WO2013037119A1 - 红细胞形态学分析装置及其方法 - Google Patents

红细胞形态学分析装置及其方法 Download PDF

Info

Publication number
WO2013037119A1
WO2013037119A1 PCT/CN2011/079710 CN2011079710W WO2013037119A1 WO 2013037119 A1 WO2013037119 A1 WO 2013037119A1 CN 2011079710 W CN2011079710 W CN 2011079710W WO 2013037119 A1 WO2013037119 A1 WO 2013037119A1
Authority
WO
WIPO (PCT)
Prior art keywords
red blood
blood cells
sample
morphological
cell
Prior art date
Application number
PCT/CN2011/079710
Other languages
English (en)
French (fr)
Inventor
丁建文
周丰良
梁光明
Original Assignee
长沙高新技术产业开发区爱威科技实业有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 长沙高新技术产业开发区爱威科技实业有限公司 filed Critical 长沙高新技术产业开发区爱威科技实业有限公司
Priority to US13/880,593 priority Critical patent/US9170256B2/en
Priority to BR112013032938-6A priority patent/BR112013032938B1/pt
Priority to PCT/CN2011/079710 priority patent/WO2013037119A1/zh
Priority to EP11872496.2A priority patent/EP2757372B1/en
Publication of WO2013037119A1 publication Critical patent/WO2013037119A1/zh

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/48—Biological material, e.g. blood, urine; Haemocytometers
    • G01N33/50—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing
    • G01N33/5005—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing involving human or animal cells
    • G01N33/5091—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing involving human or animal cells for testing the pathological state of an organism
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/48—Biological material, e.g. blood, urine; Haemocytometers
    • G01N33/483—Physical analysis of biological material
    • G01N33/487—Physical analysis of biological material of liquid biological material
    • G01N33/49—Blood
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
    • G01N15/10—Investigating individual particles
    • G01N15/14—Optical investigation techniques, e.g. flow cytometry
    • G01N15/1429—Signal processing
    • G01N15/1433—Signal processing using image recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
    • G06V20/695—Preprocessing, e.g. image segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
    • G06V20/698—Matching; Classification
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
    • G01N15/01—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials specially adapted for biological cells, e.g. blood cells
    • G01N2015/011—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials specially adapted for biological cells, e.g. blood cells with lysing, e.g. of erythrocytes
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
    • G01N15/10—Investigating individual particles
    • G01N15/14—Optical investigation techniques, e.g. flow cytometry
    • G01N2015/1493—Particle size
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
    • G01N15/10—Investigating individual particles
    • G01N15/14—Optical investigation techniques, e.g. flow cytometry
    • G01N2015/1497—Particle shape

Definitions

  • Red blood cell morphology analysis device and method thereof
  • the invention relates to a red blood cell morphology analysis device and a method thereof in a sample, in particular to a red blood cell morphology analysis device and a method thereof capable of automatically identifying the type and source of red blood cells contained in a sample.
  • the red blood cell parameters of the existing red blood cell morphological analysis are obtained by calculating the average value, for example, the morphological classification parameter of anemia, mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH).
  • Mean corpuscular hemoglobin concentration (MCHC) is calculated based on the results of red blood cells, hemoglobin concentration and hematocrit. It is not the actual measured value, so the red blood cell count, hemoglobin content, and hematocrit data are measured. Must be accurate, otherwise the resulting morphological classification parameters have large errors.
  • the technical problem to be solved by the present invention is to provide a statistical method for expressing the source and properties of red blood cells in a sample by using morphological characteristic parameters and a neural network for the purpose of determining whether the red blood cell morphology is normal or not.
  • a red blood cell morphological analysis device and method for assisting in identifying the source and properties of red blood cells is provided.
  • a red blood cell morphology analysis device which comprises:
  • the low magnification lens of the automatic microscope scans a sample of the set area, marks the identified target area, and simultaneously scans the marked area by the high magnification lens of the automatic microscope;
  • a camera or CCD component performs image information acquisition on the marked area
  • an image digitizer for analyzing the image the image digitizer firstly collects the image according to The red blood cells contained are segmented and positioned, and then the segmented images are digitized, that is, four types of morphological characteristic parameters of size, shape, chromaticity and texture of each red blood cell are extracted;
  • a feature fusion device based on fuzzy clustering, which is used to normalize the dimensionality, shape, chromaticity and texture of the four types of multi-dimensional morphological parameters of the red blood cells separated by the above steps. Four eigenvalues of size, shape, chromaticity and texture are obtained, and statistical calculation and statistical expression are respectively performed according to the size, shape, chromaticity and texture characteristics of all red blood cells in the specimen, which provides real and objective analysis for the red blood cell type and source in the sample. in accordance with;
  • the control unit is connected to the above automatic microscope, camera or CCD component, image digitizer and output device to control the action of the automatic microscope, camera or CCD component, image digitizer and output device.
  • the output device expresses a combination of characteristic parameter data representing the size, shape, chroma and texture of the central region of the red blood cells characteristic of various types of red blood cells, and the method for analyzing and identifying red blood cells in the sample is to judge the sample according to the change of the characteristics of the central region of the red blood cells.
  • the morphological analysis method of red blood cells in the sample is based on clinically confirmed data of different types of red blood cells with different morphological characteristics, and the classification and counting are performed according to this, and then according to a certain morphological characteristic parameter of each type of red blood cells in the sample.
  • the ratio of the data combination to the data combination of the same morphological characteristic parameters of the total red blood cells is statistically processed and expressed in a graphical or data manner.
  • a combination of two or more types of red blood cell morphological parameters of each type of red blood cells in the sample is statistically analyzed, and a multi-parameter analysis result of each type of red blood cells in the sample is obtained, and the red blood cell morphology is determined according to the change. Learning changes, visually expressing the type of red blood cell morphological changes in the sample through graphs and data.
  • red blood cell analysis using this device has different clinical significance in different samples.
  • the device analyzes the result that the single red blood cell is small in size and low in color.
  • the total red blood cell morphological characteristic parameter is combined with the parameter indicating the size as the abscissa, and the characteristic parameter data indicating the chromaticity is the morphological analysis pattern of the ordinate.
  • the red blood cell distribution width is shifted to the left, and the red blood cell distribution area is shifted downward, and the expression is In order to disperse the sinking red blood cell morphological map to the left, such a pattern appears in the blood sample to suggest a type of anemia.
  • red blood cell distribution width increased Offset
  • red blood cell distribution area moves up, and is expressed as a distribution pattern of red blood cells that are scattered to the right. This type of pattern appears in the blood sample to suggest another type of anemia. If it appears in the urine sample and it accounts for a certain proportion, it means It is derived from non-renal red blood cells.
  • the invention also discloses a red blood cell morphology analysis method, which comprises the following steps:
  • Step 1 Scan the sample of the set area with a low-magnification lens of the automatic microscope, mark the target area to be found, and simultaneously scan the sample of the marked area by the high-magnification lens of the automatic microscope;
  • Step 2 Collect image information of the marked area sample with a camera or CCD element
  • Step 3 Using the image digitizer to first segment and locate the collected image according to the contained cells, and then digitizing the segmented image, that is, extracting the morphological characteristic parameters of each cell, using size, shape, chromaticity, and Texture four types of features to describe each cell;
  • Step 4 input the four types of morphological characteristic parameters of the size, shape, chroma and texture of each cell obtained in the above steps into a classifier based on a neural network, and the classifier extracts red blood cell size from various cells, Shape, Chroma and Replacement Page (Rule 26) Texture four types of morphological characteristic parameters;
  • Step 5 Input the four types of morphological characteristic parameters of the size, shape, chromaticity and texture of the red blood cells separated in step 4 into a feature fusion device based on fuzzy clustering, and each type of multidimensional morphology is obtained by the feature fusion device.
  • the feature parameters are normalized to obtain a one-dimensional feature vector;
  • Step 6 Display the normalized feature quantities of each type of red blood cells in each specimen through the output device, and obtain a statistical chart of each type of normalized characteristic parameters.
  • the above method for morphological analysis of red blood cells further comprises the following steps: according to the ratio of the number of red blood cells in the sample to the total number of red blood cells in the sample, statistically processing, expressing in a graphical or data manner, analyzing and identifying red blood cells in the sample;
  • the above method for morphological analysis of red blood cells further comprises the step 8: for each type of normalized morphological feature vector, by giving a feature value threshold, calculating red blood cells above or below the threshold account for the total red blood cell sample.
  • the proportion of the morphological characteristic parameter data combination is statistically processed and expressed graphically or by data to provide an objective basis for analyzing and identifying red blood cells in the sample.
  • the classifier based on the neural network includes a feedback process for refining, classifying, and supplementing characteristic parameters of the classified suspicious targets and identifying the wrong targets, and establishing corresponding mathematical models for performing neural networks. Training, the neural network automatically learns and memorizes the refined, classified, and supplemented feature parameters into the model database, and then returns to the neural network-based classifier for cell classification.
  • the normalized size feature vector obtained in step 6 expresses the characteristic parameter data combination of the representative size of each type of red blood cell, and the red blood cell analysis and identification method in the sample can be performed according to the red blood cell size distribution and the normal red blood cell size distribution.
  • the comparison determines the source of red blood cells or the type of red blood cells in the sample according to the direction and extent of the deviation.
  • the normalized shape feature vector obtained in step 6 expresses the characteristic parameter data combination of the representative shape of each type of red blood cell, and the analysis and identification method of the red blood cell in the sample can be judged according to the change of the red blood cell shape characteristic parameter.
  • the normalized chroma feature vector obtained in the step ⁇ expresses the characteristic parameter data combination of the representative chromaticity unique to each type of red blood cell, and the analysis and identification method of the red blood cell in the sample can be based on the red blood cell color and the normal red blood cell color.
  • the degree of sputum comparison determines the source of red blood cells or the type of red blood cells in the sample according to the direction and extent of the deviation.
  • the normalized texture feature vector obtained in the step ⁇ expresses the characteristic parameter data combination of the representative texture of each type of red blood cell, and the analysis and identification method of the red blood cell in the sample can be judged according to the change of the red blood cell texture characteristic parameter.
  • the present invention takes a sample (blood, urine) under a microscope and amplifies the morphological image of each cell in the sample by CCD and processes each cell by digital image processing.
  • the morphological characteristic parameters are obtained by separating the parameters into the neural network based classifier to separate the red blood cells, and then normalizing the various red blood cell morphological parameter data by the feature fusion device based on the fuzzy clustering. Perform statistical analysis on each type of normalized parameters obtained, or perform comprehensive statistical analysis based on several types of parameters, and express them in a graphical or numerical form to determine whether the morphology of red blood cells is normal or not. Detection of morphological red blood cells can identify the source and nature of red blood cells.
  • the invention is based on the original urine cell identification, taking into account that the software may have errors in the identification of individual targets, we introduce a replacement page (Article 26)
  • the morphological parameters and statistical analysis methods are used.
  • the method can automatically analyze the red blood cell source in the hematuria sample according to the statistical analysis of all target parameters in a sample. This method reduces the influence of errors caused by individual target recognition errors by using the overall target in the sample for statistical judgment. It is an application innovation of statistical methods in the analysis of morphological parameters of urine red blood cells.
  • Fig. 1 is a schematic view showing the structure of a red blood cell morphology analyzing device of the present invention.
  • FIG. 2 is a schematic view showing the operation flow of the erythrocyte morphology analysis method of the present invention.
  • Figure 3 is a schematic diagram of a PCA-weighted feature fusion algorithm.
  • Fig. 4 is a schematic view showing the distribution width and peak value mentioned in the present invention.
  • Figure 5 is a normal morphological red blood cell image taken with a digital CCD.
  • Fig. 0J is a statistical graph of the characteristic erythrocyte size characteristic parameter, and its distribution width C L , whose peak value is b in the normal range, Dl ⁇ b ⁇ D2.
  • Figure 0.2- Figure 6.9 is another representation of Figure 6.1.
  • Fig. 7 is a statistical graph of the characteristic parameters of the red blood cell shape of normal morphology, the distribution of which is concentrated, the distribution width is small, and the frequency value corresponding to the peak value is C>H (60%).
  • Fig. 8 is a statistical graph of the morphological characteristic parameters of normal erythrocytes, showing a single narrow peak with a concentrated distribution, and the peak corresponding frequency value C>H (60%).
  • Fig. 9 is a statistical graph of the characteristic parameters of the normal morphology red blood cell texture, whose peak corresponds to the X-axis value b in the normal range, Wl ⁇ b ⁇ W2.
  • Fig. 10 is a scatter plot for the comprehensive analysis of the size and chromaticity characteristic parameters of normal erythrocytes, and the distribution of normal morphological red blood cells is between 75 ⁇ X ⁇ 125 and 20 ⁇ Y ⁇ 40.
  • Figure 1 1 shows the image of the spore-shaped red blood cells taken by the digital CCD.
  • the spore red blood cells have vesicles protruding in the outer membrane of the red blood cells or the cells are spore-like changes. The volume is uneven, the shape of the spore is changed, and the chromaticity is usually shallow.
  • the expression of the characteristic parameter chart is shown in Figure 12-15. Normally shaped red blood cells are shown in dotted lines for comparison.
  • Fig. 12 is a statistical graph of the characteristic parameters of the spore-shaped red blood cells.
  • the size distribution parameter ⁇ is large and the distribution is not concentrated because of the uneven size and the large tendency due to spores. More than one peak, the frequency at the peak is correspondingly reduced.
  • Its distribution width a>L from the figure, the width is larger than the distribution width a of the normal red blood cells, indicating that the red blood cell size is unevenly distributed.
  • b>D2 part of the red blood cell size is too large.
  • Figure 13 is a statistical graph of the shape characteristic parameters of the spore-shaped red blood cells, whose peak corresponding frequency value C is smaller than the normal morphology red blood cells, C ⁇ H (00%) o
  • Figure 14 is a statistical graph of the characteristic parameters of spore-shaped red blood cells.
  • the mean value of the spore-shaped red blood cells is significantly reduced, and the peak corresponding frequency value C is significantly smaller than that of normal-type red blood cells, C ⁇ H, and the variation of the color distribution of the spore-shaped red blood cells. Not large, the cells are evenly distributed with a narrow single peak. .
  • Fig. 15 is a statistical graph of the characteristic parameters of the spore-shaped red blood cell texture.
  • the texture of the spore red cell is more complex than that of the normal red blood cell, and the texture characteristic parameter statistical graph, the texture peak corresponds to the X-axis value b is larger, b>W2.
  • Figure 10 is a scatter plot of the size and chromaticity characteristics of spore-shaped red blood cells.
  • the erythrocyte chromaticity is significantly lower, and the red blood cell size range is expanded, mainly distributed between 80 ⁇ X ⁇ 160.
  • Fig. 17 is a view showing an image of a size uneven red blood cell taken by a digital CCD. Size is not red blood cells, refers to the diameter replacement page between red blood cells (Article 26) More than double the situation, common in a variety of proliferative anemia and megaloblastic anemia, volume, size, light color, shape is normal. The expression of its characteristic parameter statistics is shown in Figure 18-22. Normally shaped red blood cells are shown in dotted lines for comparison.
  • Fig. 18 is a statistical graph of the size parameter of the red blood cell size with uneven size, and its distribution width a>L.
  • the width of the red blood cell distribution width a is larger than that of the normal red blood cell distribution, indicating that the red blood cell size is unevenly distributed.
  • b>D2 part of the red blood cell size is too large.
  • Fig. 19 is a statistical graph of the characteristic parameters of the shape of the red blood cells with uneven size, and the peak corresponding frequency value C is smaller than the normal shape red blood cells, C ⁇ H (60%).
  • Figure 20 is a statistical graph of the chromaticity characteristic parameters of the unequal red blood cells.
  • the mean erythrocyte chromaticity was higher, and the peak corresponding frequency value C was significantly smaller than that of normal erythrocytes, C ⁇ H (60%). Its distribution width increases significantly toward the chromaticity value.
  • Figure 21 is a statistical graph of the characteristic parameters of the size-heterogeneous red blood cell texture.
  • Figure 22 is a scatter plot of the size and chromaticity characteristic parameters of the unequal-sized red blood cells. The figure shows that the chromaticity and size distribution range of red blood cells is large, 5 ⁇ X ⁇ 30 and 40 ⁇ Y ⁇ 150.
  • Figure 23 is a scatter plot of the comprehensive analysis of the characteristic parameters of small-volume hypochromic red blood cells (anemia or renal).
  • Figure 24 is a graph showing the ratio of various cells in the sample to the total number of red blood cells.
  • Figure 25 is a statistical diagram of the same chromaticity characteristic parameter.
  • Figure 20 is a variant representation of Figure 24.
  • Figure 27 is a statistical diagram expressed in a multi-image arrangement.
  • FIG. 1 is a schematic illustration of a red blood cell morphology analysis apparatus of the present invention.
  • the red blood cell morphology analysis apparatus of the present invention comprises:
  • An automatic microscope 1 the low magnification lens of the automatic microscope 1 first scans the sample of the set area, and marks the found target area, and simultaneously marks the marked area by the high magnification lens of the automatic microscope 1 The sample is scanned;
  • a camera or CCD element 2 the camera or CCD element 2 performs image information acquisition on the marked area samples
  • an image digitizer 3 for generating a digital representation of the image, the image digitizer 3 first segmenting and positioning the acquired image according to the contained cells, and then digitizing the segmented image, that is, extracting each Four types of morphological characteristics of cell size, shape, color and texture;
  • a feature fusion device based on fuzzy clustering.
  • the feature fusion device is used to normalize each type of multi-dimensional morphological feature parameters obtained in the above steps to provide a statistical classification basis for red blood cells;
  • an output device 6 the output device 6 can include a monitor and a printer for visually displaying the detection result; g. a control unit 5, the control unit 5 is respectively connected to the above-mentioned automatic microscope 1, camera or CCD element 2, image The digitizer 3 and the output device 0 control the operation of the automatic microscope 1, the camera or the CCD element 2, the image digitizer 3, and the output device.
  • the neural network used in the classifier 7 is a BP blue layer neural network based on the RDROP algorithm, and the replacement page (Article 26)
  • the three-layer neural network includes an input layer, an output layer, and a hidden intermediate layer. And the data of the neural network is expandable and self-memory.
  • This neural network is used for expert system training and sample target recognition.
  • the classifier can also employ other types of neural networks. '
  • the neural network of the present invention has a plurality of input nodes, each of which expresses a certain morphological characteristic parameter of the cell to be tested.
  • the extraction methods of the morphological characteristic parameters and the classification status of the morphological characteristic parameters are as follows:
  • the morphological features of the red blood cells are manually classified by experts to classify the formed parts in the sample images, and the classification semantic model is established accordingly. Based on this, a classification mathematical model is established to define various morphological characteristics of red blood cells. There are four categories of features, including size features, shape features, chroma features, and texture features.
  • the present invention extracts up to one hundred multidimensional target features, the following are only representative features:
  • size characteristics including area, perimeter, equivalent diameter, long axis, short axis, average radius, etc.
  • Shape features are used to describe the shape of the target, including circular rate, eccentricity, regional center of gravity, curvature, regional chord distribution, morphological description, correlation feature, boundary-fit polygonal shape descriptor, Fourier coefficient morphological feature Vector, convex hull-based morphological description related features, circumscribed rectangle-based feature description, shape-based feature description based on constant-distance features, and shape features based on region skeleton extraction.
  • the chord distribution morphological descriptor, the convex trait-based morphological description related feature, the circumscribed rectangle based feature description, the constant distance feature based shape feature description, and the region skeleton extraction based shape feature are several proposed features in the present invention. a new method of describing morphology;
  • Chromatic characteristics including color histogram based on HSV target area, feature extraction based on probability window, and color distance;
  • Texture features are multi-scale texture features based on wavelet transform domain: they include multi-scale wavelet energy ratio, multi-scale wavelet standard deviation, texture features of integrated co-occurrence matrix, and Zernike moment feature description of fusion texture spectrum.
  • the operation steps of the erythrocyte morphology analysis method of the present invention are as follows:
  • Step 1 Scan the sample of the set area with the low magnification lens of the automatic microscope, mark the target area, and scan the marked area by the high magnification lens of the automatic microscope 1;
  • Step 2 Image information is collected on the marked area by camera or CCD element 2, Figure 5, Figure 1 1 and Figure 17 are collected red blood cell images; normal form red blood cells, cell size is relatively uniform, cell body is normal or large, Hemoglobin is abundant, no spores are formed, and the cell membrane is intact.
  • the statistical graphs of various parameters are basically normal distribution, and the distribution area is more concentrated.
  • Step 3 Using the image digitizer 3 to segment and locate the acquired image according to the contained cells, and then digitizing the segmented image, that is, extracting the morphological characteristic parameters of each cell, using size, shape, chromaticity, and Texture four types of features to describe each cell
  • Step 4 input the four types of morphological characteristic parameters of the size, shape, chromaticity and texture of each cell obtained in step 3 into a classifier based on a neural network, and the red blood cells are separated from the cells by the classifier;
  • the table is an example of some characteristic parameters of a different kind of red blood cells:
  • the classifier based on the neural network includes a feedback process, which is a suspicious target for classification (26) Standardize and identify the wrong target to refine, classify and supplement the characteristic parameters, and establish corresponding mathematical models to train the neural network.
  • the neural network automatically learns and memorizes the refined, classified and supplemented characteristic parameters into the model database. Return to the neural network based classifier for cell sorting.
  • Step 5 input the morphological characteristic parameters of the red blood cells separated in step 4 into a feature fusion device based on fuzzy clustering, and perform normalization processing by the feature fusion device to obtain a one-dimensional feature vector;
  • the specific operation of the feature fuser is to calculate a normalized feature value by using each of the initially obtained feature parameters through a PCA-weighted feature fusion algorithm.
  • Figure 3 is a schematic diagram of the PCA-weighted feature fusion algorithm.
  • the training sample ⁇ -dimensional feature vector is input, and the n-dimensional feature vector can be a size, a shape, a texture, and a chrominance feature vector.
  • the feature extraction is performed by the PCA algorithm, and the feature extraction is divided into K-L feature pressure dimension and principal component feature selection.
  • the size feature, the shape feature, the texture feature, and the chromaticity feature are respectively selected in the principal component feature vector space obtained after feature extraction.
  • the eigenvectors are normalized to the feature subspaces, and then the weights of the eigenvectors are weighted and fused. Each feature subspace is weighted and fused to obtain a corresponding feature quantity, which is then merged into a one-dimensional eigenvector.
  • the goal of PCA is to extract the m-dimensional principal component eigenvectors of the original feature space F by the orthogonal vector matrix.
  • the principle is that the original feature vector;, / 2 , L is projected along the direction; PCA will maximize the energy obtained, then _; called the first principal component; under the condition of X, orthogonal, original
  • the following cost function can be utilized and defined to evaluate the discriminative ability of the feature, and the evaluation function is as shown in equation (5).
  • a larger value indicates that the distinguishing ability of the feature is stronger, and J p can be used to represent the weight of each vector.
  • the weight vector corresponding to each vector is (J, , J 2 , L , J d ), and ⁇ is the feature dimension of a certain feature subspace.
  • the feature subspace can be fused by matrix transformation to obtain one-dimensional eigenvalues.
  • the matrix transformation is as in equation (6):
  • the element fl in (aographya 2 , L , ) 7 ' indicates that the i-th training sample corresponds to the feature value after dimension reduction of a certain feature subspace.
  • the feature values are normalized to between 0 and 1, Each type of feature can be represented in segments.
  • Step 6 Display the normalized feature quantity of each type of red blood cells in each specimen through the output device, and obtain the statistical graph of each type of normalized characteristic parameter; Replacement page (Article 26) Step ⁇ . ⁇
  • the statistical charts of the four types of characteristic parameters of sample size, shape, chromaticity and texture are obtained respectively.
  • the statistical distribution maps of the four types of characteristic parameters are compared with the normal red blood cell size distribution, and judged according to the direction and degree of deviation.
  • the source of red blood cells or the type of red blood cells in the sample are compared with the normal red blood cell size distribution, and judged according to the direction and degree of deviation.
  • Sample size, shape, chroma, texture Four types of characteristic parameters Statistical data graph expression is not limited, it can be histogram, distribution map, scatter graph, graph, histogram, area graph, pie chart, scatter plot, ring Diagrams, radar diagrams, bubble diagrams, cylinder diagrams, etc.
  • Steps 6.1.1 Use the normalized size feature vector after fusion to make a red blood cell size distribution characteristic curve as shown in Fig. 6J-Fig. 0.9, Fig. 12 and Fig. 18.
  • the abscissa is the size value, and the ordinate is the frequency at which the corresponding size value appears.
  • the red blood cell size characteristic mainly reflects the size distribution of red blood cells in the sample.
  • the red blood cell size distribution characteristic curve is shifted to the left, and the curve is shifted to the right.
  • the red curve changes to the left.
  • the peaks appear in the curve, indicating that the red blood cells are large and small, and there is heterogeneity. If the peak is small, the cell volume is small; the peak value is large, and the cell volume is large. In the case where a plurality of peaks appear, it indicates that cells having a distinctly uneven size are mixed, and the cell size distribution can be obtained by comparing the multi-peak value with the reference value.
  • the size of the distribution width of the frequency indicates the degree of concentration of the cell size.
  • the source of red blood cells or the type of red blood cells in the sample is judged based on the direction and extent of the deviation based on the distribution of the sample red blood cell size and the size distribution of the normal red blood cells.
  • Figure 6.1 shows a statistical plot of the size characteristic parameters of a urine sample containing normal red blood cells, shown here as a curve.
  • the statistical graph of the characteristic parameters can be expressed in various forms of statistical graphs, as shown in Figure 6.2- Figure 6.9, which are statistical graphs of normal-form red blood cell size characteristic parameters.
  • ⁇ represents the width of the distribution in the chart.
  • Set the distribution width threshold L When the distribution width a > L, it can be determined that the sample red blood cells are uneven red blood cells, and the system prompts the sample to be suspected of uneven size red blood cells.
  • b is the X-axis value corresponding to the peak of the statistical distribution.
  • the b value changes, in the case where the chart is expressed by the curve, it appears as the left or right shift of the curve.
  • D l the peak minimum value
  • D2 the maximum threshold D2.
  • Dl ⁇ b ⁇ D2 the red blood cell size is within the normal range.
  • b ⁇ D l the curve shifts to the left, the system prompts the sample to be suspected of small red blood cells; when b>D2, the curve shifts to the right, and the system prompts the sample to be suspected of red blood cells.
  • Figure 12 and Figure 18 are statistical diagrams of the red cell size characteristics.
  • Fig. 12 shows the size characteristics of the spore-shaped red blood cells, which have vesicles protruding from the outer membrane of the spore-shaped red blood cells or changes in the mold of the cells.
  • Figure 18 illustrates the size-differentiated red blood cell size characteristics.
  • the red blood cell size distribution width of the two graphs is a>L. From the graph, the width is larger than the distribution width a of the normal red blood cells, indicating that the red blood cell size distribution is uneven. b>D2, part of the red blood cell size is too large.
  • Step 0.1 .2 uses the normalized shape feature vector after fusion to make a red blood cell shape distribution characteristic curve, as shown in FIG. 7, FIG. 13 and FIG.
  • the abscissa is the shape feature value, and the ordinate is the frequency at which the corresponding feature value appears.
  • the red blood cell shape distribution characteristic index mainly reflects the distribution of deformed red blood cells. Use a combination of characteristic parameter data representing the shape (such as circular rate, box rate, chord distribution symmetry, lenticular, endotomy, etc.) for analysis.
  • characteristic parameter data representing the shape (such as circular rate, box rate, chord distribution symmetry, lenticular, endotomy, etc.) for analysis.
  • the morphology of normal red blood cells is a biconcave disk shape, and the morphology of the deformed red blood cells is spore-like, oral shape, and the like.
  • the source of red blood cells or the type of red blood cells in the sample is judged based on changes in the characteristic parameters of the red blood cell shape.
  • Figure 7 is a statistical plot of the shape characteristic parameters of a urine sample containing normal red blood cells, shown here as a curve.
  • Fig. 13 and Fig. 9 are statistical diagrams of the shape characteristics of red blood cells.
  • Figure 3 is a characteristic map of the shape of the bud-shaped red blood cells. There are vesicles protruding in the outer membrane of the spore-shaped red blood cells or the cells are spore-like changes.
  • Figure 19 illustrates the size-invariant red blood cell shape characteristics. The red blood cell shape distribution width of the two images does not change much compared with normal red blood cells. The peak corresponding frequency value C is smaller than the normal shape red blood cells, C ⁇ H.
  • Step 6.1.3 Use the fused normalized chromaticity feature vector to make a red blood cell chromaticity distribution characteristic curve, as shown in Fig. 8, Fig. 14, and Fig. 20.
  • the abscissa is the chromaticity value, and the ordinate is the frequency at which the corresponding chromaticity value appears.
  • the red blood cell chromaticity distribution mainly reflects the red blood cell hemoglobin loss. Analyze using a combination of characteristic parameter data (such as hue, saturation, etc.) that represents chromaticity.
  • characteristic parameter data such as hue, saturation, etc.
  • the source of red blood cells or the type of red blood cells in the sample is judged based on the direction and degree of deviation according to the erythrocyte chromaticity and the normal erythrocyte chromaticity.
  • the color After erythrocyte hemoglobin, the color will become lighter. In this case, the erythrocyte chromaticity distribution characteristic frequency histogram curve will shift to the left and the peak will become smaller. The red blood cells will lose water in hypertonic urine to form wrinkled red blood cells, and the chromaticity will change. Deep, in such cases, the chroma frequency histogram curve will shift to the right and the peak will become larger. This kind of situation is not an abnormal range, and it is necessary to pay attention to the discrimination judgment.
  • Figure 8 is a statistical plot of the chromaticity characteristic parameters of a urine sample containing normal red blood cells, shown here as a curve.
  • the chromaticity characteristic statistical curve when the erythrocyte chromaticity is closer to the normal erythrocyte chromaticity, the distribution is more concentrated, and the peak value corresponding to the frequency value C is larger (0 ⁇ C ⁇ 100%).
  • C ⁇ H the system indicates that the erythrocyte chromaticity in the sample deviates from the ideal erythrocyte chromaticity.
  • b in Fig. 4 is the X-axis value corresponding to the peak of the distribution of the statistical map.
  • the value of b changes, in the case where the graph is expressed by the curve, it appears as the left or right shift of the curve, that is, the change of the chromaticity of the red blood cell. .
  • Sl ⁇ b ⁇ S2 the red color chromaticity is within the normal range.
  • 1 ⁇ 51 the curve shifts to the left, indicating that the cell chroma is low.
  • b>S2 the curve shifts to the right, indicating that the cell color is high.
  • Fig. 4 is a statistical diagram of the erythrocyte chromaticity characteristics.
  • the dotted line indicates the normal red blood cell chromaticity, showing a single narrow peak, indicating that the chromaticity of the cells is very uniform and there is no loss of cytoplasm.
  • the color of red blood cells is related to the hemoglobin content.
  • Figure 14 depicts the chromaticity characteristics of spore-shaped red blood cells.
  • Figure 20 illustrates the unequal red blood cell chromaticity characteristics.
  • the mean red blood cell chromaticity of both images is different from that of normal erythrocytes.
  • Step 0.] .4 Use the fused normalized texture feature vector to make the red blood cell texture distribution characteristic curve, as shown in Figure 9, Figure 15, and Figure 21.
  • the abscissa is the texture value, and the ordinate is the frequency at which the corresponding texture value appears.
  • the red blood cell texture distribution feature uses a combination of characteristic parameter data (such as irregular texture, shrinkage, etc.) that represents the red cell central region color ladder (such as the central pale area enlargement, the central pale area disappears, the central area chroma enhancement, etc.) and the representative texture. analysis. According to the red blood cell texture and the normal red blood cell texture, the red blood cells in the sample are judged according to the direction and degree of deviation.
  • characteristic parameter data such as irregular texture, shrinkage, etc.
  • the red cell central region color ladder such as the central pale area enlargement, the central pale area disappears, the central area chroma enhancement, etc.
  • Figure 9 which is a statistical diagram of the texture characteristic parameters of a urine sample containing normal red blood cells, is shown here in the form of a curve.
  • ⁇ represents the width of the distribution in the chart. The smaller the difference in the texture of the cells in the sample, the smaller the distribution width ⁇ of the statistical graph; the greater the difference in the texture of the cells, the larger the distribution width ⁇ of the statistical graph.
  • the b in Fig. 4 is the X-axis value corresponding to the peak of the distribution of the statistical map.
  • the b value changes, in the case where the graph is expressed by the curve, it appears as the left or right shift of the curve, that is, the intensity change of the red blood cell texture.
  • Let the peak value correspond to the X-axis minimum threshold W1 and the maximum threshold W2.
  • Wl ⁇ b ⁇ W2 the red blood cell texture is in the normal range.
  • 13 ⁇ 1 the sample texture is weak and the curve is shifted to the left; when b>W2, the sample texture is thicker and the curve is shifted to the right.
  • Figure 15 illustrates the spore-shaped red blood cell texture features.
  • Figure 21 illustrates the unequal red cell texture features. The mean value of the red blood cell texture of the two images is different from that of normal erythrocytes.
  • Step 6.2 combine the normalized feature vectors of two or more categories, and comprehensively analyze by statistical methods to obtain a multi-parameter analysis chart;
  • red blood cell size and chromaticity For example, a combination of two normalized feature vectors, red blood cell size and chromaticity, is shown in Figure 10, Figure 16, Figure 22, and Figure 23.
  • the abscissa is the red blood cell size value
  • the ordinate is the corresponding red blood cell color value, and a scatter plot is made.
  • Figure 10 is a comprehensive analysis of the size and chroma of normal-form red blood cells.
  • the distribution of normal-form red blood cells is between 75 ⁇ X ⁇ 125 and 20 ⁇ ⁇ ⁇ 40.
  • Figure 16 is a comprehensive analysis of the size and chroma of the spore-shaped red blood cells.
  • the red blood cell color is significantly lower, and the red blood cell size range is expanded, mainly distributed between 80 ⁇ X ⁇ 100.
  • the analytic red blood cell analysis showed that the chromaticity and size distribution of red blood cells were larger, 5 ⁇ ⁇ ⁇ 30 and 40 ⁇ ⁇ ⁇ 150.
  • the red blood cell analysis method of the device has different clinical significance in different samples.
  • the result of the device analysis is that the single red blood cell is small in size and chroma.
  • the total red blood cell morphological characteristic parameter is combined with the parameter indicating the size as the abscissa, and the characteristic parameter data indicating the chromaticity is the morphological analysis pattern of the ordinate.
  • the red blood cell distribution width is shifted to the left, and the red blood cell distribution area is moved downward. It is expressed as a red-cell morphological distribution pattern that is scattered to the left. Such a pattern appears to indicate anemia in a blood sample. If it appears in a urine sample and accounts for a certain proportion, it means that it is derived from renal red blood cells.
  • Step 7 is statistically processed according to the proportion of each type of red blood cells in the sample to the total number of red blood cells, and expressed in a graphical or data manner, and the red blood cells in the sample are analyzed and identified.
  • Step 7 According to a normalized one-dimensional morphological feature vector of each type of red blood cell in the sample, by giving a characteristic quantity threshold, the red blood cell above or below the threshold is calculated to account for the same morphological characteristic parameter data of the sample total red blood cell.
  • the proportion of the combination is statistically processed and expressed graphically or in data.
  • Figure 25 is a statistical diagram of the same chromaticity characteristic parameter. In the figure, using the normalized chrominance one-dimensional eigenvector after fusion, the chromaticity is given a threshold value of H to determine the low-color red blood cells, and the ratio of the cell chromaticity in the sample to the threshold H is 40%. That is, the low pigment red blood cells in the sample accounted for 40% of the total red blood cells in the sample.
  • Figure 20 is a variant representation of Figure 25.
  • the above graphic or data mode expression is displayed on the output device for reference by relevant personnel.
  • the output device may be a display window on the red blood cell morphology analysis device of the present invention, or may be a display connected to the device, and a display connected to the network for remote consultation, or expressed in a printed manner to facilitate the doctor. analysis.
  • the chart can be expressed in a multi-image arrangement, as shown in Figure 27. It can also be converted to a single image display. Additionally, the invention relates to a sample which may be a urine sample or a blood sample. It should be specially pointed out here that the blood sample is not smeared, but is diluted at a certain multiple for analysis, so that some of the cells in the sample are not destroyed by the smear.
  • the invention requires the sample to be fresh, preferably the sample is taken within 2 hours, and the sample does not need to be dyed, but is not limited to no dyeing, so that expensive reagents are not needed in the test, so economical, pollution-free, environmentally friendly .

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Chemical & Material Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Immunology (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Hematology (AREA)
  • Molecular Biology (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Pathology (AREA)
  • Urology & Nephrology (AREA)
  • Food Science & Technology (AREA)
  • Medicinal Chemistry (AREA)
  • Biophysics (AREA)
  • Ecology (AREA)
  • Signal Processing (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Dispersion Chemistry (AREA)
  • Physiology (AREA)
  • Cell Biology (AREA)
  • Microbiology (AREA)
  • Biotechnology (AREA)
  • Tropical Medicine & Parasitology (AREA)
  • Investigating Or Analysing Biological Materials (AREA)
  • Image Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)

Abstract

一种红细胞形态学分析装置及方法。该方法包括将样本置于自动显微镜下放大后由CCD采取样本中各细胞形态学图像，并经图像数字转换器数字化后，进行图像分割定位和目标特征参数提取，通过建立在神经网络基础上的分类器分离出各红细胞的形态学特征参数，再通过建立在模糊聚类基础上的特征融合器对各类红细胞形态特征参数数据进行归一化处理，对得到的每一类归一化参数分别进行统计分析，或根据几类参数进行综合统计分析，并以图形或数表的方式表达出来。由此来判断红细胞的形态是否正常，通过对各类异常形态红细胞的检测可以鉴定红细胞来源和性质。

Description

红细胞形态学分析装置及其方法 技术领域
本发明涉及一种样本中红细胞形态学分析装置及其方法， 特别是一种能够自动鉴定样本 中所含红细胞的类型和来源的红细胞形态学分析装置及其方法。
背景技术
早在 1852年就有人开始设计红细胞的计数方法， 1855年发明了用于计数红细胞的计 数板。 〗947年美国科学家库尔特（W . H . Coulter )发明了用电阻法计数粒子的专利技术。 并在 1956 年将这一技术应用于红细胞计数获得成功。 随着科学技术的日新月异， 各种新的 检测红细胞数目的手段不断涌现。 目前应用的研究幵发手段主要有： 电容式、 光电式与激光 式、 离心式、 电阻式及各种方式的联合式。 到目前为止已经有许多国家开始生产各种类型的 红细胞分析仪， 经过半个多世纪的发展， 这种仪器已经有了非常明显的进步。 红细胞的分类 变得细致， 计数变得越来越精确。 但是， 现有红细胞形态学分析的红细胞参数都是通过计算 平均值来获得， 比如用作贫血的形态学分类参数平均红细胞体积 （mean corpuscular volume, MCV) 平均红细胞血红蛋白含量 (mean corpuscular hemoglobin, MCH) , 平 均红细胞血红蛋白浓度 (mean corpuscular hemoglobin concentration, MCHC)都是根 据红细胞、 血红蛋白浓度和红细胞比容结果， 计算而得， 不是实际测得的数值， 这样红细胞 数、 血红蛋白含量、 红细胞比容的测定数据必须准确， 否则得到的形态学分类参数的误差很 大。
人工镜检是经典的检测方法， 其通过显微镜， 使用目镜上的测微尺来人工测量每个红细 胞直径大小， 再对数据进行分析， 进而作出判断。 但人工镜检方法工作人员工作量大， 由于 疲劳容易引起误判； 工作效率低， 速度慢， 有可能延误病人的诊断。
因此， 如何使红细胞计数更快、 更准、 更节约成本， 是目前医院临床检验中面临的问题。 发明内容
本发明所要解决的技术问题是， 针对现有技术不足， 提供一种运用形态学特征参数及神 经网络对样本中红细胞的来源和性质进行统计学方法表达， 以供相关人员参考判断红细胞形 态是否正常， 辅助鉴定红细胞来源和性质的红细胞形态学分析装置及其方法。
为解决上述技术问题， 本发明所采用的技术方案是： 一种红细胞形态学分析装置， 其包 括：
a . 一自动显微镜， 该自动显微镜的低倍物镜头对所设置区域的样本进行扫描， 对发现 的目标区域予以标记， 并同时由该自动显微镜的高倍物镜头对已标记区域进行扫描；
b . 一摄像机或 CCD元件， 该摄像机或 CCD元件对已标记区域进行图像信息采集； c . 一用以对上述图像进行分析处理的图像数字转换器， 该图像数字转换器先将采集的 图像根据所含的红细胞进行分割定位， 再对分割后的图像进行数字化处理， 即提取各红细胞 的大小、 形状、 色度及纹理四类形态学特征参数；
d. —建立在神经网络基础上的分类器， 用于根据上述步骤取得的各细胞的大小、 形状、 色度及纹理四类形态学特征参数， 从各类细胞中分离出红细胞； 替换页 （细则第 26条) e . 一建立在模糊聚类基础上的特征融合器， 该特征融合器用于将上述步骤分离出的红 细胞的大小、 形状、 色度及紋理四类多维形态学特征参数进行归一化降维处理得到大小、 形 状、 色度及纹理 4个特征值， 再根据标本中所有红细胞的大小、 形状、 色度、 纹理特征分别 进行统计计算和统计图形表达， 为分析样本中红细胞类别及来源提供真实客观依据；
f. 一输出设备， 用于直观显示检测结果；
g . 控制单元， 该控制单元分别连接上述自动显微镜、 摄像机或 CCD元件、 图像数字 转换器及输出设备， 以控制该自动显微镜、 摄像机或 CCD元件、 图像数字转换器及输出设备 动作。
该输出设备表达出各种类型红细胞所特有的代表红细胞中央区大小、 形状、 色度和纹理 '的特征参数数据组合， 样本中红细胞的分析鉴定方法， 是根据红细胞中央区特征的改变来判 断样本中红细胞的来源或红细胞的类型。
样本中红细胞形态学分析方法， 是参照临床确认的代表不同意义的各种类型红细胞具有 不同的形态学特征参数数据， 据此进行识别分类计数， 再根据样本中各类型红细胞某一形态 学特征参数数据组合占总红细胞的同类形态学特征参数数据组合的比例经统计学处理后以图 形或数据方式表达。
作为其中的分析方法和检测手段， 将样本中各类型红细胞的两种以上的红细胞形态学特 征参数组合用统计学方法综合分析， 得到样本中各类型红细胞多参数分析结果， 根据其改变 判断红细胞形态学变化， 通过图形和数据直观的表达出样本中红细胞形态学变化类型。
利用本装置进行红细胞分析所表达的相同的结果，在不同的样本中具有不同的临床意义， 如样本中出现小体积低色素红细胞时， 本装置分析出的结果是单个红细胞体积小、 色度低， 总红细胞形态特征参数以表示大小的参数组合为横坐标， 表示色度的特征参数数据为纵坐标 的形态学分析图形表达为红细胞分布宽度增加往左偏移， 红细胞分布区域向下移， 表达为向 左分散向下沉的红细胞形态分布图， 此类图形在血液样本中出现提示一种类型贫血， 如果在 尿液样本中出现并占一定比例则意味着来源于肾性红细胞； 如样本中出现大体积高色素红细 胞时， 本装置分析出的结果是单个红细胞体积大、 色度高， 总红细胞形态特征参数以表示大 小的参数组合为横坐标， 表示色度的特征参数数据为纵坐标的形态学分析图形表达为红细胞 分布宽度增加往右偏移， 红细胞分布区域向上移， 表达为向右分散向上升的红细胞形态分布 图， 此类图形在血液样本中出现提示另一种类型贫血， 如果在尿液样本中出现并占一定比例 则意味着来源于非肾性红细胞。
本发明还公开了一种红细胞形态学分析方法， 其包括下列步骤：
步骤 1 : 用自动显微镜的低倍物镜头对所设置区域的样本进行扫描，对发现的目标区域予 以标记， 并同时由该自动显微镜的高倍物镜头对己标记区域样本进行扫描；
步骤 2: 用摄像机或 CCD元件对己标记区域样本进行图像信息采集；
步骤 3:用图像数字转换器对采集的图像先根据所含的细胞进行分割定位，再对分割后的 图像进行数字化处理， 即提取各细胞的形态学特征参数， 用大小、 形状、 色度以及紋理四类 特征来描述各细胞；
步骤 4: 将上述步骤取得的各细胞的大小、形状、色度及纹理四类形态学特征参数输入建 立在神经网络基础上的分类器， 由该分类器从各类细胞中分离出红细胞大小、 形状、 色度及 替换页 （细则第 26条) 纹理四类形态学特征参数；
步骤 5: 将步骤 4分离出来的红细胞的大小、 形状、 色度以及纹理四类形态学特征参数输 入建立在模糊聚类基础上的特征融合器， 由该特征融合器将每一类多维形态学特征参数进行 归一化处理， 得到一维特征向量；
步骤 6:将每个标本中所有红细胞的每一类归一化特征量经输出设备显示出来， 即得出每 一类归一化特征参数的统计学图表。
上述红细胞形态学分析方法，还包括步骤 7:根据样本中各类型红细胞占总红细胞数的比 例经统计学处理， 以图形或数据方式表达， 对样本中红细胞进行分析鉴定；
上述红细胞形态学分析方法， 还包括步骤 8: 对每一类经过归一化的形态学特征向量，通 过给出一个特征量阔值， 计算高于或者低于阀值的红细胞占样本总红细胞同类形态学特征参 数数据组合的比例， 经统计学处理后以图形或数据方式表达， 以提供对样本中红细胞进行分 析鉴定的客观依据。
该建立在神经网络基础上的分类器包括一反馈过程， 该反馈过程是对分类出来的可疑目 标及识别错误目标进行细化、 分类、 补充特征参数， 并建立相应的数学模型， 对神经网络进 行训练， 神经网络自动学习并记忆该些细化、 分类、 补充的特征参数进入模型数据库， 再返 回基于神经网络的分类器进行细胞分类。
步骤 6中得出的归一化的大小特征向量表达出各种类型红细胞所特有的代表大小的特征 参数数据组合， 样本中红细胞分析鉴定方法， 可根据红细胞大小分布与正常的红细胞的大小 分布进行比对， 根据偏离的方向和程度来判断样本中红细胞的来源或红细胞的类型。
步骤 6中得出的归一化的形状特征向量表达出各种类型红细胞所特有的代表形状的特征 参数数据组合， 样本中红细胞的分析鉴定方法， 可根据红细胞形状特征参数的变化来判断样 本中红细胞的来源或红细胞的类型。
步骤 ό中得出的归一化的色度特征向量表达出各种类型红细胞所特有的代表色度的特征 参数数据组合， 样本中红细胞的分析鉴定方法， 可根据红细胞色度与正常的红细胞色度迸行 比对根据偏离的方向和程度来判断样本中红细胞的来源或红细胞的类型。
步骤 ό中得出的归一化的纹理特征向量表达出各种类型红细胞所特有的代表纹理的特征 参数数据组合， 样本中红细胞的分析鉴定方法， 可根据红细胞纹理特征参数的变化来判断样 本中红细胞的来源或红细胞的类型。
对样本中红细胞的分析鉴定，'单独采用其中任意一条来分析， 或综合采用其中的多条来 分析。
与现有技术相比， 本发明所具有的有益效果为： 本发明将样本 （血、 尿） 置于显微镜下 放大后由 CCD采取样本中各细胞形态学图像并通过数字图像处理后得到各细胞的形态学特 征参数， 将参数输入建立在神经网络基础上的分类器分离出各红细胞， 再通过建立在模糊聚 类基础上的特征融合器对各类红细胞形态特征参数数据进行归一化处理、 对得到的每一类归 一化参数分别进行统计分析， 或根据几类参数进行综合统计分析， 以图形或数表的方式表达 出来， 以此来判断红细胞的形态是否正常， 通过对各类异常形态红细胞的检测可以鉴定红细 胞来源和性质。
本发明在原有尿液细胞识别基础上， 考虑到软件对个别目标识别可能存在误差， 我们引 替换页 （细则第 26条) 入了形态学参数及其统计分析方法， 采用该方法， 仪器可根据一个样本中所有目标参数的统 计分析， 自动分析血尿样本中的红细胞来源。 这种方法由于采用样本内全体目标进行统计判 断， 降低了个别目标识别错误带来的误差影响。 是统计学方法在尿液红细胞形态学参数分析 上的一个应用创新。
附图说明
图 1为本发明红细胞形态学分析装置结构示意图。
图 2为本发明红细胞形态学分析方法的操作流程示意图。
图 3为 PCA加权的特征融合算法原理图。
图 4为本发明中提到的分布宽度、 峰值示意图。
图 5·为数码 CCD拍摄的正常形态红细胞图像。
图 0J为正常形态红细胞大小特征参数统计曲线图， 其分布宽度 C L, 其峰值为 b在正 常范围内， Dl <b<D2。
图 0.2-图 6.9为图 6.1的另一种表示。
图 7为正常形态红细胞形状特征参数统计曲线图， 其分布集中， 分布宽度小， 峰值对应 的频数值 C>H ( 60%)。
图 8为正常形态红细胞色度特征参数统计曲线图， 呈单一窄峰， 其分布集中， 峰值对应 的频数值 C>H ( 60%)。
图 9为正常形态红细胞纹理特征参数统计曲线图,其峰值对应 X轴数值 b在正常范围内， Wl <b<W2。
图 10为正常形态红细胞的大小和色度特征参数综合分析散点图，正常形态红细胞集中分 布在 75<X<125并且 20<Y<40之间。
图 1 1为数码 CCD摄取的芽孢形红细胞图像， 芽孢红细胞， 在红细胞外膜有小泡突出或 细胞呈霉菌孢子样改变。 体积不均、 有芽孢的变形状、 色度通常浅。 其特征参数统计图的表 达， 如图 12- 15所示。 图中正常形红细胞以虛线方式表示以进行比较。
图 12为芽孢形红细胞大小特征参数统计曲线图， 与正常形红细胞比较，因为大小不均且 因为芽孢而有偏大趋势， 其大小特征参数统计图分布宽度 α大， 分布不集中； 并形成 1个以 上峰值， 峰值处频数相应降低。 其分布宽度 a>L, 从图上看宽度比正常形红细胞分布宽度 a 大， 表示红细胞大小分布不均。 b>D2, 部分红细胞大小偏大。
图 13为芽孢形红细胞形状特征参数统计曲线图， 其峰值对应频数值 C小于正常形态红 细胞， C<H (00%)o
图 14为芽孢形红细胞色度特征参数统计曲线图，其芽孢形红细胞色度均值显著降低，其 峰值对应频数值 C均明显小于正常形态红细胞， C<H, 且芽孢形红细胞色度分布宽度变化 不大， 细胞分布均匀呈较窄单峰。。
图 15为芽孢形红细胞纹理特征参数统计曲线图，芽孢红细胞纹理较正常红细胞复杂，其 紋理特征参数统计图， 紋理峰值对应 X轴数值 b较大， b>W2。
图 10 为芽孢形态红细胞的大小和色度特征参数综合分析散点图， 其红细胞色度明显偏 低， 而且红细胞大小范围扩大， 主要分布在 80<X<160之间。
图 17为数码 CCD摄取的大小不均红细胞图像。大小不一形红细胞， 指红细胞之间直径 替换页 （细则第 26条) 相差一倍以上的情况,常见于各种增生性贫血及巨幼细胞性贫血， 体积大小不一， 色度浅， 形 状正常。 其特征参数统计图的表达， 如图 18-22所示。 图中正常形红细胞以虚线方式表示以 进行比较。
图 18为大小不均形红细胞大小特征参数统计曲线图， 其分布宽度 a>L, 从图上看宽度比 正常形红细胞分布宽度 a大， 表示红细胞大小分布不均。 b>D2, 部分红细胞大小偏大。
图 19为大小不均形红细胞形状特征参数统计曲线图， 其峰值对应频数值 C小于正常形 态红细胞， C<H (60%)。
图 20为大小不均形红细胞色度特征参数统计曲线图。大小不均红细胞色度均值偏高，其 峰值对应频数值 C均明显小于正常形态红细胞， C<H (60%)。 其分布宽度往色度值较大方 向明显增大。
图 21为大小不均形红细胞纹理特征参数统计曲线图。
图 22为大小不均形红细胞的大小和色度特征参数综合分析散点图，图中显示红细胞的色 度和大小分布范围大， 5<X<30且 40<Y<150。
图 23为小体积低色素红细胞 （贫血或肾性）特征参数综合分析散点图。
图 24为样本中各种细胞占总红细胞数的比例统计图。
图 25为一样本色度特征参数统计图。
图 20为图 24的一个变化表达形式。
图 27为以多图排列的方式表达的统计图。
具体实施方式
图 1为本发明红细胞形态学分析装置的示意图。 如图所示， 本发明红细胞形态学分析装 置包括：
a . 一自动显微镜 1， 该自动显微镜 1的低倍物镜头先对所设置区域的样本进行扫描， 并对发现的目标区域予以标记， 并同时由该自动显微镜 1 的高倍物镜头对已标记区域样本进 行扫描；
b . 一摄像机或 CCD元件 2， 该摄像机或 CCD元件 2对已标记区域样本进行图像信 息采集；
c .一用以产生上述图像的数字表示的图像数字转换器 3, 该图像数字转换器 3先将采 集的图像根据所含细胞进行分割定位， 再对分割后的图像进行数字化处理， 即提取各细胞的 大小、 形状、 色度及紋理四类形态学特征参数；
d.一建立在神经网络基础上的分类器，用于根据上述步骤取得的各细胞的大小、形状、 色度及纹理四类形态学特征参数， 从各类细胞中分离出红细胞；
e. 一建立在模糊聚类基础上的特征融合器 该特征融合器用于将上述步骤取得的每 一类多维形态学特征参数进行归一化处理， 以提供红细胞统计分类依据；
f. 一输出设备 6， 该输出设备 6可包括监控器及打印机， 用于直观显示检测结果； g. 一控制单元 5, 该控制单元 5分别连接上述自动显微镜 1、 摄像机或 CCD元件 2、 图像数字转换器 3及输出设备 0， 以控制该自动显微镜 1、 摄像机或 CCD元件 2、 图像数字 转换器 3及输出设备 ό动作。
在本实施例中， 该分类器 7 中采用的神经网络是基于 RDROP算法的 BP兰层神经网络， 替换页 （细则第 26条) 该三层神经网络包括一输入层， 一输出层和一隐藏着的中间层。且该神经网络的数据可扩充， 具有自记忆性。 该神经网络用于专家系统训练和样本目标识别。 当然， 该分类器也可采用其 他类型的神经网络。 '
关于神经网络， 已经有很多资料对其进行了详细介绍， 因而在此不作过多叙述。
本发明神经网络具有多个输入节点， 每一输入节点表达待测细胞的某一形态学特征参数。 该些形态学特征参数的提取方法及形态学特征参数分类状态说明如下：
1 .特征提取方法
首先通过大量样本图片， 由专家人工依据红细胞形态学特征对样本图片中的有形成分进 行分类， 依此建立分类语义模型， 在此基础上建立分类数学模型， 从而定义红细胞的各种形 态学特征， 一共有四大类特征， 包括大小特征、 形状特征、 色度特征、 纹理特征。
2. 特征分类
本发明提取了多达一百多维目标特征， 以下仅为代表性特征描述：
2.1大小特征， 包括面积、 周长、 等效直径、 长轴、 短轴、 平均半径等；
2.2形状特征： 形状特征用于描述目标的形态， 主要包括圆率、 离心率、 区域重心、 曲 率、 区域弦分布形态描述子相关特征、 边界拟合多边形形态描述子、 傅里叶系数形态描绘特 征矢量、 基于凸包的形态描述相关特征、 基于外接矩形的特征描述、 基于不变距特征的形状 特征描述和基于区域骨架提取的形状特征共 26个特征。其中区域弦分布形态描述子、基于凸 包的形态描述相关特征、 基于外接矩形的特征描述、 基于不变距特征的形状特征描述和基于 区域骨架提取的形状特征等几类特征是本发明中提出的描述形态的新方法；
2.3色度特征： 包括基于 HSV目标区域颜色直方图、 基于概率窗的区域目标主色特征提 取、 色彩距离；
2.4纹理特征， 是基于小波变换域的多尺度紋理特征： 其包括多尺度小波能量比例、多尺 度小波标准差、 综合共生矩阵的纹理特征、 融合纹理谱的 Zernike矩特征描述。
具体来说， 如图 2所示， 本发明红细胞形态学分析方法的操作步骤为：
步骤 1：用自动显微镜〗的低倍物镜头先对所设置区域的样本进行扫描，对发现的目标区 域予以标记， 并同时由该自动显微镜 1的高倍物镜头对已标记区域进行扫描；
步骤 2: 用摄像机或 CCD元件 2对已标记区域进行图像信息采集， 图 5、 图 1 1、 图 17 即为采集到的红细胞图像； 正常形态红细胞， 细胞大小较为一致， 胞体正常或偏大， 血红蛋 白丰富，无芽孢形成， 细胞膜完整。 各类参数统计曲线图基本上为正态分布， 分布区域较集 中。
步骤 3: 用图像数字转换器 3对采集的图像先根据所含细胞进行分割定位， 再对分割后 的图像进行数字化处理， 即提取各细胞的形态学特征参数， 用大小、 形状、 色度以及纹理四 类特征来描述各细胞
步骤 4: 将步骤 3取得的各细胞的大小、 形状、 色度及纹理四类形态学特征参数输入建 立在神经网络基础上的分类器， 由该分类器从各类细胞中分离出红细胞； 下表为不同种类的 某一个红细胞的部分特征参数举例表：
6
替换页 （细则第 26条)
Figure imgf000009_0001
该建立在神经网络基础上的分类器包括一反馈过程， 该反馈过程是对分类出来的可疑目 替 （ 26 ) 标及识别错误目标进行细化、 分类、 补充特征参数， 并建立相应的数学模型， 对神经网络进 行训练， 神经网络自动学习并记忆该些细化、 分类、 补充的特征参数进入模型数据库， 再返 回基于神经网络的分类器进行细胞分类。
步骤 5:将步骤 4分离出的红细胞的形态学特征参数输入建立在模糊聚类基础上的特征融 合器， 由该特征融合器进行归一化处理， 得到一维特征向量；
该特征融合器的具体操作实际就是将初始得到的每一类特征参数通过基于 PCA加权的特 征融合算法计算得出一个归一化的特征值。
下面具体说明介绍该 PCA加权的特征融合算法：
1 . 图 3为该 PCA加权的特征融合算法原理图。 首先， 输入训练样本 π维特征向量， 该 n 维特征向量可以是大小、 形状、 紋理、 色度特征向量。 然后， 通过 PCA算法进行特征抽 取， 特征抽取分为 K-L特征压维和主分量特征选择。在特征抽取后得到的主分量特征向量 空间中分别选出大小特征、 形状特征、 纹理特征、 色度特征。 接着对这些特征子空间进行 特征向量归一化， 再联合各维向量的权重进行加权融合， 各特征子空间通过加权融合后得 到相应的一个特征量， 再将其融为一维特征向量。
2. 具体而言， 基于 PCA的特征选择算法原理为：
设输入原始特征空间 = y;,y ,L ,/„} , 将其通 ϋΚ- LIE交变换得到对应特征值从大到小 排列的正交特征向量 间 {x ，L ,x , 各特征向量分别对应特征值 (A, ，L , Λ„)。 通过累加 贡献率得到压维后的特征向量空间， 累加贡献率如式 （1 ) :
Figure imgf000010_0001
若 em > e ( e为贡献率阈值， 0 < e≤l )则选择满足条件的最小 mO≤N)。 取 ,x„} 前 m压维后的特征向量作为特征向量空间 = {^ ,x 。
PCA的目标就是通过该正交向量矩阵 来提取出原始特征空间 F的 m维主分量特征向量。 其原理是将原始特征向量 ;，/2,L 沿着; 方向投影时， PCA将使得到的 能量最大， 这时 _；称为第一主分量； 在与 X,正交的条件下， 原始特征空间在 上的投影， 使得/ 2能量最大， 这时称为第二主分量， 以此类推可得到主分量特征空间 r = ^, /2，L ,/„}。
3. 特征归一化
在主分量特征空间 y分别选出色度、 纹理、 形状、 大小特征子空间 、 γ, , ys、 Ye , 子 空间表达式如式 （2):
Figure imgf000010_0002
替 （ 第 26条) 分别对色度、纹理、形状、大小特征子空间内的各维特征向量归一化， 归一化公式如式（3):
Figure imgf000011_0001
Μ λ ¾ w • max + pt. nun 其中 为第 p维特征向量中第 i个特征值， ^和 ^分别为第 P维特征向量中最大 和最小特征值。 通过式 （3) 得到归一化后的特征子空间 F 、 J 、 4、 Y∞。
4. 基于权函数的特征融合
由于归一化的特征子空间是在主分量特征空间中提取出来的， 可知每一维特征向量的贡 献能力有大小之分， 因此所占权重也有所不同。 可计算特征向量的均值和标准方差来描述特 征向量的贡献能力， 计算公式如式 （4):
σ„ =
Figure imgf000011_0002
工 ¾、中 和 σρ分别表示第 ρ维向量的均值和标准方差 t
可利用 和 定义下面的代价函数评估特征的鉴别能力，评估函数如式 （5)
Figure imgf000011_0003
越大表明该特征的鉴别能力越强， 可用 Jp来表示各向量的权重。则各个向量对应的权 重向量为 (J, , J2，L , Jd ) , ί为某一特征子空间的特征维数。可通过矩阵变换将特征子空间融合 得到一维特征值。 矩阵变换如式 （6) ：
(6)
Figure imgf000011_0004
(a„a2,L , )7'中的元素 fl,表示第 i个训练样本对应与某一特征子空间降维后的特征值。 该特征值都归一化到 0~1之间， 可分段表示各类型的特征。
通过该方法可得到融合后的归一化色度、 紋理、 形状、 大小特征的一维特征向量空间 ，。,，α·,Α)7。
步骤 6:将每个标本中所有红细胞的每一类归一化特征量经输出设备显示出来， 即得出每 一类归一化特征参数的统计学图表； 替换页 （细则第 26条) 步骤 ό. Ι
分别得出样本大小、 形状、 色度、 纹理四类特征参数统计学图表； 根据作出的四类特征 参数统计学分布图与正常的红细胞的大小分布进行比对， 根据偏离的方向和程度来判断样本 中红细胞的来源或红细胞的类型。
样本大小、 形状、 色度、 纹理四类特征参数统计学数据图表达方式不限定， 可以是直方 图、 分布图、 散射图、 曲线图、 直方图、 面积图、 饼图、 散点图、 环形图、 雷达图、 气泡图、 圆柱图等。
步骤 6.1 .1 使用融合后的归一化的大小特征向量， 作出红细胞大小分布特征曲线， 如图 6J -图 0.9、 图 12及图 18所示。 横坐标为大小值， 纵坐标为对应大小值出现的频数。
红细胞大小特征主要反映样本中红细胞的大小分布情况。 红细胞大小分布特征曲线左移 为小红细胞性改变， 曲线右移为大红细胞性改变， 曲线出现多个峰值即表明红细胞有大有小， 存在不均一性。 若峰值小， 则细胞体积小； 峰值大， 细胞体积大。在出现多个峰值的情形下， 则表示有明显大小不均混合的细胞，通过将多峰值与参考值比较可以得出细胞大小分布情况。 频数的分布宽度的大小表明细胞大小的集中程度。 根据样本红细胞大小分布与正常的红细胞 的大小分布进行比对根据偏离的方向和程度来判断样本中红细胞的来源或红细胞的类型。
如图 6.1 , 为一个包含正常红细胞的尿液样本的大小特征参数的统计图， 这里以曲线的方 式显示。但特征参数统计曲线图可以以多种形式的统计图表达， 如图 6.2-图 6.9所示， 均为正 常形态红细胞大小特征参数统计图。
为了更方便说明统计图， 如图 4所示 , α表示统计图中的分布宽度。 就大小特征统计曲线 而言， 当红细胞大小越相近， 分布宽度 α越小。 红细胞大小越不均一， 分布宽度 α越大。 设 定分布宽度阀值 L, 当分布宽度 a > L 的时候， 可以判定该样本红细胞为大小不均红细胞， 系统提示样本疑似大小不均红细胞。
b为统计图分布峰值对应的 X轴数值， 当 b值变化时， 在统计图用曲线表达的情况下， 表现为曲线的左移或右移。 设峰值最小阔值 D l， 最大阔值 D2。 当 Dl <b<D2时， 红细胞 大小在正常范围内。 当 b<D l， 曲线左移， 系统提示样本疑似小红细胞； 当 b>D2, 曲线右 移， 系统提示样本疑似大红细胞。
图 12、 图 18为红细胞大小特征统计图例图。 图 12表述芽孢形红细胞大小特征， 在芽孢 形红细胞外膜有小泡突出或细胞呈霉菌孢子样改变。 图 18表述大小不均红细胞大小特征。 2 个图的红细胞大小分布宽度 a>L, 从图上看宽度比正常形红细胞分布宽度 a大， 表示红细胞 大小分布不均。 b>D2, 部分红细胞大小偏大。
步骤 0.1 .2使用融合后的归一化的形状特征向量，作出红细胞形状分布特征曲线，如图 7、 图 13及图 19所示。 横坐标为形状特征值， 纵坐标为对应特征值出现的频数。
红细胞形状分布特征指标主要反映畸形红细胞分布情况。 使用代表形状的特征参数数据 组合 （如圆率、 方框率、 弦分布对称、 凸孢、 内切等） 来分析。 正常红细胞的形态是双凹圆 盘状， 畸形红细胞的形态是芽孢样、 口型状等。 根据红细胞形状特征参数的变化来判断样本 中红细胞的来源或红细胞的类型。
如图 7, 为一个包含正常红细胞的尿液样本的形状特征参数的统计图， 此处以曲线的方 式显示。
10
替换页 （细则第 26条） 就形状特征统计曲线而言， 当红细胞形态越接近理想标准红细胞形态， 分布越集中， 分 布宽度越小， 峰值对应的频数值 C越大 （0<C<100%)。 设形状特征统计曲线的峰值对应频 数阀值为 ^1， 此处以 Η=ό07。为例， 当 C<H时， 系统提示样本形态疑似较多偏离理想红细胞 形态。
图 13及图〗9为红细胞形状特征统计图例图。 图〗3是芽抱形红细胞形状特征图， 在芽 孢形红细胞外膜有小泡突出或细胞呈霉菌孢子样改变。 图 19表述大小不均红细胞形状特征。 两图的红细胞形状分布宽度与正常红细胞相比变化不大。而峰值对应频数值 C均小于正常形 态红细胞， C<H。
步骤 6.1.3使用融合后的归一化的色度特征向量，作出红细胞色度分布特征曲线， 如图 8、 图 14及图 20所示。 横坐标为色度值， 纵坐标为对应色度值出现的频数。
红细胞色度分布特征主要反映红细胞血色素丢失情况。 使用代表色度的特征参数数据组 合 （如色调、 饱和度等） 来分析。 根据红细胞色度与正常的红细胞色度进行比对根据偏离的 方向和程度来判断样本中红细胞的来源或红细胞的类型。
红细胞失血色素后色度会变浅， 此类情况红细胞色度分布特征频数直方图曲线会发生左 移， 峰值变小； 红细胞在高渗尿液中易失水形成皱缩红细胞， 色度会变深， 此类情况色度频 数直方图曲线会发生右移， 峰值变大。 这类情况不属于异常范围， 需注意鉴别判断。
如图 8， 为一个含有正常红细胞的尿液样本的色度特征参数的统计图， 此处以曲线的方 式显示。 就色度特征统计曲线而言， 当红细胞色度越接近正常红细胞色度， 分布越集中， 峰 值对应的频数值 C越大 （0<C<100%)。 设色度特征统计曲线的峰值对应频数阀值为 H, 此 处以 H=60%为例， 当 C<H时， 系统提示样本中红细胞色度较多偏离理想红细胞色度。 就色 度特征统计曲线而言， 当红细胞色度分布越接近， 分布宽度越小。
图 4中的 b为统计图分布峰值对应的 X轴数值， 当 b值变化时，在统计图用曲线表达的 情况下， 表现为曲线的左移或右移， 也就是红细胞色度的深浅变化。 色度越深， b值越大； 色度越浅， b值越小。 设峰值对应 X轴最小阀值 S1 , 最大阀值 S2。 当 Sl <b<S2时， 红细 胞色度在正常范围内。 当 1^<51 , 曲线左移， 表示细胞色度偏低， 当 b>S2, 曲线右移， 表示 细胞色度偏高。
图〗 4、 图 20为红细胞色度特征统计图例图。 虚线表示正常红细胞色度， 呈单一窄峰， 说明细胞的色度非常均匀， 没有细胞浆的丢失。 红细胞的色度与血红蛋白含量有相关性。 图 14表述芽孢形红细胞色度特征。 图 20表述大小不均红细胞色度特征。 两图的红细胞色度 均值与正常形态红细胞相比均有变化。 图 W芽孢形红细胞色度均值显著降低， 图 20大小不 均红细胞色度均值偏高。其峰值对应频数值 C均明显小于正常形态红细胞， C<H。 图 14芽 孢形红细胞色度分布宽度变化不大，细胞分布均匀呈较窄单峰。图 21大小不均红细胞分布宽 度往色度值较大方向明显增大。
步骤 0.】 .4使用融合后的归一化的紋理特征向量，作出红细胞纹理分布特征曲线，如图 9、 图 15及图 21所示。 横坐标为紋理值， 纵坐标为对应紋理值出现的频数。
红细胞纹理分布特征使用代表红细胞中央区色梯（如中央苍白区扩大、中央苍白区消失、 中央区色度增强等）及代表纹理的特征参数数据组合（如出现不规则纹理、 皱缩等）来分析。 根据红细胞纹理与正常的红细胞纹理进行比对根据偏离的方向和程度来判断样本中红细胞的
Π 来源或红细胞的类型。
如图 9， 为一个包含正常红细胞的尿液样本的纹理特征参数的统计图， 此处以曲线的方 式显示。 就纹理特征统计曲线而言， 当红细胞的紋理越丰富， 其纹理值越大。 如图 4所示 , α 表示统计图中的分布宽度。 当样本中细胞的纹理变化差别越小， 统计图的分布宽度 α越小； 细胞的纹理变化差别越大， 统计图的分布宽度 α越大。
图 4中的 b为统计图分布峰值对应的 X轴数值， 当 b值变化时，在统计图用曲线表达的 情况下， 表现为曲线的左移或右移， 也就是红细胞纹理的强度变化。 纹理越强， 其能量越大， b值越大；纹理越浅， b值越小。设峰值对应 X轴最小阔值 W1，最大阔值 W2。当 Wl <b<W2 时， 红细胞纹理在正常范围内。 当13<\^1 , 表示样本纹理细弱， 曲线左移； 当 b>W2, 表示 样本纹理较粗， 曲线右移。
图 15表述芽孢形红细胞纹理特征。 图 21表述大小不均红细胞紋理特征。 两图的红细胞 纹理均值与正常形态红细胞相比均有变化。
步骤 6.2将二种以上类别的归一化特征向量组合， 用统计学方法综合分析， 得到多参数 分析图表；
以红细胞大小和色度这两种归一化特征向量组合进行综合分析为例， 如图 10、 图 16、 图 22及图 23所示。 横坐标为红细胞大小值， 纵坐标为对应红细胞色度值， 作出散点图。
图 10 为正常形态红细胞的大小和色度综合分析图中， 正常形态红细胞集中分布在 75<X<125并且 20<丫<40之间。
图 16为芽孢形态红细胞的大小和色度综合分析图中，红细胞色度明显偏低，而且红细胞 大小范围扩大， 主要分布在 80<X<100之间。 在图 22大小不均形红细胞综合分析图中， 红 细胞的色度和大小分布范围更大， 5<Χ<30且 40<丫<150。
利用本装置进行红细胞分析方法所表达的相同的结果， 在不同的样本中具有不同的临床 意义， 如样本中出现小体积低色素红细胞时， 本装置分析出的结果是单个红细胞体积小、 色 度低， 总红细胞形态特征参数以表示大小的参数组合为横坐标， 表示色度的特征参数数据为 纵坐标的形态学分析图形表达为红细胞分布宽度增加往左偏移， 红细胞分布区域向下移， 表 达为向左分散向下沉的红细胞形态分布图， 此类图形在血液样本中出现提示贫血， 如果在尿 液样本中出现并占一定比例则意味着来源于肾性红细胞。
如图 23所示， 当血液样本中 Ρ% (50 Ρ 100)的红细胞分布在 Χ<75并且 Υ<20区间， 系统提示样本疑似贫血； 当尿液样本中 Ρ% (50 Ρ 100)的红细胞分布在 Χ<75并且丫 <20 区间， 系统提示样本疑似肾病。
步骤 7根据样本中各类型红细胞占总红细胞数的比例经统计学处理， 以图形或数据方式 表达， 对样本中红细胞进行分析鉴定。
如图 24所示，通过对样本中细胞进行识别后分类计数，得出样本中各种细胞占其中总红 细胞数的比例。 在通过图形或数据的表达情形下， 方便操作者以此数据为依据， 对样本情形 作出判断。 此处以统计学饼图的方式表达为例。
步骤 7根据样本中各类型红细胞的一个归一化的一维形态学特征向量， 通过给出一个特 征量阀值， 计算高于或者低于阀值的红细胞占样本总红细胞同类形态学特征参数数据组合的 比例， 经统计学处理后以图形或数据方式表达。 图 25为一样本色度特征参数统计图。 图中， 使用融合后的归一化的色度一维特征向量， 设色度给定阀值为 H来判断低色素红细胞， 样本中细胞色度低于阀值 H的比例为 40%。 也 就是说， 该样本中低色素红细胞占样本中总红细胞数的 40%。 图 20为图 25的一个变化表达 形式。
上述图形或数据方式表达在输出装置上显示， 供相关人员参考。 该输出装置可以是本发 明红细胞形态学分析装置上的显示窗口， 也可以是与装置相连接的显示器， 及与网络连接用 于远程会诊的显示器， 或者是以打印的方式表达出来， 以便于医生分析。
该统计图可以以多图排列的方式表达， 如图 27所示。 也可以转换为单图方式显示。 另外， 本发明涉及样本可以是尿液样本、 血液样本。 这里需要特别指出的是， 血液样本 不做涂片处理， 而是以一定的倍数进行稀释后进行分析， 这样就不会因为涂片而破坏样本中 的部分细胞。 本发明要求样本新鲜， 最好是取得样本 2小时内进行检査， 样本无须染色， 但 不限于不染色， 这样在检验中不需用到昂贵的试剂， 因而经济节约， 无污染， 有利于环保。
13
替换页 （细则第 26条)

