Blood cell classification method and related equipment

Abstract

The present application provides a blood cell classification method, in which the impedance pulse signal of the blood sample to be classified is input into a trained target neural network classification model. The target neural network classification model is trained by a large number of impedance pulse signal samples with classification result label information. It has a strong pulse morphology feature extraction capability, so that the classification result of the impedance pulse signal can be determined based on the morphological characteristics of the impedance pulse signal of the blood sample to be classified. Compared with the existing technology, the classification result is more accurate, and it can identify the impedance pulse signal caused by interference factors, and can perform identification and classification based on the impedance pulse signal, with higher real-time performance. In addition, the present application also provides a blood cell analyzer and a storage medium.

Description

Technical Field

[0001] The present application relates to the technical field of medical equipment, and more specifically, to a blood cell classification method and related equipment. Background Art

[0002] A hematology analyzer is a commonly used medical device that detects blood cell information, such as the type and number of each type of blood cell. Currently, both three- and five-differentiation hematology analyzers primarily use the pinhole impedance method to detect blood cells.

[0003] The main process of this detection method is to pass the blood cells to be tested one by one through an impedance detection channel shaped like a small hole. The channel is filled with a conductive diluent. Because blood cells are poor conductors, they cause a change in the resistance at both ends of the channel when passing through the small hole. The blood cell analyzer converts this resistance change into a pulse signal. The pulse recognition algorithm identifies the valid pulse signal and then analyzes the amplitude and pulse width of the valid pulse to determine the blood cell type in the blood to be tested. However, this method has low classification accuracy. Summary of the Invention

[0004] To this end, the present application provides a blood cell analysis method to improve the accuracy of blood cell classification. In order to achieve the above-mentioned invention objectives, the present application provides the following technical solutions:

[0005] In a first aspect, the present invention provides a cell classification method, comprising:

[0006] obtaining a target impedance pulse signal of a blood sample to be classified;

[0007] Obtaining a pre-trained target neural network classification model, wherein the target neural network classification model is obtained by training multiple groups of blood training samples using a machine learning algorithm, and each group of blood training samples includes an impedance pulse signal and label information for representing a classification result of the impedance pulse signal;

[0008] The target impedance pulse signal is input into the target neural network classification model to obtain a classification result of the target impedance pulse signal output by the target neural network classification model, wherein the classification result includes blood cell type or interference.

[0009] In a second aspect, an embodiment of the present application provides a blood analyzer, comprising:

[0010] a sample collection module, for obtaining a blood sample;

[0011] a detection module, configured to obtain a target impedance pulse signal of the blood sample;

[0012] a processor configured to obtain at least a pre-trained target neural network classification model, the target neural network classification model being obtained by training a plurality of groups of blood training samples using a machine learning algorithm, wherein each group of blood training samples includes an impedance pulse signal of blood cells and label information representing a classification result of the impedance pulse signal; and input the target impedance pulse signal into the target neural network classification model to obtain a classification result of the target impedance pulse signal output by the target neural network classification model, the classification result including a blood cell type or interference;

[0013] A human-computer interaction module is configured to output the classification result.

[0014] In a third aspect, an embodiment of the present application provides a readable storage medium having a computer program stored thereon. When the computer program is loaded and executed by a processor, the blood cell classification method as described in any one of the first aspects is implemented.

[0015] The blood cell classification method provided in the embodiments of the present application inputs the impedance pulse signal of the blood sample to be classified into a trained target neural network classification model. The target neural network classification model is trained from a large number of impedance pulse signal samples with classification result label information. It has a strong pulse morphology feature extraction capability, and can thus determine the classification result of the impedance pulse signal based on the morphological characteristics of the impedance pulse signal of the blood sample to be classified. Compared with the existing technology, the classification result is more accurate. In addition, the embodiments of the present application can identify impedance pulse signals caused by interference factors and can perform identification and classification based on impedance pulse signals, which is also more real-time. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 A schematic diagram of a flow chart of a blood cell classification method provided in an embodiment of the present application;

