaller.

In the type shown in FIG. 3, the diagonal plot points are not straight lines. Therefore, the estimated normal distribution does not fit the given distribution,
The DEFF value is increasing. Many granularities of this type exist and are easily flagged by the DEFF value. Looking at the pattern, it seems more like a lognormal distribution than a normal distribution.

In the type shown in FIG. 4, the diagonal plot point has two straight line parts. This shows that the given distribution is the sum of two normal distributions, and as shown in the figure, the two distributions are clearly separated. DIES 1 and DIST 2 in the figure indicate the respective red blood cell equivalent values. Like other types, this type also has a large DEFF value, so it can be easily flagged.

The three types of red blood cell particle size have been described above, and since there is a suspicion that the particle size of the type has a lognormal distribution, we applied lognormal distribution estimation. The results are shown in FIG. As you can see from the figure, the estimated distribution almost completely fits the given distribution. The DEFF value is about 1/4 to 1/6 compared to the normal distribution estimation. this is,
type, which suggests that the RBC granularity is log-normally distributed.

Many specimens that seemed to belong to the type were found within the RBC particle size. As a means of distinguishing between types, we use normal distribution estimation.
I compared the DEFF values estimated by the lognormal distribution of DEFF values. The results were as follows.

Similar to the type: 52.8% Similar to the type: 47.2% Total: 100.0% From this, it is estimated that the number of specimens of each type is approximately the same. Until now,
I have thought that the RBC particle size is close to a normal distribution, but
At this point, I have no choice but to change my way of thinking. It is unclear what is the reason for the division into different types. However, it is considered necessary to investigate whether this is unique to the specimen or is due to the measurement principle or method. For reference, normal distribution estimation and lognormal distribution were performed for all RBC particle sizes, and the distribution of DEFF values was investigated. The results are shown in FIGS. 6 and 8. That is, FIG. 6 shows the distribution of DEFF of red blood cells by the normal distribution automatic Bhattakarya method. Note that this is a case where the window is 5 and n (number of samples) is 2.185.
If DEFF≧60 is considered abnormal, the abnormal sample ratio is 16%
Therefore, if DEFF≧100 is considered abnormal, the ratio of abnormal specimens is 6.5%. FIG. 7 shows the distribution of DEFF by the lognormal distribution automatic Bhattakarya method. Note that window: 5, n2:. This is the case for 185. If DEFF≧60 is considered abnormal, the abnormal specimen ratio is 20%, and if DEFF≧100 is considered abnormal, the abnormal specimen ratio is 6.9%.

As is clear from FIGS. 6 and 7, both distributions show a clean logarithmic distribution, and the absolute values of the DEFF values are almost the same. This allows RBC
Regarding particle size, we have reached the conclusion that either normal distribution estimation or lognormal distribution estimation can be used.

Based on the above, it would be best to use the following procedure as a classification flagging method for RBC particle size distribution. First, a normal distribution is estimated, and the type with the smallest DEFF value is determined. Next, a lognormal distribution is estimated, and the type with the smallest DEFF value is determined. Then, in both estimation methods, those with large DEFF values are classified into types as being an overlap of multiple distributions.

Most of the distributions belonging to this type are the sum of two normal distributions covering approximately the same area, as shown in the example shown in FIG. this is,
It would be natural to think of it as a mixture of two types of red blood cells.

Furthermore, when applying the multiple distribution Bhattakarya method assuming that the platelet (PLT) particle size distribution follows a lognormal distribution, it was found that the PLT particle size distribution sample, which was considered to be a normal sample, almost completely fit the lognormal distribution. Figure 8 shows an example of the analysis.

In FIG. 8, the given grain size is shown as a solid line, the estimated lognormal distribution is shown as a dashed line, and the difference in the natural logarithm of the frequencies according to the Bhattacharya method is shown as a diagonal plot. The solid line and dashed line almost overlap, and DEFF
Since also has a small value, it can be concluded that the given distribution is a lognormal distribution. Since the diagonal plot points are also almost straight lines except for the slope on the right side, this assumption is considered to be correct in mathematics.

Next, when we analyzed many PLT particle size distributions under the assumption that the PLT particle size distribution is a lognormal distribution, we found that some samples did not fit the lognormal distribution. Described in ~(3). The DEFF value was used as the classification parameter.

(1) Case where microcyte fogging is observed Figure 9 shows an analysis example of PLT particle size distribution in which the DEFF value is considered to be apparently large due to microcyte fogging. As is clear from the figure, the PLT particle size distribution area and the RBC particle size distribution area largely overlap. Assuming that there is no influence of the RBC grain size at the window position shown in the figure, the estimated lognormal distribution shown by the dashed line indicates the grain size distribution of the PLT. By calculating the area of this distribution, an estimated PLT value can be calculated. This value is
Obtained by separating PLT and RBC with a cursor
It is considered to be more reliable than the PLT value.
In particular, this difference has a large effect on samples with a low PLT value of 50,000 as shown in FIG.