Claims

权 利 要 求 书
1、 一种红细胞形态学分析装置， 其特征在于包括：
a . 一自动显微镜， 该自动显微镜的低倍物镜头对所设置区域的样本进行扫描， 对发现 的目标区域予以标记， 并同时由该自动显微镜的高倍物镜头对已标记区域进行扫描；
b . 一摄像机或 CCD元件， 该摄像机或 CCD元件对已标记区域进行图像信息采集； c .一用以对上述图像进行分析处理的图像数字转换器，该图像数字转换器先将采集的 图像根据所含的细胞进行分割定位， 再对分割后的图像进行数字化处理，提取各细胞的大小、 形状、 色度及纹理四类形态学特征参数；
d . 一建立在神经网络基础上的分类器，用于根据上述歩骤取得的各细胞的大小、形状、 色度及纹理四类形态学特征参数对细胞进行分类， 以从各类细胞中分离出红细胞；
e .一建立在模糊聚类基础上的特征融合器，该特征融合器用于将上述步骤分离出的红 细胞的大小、 形状、 色度及纹理四类多维形态学特征参数进行归一化降维处理得到大小、 形 状、 色度及纹理 4个特征值， 再根据标本中所有红细胞的大小、 形状、 色度、 纹理特征分别 进行统计计算和统计图形表达， 为分析样本中红细胞类别及来源提供真实客观依据；
f. 一输出设备， 用于直观显示检测结果；
g . 一控制单元， 该控制单元分别连接上述自动显微镜、摄像机或 CCD元件、 图像数字 转换器及输出设备， 以控制该自动显微镜、 摄像机或 CCD元件、 图像数字转换器及输出设备 动作。
2、 根据权利要求 1所述的红细胞形态学分析装置， 其特征在于， 该输出设备表达出各种 类型红细胞所特有的代表红细胞中央区色梯、 纹理的特征参数数据组合， 该样本中红细胞的 分析鉴定方法是根据红细胞中央区的特征改变来判断样本中红细胞的来源或红细胞的类型。
3、 根据权利要求 1所述的红细胞形态学分析装置， 其特征在于， 该样本中红细胞形态学 分析方法是参照临床确认的代表不同意义的各种类型红细胞具有不同的形态学特征参数数 据， 据此进行识别分类计数， 再根据样本中各类型红细胞某一形态学特征参数数据组合占总 红细胞的同类形态学特征参数数据组合的比例经统计学处理后以图形或数据方式表达。
4、 根据权利要求 1所述的红细胞形态学分析装置， 其特征在于， 该样本中红细胞形态学 分析方法是将样本中各类型红细胞的两种以上的红细胞形态学特征参数组合用统计学方法综 合分析， 得到样本中各类型红细胞多参数分析结果， 根据其改变判断红细胞形态学变化， 通 过图形和数据直观的表达出样本中红细胞形态学变化类型。
5、 根据权利要求 1所述的红细胞形态学分析装置， 其特征在于， 利用本装置进行红细胞 分析所表达的相同的结果， 在不同的样本中具有不同的临床意义， 如样本中出现小体积低色 素红细胞时， 本装置分析出的结果是单个红细胞体积小、 色度低， 总红细胞形态特征参数以 表示大小的参数组合为横坐标， 表示色度的特征参数数据为纵坐标的形态学分析图形表达为 红细胞分布宽度增加往左偏移， 红细胞分布区域向下移， 表达为向左分散向下沉的红细胞形 态分布图， 此类图形在血液样本中出现提示一种类型贫血， 如果在尿液样本中出现并占一定 比例则意味着来源于肾性红细胞； 如样本中出现大体积高色素红细胞时， 本装置分析出的结 果是单个红细胞体积大、 色度高， 总红细胞形态特征参数以表示大小的参数组合为横坐标， 表示色度的特征参数数据为纵坐标的形态学分析图形表达为红细胞分布宽度增加往右偏移， 红细胞分布区域向上移， 表达为向右分散向上升的红细胞形态分布图， 此类图形在血液样本 中出现提示另一种类型贫血， 如果在尿液样本中出现并占一定比例则意味着来源于非肾性红 细胞。
6、 一种红细胞形态学分析方法， 其特征在于， 包括下列歩骤：
歩骤 1:用自动显微镜的低倍物镜头对所设置区域的样本进行扫描，对发现的目标区域予 以标记， 并同时由该自动显微镜的高倍物镜头对已标记区域样本进行扫描；
歩骤 2： 用摄像机或 C C D元件对已标记区域样本进行图像信息采集；
歩骤 3:用图像数字转换器对采集的图像先根据所含的细胞进行分割定位，再对分割后的 图像进行数字化处理， 提取各细胞的形态学特征参数， 用大小、 形状、 色度以及纹理四类特 征来描述各细胞；
步骤 4: 将上述步骤取得的各细胞的大小、形状、色度及纹理四类形态学特征参数输入建 立在神经网络基础上的分类器， 由该分类器从各类细胞中分离出红细胞；
步骤 5: 将步骤 4分离出来的红细胞的大小、 形状、 色度以及纹理四类形态学特征参数输 入建立在模糊聚类基础上的特征融合器， 由该特征融合器将每一类多维形态学特征参数进行 归一化处理， 得到一维特征向量；
步骤 6:将每个标本中所有红细胞的每一类归一化特征量经输出设备显示出来， 即得出每 一类归一化特征参数的统计学图表。
7、 根据权利要求 6所述的红细胞形态学分析方法， 其特征在于， 还包括步骤 7: 根据样 本中各类型红细胞占总红细胞数的比例经统计学处理， 以图形或数据方式表达， 对样本中红 细胞进行分析鉴定。
8、 根据权利要求 6所述的红细胞形态学分析方法， 其特征在于， 还包括步骤 8: 对每一 类经过归一化的形态学特征向量， 通过给出一个特征量阀值， 计算高于或者低于阀值的红细 胞占样本总红细胞同类形态学特征参数数据组合的比例， 经统计学处理后以图形或数据方式 表达， 以提供对样本中红细胞进行分析鉴定的客观依据。
9、 根据权利要求 6所述的红细胞形态学分析方法， 其特征在于， 该建立在神经网络基础 上的分类器包括一反馈过程，该反馈过程是对分类出来的可疑目标及识别错误目标进行细化、 分类、 补充特征参数， 并建立相应的数学模型， 对神经网络进行训练， 神经网络自动学习并 记忆该些细化、 分类、 补充的特征参数进入模型数据库， 再返回基于神经网络的分类器进行 细胞分类。
10、 根据权利要求 6记载的红细胞形态学分析方法， 其特征在于， 步骤 6中得出的归一化 的大小特征向量表达出各种类型红细胞所特有的代表大小的特征参数数据组合， 样本中红细 胞分析鉴定方法， 可根据红细胞大小分布与正常的红细胞的大小分布进行比对， 根据偏离的 方向和程度来判断样本中红细胞的来源或红细胞的类型。
11、 根据权利要求 6记载的红细胞形态学分析方法， 其特征在于， 步骤 6中得出的归一化 的形状特征向量表达出各种类型红细胞所特有的代表形状的特征参数数据组合， 样本中红细 胞的分析鉴定方法， 可根据红细胞形状特征参数的变化来判断样本中红细胞的来源或红细胞 的类型。
12、 根据权利要求 6记载的红细胞形态学分析方法， 其特征在于， 歩骤 6中得出的归一化 的色度特征向量表达出各种类型红细胞所特有的代表色度的特征参数数据组合， 样本中红细 胞的分析鉴定方法， 可根据红细胞色度与正常的红细胞色度进行比对根据偏离的方向和程度 来判断样本中红细胞的来源或红细胞的类型。
13、 根据权利要求 6记载的红细胞形态学分析方法， 其特征在于， 步骤 6中得出的归一化 的纹理特征向量表达出各种类型红细胞所特有的代表纹理的特征参数数据组合， 样本中红细 胞的分析鉴定方法， 可根据红细胞纹理特征参数的变化来判断样本中红细胞的来源或红细胞 的类型。
14、 根据权利要求 10~13记载的红细胞形态学分析方法， 其特征在于， 对样本中红细胞 的分析鉴定， 单独采用其中任意一条来分析， 或综合采用其中的多条来分析。
PCT/CN2011/079710 2011-09-16 2011-09-16 红细胞形态学分析装置及其方法 WO2013037119A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
US13/880,593 US9170256B2 (en) 2011-09-16 2011-09-16 Device and method for erythrocyte morphology analysis
BR112013032938-6A BR112013032938B1 (pt) 2011-09-16 2011-09-16 Dispositivo e métodos para realizar análise morfológica para eritrócitos
PCT/CN2011/079710 WO2013037119A1 (zh) 2011-09-16 2011-09-16 红细胞形态学分析装置及其方法
EP11872496.2A EP2757372B1 (en) 2011-09-16 2011-09-16 Device and method for erythrocyte morphology analysis