[0018] Figure 2 An example diagram of a blood cell classification method provided in an embodiment of the present application;

[0019] Figure 3a-3b Schematic diagrams of two application scenarios of the blood cell classification method provided in the embodiments of the present application;

[0020] Figure 4A schematic diagram of a process for training a target neural network classification model provided in an embodiment of the present application;

[0021] Figures 5a-5c Three examples of impedance detection signals of blood training samples provided in an embodiment of the present application;

[0022] Figure 6 A structural schematic diagram of a blood cell analyzer provided in an embodiment of the present application;

[0023] Figure 7 Another structural schematic diagram of the blood cell analyzer provided in an embodiment of the present application;

[0024] Figure 8 A structural schematic diagram of the control module of the blood cell analyzer provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0026] Human blood contains various cells, including red blood cells, white blood cells, and platelets. A blood cell analyzer can detect the types of blood cells present in a blood sample. Current blood cell analyzers, such as three-differentiation and five-differentiation blood cell analyzers, often use the pinhole impedance method to classify blood cell types. This method involves passing a blood sample through a small hole between two electrodes. Because blood cells are poor conductors, the passage of a blood cell causes a change in resistance across the two electrodes. The detection circuit converts the resistance signal into a voltage signal, generating a pulse signal. Each blood cell represents one pulse signal.

[0027] Currently, pulse recognition algorithms are used to identify pulse signals and determine their classification results. Specifically, valid pulse signals are first determined based on their rising and falling trends over a period of time. The amplitude and pulse width of these valid pulse signals are then calculated. Since different blood cell types have different amplitude and pulse width ranges, the blood cell type corresponding to the amplitude and pulse width of the valid pulse signal is determined within these ranges, thereby achieving blood cell classification.

[0028] After studying the above method, the inventors of this application found that the prior art has at least the following problems:

[0029] First, the identification of valid pulse signals depends solely on the rising and falling trends of the signals. However, the detection environment of the pinhole impedance method is relatively complex. Interference factors such as bubbles in the dilution flow in the pinhole channel and electrical signals received by the detection circuit are very likely to generate pulse signals. If the rising and falling trends of these pulse signals meet the identification criteria, they will be identified as valid pulse signals, which will affect the accuracy of the classification results.

[0030] Second, the classification criteria for valid pulse signals are artificially set to pulse amplitude and pulse width. However, these two pieces of information are incomplete, resulting in inaccurate pulse signal classification. Pulse signals generated by many interfering factors may also meet these classification criteria, making it difficult to distinguish interfering pulse signals from blood cell pulse signals. Furthermore, some blood cells in a critical state may not be accurately classified, resulting in low classification accuracy.

[0031] To address at least one of the aforementioned technical issues, the inventors of this application have proposed a blood cell classification method that uses a trained neural network model to classify blood cell samples. Because neural network models have strong feature extraction and analysis capabilities, they can produce more accurate classification results. The following describes the method of this application through specific implementations and accompanying drawings.

[0032] See Figure 1 , which shows a schematic process of an embodiment of a blood cell classification method, specifically including steps 101-103.

[0033] S101: Obtain a target impedance pulse signal of a blood sample to be classified.

[0034] The blood sample to be classified is processed by a reagent containing a hemolytic agent and then sent to the impedance channel of the blood analyzer. The impedance detection circuit of the impedance channel detects the blood sample to be classified according to the above-mentioned small hole impedance method to obtain an impedance pulse signal of the blood sample to be classified. A specific example of the impedance pulse signal can be seen in Figure 2 , the signal diagram includes multiple impedance pulse signals.

[0035] It should be noted that, to facilitate differentiation from other subsequent impedance pulse signals, the impedance pulse signal of the blood sample to be classified is referred to herein as the target impedance pulse signal. Furthermore, the impedance pulse signal is not limited to that obtained using the pinhole impedance method; any method capable of detecting the impedance pulse signal of a blood sample may be used.

