JPS6337346B2 - - Google Patents

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention uses an image of a stained blood specimen to
The present invention relates to an apparatus that identifies reticulocytes and automatically performs classification based on the difference in density of reticular zones in reticulocytes (Heilmeier classification).

[Background of the invention] New methylene blue
Reticulocytes are young red blood cells in which purplish-blue granular material and string-like material (reticular bodies) are entangled in greenish-yellow red blood cells in fresh blood specimens that have been stained with supravital stains such as . Nucleated erythroblasts enucleate within the bone marrow and emerge into the peripheral blood, but in the process they inevitably pass through the reticulocyte stage. An increase or decrease in the reticulocyte ratio in peripheral blood indicates an increase or decrease in immature red blood cells, that is, an increase or decrease in erythropoietic ability. Furthermore, reticulocytes are classified according to differences in the density of the reticulum (Heilmeyer classification...see Masafumi Komiya's "Illustrated Guide to How to View Blood"), and the number of reticulocytes belonging to each classification is calculated to determine whether normal or abnormal red blood production ability is present. More detailed diagnosis becomes possible.

This reticulocyte classification test is a test in which collected blood is subjected to supravital staining, smeared on a glass slide, dried, and then reticulocytes are detected under a microscope, classified, and counted. Similar to the white blood cell classification test, this test is performed visually under a microscope by a laboratory technician, and the work is complicated and causes a lot of eye fatigue, so there is a strong desire to automate the test.

[Purpose of the invention]

The present invention is based on the above facts and aims to provide an apparatus for automating reticulocyte classification tests.

[Summary of the invention]

The automatic reticulocyte classification device of the present invention includes a means for obtaining a density histogram of a reticulocyte image extracted from an image of a stained blood sample, and a predetermined high concentration in this histogram that is comparable to the density of the reticulum. It is characterized by having means for determining the concentration width of a portion, means for determining the sum of frequencies of the high concentration portion, and means for performing Heilmeyer classification of reticulocytes using the sum of the frequency values and the concentration width. shall be.

That is, the reticular bands in reticulocytes vary in shape and direction, and may even exist overlapping each other. Therefore, the maximum density of the red blood cell image or the area of the portion with a certain density or higher may not accurately indicate the density of the reticular zone. As mentioned above, if classification is performed using both the concentration width of a predetermined high-density portion of the concentration histogram and its total frequency (corresponding to the area of the portion with a concentration equal to or higher than that of the reticular body in each red blood cell image), the reticular body Regardless of direction or overlap, Heilmeier classification equivalent to visual classification by an expert is possible.

Specifically, the square root of the square sum of the density range and the total frequency is used as a parameter, and Heilmeyer classification is performed by comparing the magnitudes of these values. In other words, classification is performed based on the distance from the origin on a plane whose two axes are the density range and the total frequency.

[Embodiments of the invention]

FIG. 1 is a diagram showing the configuration of an embodiment of the present invention. Light from a light source 1 is guided to a blood sample 3 placed on the stage of a microscope 2, and the transmitted light from the blood sample 3 is guided to a color television camera 4, where red (R), green (G), and blue light are transmitted.
Separate (B) into its components. A/D these color signals
The converter 5 converts it into a digital signal and stores it in the IC memory 6. For example, the R signal is stored in the IC memory 6a, the B signal is stored in the IC memory 6b, and the G signal is stored in the IC memory 6a.
The data is stored in the memory 6c.

Generally, there are from several to a dozen red blood cells in an image obtained from a television camera, so
In order to process each red blood cell individually, it is necessary to label each red blood cell. Creation of labeled mask images will now be described.

The B image component is read out from the IC memory 6b and the histogram creation circuit 8 creates a density histogram as shown in FIG. As can be seen from this figure, each blood cell component BL is distributed in the high concentration area, and the background component BG is distributed in the low concentration area, so the boundary value between the two, that is, the concentration value S(g ) is the threshold value for separating the blood cell part and the background part. 9 is this concentration threshold S(g)
This is a threshold detection circuit that detects. Next, the above-mentioned B image component is read out from the IC memory 6b, and a binarized image is created in the comparator circuit 10 with the above-mentioned density value S(g) set as a threshold level. For example, a binarized pattern is created in which blood cell portions are set to "1" and other parts are set to "0" and stored in the IC memory 6d.
This binarized pattern is read out again from the IC memory 6d, each blood cell image is labeled using the labeling circuit 11, and is stored again in the IC memory 6d. The labeled mask image is thus stored in the IC memory 6d.