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2011/079710 WO2013037119A1 (zh) 2011-09-16 2011-09-16 红细胞形态学分析装置及其方法

Publications (1)

Publication Number Publication Date
WO2013037119A1 true WO2013037119A1 (zh) 2013-03-21

Family

ID=47882538

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2011/079710 WO2013037119A1 (zh) 2011-09-16 2011-09-16 红细胞形态学分析装置及其方法

Country Status (4)

Country Link
US (1) US9170256B2 (zh)
EP (1) EP2757372B1 (zh)
BR (1) BR112013032938B1 (zh)
WO (1) WO2013037119A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015058353A1 (zh) * 2013-10-22 2015-04-30 爱威科技股份有限公司 一种红细胞形态学分析结果表示方法
TWI754945B (zh) * 2019-11-27 2022-02-11 靜宜大學 利用光動力學技術之人工智慧的細胞檢測方法及其系統

Families Citing this family (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104881865B (zh) * 2015-04-29 2017-11-24 北京林业大学 基于无人机图像分析的森林病虫害监测预警方法及其系统
EP3335194A4 (en) * 2015-08-12 2019-04-24 Molecular Devices, LLC SYSTEM AND METHOD FOR AUTOMATICALLY ANALYZING PHENOTYPIC RESPONSES OF CELLS
EP3859425B1 (en) * 2015-09-17 2024-04-17 S.D. Sight Diagnostics Ltd. Methods and apparatus for detecting an entity in a bodily sample
CA3018536A1 (en) 2016-03-30 2017-10-05 S.D. Sight Diagnostics Ltd Distinguishing between blood sample components
US11307196B2 (en) 2016-05-11 2022-04-19 S.D. Sight Diagnostics Ltd. Sample carrier for optical measurements
RU2629609C1 (ru) * 2016-08-15 2017-08-30 Федеральное Государственное Бюджетное Образовательное Учреждение Высшего Образования "Нижегородская Государственная Сельскохозяйственная Академия (ФГБОУ ВО НГСХА) Способ прижизненной оценки действия реагентов на образование стоматоцитов
CN106290081A (zh) * 2016-08-16 2017-01-04 江苏康尚生物医疗科技有限公司 一种区分粒子群落的方法及粒子分析仪
US10311573B2 (en) * 2017-05-02 2019-06-04 Techcyte, Inc. Training and machine learning classification of mold in digital microscopy images
GB2566009A (en) * 2017-08-23 2019-03-06 Epigem Ltd Classification, screening and diagnostic methods and apparatus
WO2019182528A2 (en) * 2017-10-23 2019-09-26 Altiparmak Gida Sanayi Ve Ticaret Anonim Sirketi Pollen classification system and related method
US11921272B2 (en) 2017-11-14 2024-03-05 S.D. Sight Diagnostics Ltd. Sample carrier for optical measurements
JP2019195304A (ja) * 2018-05-10 2019-11-14 学校法人順天堂 画像解析方法、装置、コンピュータプログラム、及び深層学習アルゴリズムの生成方法
WO2020047468A1 (en) * 2018-08-30 2020-03-05 Becton, Dickinson And Company Characterization and sorting for particle analyzers
JP7381003B2 (ja) 2019-04-26 2023-11-15 学校法人順天堂 疾患解析を支援する方法、装置、及びコンピュータプログラム、並びにコンピュータアルゴリズムを訓練する方法、装置、及びプログラム
CN110503648B (zh) * 2019-08-30 2022-03-01 中国人民解放军陆军军医大学第二附属医院 一种细胞图像调整方法及装置
CN110889839B (zh) * 2019-11-27 2022-11-25 贵州中烟工业有限责任公司 一种打叶复烤烟片片形的表征方法及系统
CN110647876B (zh) * 2019-11-28 2020-06-26 北京小蝇科技有限责任公司 一种血涂片全视野智能分析方法
US11178336B1 (en) * 2020-05-06 2021-11-16 International Business Machines Corporation Altering device capture settings through user feedback
CN111832389B (zh) * 2020-05-25 2022-12-09 中国人民解放军陆军军医大学第二附属医院 一种骨髓细胞形态学自动检测系统的计数及分析方法
DE102020123505A1 (de) * 2020-09-09 2022-03-10 Carl Zeiss Microscopy Gmbh Mikroskopiesystem und verfahren zum erzeugen eines übersichtsbildes
CN112614104B (zh) * 2020-12-22 2023-07-14 湖南伊鸿健康科技有限公司 一种红细胞重叠的分割计数方法及终端
CN113205498B (zh) * 2021-04-30 2022-09-09 天津深析智能科技发展有限公司 一种细胞形态扫描区域选择方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1037035A (zh) * 1988-04-08 1989-11-08 神经医学系统公司 基于神经网络的自动细胞试样分类系统和方法
WO1991020048A1 (en) * 1990-06-21 1991-12-26 Applied Electronic Vision, Inc. Cellular analysis utilizing video processing and neural network
EP0575091A2 (en) * 1992-06-19 1993-12-22 Toa Medical Electronics Co., Ltd. Method and device for analyzing particles
CN1383522A (zh) * 2000-04-24 2002-12-04 国际遥距成象系统公司 多个神经网络的成像设备和方法
CN101713776A (zh) * 2009-11-13 2010-05-26 长春迪瑞实业有限公司 一种基于神经网络的尿液中有形成分识别分类方法
CN102359938A (zh) * 2011-09-16 2012-02-22 长沙高新技术产业开发区爱威科技实业有限公司 红细胞形态学分析装置及其方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4129854A (en) * 1976-10-25 1978-12-12 Hitachi, Ltd. Cell classification method
WO1996009600A1 (en) * 1994-09-20 1996-03-28 Neopath, Inc. Apparatus for identification and integration of multiple cell patterns
US7106502B1 (en) * 2003-08-21 2006-09-12 The United States Of America As Represented By The Administrator Of National Aeronautics And Space Administration Operation of a Cartesian robotic system in a compact microscope imaging system with intelligent controls
US20130094750A1 (en) * 2011-10-12 2013-04-18 Tolga Tasdizen Methods and systems for segmentation of cells for an automated differential counting system

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1037035A (zh) * 1988-04-08 1989-11-08 神经医学系统公司 基于神经网络的自动细胞试样分类系统和方法
WO1991020048A1 (en) * 1990-06-21 1991-12-26 Applied Electronic Vision, Inc. Cellular analysis utilizing video processing and neural network
EP0575091A2 (en) * 1992-06-19 1993-12-22 Toa Medical Electronics Co., Ltd. Method and device for analyzing particles
CN1383522A (zh) * 2000-04-24 2002-12-04 国际遥距成象系统公司 多个神经网络的成像设备和方法
CN101713776A (zh) * 2009-11-13 2010-05-26 长春迪瑞实业有限公司 一种基于神经网络的尿液中有形成分识别分类方法
CN102359938A (zh) * 2011-09-16 2012-02-22 长沙高新技术产业开发区爱威科技实业有限公司 红细胞形态学分析装置及其方法

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
LIANG, GUANGMING: "Intelligent recognition key technology for corporeal components in body fluid cells", CHINESE DOCTORAL DISSERTATIONS, vol. 7, 15 July 2009 (2009-07-15), pages 2 - 10, 113-114, 119-120, 128-131, XP008171958 *
LIU, BIN ET AL.: "Research of automatic recognition system for blood cell imaging", COMPUTER ENGINEERING, vol. 29, no. 1, January 2003 (2003-01-01), pages 174 - 175, 188, XP008171878 *
NICHOLAS E. ROSS ET AL.: "Automated image processing method for the diagnosis and classification of malaria on thin blood smears", MED BIOL ENG COMPUT, vol. 44, no. 5, 8 April 2006 (2006-04-08), pages 427 - 436, XP019833983 *
ZHANG, ZANCHAO: "The study of key technology for full automatic urinary analysis system", CHINESE DOCTORAL DISSERTATIONS, vol. 9, 15 September 2008 (2008-09-15), pages 15 - 21, 47-51, 69-76, 85-91, 98-100, XP008171959 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015058353A1 (zh) * 2013-10-22 2015-04-30 爱威科技股份有限公司 一种红细胞形态学分析结果表示方法
TWI754945B (zh) * 2019-11-27 2022-02-11 靜宜大學 利用光動力學技術之人工智慧的細胞檢測方法及其系統

Also Published As

Publication number Publication date
US9170256B2 (en) 2015-10-27
EP2757372A4 (en) 2015-05-20
US20140185906A1 (en) 2014-07-03
EP2757372B1 (en) 2023-08-23
BR112013032938B1 (pt) 2021-06-29
BR112013032938A2 (pt) 2017-01-24
EP2757372A1 (en) 2014-07-23

Similar Documents

Publication Publication Date Title
WO2013037119A1 (zh) 红细胞形态学分析装置及其方法
CN102359938B (zh) 红细胞形态学分析装置及其方法
JP7465914B2 (ja) 生体粒子の分類のシステム及び方法
US6947586B2 (en) Multi-neural net imaging apparatus and method
Madhukar et al. New decision support tool for acute lymphoblastic leukemia classification
Mohapatra et al. Automated leukemia detection using hausdorff dimension in blood microscopic images
CN103020639A (zh) 一种白细胞自动识别计数方法
Forero et al. Automatic identification techniques of tuberculosis bacteria
KR100303608B1 (ko) 혈구세포자동인식방법및장치
CN105404887B (zh) 一种基于随机森林的白细胞五分类方法
CN107730499A (zh) 一种基于nu‑支持向量机的白细胞分类方法
CN103679184A (zh) 一种基于相关向量机的白细胞自动识别方法
WO2006113979A1 (en) Method for identifying guignardia citricarpa
CN113658174A (zh) 基于深度学习和图像处理算法的微核组学图像检测方法
Angulo et al. Ontology-based lymphocyte population description using mathematical morphology on colour blood images
Fitri et al. A comparison of platelets classification from digitalization microscopic peripheral blood smear
Anilkumar et al. Colour based image segmentation for automated detection of leukaemia: a comparison between CIELAB and CMYK colour spaces
Grochowski et al. Machine learning system for automated blood smear analysis
CN104573668B (zh) 基于光谱反射率的面部色泽适应性自动识别方法
KR20200136004A (ko) 세포 샘플 내에서 적어도 하나의 기형을 가지는 세포를 검출하기 위한 방법
CN202393695U (zh) 红细胞形态学分析装置
Nithya et al. Detection of Anaemia using Image Processing Techniques from microscopy blood smear images
Arunachalam Applications of Machine learning and Image processing techniques in the detection of leukemia
CA2377602C (en) Multi-neural net imaging apparatus and method
EP1301894B1 (en) Multi-neural net imaging apparatus and method

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 13880593

Country of ref document: US

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 11872496

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2011872496

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 2013/15380

Country of ref document: TR

NENP Non-entry into the national phase

Ref country code: DE

REG Reference to national code

Ref country code: BR

Ref legal event code: B01A

Ref document number: 112013032938

Country of ref document: BR

ENP Entry into the national phase

Ref document number: 112013032938

Country of ref document: BR

Kind code of ref document: A2

Effective date: 20131220