[0036] Since impedance pulse signals may be generated by interference factors, and impedance pulse signals in different forms may correspond to the same or different types of blood cells, they need to be accurately identified by a neural network model.

[0037] S102: Obtain a pre-trained target neural network classification model, where the target neural network classification model is obtained by training multiple groups of blood training samples using a machine learning algorithm, and each group of blood training samples includes an impedance pulse signal of blood cells and label information for representing a classification result of the impedance pulse signal.

[0038] Among them, the blood training samples used for training include impedance pulse signals of blood cells and interference factors, and each impedance pulse signal has its own corresponding label information, and the label information is used to indicate the classification result of the impedance pulse signal, that is, what type of blood cells the impedance pulse signal is generated by. In a specific embodiment, the label information may include interference (Interference) and blood cell types such as red blood cells (red blood cell, referred to as RBC), platelets (Platelet, referred to as PLT), etc. It should be noted that if the label information is interference, it means that the impedance pulse signal is caused by interference factors. The target neural network classification model is pre-trained by the initial neural network model, and its specific training method can be found in the following description.

[0039] S103: Inputting the target impedance pulse signal into the target neural network classification model to obtain a classification result of the target impedance pulse signal output by the target neural network classification model, where the classification result includes blood cell type and interference.

[0040] The target impedance pulse signal is input into the target neural network classification model, which classifies and identifies the target impedance pulse signal to determine the classification result of each impedance pulse signal. It should be noted that there are multiple target impedance pulse signals, and the target neural network classification model identifies each target impedance pulse signal separately.

[0041] It is understood that the types of blood training sample label information will determine the types of classification results output by the trained target neural network classification model. In one embodiment, the classification results of the target impedance pulse signal include: interference and blood cell types such as red blood cells and platelets.

[0042] Interference can be caused by a variety of factors, such as interference from external electrical signals on the impedance detection circuit, failure of the impedance detection circuit's own components, microbial interference in the detection channel, impurities introduced into the blood sample to be classified during the sampling process, bubbles contained in the diluent flow in the detection channel, and so on. There are many interference factors, and the impedance pulse signals of interference factors caused by different reasons may manifest differently. Existing recognition algorithms may identify certain interference factors as blood cell types, thereby affecting the accuracy of blood cell classification. However, the blood training samples of the target neural network classification model take into account the existence of interference factors and can identify the impedance pulse signals caused by interference factors, thereby improving the accuracy of the classification results.

[0043] Furthermore, the impedance pulse signal has pulse morphology characteristics, and different types of impedance pulse signals have different pulse morphology characteristics. The target neural network classification model analyzes the pulse morphology characteristics of the target impedance pulse signal to obtain a classification result for the target impedance pulse signal. In one embodiment, the pulse morphology characteristics include any combination of the following: pulse position, pulse peak value, pulse width, pulse peak-to-width ratio, pulse rise time, pulse fall time, pulse line slope, pulse baseline information, total pulse area information, area difference before and after the pulse peak, and noise information.

[0044] Pulse position (Pos) indicates the position of the target impedance pulse signal in the complete detection signal of the blood sample to be classified. For example, the impedance detection signal is collected for 10s at a sampling rate of 1MHz to obtain 10e 6 sampling points, the peak value of a target impedance pulse signal is located at the 1e 6 The sampling point is 1e 6 / 10e 6=0.1. Pulse peak value (Peak) indicates the peak value of the target impedance pulse signal; taking the impedance pulse signal collected by the 12-bit impedance detection circuit as an example, the peak value is between 0 and 4095. Pulse width (Width) indicates the width of the target impedance pulse signal, which can be expressed in the number of sampling points. Pulse peak-to-width ratio (Ratio) indicates the ratio of the peak value of the target impedance pulse signal divided by the width. Pulse rise time (R-time) indicates the time information of the target impedance pulse signal from a steady rise to the peak value. Pulse fall time (D-time) indicates the time information of the target impedance pulse signal from a peak value to a steady state. Pulse line slope (K) indicates the slope of the peak value of the target impedance pulse signal. Pulse baseline information (Base) indicates the baseline size taken by the target impedance pulse signal during identification. Pulse total area information (Area) indicates the integral of the difference between the target impedance pulse signal and the baseline. The area difference before and after the pulse peak (AreaDiff) represents the difference between the area of the first half of the pulse at the peak position of the target impedance pulse signal and the area of the second half of the pulse, as a ratio to the total area. Noise information (Noise) represents the effective value of the noise in the data segment without the target impedance pulse signal. It should be noted that the above pulse morphology features are only examples, and the target neural network classification model can use other forms of features as classification criteria, which are not specifically limited in the embodiments of this application.

[0045] The target neural network classification model includes not only the types of pulse morphology features to be extracted, but also the weights for each pulse morphology feature and a number of classification thresholds equal to the number of classification results (each classification threshold represents a classification result). It should be noted that these model parameters are determined during the training process of the target neural network classification model. After the training process is completed, the model parameters of the target neural network classification model are finalized. The model parameters can be used to construct a mathematical model for classification. The mathematical model represents the correlation between the pulse morphology features and the final classification results.

[0046] In a specific embodiment, the target neural network classification model implements classification by executing the following steps: extracting multiple pulse morphology features from the target impedance pulse signal; performing weighted calculation on each pulse morphology feature and the weight corresponding to the pulse morphology feature to obtain a eigenvalue weighted result; comparing the eigenvalue weighted result with a preset classification threshold (i.e., the classification threshold obtained after training is completed) to determine the classification threshold corresponding to the eigenvalue weighted result. For ease of description, the classification threshold can be referred to as the target classification threshold; identifying the classification result of the target classification threshold as the classification result of the target impedance pulse signal. It should be noted that each classification threshold can be a numerical range. After obtaining the eigenvalue weighted result, it is determined to which numerical range the eigenvalue weighted result corresponds. The numerical range is the target classification threshold. Since each classification threshold represents a classification result, the classification result of the target impedance pulse signal is determined based on the classification result corresponding to the target classification threshold. For example, the classification threshold [a1-a2] represents red blood cells, the classification threshold [a3-a4] represents platelets, and the classification threshold [a5-a6] represents interference; assuming that the target classification threshold corresponding to the target impedance pulse signal is [a3-a4], it can be determined that the classification result of the target impedance pulse signal is platelets.

[0047] It can be understood that the target neural network classification model is a neural network model. Its advantage lies in that it can mine the morphological features of the impedance pulse signal corresponding to each classification result from a large number of blood training samples, and find the optimal morphological features for distinguishing different classification results. Compared with the existing recognition algorithms that only use pulse amplitude and width for distinction, the morphological features are richer and more accurate, so that it can accurately locate and even amplify the differences between blood cells and blood cells, and between blood cells and interference factors, and thus the classification results obtained based on the morphological features are more accurate.

[0048] Furthermore, in the embodiment of the present application, the signal input into the target neural network classification model is an impedance pulse signal, that is, after the blood to be classified is sent into the cell analyzer and the impedance detection circuit obtains the impedance detection signal of the blood to be classified, the impedance detection signal can be input into the target neural network classification model in real time, and the model identifies and classifies the impedance pulse signal in the impedance detection signal. However, the existing recognition algorithm first needs to obtain a complete impedance detection signal, and then identify the effective pulse signal from the complete impedance detection signal, and then generate a distribution histogram based on the amplitude of the effective pulse signal. Only by analyzing the distribution histogram can the type of blood cells corresponding to the effective pulse signal be determined. Compared with the prior art, the blood cell classification method of the embodiment of the present application does not need to obtain a complete impedance detection signal. During the processing of a blood sample, when a certain impedance pulse signal is detected, the type of impedance pulse signal generated before the impedance pulse signal is also identified, which is more real-time.

[0049] As can be seen from the above technical solutions, the blood cell classification method provided by the embodiment of the present application inputs the impedance pulse signal of the blood sample to be classified into the trained target neural network classification model. The target neural network classification model is trained by a large number of impedance pulse signal samples with classification result label information. It has a strong pulse morphology feature extraction capability, so that the classification result of the impedance pulse signal can be determined based on the morphological characteristics of the impedance pulse signal of the blood sample to be classified. Since the pulse morphology features are richer and more comprehensive, the classification result is more accurate than the existing technology. In addition, the embodiment of the present application can identify the impedance pulse signal caused by interference factors, and can perform identification and classification based on the impedance pulse signal, and the real-time performance is also higher.

[0050] See Figure 2 , which shows a schematic diagram of an application process of the cell classification method. As shown in Figure 2, assuming that part of the impedance detection signal of the blood sample to be classified includes multiple target impedance pulse signals, the part of the impedance detection signal is input into the target neural network classification model for identification. The target neural network classification model can extract the pulse morphology feature x of the target impedance pulse signal from it. i It should be noted that the pulse shape characteristics can be multiple, although only one x is shown in the figure i , but it does not mean that there is only one pulse morphology feature, it is just an example of a feature. The target neural network classification model calculates the pulse morphology feature according to the classification method described above and obtains the corresponding classification result y i And output. Figure 2 As shown, some target impedance pulse signals in the impedance detection signal are marked as blood cells, some target impedance pulse signals are marked as platelets, and some target impedance pulse signals are marked as interference.

[0051] In practical applications, the classification results can be further used in various scenarios related to blood cell processing. Two specific application scenarios are provided below as examples for illustration.

[0052] Scenario 1: Provides a prompt for the interference situation based on the classification results. Figure 3a As shown, after the target neural network classification model obtains the classification result, it can perform statistics on the target impedance pulse signal of the interference type, and output corresponding prompt information based on the statistical related information.

[0053] Specifically, if the classification result of the target impedance pulse signal includes interference, relevant information of the target impedance pulse signal of the interference type is collected; the relevant information includes distribution information and / or morphological information; and prompt information corresponding to the relevant information is output based on the relevant information.

[0054] As mentioned above, the classification results obtained in the embodiments of the present application may include interference. In this case, when the classification results indicate that interference actually exists, relevant information about the interference can be counted. For example, the distribution information of the interference in the complete impedance detection signal of the blood sample to be classified, such as the number and location, etc., can be counted; or the morphological information of the target impedance pulse signal corresponding to the interference, such as the amplitude, etc., can also be counted. Based on the relevant information counted, a prompt can be given for the interference situation. For example, it can be determined whether the distribution position of the interference meets the preset distance standard. If it does, it indicates whether the interference distribution is too dense. For example, it can be determined whether the number of interferences exceeds the preset alarm number threshold. If so, it indicates that there is too much interference. If the interference is too much or too dense, it means that there are interference factors in the detection environment of the blood sample to be classified, and relevant prompt information can be output to prompt medical staff to perform interference detection on the detection environment of the blood cell analysis instrument used for detection to eliminate the relevant interference factors.

[0055] Scenario 2: Count different types of blood cells based on the classification results. Figure 3b As shown, after the target neural network classification model obtains the classification result, each type of blood cells is counted, so that the measurement value of each type of blood cells can be given in the blood cell analysis report.

[0056] Specifically, in addition to interference, embodiments of the present application can identify different blood cell types. If the classification result of the target impedance pulse signal includes blood cell type, the number of target impedance pulse signals of the same blood cell type in the blood to be classified is counted; the number of target impedance pulse signals is determined as the number of blood cells of the same blood cell type. For example, the number of red blood cells is A, the number of platelets is B, and so on.

[0057] It should be noted that the above two application scenarios are merely examples, and the blood cell classification results can also be applied to any other blood cell processing scenarios that can be expected by those skilled in the art.

[0058] The above describes how to use the target neural network classification model. The following describes the training process of the target neural network classification model in detail.

[0059] like Figure 4 As shown, the training steps of the target neural network classification model include S401-S403.

[0060] S401: Obtain multiple groups of blood training samples, each group of blood training samples includes an impedance pulse signal and label information for representing a classification result of the impedance pulse signal, where the classification result includes a blood cell type or interference.

[0061] To further refine the trained target neural network classification model, a large number of blood training samples can be collected in advance to form a learning sample library. Similar to step S201 in the aforementioned embodiment, each set of blood training samples includes label information for the impedance pulse signal and the classification result. It should be noted that the label information can be manually annotated or automatically identified by the algorithm.

[0062] Exemplarily, the impedance pulse signals of the blood training samples include three categories: impedance pulse signals generated by blood samples in the absence of interference factors (for the sake of convenience of description, referred to as first-category pulse signal samples), impedance pulse signals generated by interference factors (for the sake of convenience of description, referred to as second-category pulse signal samples), and impedance pulse signals generated by blood samples with interference factors (for the sake of convenience of description, referred to as third-category pulse signal samples).

[0063] like Figure 5a-5c The following are examples of pulse signal samples of the first, second, and third categories, respectively. Each type of pulse signal sample is labeled, which is the label information of the classification result of the impedance pulse signal. For the first type of pulse signal samples, the existing pulse recognition and blood cell classification algorithm can be used to locate the position of the impedance pulse signal and calculate the classification result of the impedance pulse signal. For the second type of pulse signal samples, the existing pulse recognition and blood cell classification algorithm can be used to directly mark the impedance pulse signal as an interference type after locating it. For the third type of pulse signal samples, the classification result of the impedance pulse signal can be manually labeled.

[0064] S402: Obtain an initial neural network classification model.

[0065] The initial neural network model can be a neural network model with classification capabilities, such as a BP (backpropagation) network or an LSTM (Long-Short Term Memory) network. The model parameters of the initial neural network classification model are determined through the training step of step S403. It should be noted that this step is not limited to being performed after step S401 and can also be performed before or simultaneously therewith.

[0066] S403: Input multiple groups of blood training samples into the initial neural network classification model to obtain label information output by the initial neural network classification model. When the relationship between the output label information and the label information of the blood training samples meets the convergence condition, the training is stopped to obtain the target neural network classification model.

[0067] The blood training samples can be input into the initial neural network classification model in batches, and the model can extract the characteristic data of the blood training samples. For example, a pulse waveform is formed by intercepting several sampling points near the position of the impedance pulse signal of the blood training sample. The characteristic data of the pulse waveform is extracted, including but not limited to position information Pos, type information Type, pulse peak Peak, pulse width Width, pulse baseline information Base, total pulse area information Area, area difference before and after the pulse peak AreaDiff, and noise information Noise. The characteristic data Din(i) of any pulse waveform i can be recorded as [Pos(i), Type(i), Peak(i), Width(i), Base(i), Area(i), AreaDiff(i), Noise(i)].

[0068] After predicting and outputting label information for a batch of blood training samples based on the feature data, the gap between the predicted label information and the labeled label information is compared. Based on the gap, the model parameters of the initial neural network classification model are adjusted. The next batch of blood training samples is then input and the training process is repeated until the initial neural network classification model converges, thereby obtaining the target neural network classification model. For example, in the training process of a BP (back propagation) network, the learning algorithm uses the steepest descent back propagation (SDBP) algorithm, calculating the output of each neuron from the first layer of the network backward and calculating the impact of the model parameter values on the total error from the last layer forward. The network model parameters are continuously adjusted through the back propagation algorithm to minimize the sum of squares of the network's total error and achieve network convergence. It should be noted that the learning algorithm can also be other, such as LM (Leverberg-Marquardt) algorithm, Momentum Backpropagation (MOBP) algorithm, Variable Learning rate Backpropagation (VLBP) algorithm, Resilient Backpropagation (RPROP) algorithm, variable gradient algorithm, quasi-Newton algorithm, etc.; the error function used for adjustment can be the mean square error function, etc.; the number of hidden layer nodes is set to 50; the maximum number of training times is set to 10000, the target error is set to 0.001; the minimum gradient 1e -6 .

[0069] The training method provided by the above embodiment can obtain a target neural network classification model for classification, which can identify interference in blood samples, thereby avoiding the influence of interference on blood cell type recognition and improving the accuracy of blood cell type recognition results.

[0070] The above describes relevant embodiments of the blood cell classification method. To ensure the application and implementation of the method in practice, the present application provides the following embodiments of blood cell classification devices.

[0071] In a specific embodiment, the blood cell classification device can be a processing device that has a pre-trained target neural network classification model pre-installed, and the target impedance pulse signal detected by the blood analyzer is input into the processing device. The processing device includes a processor, which is configured to obtain a target neural network classification model, wherein the target neural network classification model is obtained by training multiple groups of blood training samples using a machine learning algorithm, and each group of blood training samples includes an impedance pulse signal of blood cells and label information for representing the classification result of the impedance pulse signal; and input the target impedance pulse signal into the target neural network classification model to obtain a classification result of the target impedance pulse signal output by the target neural network classification model, wherein the classification result includes blood cell type or interference. The processing device also includes a human-computer interaction module, which is configured to output the classification result.

[0072] In another specific embodiment, the blood cell classification device may be a blood analyzer, which can detect a blood sample to be classified to obtain a target impedance pulse signal, and can classify the target impedance pulse signal to obtain and output a classification result.

[0073] This application provides an embodiment of a blood analyzer, see Figure 6 , specifically including: a sample collection module 601, a detection module 602, a processor 603 and a human-computer interaction module 604:

[0074] The sample collection module 601 is used to obtain a blood sample;

[0075] A detection module 602 is used to obtain a target impedance pulse signal of a blood sample to be classified;

[0076] Processor 603 is configured to at least obtain a pre-trained target neural network classification model, the target neural network classification model being obtained by training a plurality of groups of blood training samples using a machine learning algorithm, wherein each group of blood training samples includes an impedance pulse signal of blood cells and label information representing a classification result of the impedance pulse signal; and input the target impedance pulse signal into the target neural network classification model to obtain a classification result of the target impedance pulse signal output by the target neural network classification model, wherein the classification result includes a blood cell type or interference;

[0077] The human-computer interaction module 604 is configured to output the classification result.

[0078] In a specific implementation, the target neural network classification model obtains the classification result of the target impedance pulse signal by analyzing the pulse morphology characteristics of the target impedance pulse signal.

[0079] In this specific implementation, the pulse morphology characteristics include any combination of the following: pulse position, pulse peak, pulse width, pulse peak-to-width ratio, pulse rise time, pulse fall time, pulse line slope, pulse baseline information, pulse total area information, area difference before and after the pulse peak, and noise information.

[0080] In a specific implementation, when the target impedance pulse signal is input into the target neural network classification model to obtain a classification result of the target impedance pulse signal output by the target neural network classification model, the processor 603 is specifically configured to:

[0081] The target impedance pulse signal is input into the target neural network classification model so that the target neural network classification model performs the following classification steps: extracting multiple pulse morphology features from the target impedance pulse signal; performing weighted calculation on each pulse morphology feature and the weight corresponding to the pulse morphology feature to obtain a characteristic value weighted result; comparing the characteristic value weighted result with a preset classification threshold to determine the corresponding target classification threshold; and identifying the classification result of the target classification threshold as the classification result of the target impedance pulse signal.

