JP3539581B2 - Particle image analysis method - Google Patents

Description

[0001]
[Industrial applications]
The present invention relates to a method for classifying particles using morphological information of particles, and more particularly to an analysis method suitable for classifying blood or urine particle images.
[0002]
[Prior art]
Conventionally, morphological examination of particles in urine has been carried out by visual inspection, in which a urine sample is centrifuged, the sediment is stained, a specimen is prepared on a slide glass, and observed under a microscope. . At this time, by keeping the concentration of centrifugation constant and the amount of the sample to be observed constant, it is possible to know what kind of sediment is contained in the original urine sample and at what concentration. . In the sediment, there are particles ranging in size from a few micrometers, such as blood cells and bacteria, to particles, hundreds of micrometers, such as cylinders. Switch and observe.
Attempts to take images of particles in a continuously flowing sample sample and analyze and classify the particles from individual particle images include JP-A-57-500995, JP-A-63-94156, and JP-A-5-59156. No. 296915 is known.
In Japanese Patent Publication No. 57-500995, a sample sample is passed through a specially shaped flow path, in which particles in the sample are caused to flow through a wide imaging area, a still image is taken with a flash lamp, and particle analysis is performed using the image. How to do is shown. When an enlarged image of sample particles is projected on a CCD camera using a microscope, a flash lamp as a pulse light source periodically emits light in synchronization with the operation of the CCD camera. The light emission time of the pulse light source is short, a still image can be obtained even when particles are continuously flowing, and the CCD camera can capture 30 still images per second.
In Japanese Patent Application Laid-Open No. 63-94156, a particle detection optical system is provided upstream of a particle image photographing area in a sample flow separately from a still image photographing system. This is a method in which the particle detection unit detects the passage of particles in advance and turns on the flash lamp at an appropriate timing when the particles reach the particle imaging region. In this method, the passage of particles is detected without periodic light emission of the pulse light source, and a still image can be taken at the same time only at that time. It does not process images without meaningless particles.
Japanese Patent Application Laid-Open No. 5-296915 discloses a method in which a particle detection optical system is further incorporated into a particle image pickup system. That is, a method of irradiating a sample sample flow with a laser beam for particle detection through a microscope condenser lens of a microscope image capturing system is described. This is characterized in that it is not necessary to separately prepare an optical system for detecting particles, and the particle detection position can be arranged as close to the particle image capturing area as possible.
Conventionally, this type of analysis method first distinguishes between a region occupied by an object (particles to be analyzed) and a background region where no particles to be analyzed exist in a captured image, and converts the captured image into a partial image having a meaning. Divided into The operation of dividing the image into partial images having this meaning is called image region division, and the method is described in, for example, "Image Processing: Kyoritsu Shuppan, 1983". A binary image to be analyzed is created by region division, and the binary image is shaped by noise removal and filling processing. The noise removal and fill processing can be performed by a combination of image contraction and dilation processing, and are described, for example, in "Introduction to Computer Image Processing: Soken Shuppan, 1985". After shaping the binary image, the same label (number) is assigned to all pixels belonging to the same connected component included in the binary image, and different labels are assigned to different connected components. Therefore, the number of labels indicates the number of connected components. Such an operation is called labeling of connected components (label processing), and is described in, for example, "Introduction to Computer Image Processing: Soken Shuppan, 1985". The connected component means a series (area) of pixel points (area 1) of the binary image. For example, all points that can reach the next point by sequentially tracing eight neighborhoods from a certain point are considered. It is called eight connected components. Next, a feature amount (a size and a shape of a region for each connected component, or density information corresponding to a region in the original image, etc.) indicating a geometric feature to be analyzed is obtained, and analysis and classification are performed. For example, in a blood cell image and the like, as described in JP-B-58-29872, each cell in an image is classified using the morphological information and density information of the cell, and the appearance cell ratio for each category is reported.
[0003]
[Problems to be solved by the invention]
However, this method has a disadvantage that classification cannot always be performed accurately. In particular, in the case of unstained cells (cells that do not stain the stain) or lightly stained (cells that do not stain the stain deeply), the target cannot be distinguished from the background because the cell region in the image cannot be completely extracted. There is a problem that the number of connected components is apparently increased due to division into small regions and detection of a large number of connected components.