(2) Example of suspected platelet (PLT) aggregation Figure 10 shows an example of analysis of PLT particle size distribution in which a drop in particle size was observed, which was thought to be due to aggregation of two to three PLTs. For these distributions, flagging is also easy because the DEFF values are several to 20 times higher than what is considered normal.

(3) Example of reliability of low platelet (PLT) values For specimens with PLT values in the thousands to tens of thousands, the effects of electrical noise, bubbles, ghosts, etc. become large, reducing accuracy and reproducibility. If the particle size distribution is such that the MDR of the main distribution is greater than or equal to the width of the window, the main distribution can be estimated with good reproducibility. If the assumption that the PLT particle size distribution is a lognormal distribution is correct, the estimated PLT value will be closer to the true value.

FIG. 11 shows an analysis example of the particle size distribution of low PLT values. It can be seen that in particle size distributions with low PLT values, there are almost no valley-like valleys, making separation using a cursor difficult. The current estimation method uses the rise curve of the PLT particle size distribution to accurately estimate the lognormal distribution.

The analysis results regarding PLT particle size have been explained above. The DEFF value used as a parameter is PLT
It was found to be an effective image for detecting particle size abnormalities. When using for flagging,
Figure 12 shows the results of investigating the distribution of DEFF values. In other words, Figure 12 shows the PLT using the lognormal distribution automatic Bhattakarya method.
It shows the distribution of DEFF. Furthermore, the window:
5, n: 2.185. If DEFF≧150 is considered abnormal, the abnormal sample ratio will be 13%, and DEFF≧
If 200 is considered abnormal, the abnormal sample ratio is 6.2%.

The distribution of DEFF values for PLT particle size distribution is
The distribution range is considerably wider than that of RBC, and the absolute value is also large. This indicates that the PLT particle size distribution is easily influenced by other particles. In particular, small red blood cells and platelet aggregate particles are the cause. Also, the influence of what Dr. Bessman calls blood cell debris cannot be ignored.

〔Effect of the invention〕

Since the present invention is configured as described above, it exhibits various excellent effects as described below.

(1) The particle size distribution of red blood cells, which is considered to be normal, can be estimated to be a completely normal distribution, and the particle size distribution, which is considered to be normal, can be estimated to be a completely lognormal distribution.

(2) If the red blood cell particle size distribution is assumed to be a normal distribution, an abnormal particle size distribution can be easily flagged, and if the platelet particle size distribution is assumed to be a lognormal distribution, an abnormal particle size distribution can be easily flagged.

(3) It is possible to separate each blood cell particle size distribution into elemental distributions and correctly estimate their composition ratios. Among these, it has an important meaning in low-value estimation (measurement) of platelet count.

(4) It becomes possible to eliminate the influence of other particles on platelet particle size distribution.

(5) New medical information can be provided by extracting characteristics of particle size distribution using the new parameters, and it is predicted that a sufficiently favorable conclusion will be obtained in case studies.

(6) The reliability of the accuracy of blood cell counts can be significantly improved.

(7) It can be applied not only to blood cell particle size but also to particle size distribution analysis of latex particles, cell particles, etc.

[Brief explanation of the drawing]

Fig. 1 is a flowchart showing an example of the blood cell particle size distribution analysis method of the present invention, Figs. 2 to 12 are graphs in Examples, and Figs. 2 to 4 show examples of red blood cell (RBC) particle size analysis. Graph, Figure 5 is a graph showing an analysis example of lognormal distribution estimation of red blood cell particle size, Figure 6 is the matching distance by normal distribution estimation
A graph showing an example of the distribution of DEFF values, Fig. 7 is a graph showing an example of the distribution of matching distance DEFF values based on lognormal distribution estimation, Fig. 8 is a graph showing an example of platelet (PLT) particle size distribution analysis, and Fig. 9 is a graph showing an example of the distribution of DEFF values based on lognormal distribution estimation. Graph showing an analysis example of PLT particle size with red blood cell cover, Figure 10 is a graph showing an analysis example of PLT particle size suspected of PLT aggregation, Figure 11 is a graph showing an analysis example of low PLT value, Figure 12 is a graph showing an example of distribution of matching distance DEFF values in lognormal distribution estimation.

Claims (1)

[Claims] 1. Converting a blood cell pulse proportional to volume into a particle size distribution and importing it into a storage device, removing noise components included in the imported particle size distribution data and smoothing it, and then analyzing the particle size distribution of the blood cells. To do this, (a) take the natural logarithm of the smoothed particle size distribution and differentiate this natural logarithm, (b) set up an appropriate window and find the linear part of the differential coefficient within this window, (c) Find the mean value and standard deviation of the estimated distribution from this straight line part, (d) substitute the found mean value and standard deviation into the normal distribution formula to normalize the estimated distribution, and (e) estimate the original particle size distribution. Find the area difference from the distribution, flag those with large area differences as abnormal particle size distributions, (f) sub-distribution by subtracting the estimated distribution from the flagged particle size distribution, and (g) if the area of the sub-distribution is small. (a) If the area of the sub-distribution is large, the sub-distribution is newly set as a raw particle size distribution and the following operations are repeated: