A leukocyte abnormal scatter plot detection method

Abstract

This invention relates to the field of medical device technology, specifically to a method for detecting abnormal white blood cell scatter plots. The method includes classifying scatter plot data using a blood cell analyzer to obtain classification results; using the classification results plus scatter plot coordinates to draw a grayscale scatter plot image for each cell type; inputting the grayscale scatter plot image into an ellipse equation recognition model to calculate the ellipse shape of each cell cluster outline to obtain recognition results; and performing anomaly analysis on the recognition results to obtain anomaly results. This method improves the accuracy of anomaly detection results and solves the problem of low accuracy in existing methods for detecting abnormal white blood cell scatter plots.

Description

Technical Field

[0001] This invention relates to the field of medical device technology, and in particular to a method for detecting abnormal white blood cell scatter plots. Background Technology

[0002] A complete blood count (CBC) is one of the most efficient and practical testing methods in clinical medicine. By analyzing the content of different cells in human blood and comparing it with the range of various cell contents in normal human blood, it can provide doctors with a basis for disease diagnosis and is an important basis for judging human health.

[0003] Flow cytometry is an important instrument for diagnosing common diseases. One key performance indicator is the differential counting of lymphocytes, neutrophils, monocytes, eosinophils, and basophils. When the human body is infected with certain diseases, the number and morphology of different types of cells in the blood change. Healthcare professionals can use cell analyzers to detect and quantify these changes in quantity and morphology as early as possible; the earlier the detection, the better the treatment. Existing cell analyzers typically count tens of thousands of cells in the sample, simultaneously extracting characteristic signals such as the volume, complexity, and refractive power of each cell to obtain a scatter plot of the counted sample.

[0004] By counting the number of scatter points in each scatter plot, the instrument can provide doctors with the location and shape of each subclass of white blood cells on the scatter plot. Based on the shape, distribution, location, and size of each type of white blood cell, doctors can determine whether there are any abnormalities in the sample, and if abnormalities are found, microscopic examination will be arranged.

[0005] Currently, the abnormality of a scatter plot is usually determined based on the position and morphology of each cell cluster in the scatter plot. After the instrument completes cell signal acquisition, generates a cell distribution scatter plot, and automatically classifies the scatter plot, it can analyze the position and size of each type of cell in the scatter plot. These position and size measurements are used to determine if the scatter plot exceeds the normal range. If it does, it indicates an abnormality, and the doctor is notified that the classification result is unreliable.

[0006] Currently, the location and size of various cells are mainly determined through traditional machine vision image processing methods. After converting the scatter plot into a binary image, each type of blood cell data is processed separately, including noise reduction, binarization, discontinuity filling, and edge extraction, to obtain the distribution shape of each type of blood cell; that is, the boundary curve of each type of blood cell. Then, an elliptic curve model is used to fit the boundary curve to obtain an elliptic curve that is most similar to the boundary curve. Some standard parameters of this elliptic curve are calculated: tilt angle, center position, major axis, and minor axis.

[0007] The above methods heavily rely on the statistical characteristics of the samples and the parameters of the edge detection algorithm. If the number of scattered points is too large or too small, or if the edge detection threshold is increased or decreased even slightly, the results of traditional image processing methods will vary greatly, reducing the accuracy of anomaly detection results. Summary of the Invention

[0008] The purpose of this invention is to provide a method for detecting abnormal white blood cell scatter plots, which aims to solve the problem of low accuracy in existing methods for detecting abnormal white blood cell scatter plots.

[0009] To achieve the above objectives, the present invention provides a method for detecting abnormal white blood cell scatter plots, comprising the following steps:

[0010] The scatter plot data was classified using a blood cell analyzer to obtain the classification results;

[0011] The grayscale scatter plot image of each type of cell was drawn by adding the scatter plot coordinates to the classification results.

[0012] The grayscale scatter plot image is input into the ellipse equation recognition model to calculate the ellipse shape of each cell cluster outline, and the recognition result is obtained.

[0013] Anomaly analysis is performed on the identification results to obtain abnormal results.

[0014] The step of inputting the grayscale scatter plot image into the elliptic equation recognition model to calculate the shape of each cell cluster outline and obtain the recognition result includes:

[0015] Obtain training samples;

[0016] Construct a network that takes a 4-channel image as input and outputs 4×5 parameters;

[0017] The training samples are divided into a training set, a test set, and a validation set in a ratio of 8:1:1.

[0018] The network is trained, tested, and validated sequentially using the training set, the test set, and the validation set to obtain an elliptic equation recognition model.

[0019] The grayscale scatter plot image is input into the ellipse equation recognition model to calculate the shape of each cell cluster outline, and the recognition result is obtained.

[0020] The acquisition of training samples includes:

[0021] The scatter coordinates and cell types of multiple cells were obtained using a blood cell analyzer.

[0022] The process iterates through each cell, and based on the cell type and the scatter coordinates, increments the gray value by 1 at the corresponding position in the grayscale image of the cell type to obtain four 256×256 cell density distribution maps for the four cell types.

[0023] The elliptical outlines of the four cell density distribution maps were manually annotated to obtain four sets of elliptical parameters;

[0024] The four cell density distribution maps are combined into four sets of four-channel images, which are used as training input images. The four sets of ellipse parameters are used as the target output values ​​of the network to obtain training samples.

[0025] The network includes a feature extraction network and an elliptic parameter detection head.

[0026] The elliptical shape includes the ellipse center, inclination angle, major axis length, and minor axis length.

[0027] The step of performing anomaly analysis on the identification results to obtain anomaly results includes:

[0028] Alarm thresholds for various cell parameters are manually set. When the center coordinates or elliptical shape of any cell in the abnormal result exceeds the alarm threshold, the grayscale scatter plot image is considered abnormal, and an abnormal result is obtained.

[0029] The step of performing anomaly analysis on the identification results to obtain anomaly results includes:

[0030] Collect normal and abnormal samples from scatter plots;

[0031] The normal samples and abnormal samples of the scatter plot are respectively input into the ellipse equation recognition model to obtain the output results;

[0032] The output is then fed into an SVM network for training to obtain an anomaly scatter plot classifier.

[0033] The identification results are input into the anomaly scatter plot classifier, which outputs the anomaly results.

[0034] This invention provides a scatter plot detection method for abnormal white blood cells. The method involves classifying scatter plot data using a blood cell analyzer to obtain classification results; using the classification results and scatter plot coordinates to draw a grayscale scatter plot image for each cell type; inputting the grayscale scatter plot image into an ellipse equation recognition model to calculate the ellipse shape of each cell cluster's outline to obtain recognition results; and performing anomaly analysis on the recognition results to obtain anomaly results. This method improves the accuracy of anomaly detection and solves the problem of low accuracy in existing scatter plot detection methods for abnormal white blood cells. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a schematic diagram of scatter coordinates and cell types.

[0037] Figure 2 This is a cell density distribution map.

[0038] Figure 3 This is a schematic diagram of four sets of ellipse parameters.

[0039] Figure 4 This is a schematic diagram of the network structure.

[0040] Figure 5 This is a flowchart illustrating the steps of a scatter plot detection method for abnormal white blood cells provided by the present invention.

[0041] Figure 6 This is a flowchart of a scatter plot detection method for abnormal white blood cells provided by the present invention. Detailed Implementation

[0042] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0043] Please see Figures 1 to 6 This invention provides a method for detecting abnormal white blood cell scatter plots, comprising the following steps:

[0044] S1 uses a blood cell analyzer to classify the scatter plot data and obtain the classification results;

[0045] Specifically, taking the white blood cell DIFF channel classification of a blood cell analyzer as an example, this channel can classify white blood cells into lymphocytes, monocytes, neutrophils, and eosinophils.

[0046] S2 uses the classification results plus scatter plot coordinates to draw a grayscale scatter plot image of each type of cell;

[0047] S3 inputs the grayscale scatter plot image into the ellipse equation recognition model to calculate the ellipse shape of each cell cluster outline and obtain the recognition result;

[0048] Specifically, the elliptical shape includes the ellipse center, inclination angle, major axis length, and minor axis length.

[0049] The specific method is as follows:

[0050] S31 obtains training samples;

[0051] The specific method is as follows:

[0052] The S311 obtains the scatter coordinates and cell types of multiple cells through a blood cell analyzer, such as... Figure 1 ;

[0053] S312 iterates through each cell, and based on the cell type and scatter coordinates, increments the gray value by 1 at the corresponding position in the grayscale image of the cell type, resulting in four 256×256 cell density distribution maps for the four cell types. Figure 2 ;

[0054] S313 manually annotated the elliptical outlines of the four cell density distribution maps, obtaining four sets of elliptical parameters (center point (x, y), tilt angle α, major axis length h, minor axis length w), as shown. Figure 3 The inclination angle 'a' is defined as the angle between the major axis of the ellipse and the x-axis, expressed in radians. When the major axis is counterclockwise along the x-axis, angle 'a' is negative. The magnitude of 'a' is... arrive

[0055] S314 combines the four cell density distribution maps into four sets of four-channel images, which are used as training input images. The four sets of ellipse parameters are used as network output target values ​​to obtain training samples.

[0056] Specifically, repeat steps S311 to S312 to obtain a sufficient number of training samples. Since this invention uses the brightness parameter of a grayscale image to represent the density of the scatter plot, the training samples carry richer features compared to binary scatter plot images.

[0057] This recognition model is not an elliptical region detection model. In fact, the elliptical regions have already been identified by the instrument's recognition algorithm (corresponding to several cell distribution maps). The model's job is to "fit" the identified scatter plot into an elliptical region, obtaining the parametric equations of the elliptical region. Therefore, this paper constructs a network that takes a four-channel scatter plot image as input and outputs four elliptical equations. This network first extracts features using a feature extraction network, and then the detection head identifies the four elliptical parameters from these features, as detailed below.

[0058] S32 constructs a network that takes a 4-channel image as input and outputs a 4×5 parameter network.

[0059] Specifically, the network includes a feature extraction network and an elliptic parameter detection head.

[0060] Feature extraction network construction: The feature extraction network serves as a scatter plot feature extractor. Various mainstream networks can be used, such as VGG, Inception, ResNet, and MobileNet. In this implementation, MobileNetV3 is used as the feature extraction unit. The input image is modified to have 4 channels to meet the requirements of this invention. In MobileNetV3, after the input image passes through four "convolution and activation" modules, the feature map resolution decreases sequentially to 1 / 2, 1 / 4, 1 / 8, and 1 / 16. In this implementation, the feature maps at scales of 1 / 4 and 1 / 16 are fused to obtain multi-scale features. After upsampling the 1 / 16 feature map, it is fused with the 1 / 4 feature map to obtain a 256×64×64 feature map.

[0061] Constructing the Ellipse Parameter Detection Head: The feature extraction network is designed to output a 256×64×64 feature map. After several convolutional processes, these features are transformed into 1024×16×16 features. These features are then subjected to fully connected computation, ultimately outputting a 4×5 result. This 4×5 result represents the ellipse contour parameters for the four cell types (outputting four sets of ellipses, each set containing five parameters, corresponding to the center coordinates x and y of the ellipse contour for the four cell types, the ellipse tilt angle α, the major axis length h, and the minor axis length w). S33 divides the training samples into a training set, a test set, and a validation set in an 8:1:1 ratio.

[0062] Specifically, the collected training samples are divided into training, testing, and validation sets in an 8:1:1 ratio. The network parameters are adjusted using the training set, and the network accuracy is calculated using the validation set, until the network accuracy meets the requirements.

[0063] Because it is necessary to consider the center coordinates, tilt angle, and major and minor axis lengths of the ellipse, the model's loss function L... loss From the position loss function L c Tilt angle loss function L a Size loss function L s composition.

[0064] in:

[0065] Loss function L c Calculate the center offset loss of the fitted ellipse:

[0066]

[0067] i indicates the cell type;

[0068] Represents the coordinates of the center point of the cell elliptical contour predicted by the model;

[0069] C i The actual coordinates of the center point of the cell's elliptical outline;

[0070] x i y i These are the x and y coordinates of the center point, respectively.

[0071] The symbol |||| represents finding the magnitude of a vector. That is, if ci and Ci are vectors, then ||ci-Ci|| is the square root of the sum of the squares of the components after subtracting the vectors.

[0072] Size loss function L s Calculate the loss for the major axis length h and minor axis length w of the ellipse:

[0073]

[0074] h i w i These represent the major axis length and minor axis length of the cell ellipse profile predicted by the model, respectively.

[0075] H i W i These represent the actual lengths of the major and minor axes of the cell's elliptical outline, respectively.

[0076] A single vertical line means taking the absolute value.

[0077] Tilt angle loss function L a Calculate the tilt angle loss using Smooth L1 Loss:

[0078] δ=a1-A1

[0079]

[0080]

[0081] a i w i These represent the major axis length and minor axis length of the cell ellipse profile predicted by the model, respectively.