Once each blood cell image is labeled, the process described below is repeated for each label. Then, the field of view of the stage of the microscope 2 is changed and image input is repeated until the total number of processed blood cells reaches the initially set number, for example, 1000.
The stage control circuit 24 performs stage movement and automatic focus adjustment of the microscope stage.

Next, a process for identifying red blood cells and reticulocytes will be explained. First, extraction of feature parameters necessary for identifying red blood cells and reticulocytes will be explained.

This feature parameter extraction is performed using the mask image, G image, and R image stored in the IC memory 6. That is, the labeled mask image is
Read from IC memory 6d, area calculation circuit 7a,
Then, the area S and circumference L of each blood cell are determined by the circumference calculation circuit 7b. Next, from the above mask image and the G image stored in the IC memory 6c,
A density histogram of the whole blood cell image within the screen is created by the histogram creation circuit 8. An example of this density histogram is shown in FIG. First, from this density histogram, the maximum value H _T of the density histogram is found, and then the density value f on the high density side corresponding to the frequency value of α percent of the maximum value H _T is found by the threshold density detection circuit 12. .

Next, the mask image is read out from the IC memory 6d and the G image is read out from the IC memory 6c, and the histogram creation circuit 8 is used to create a density histogram for each blood cell in the G image as shown in FIGS. 4A and 4B. seek. FIG. 4A shows a concentration histogram of red blood cells, and FIG. 4B shows a concentration histogram of reticulocytes. As can be seen from both figures, the histograms of the high-density portions of the two are clearly different, and by utilizing this property, it is possible to distinguish between the two. Using this density histogram, the maximum value H ₀ of the density histogram is determined by the threshold density detection circuit 13, and then the density value d on the high density side corresponding to the frequency value of α percent of the maximum value H ₀ , and the maximum value H 0 are determined. Find the density value e. Here, the above value of α is a value determined statistically and varies depending on the staining conditions of the blood specimen, but good results have been obtained with a value of 20 to 30%. Similarly, the R image is read from the IC memory 6a, a density histogram is created for each blood cell image in the R image, and a density value corresponding to α percent of the maximum histogram and a maximum density value are determined. G
A portion with a higher density than the above density value d determined from the image corresponds to the reticulum of reticulocytes in an actual blood cell image. Therefore, with this density value d set to the threshold level, the comparator circuit 14 detects the reticular body using the G image in the IC memory 6c, and considers the part darker than the density d to be the reticular part. “1”,
The other parts are set to "0" and stored in the IC memory 6b. Next, the reticular body image thus obtained is read out again from the IC memory 6b, and the reticular body perimeter L _R is determined by the perimeter calculating circuit 7b.

Feature parameters other than those mentioned above are obtained by the parameter extraction circuit 15. This parameter extraction circuit 15
A detailed explanatory diagram is shown in FIG. Characteristic parameters are determined using the density histogram H(g) of the G image component or R image component for each blood cell image unit, the previously determined density values d and f, and the maximum density value e. 30
is an adder that converts the density histogram H(g) into the density value d
to the maximum density value e to obtain S _de = _e 〓 ^d H(g). On the other hand, 31 is a subtractor, D _de =
Find ed. 32 and 33 are logarithmic converters for obtaining logS _de and logD _de , respectively. S _de is the area (shaded area) from the density value d to the maximum density value e of the density histogram for each blood cell image shown in Figures 4A and 4B, and is comparable to the density of the reticular body in each red blood cell image. Corresponds to the area of the high concentration part. D _de is the density difference between the highest density value e and the density value d, that is, the density width of the high density portion.

Similarly, 34 is an adder, and the density histogram H
(g) is added from the density value f to the maximum density value e,
This is to find S _fe = _e 〓 ^f H(g). On the other hand, 35 is a subtracter to obtain D _fe =e−f. 36 and 3
7 is a logarithmic converter, which calculates logS _fe and logD _fe , respectively.

38 is an arithmetic circuit that calculates the square root of the sum of the squares of D _de and D _de , that is, D _is =√ _de ² + _de ² . This parameter is
These are parameters for performing the Heilmeyer classification of reticulocytes, which is the main part of the present invention.

Heilmeier's classification described above distinguishes reticulocytes into 5 types based on the density of linear granules.
Type 4, type 3, type 2, type 1, from sparse reticular body.
Represented by type 0 (“Illustrated view of blood cells” by Masafumi Komiya)
Nanzando).