An object of the present invention is to provide a stable particle image analysis method by describing the number of particles to be optimally analyzed / classified and the state of particles in the image for a plurality of connected components in an image. .
[0004]
[Means for Solving the Problems]
The above purpose is to first obtain the distribution state of the whole particles from the number of divided regions in the image and the position information in the image for each region, and to classify and classify from the area of the particle region having the maximum value Achieved by determining the number of particles.
More specifically, a step (S10 in FIG. 3) of dividing a region for each classification target from the captured image by the region dividing circuit (105 in FIG. 2) under the control of the computer (109 in FIG. 2); Under the control of the computer, the image processing circuit (106 in FIG. 2) performs labeling processing on the connected components to be processed in the divided area, and obtains a feature amount (for example, area or the like) for each label (S20 in FIG. 3). , 30), a step of determining an identification method based on the number of labels and the feature amount (40 in FIG. 3) in an identification control circuit (111 in FIG. 2) under the control of a computer, and the identification control circuit A step of performing classification and classification for each particle by the above-mentioned method (50 in FIG. 3), and performing, by a computer, addition by category of each particle or addition by rank of each image in accordance with the result of the classification and classification; Particles in the image A step (S60 of FIG. 3) to output the appearance status. In the step of determining the number of divided areas, when the number of connected components of the area extracted in one image (the number of labels given in the labeling process in S20) is plural, the identification classification in one image is performed. The number of connected components of the region to be performed is limited, or the identification and classification of all particles in the detected region is performed only when the number of connected components of the region extracted in one image is equal to or less than a predetermined value. Identification and classification are performed in the order of the size of the connected components of the regions extracted in the image (for example, the number of processes is set in the order of the area of the label). In the step of outputting the appearance status, the classification result of each particle (the number of appearance particles for each category) is output, and the number of images in which a plurality of particles exist is output for each rank according to the number of labels.
[0005]
[Action]
According to the present invention, the identification control circuit determines the identification method based on the number of connected components of the region extracted in advance and the feature amount (spatial spread) of the region for each connected component in accordance with a computer instruction (pre-identification processing). ), Identification is performed in particle units or image units in that manner. Further, the computer describes the identification result in image units and particle units, and outputs the result to an output device. As a result, the operator can accurately grasp the appearance of particles in the entire image of one sample.
That is, taking a cylinder as an example, a stained cylinder in a captured image is binarized by a region division process, and a binary image is shaped as shown in FIG. 1A. In this case, the number of connected components to be analyzed in the binary image is 1 (the number of labels is 1) and can be identified by obtaining information on the region 10 to be processed. When the detection particles are lightly stained, as shown in FIG. 1 (b), the whole is not uniformly divided into regions, and a portion having a low concentration (a portion that is difficult to distinguish from the background) is divided into a plurality of regions. The number of labels (the number of connected components) obtained by the labeling process is plural. In such a case, the number of targets to be processed is determined and classified based on the distribution state of each label and the size of the detection area, and the number of detected particles in the image is stored. In addition, when the dye was not stained, the entire cylinder was divided into minute regions as shown in FIG. 1 (c), and the number of detection regions by the labeling process was labeled much more than in FIG. 1 (b). Will be. In such a case, since the spread of the distribution on the image has a certain spread, the result of the image unit is reported rather than the classification result of the labeled individual particles. This means that when the number of detected particles is small in the image, all labeled detection particles are to be classified, and when there are many detected particles in the image, the classification of image units or the cell unit depending on the distribution state of detected particles. This is done by deciding whether to classify. Thereby, the classification of the particle images can be performed accurately and stably.