[0082] The model's loss function L loss We obtain the result by weighted sum of the three:

[0083]

[0084] γ1, γ2, and γ3 are weights, with a default value of 1.

[0085] S34 uses the training set, the test set, and the validation set to train, test, and validate the network in sequence to obtain an elliptic equation recognition model;

[0086] S35 inputs the grayscale scatter plot image into the ellipse equation recognition model to calculate the shape of each cell cluster outline and obtain the recognition result.

[0087] S4 performs anomaly analysis on the identification results to obtain abnormal results.

[0088] Specifically, there are two ways to perform anomaly analysis on the identification results and obtain anomaly results. The first way is to manually set alarm thresholds for various cell parameters. When the center coordinates or elliptical shape of any cell in the anomaly results exceed the alarm threshold, the grayscale scatter plot image is considered abnormal, and an anomaly result is obtained.

[0089] The second method is as follows: collect normal scatter plot samples and abnormal scatter plot samples; input the normal scatter plot samples and the abnormal scatter plot samples into the elliptic equation recognition model respectively to obtain the output results; input the output results into the SVM network for training to obtain an abnormal scatter plot classifier; input the recognition results into the abnormal scatter plot classifier to output abnormal results.

[0090] Anomaly scatter plot identification and processing: When the scatter plot of a sample is identified as abnormal, the system will notify the instrument that the test result is abnormal. The instrument will display the alarm result in the test report of the corresponding sample and can decide whether to automatically retest the sample based on its own situation.

[0091] The above-disclosed embodiments are merely preferred embodiments of the scatter plot detection method for abnormal white blood cells of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art can understand that implementing all or part of the above embodiments and making equivalent changes in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method of detecting white blood cell abnormal scattergram, characterized by, Includes the following steps: The scatter plot data was classified using a blood cell analyzer to obtain the classification results; The grayscale scatter plot image of each type of cell was drawn by adding the scatter plot coordinates to the classification results. The grayscale scatter plot image is input into the ellipse equation recognition model to calculate the ellipse shape of each cell cluster outline, and the recognition result is obtained. Anomaly analysis is performed on the identification results to obtain abnormal results; The step of inputting the grayscale scatter plot image into the elliptic equation recognition model to calculate the shape of each cell cluster outline and obtain the recognition result includes: Obtain training samples; Construct a network that takes a 4-channel image as input and outputs 4×5 parameters; The training samples are divided into a training set, a test set, and a validation set in a ratio of 8:1:

1. The network is trained, tested, and validated sequentially using the training set, the test set, and the validation set to obtain an elliptic equation recognition model. The grayscale scatter plot image is input into the ellipse equation recognition model to calculate the shape of each cell cluster outline and obtain the recognition result. The acquisition of training samples includes: The scatter coordinates and cell types of multiple cells were obtained using a blood cell analyzer. The process iterates through each cell, and based on the cell type and the scatter coordinates, increments the gray value by 1 at the corresponding position in the grayscale image of the cell type to obtain four 256×256 cell density distribution maps for the four cell types. The elliptical outlines of the four cell density distribution maps were manually annotated to obtain four sets of elliptical parameters; The four cell density distribution maps are combined into four sets of four-channel images, which are used as training input images. The four sets of ellipse parameters are used as network output target values ​​to obtain training samples. The network includes a feature extraction network and an elliptic parameter detection head; The elliptical shape includes the ellipse center, inclination angle, major axis length, and minor axis length.

2. The method for detecting abnormal white blood cell scatter plots as described in claim 1, characterized in that, The anomaly analysis of the identification results to obtain anomaly results includes: Alarm thresholds for various cell parameters are manually set. When the center coordinates or elliptical shape of any cell in the abnormal result exceeds the alarm threshold, the grayscale scatter plot image is considered abnormal, and an abnormal result is obtained.

3. The method for detecting abnormal white blood cell scatter plots as described in claim 1, characterized in that, The anomaly analysis of the identification results to obtain anomaly results includes: Collect normal and abnormal samples from scatter plots; The normal samples and abnormal samples of the scatter plot are respectively input into the ellipse equation recognition model to obtain the output results; The output is then fed into an SVM network for training to obtain an anomaly scatter plot classifier. The identification results are input into the anomaly scatter plot classifier, which outputs the anomaly results.

Images