The purpose of logarithmically transforming the parameters S _de , D _de , S _fe and D _fe in the parameter extraction circuit 15 described above is as follows. Here, the parameter
This will be explained using logS _de and logD _de as examples. The distribution of the parameters S _de and D _de for red blood cells before conversion is concentrated near the origin, so the center of the distribution is located close to the origin and shows a long tail toward the reticulocyte region. In this state, the above parameters S _de and
When red blood cells and reticulocytes are distinguished using the magnitude relationship of D _de , red blood cells located in a boundary area where red blood cells and reticulocytes coexist may be determined to be reticulocytes. Therefore, the relationship between the parameters logS _de and logD _de obtained by logarithmically transforming the above parameters S _de and D _de and red blood cells and reticulocytes is as shown in FIG. 6. In Figure 6, the above parameter logD _de is set on the horizontal axis, the above parameter logD _de is set on the vertical axis, and the above parameter logS _de is set on the vertical axis.
The area above the straight line LR (discriminant function) defined by logD _de and logS _de is the area where reticulocytes exist, and the area below the line LR is the area where red blood cells exist. Therefore, the identification circuit 16 identifies blood cell images having a value larger than the above-mentioned straight line LR as reticulocytes, and blood cell images having a smaller value as red blood cells. When the parameters S _de and D _de are transformed logarithmically in this way, the center of the distribution of red blood cells moves away from the origin and approaches the shape of a normal distribution.
Since the probability of red blood cells occurring is 100 times greater than that of reticulocytes, all red blood cells in the boundary area where red blood cells and reticulocytes coexist are identified as red blood cells.
Therefore, the accuracy of identifying red blood cells can be 99.9% or higher. The identification accuracy for reticulocytes is about 90%.

Also, from the concentration histogram of the whole blood cell image in one screen, the concentration value f at the α percent point is calculated,
The reason for determining the feature parameters logS _fe and logD _fe is as follows. For types 3 to 4 of the Heilmeyer classification above, the concentration histogram H(g) is the 7th
It will look like Figure A. On the other hand, 0 of Heilmayr classification
When the reticulocyte reticulum concentration is high and the area is large, as in type 2, the concentration histogram H(g) becomes as shown in FIG. 6B. Therefore, the concentration value of the α percent point obtained from the concentration histogram H(g) in blood cell image units is set at the d points in FIG. 6A and B. Therefore, the parameters logS _de and logD _de described above are , there is a possibility that the characteristics of reticulocytes cannot be extracted. In other words, when red blood cells and reticulocytes are identified using the above parameters logS _de and D _de , there is a possibility that some of the red blood cells that are determined to be red blood cells include reticulocytes of Heilmeyer classification types 0 to 2. Red blood cells and reticulocytes may not be accurately distinguished. In order to solve this problem, 1
The above problem is solved by finding the density value f at the α percent point from the density histogram of the whole blood cell image within the screen and using the density value averaged over one screen. In other words, Fig. 4A showing the concentration histogram of red blood cells and the Heilmeyer classification 0~
As can be seen from Figure 7B, which shows the concentration histogram of type 2 reticulocytes, the histograms at the concentration higher than the above concentration value f are clearly different, and using the parameters logS _fe and logD _fe , it is possible to distinguish between the two. This makes it possible to do so. The relationship between the above parameters logS _fe and logD _fe and red blood cells and reticulocytes of Heilmeyer classification types 0 to 2 is determined as shown in FIG. 8. Here, in Fig. 8, the above parameter logS _fe is set on the horizontal axis, and the above parameter is set on the vertical axis.
The value line _LR ′ (discriminant function) determined by these parameters logS _fe value and logD _fe value is the current area of reticulocytes, and the area below the line LR′ is the area where red blood cells exist. Become. Therefore, blood cells with a value larger than the above-mentioned straight line LR' are identified as reticulocytes, and blood cells with a smaller value are identified as red blood cells.

Similarly, the parameter extraction circuit 15 is used to extract characteristic parameters for the R image, but since the amount of characteristic parameters obtained from the R image is almost the same as that for the G image, various parameters are not necessary. , only the parameters S _de and D _de need to be found.

Next, using the various feature parameters mentioned above,
The discrimination circuit 16 in FIG. 1 classifies red blood cells, reticulocytes and others, and counters 18 and 1 respectively
Count at 9 and 20. This operation is repeated until the total number of red blood cells and reticulocytes reaches a predetermined number. Next, those determined to be reticulocytes by the identification circuit 16 are also reclassified by the Heilmeyer classification identification circuit 17, and the results are sent to a predetermined counter 2.
Count at 1, 22, 23.

25 is a control circuit that controls the entire automatic reticulocyte counting device in the present invention.