[0006]
【Example】
Hereinafter, the present invention will be described in detail with reference to examples. FIG. 2 shows an apparatus configuration for carrying out the present invention. First, particles in a fluid are irradiated by a pulse lamp, expanded by an optical expansion unit, and converted into an analog electric signal by a photoelectric converter 101 via a color filter. The output of the photoelectric converter 101 is guided to an image input unit 102 and converted into a digital signal. The output of the image input unit 102 is stored in the image memory 103 via the image data bus 107. At this time, the image memory 103 is composed of a plurality of frames and stores an image corresponding to each color filter, a calculated image, and the like. The output of the image input unit 102 is guided to a region dividing circuit 105 via an image data bus 107, and a target region is extracted by a region dividing method using density information, and the region divided image is output to an image memory 104 and stored. The image stored in the image memory 104 is guided to the image processing circuit 106, where the image processing circuit 106 performs filtering processing such as thickening and thinning of the image and labeling processing, and stores the processed image in the image memory 104. Further, the image processing circuit 106 obtains a characteristic amount for each area of the image subjected to the labeling processing, and transfers the characteristic amount to the identification control circuit 111 via the CPU bus 108. The discrimination control circuit 111 performs a discrimination process based on the feature amount of each area according to an instruction from the computer 109, and transfers a discrimination result to the CPU 109. The computer 109 outputs the processing result to the output device 110 when the processing of the specified amount is completed. The above-described processing control and data transfer are all controlled by the computer 109 via the CPU bus 109. The content of the image memory 103 is stored in the image storage device 112 according to the identification result. Here, the image storage device 112 may be a semiconductor memory device capable of storing digital signals, a magnetic or magneto-optical storage device, a digital video tape recording device, or a video tape storage device that converts a digital signal into an analog signal. .
[0007]
Hereinafter, a processing procedure performed by the computer 109 will be described with reference to FIGS. 3 is a flowchart showing the processing procedure of the computer 109, FIGS. 4 to 6 are diagrams showing a detailed flow of the process in FIG. 3, and FIG. 5 is a diagram showing one index of the distribution state.
First, the computer 109 extracts a target area in the image from the image after the image input by the area dividing process using the density information (step S10). To extract the target area in the image, a density histogram of the image is created, the average density of the background and the peak near the target average density are detected on the density histogram, and the bottom of the valley between these two peak positions is detected. And a binary image is created by binarization using the density value at the bottom of the valley as a threshold value.
[0008]
Thereafter, the image processing circuit 106 is instructed to perform a binary image shaping process by a noise removal and a filling process, and the connected components are labeled by a connected component labeling process (step S20). The noise removal and filling processing of the binary image are performed by repeating the contraction and dilation processing of the image (dilation → contraction → contraction → dilation is performed as one operation, and this is repeated).
[0009]
Next, the feature amount of the region for each connected component (label) is calculated (step S30). The feature amount referred to here refers to, for example, the area of the region on the image, the perimeter, the circularity, the average density of each wavelength component, and the like, as described in “Image Processing: Kyoritsu Shuppan, 1983”.
[0010]
Next, in the pre-identification processing, the identification method is determined by the identification control circuit 111 based on the feature amount obtained in step 30 (step S40). As shown in FIG. 4, details of step S40 are rearranged (sorted) by the size of the area information among the feature amounts obtained in step S30 (step S41), and then a statistic is calculated (step S42). The statistics obtained here are, for example, the angle θ from the deviation values σx and σy of the number of labels and the horizontal deviation value σx and the vertical deviation value σy of the coordinates as shown in FIG. Determined by 1).
θ = tan ⁻¹ (σy / σx) (1)
Next, an identification method is determined from the statistics obtained in step S42 (step S43). First, when the number of labels is less than k0, a method of performing particle unit identification for processing all labels is adopted. When the number of labels is k0 or more and less than k1, the processing number is set according to the maximum area. For example, when the maximum area is s or less and the angle θ is t0 to t1, m1 processing is performed in the descending order of the area. In cases other than t1, m2 processing is performed. If the maximum area is s or more, m3 processing is performed. When the number of labels is equal to or larger than k1, identification of particle units is not performed as identification of image units. Here, k0 is a value determined experimentally and is usually set to 10, and k1 is set to 20. S is a value determined by the area dividing process, and is usually set to about 1000. The angle ranges t0 and t1 are set to 0.5 and 0.7, respectively. The processing numbers m1, m2, and m3 of the particles are set to 10, 5, and 3, respectively.
[0011]
Next, the identification is performed by the identification control circuit 111 according to the method determined in step S40 (step S50). As shown in FIG. 6, the details of the identification processing are as follows. First, when it is determined that the identification is not a particle unit (step S51), the processing is skipped if the identification is not a particle unit. In the case of the particle unit identification, each specified particle is identified (steps S52 and S53).
[0012]
Next, the computer 109 stores and updates a table as shown in FIG. 7 based on the identification result of step S50, and causes the output device 110 to output the table (S60). For example, in the case of particle-based identification, the computer 109 performs addition for each category according to the identification result, and also adds rank according to the number of labels of the image. In the case of image unit identification, addition is performed for each rank according to the number of labels. The rank of each image is set according to the number of labels. For example, rank A has 1 label, rank B has 2 to 5 labels, rank C has 5 to 9 labels, and rank D has 10 to 19 labels. Set as follows. At this time, it is also possible to instruct the image storage device 112 to store the contents of the image memory 103 according to the identification result of each image.
The above is the processing of one input image. One sequence is completed by performing this process for a predetermined number of images, and a series of processes is completed by outputting a table as shown in FIG. As described above, by holding the information of each image and grasping the configuration of one sample as a whole, the reliability of the data is enhanced and the analysis result is useful.
[0013]
【The invention's effect】
According to the present invention, even if a plurality of particles are present in the image, or even if unstained cells or lightly stained cells having no density difference from the background portion are divided into a plurality of portions by the region dividing process, the entire image is Since it can be described, efficient particle analysis can be stably performed.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating the principle of the present invention.
FIG. 2 is a configuration diagram of a particle image analyzer according to one embodiment of the present invention.
FIG. 3 is a flowchart illustrating an identification process according to an embodiment of the present invention.
FIG. 4 is a flowchart illustrating details of pre-identification processing according to an embodiment of the present invention.
FIG. 5 is an explanatory diagram for describing the spread of particles in one embodiment of the present invention.
FIG. 6 is a flowchart illustrating details of an identification process according to an embodiment of the present invention.
FIG. 7 is a diagram for explaining an identification result in one embodiment of the present invention.
[Explanation of symbols]
10: basic particle image, 101: photoelectric converter, 102: image input unit, 103, 104: image memory, 105: area dividing circuit, 106: image processing circuit, 107: image data bus, 108: CPU bus, 109: computer, 110: output device, 111: identification control circuit, 112: image storage device.

Claims (6)

Flowing a particle sample suspended in a liquid into a flow cell, detecting particles passing through a particle detection region in the flow cell, capturing a still image of the detected particles passing through the imaging region in the flow cell, and capturing the captured particle image In a particle image analysis method for performing morphological classification of particles by image analysis of
A first step of extracting an area of each classification target images captured by the area dividing,
A second step of obtaining a feature quantity for each region including a position and area on the image of the number and the area of the region extracted in the first step,
A third step of determining, based on the number of regions obtained in the second step and the feature amount of each region, an identification scheme for said first region extracted in step,
And a fourth step of performing classification and classification on the area in accordance with the classification method determined in the third step . If the said number of areas obtained in the second step is a value k0 more defined, in the third step, claims and determines the detection method to limit the realm that make their identification and categorization 1 Analysis method of the described particle image. If the said number of areas obtained in the second step is less than the prescribed value k0, wherein in the third step, determining the identification system in which the identification and categorization for realm all extracted in the first step 2. The method for analyzing a particle image according to claim 1, wherein: When the number of regions obtained in the second step is equal to or more than a prescribed value k0 and less than k1, in the third step, the number of areas determined by the maximum area and the position obtained in the second step is equal to the number of areas . analyzing method according to claim 1, wherein the particle image and determining the identification method of performing have magnitude order identification classification for the area. A method for analyzing a particle image according to any one of claims 2 to 4,
As a result of the classification performed in the fourth step according to the classification method determined in the third step, each area is set in accordance with an addition value for each category in which each area is classified and classified, and the number of areas. 5. A method for analyzing a particle image, comprising: a fifth step of outputting an added value for each rank of an image. When the number of regions obtained in the second step is equal to or more than a prescribed value k1, in the third step, addition is performed for each rank of an image set according to the number of regions, and the added value for each rank is calculated. 2. The particle image analysis method according to claim 1, wherein the method is determined as a classification and identification method for each image to be output.

Images