Reticulocyte identification in the identification circuit 16 is performed by multi-stage identification branch logic. First, in the first step, connected red blood cells, large and small dust, etc. are removed using the parameters of the area S and perimeter L of each blood cell image. Next, in the second stage, from the G image
2 using feature parameters of logS _de and logD _de
Distinguish red blood cells and reticulocytes using the following discrimination function. That is, as shown in Fig. 6, a secondary discrimination function LR is determined, and blood cells whose parameters logS _de and logD _de have larger values than the above function LR are determined to be reticulocytes, and blood cells whose values are smaller than the above function LR are determined to be red blood cells. do. Among those determined to be red blood cells in the second stage,
Heilmeier classification 0~ as shown in Figure 7B
Since type 2 reticulocytes may be included, in the third step, the feature parameters of logS _fe and logD _fe are used to detect the above reticulocytes and classify them into Heilmeier's classification 0-2. Count into molds.
On the other hand, for those determined to be reticulocytes in the second stage, the reticulocytes are determined using characteristic parameters from the G and R images, such as the area S _de of the reticulum, the perimeter L _R , and the concentration difference D _de . and pseudoreticulocyte, which is a red blood cell with scratch-like wrinkles. Although the pseudo reticulocytes appear as reticulocytes, the reticulocytes have a relatively small area, a simple shape, that is, a short circumference, and a large concentration compared to reticulocytes. These features all depend on the feature parameters S _de , L _R and
The fourth step is to distinguish between _{reticulocytes} and pseudoreticulocytes using these parameters. Blood cells other than reticulocytes are classified as other and excluded from processing.

Those determined to be reticulocytes in the fourth stage are then subjected to Heilmayr classification by a Heilmayr classification identification circuit 17. That is, when the relationship between the parameters S _de and D _de obtained from the G image and the reticulocytes according to the Heilmeyer classification accurately classified by visual inspection is determined, the result is as shown in FIG. Here, Figure 9 shows
The above parameter S _de is set on the horizontal axis, and the above parameter D _de is set on the vertical axis. As is clear from the figure, reticulocytes _H4 of Heilmeyer type 4 are close to the origin, and reticulocytes of Heilmeyer types 0 to 2 are close to the origin.
H ₀ -H ₂ are far from the origin, and Heilmeyer type 3 reticulocyte H ₃ is located between H ₄ and H ₀ - ₂ above. Therefore, by using the characteristic parameter D _is = √ _de ² + _de ² corresponding to the distance from the origin, it becomes possible to automatically classify Heilmeyer based on the magnitude relationship of this parameter D _is . In other words, the distance from the origin determined by the value of the parameter D _is above
The values of L ₁ and L ₂ are set in advance in the Heilmeyer classifier, and the individual reticulocyte images obtained at 38 in Figure 5 are
Comparing the value of D _is with L ₁ and L ₂ , reticulocytes with D _is ≦L ₁ are Heilmeyer type 4 (H ₄ ),
Reticulocytes with L ₁ <D _is ≦L ₂ are Heilmeyer type 3 (H ₃ ), and reticulocytes with D _is >L ₂ are Heilmeyer types 0 to 2 (H ₀ to ₂ ). It will be judged. This result is calculated by counter 2, respectively.
It is counted as 1, 22, 23.

In this example, reticulocytes of Heilmeyer classification types 0 to 2 are very small, totaling 0.1% or less in normal people, so subclassification into types 0 to 2 is not performed.

The above-described identification circuit 16 and Heilmeyer classification identification circuit 17 can be easily constructed using known circuit components, but it is also possible to perform identification using a computer.

〔Effect of the invention〕

As described above, according to the present invention, it is possible to achieve the effect that the Heilmeyer classification of reticulocytes, which has conventionally been performed by visual inspection of microscopic images, can be performed automatically and with high precision.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of an automatic reticulocyte classification device according to the present invention, FIG. 2, FIG. 3, FIG. 4A, FIG. 4B, FIG. 6, and FIG. A,
7B, FIG. 8, and FIG. 9 are diagrams for explaining the operation of the configuration shown in FIG. 1, and FIG. 5 shows the configuration of an embodiment of the main part shown in FIG. 1. It is a diagram.

Claims (1)

[Claims] 1. In a device that performs Heilmeier classification of reticulocytes from the density histogram of a reticulocyte image extracted from a blood sample image, the density range of a predetermined high-density portion comparable to the concentration of the reticulum in the density histogram of each reticulocyte image is calculated. a means for determining the total frequency of the high concentration portion, a means for determining the sum of the frequencies of the high concentration portion, a square root of the sum of the squares of the concentration width and the total frequency, and a comparison between the set predetermined value and the value of the square root to determine the reticulocyte count. 1. An automatic reticulocyte classification device comprising means for performing Heilmeyer classification of reticulocytes.