[0082] In a specific implementation, the processor 603 is also configured to: if the classification result of the target impedance pulse signal includes interference, then statistically analyze the relevant information of the target impedance pulse signal of the interference type; the relevant information includes distribution information and / or morphological information; based on the relevant information, generate prompt information corresponding to the relevant information; the human-computer interaction module is also used to output the prompt information.

[0083] In a specific implementation, the processor 603 is further configured to: if the classification result of the target impedance pulse signal includes the blood cell type, count the number of target impedance pulse signals of the same blood cell type; and determine the number of target impedance pulse signals as the number of blood cells of the same blood cell type.

[0084] In this specific implementation, the blood cell types include: red blood cells or platelets.

[0085] In a specific implementation, the processor 603 is further configured to: obtain multiple groups of blood training samples, each group of blood training samples includes an impedance pulse signal and label information for representing a classification result of the impedance pulse signal, the classification result includes a blood cell type or interference; obtain an initial neural network classification model; input the multiple groups of blood training samples into the initial neural network classification model to obtain label information output by the initial neural network classification model, and when the relationship between the output label information and the label information of the blood training samples meets a convergence condition, stop training to obtain a target neural network classification model.

[0086] In this specific implementation, the impedance pulse signal of the blood training sample includes: an impedance pulse signal generated by the blood sample without interference factors, an impedance pulse signal generated by interference factors, and an impedance pulse signal generated by the blood sample with interference factors.

[0087] In addition, the present application also provides a readable storage medium on which a computer program is stored, characterized in that when the computer program is loaded and executed by a processor, the steps in any of the above-mentioned blood cell classification method embodiments are implemented.

[0088] See Figure 7 , the present application embodiment also provides a structural diagram of a blood cell analyzer. Figure 7 As shown, the blood analyzer 700 includes at least a sampling device 710 , a sample preparation device 720 , a detection device 730 , a control device 740 and a display device 750 .

[0089] The sampling device 710 has a pipette (e.g., a sampling needle) with a pipette nozzle and a driving unit for driving the pipette to quantitatively absorb the blood sample to be tested through the pipette nozzle. For example, the sampling needle is driven by the driving unit to move to absorb the blood sample to be tested from a sample container containing the blood sample.

[0090] The sample preparation device 720 includes at least one reaction cell and a reagent supply device (not shown). The at least one reaction cell is used to receive the blood sample to be tested drawn by the sampling device 710, and the reagent supply device provides a processing reagent to the at least one reaction cell. The blood sample to be tested drawn by the sampling device 710 and the processing reagent provided by the reagent supply device are mixed in the reaction cell to prepare a sample solution to be tested.

[0091] The detection device 730 is used to detect the sample liquid prepared by the sample preparation device 720 to obtain blood cell-related data. In some embodiments, the detection device 730 includes an impedance detection unit, and the blood cell-related data includes a target impedance pulse signal. The target impedance pulse signal can be amplified by an amplifier and transmitted to the control device 740. The control device 740 processes the target impedance pulse signal to obtain a classification result of the sample liquid, such as blood cell type (red blood cells or platelets) or interference.

[0092] like Figure 8 As shown, the control device 740 includes at least a processing component 741, RAM 742, ROM 743, a communication interface 744, a memory 746, and an I / O interface 745. The processing component 741, RAM 742, ROM 743, the communication interface 744, the memory 746, and the I / O interface 745 communicate via a bus 747. The processing component can be a CPU, a GPU, or other chip with computing capabilities. The memory 746 contains various computer programs such as an operating system and application programs for execution by the processor component 741, as well as the data required to execute the computer programs. In addition, during the blood sample analysis process, any data that needs to be stored locally can be stored in the memory 746. The I / O interface 745 is composed of serial interfaces such as USB, IEEE1394, or RS-232C, parallel interfaces such as SCSI, IDE, or IEEE1284, and analog signal interfaces composed of D / A converters and A / D converters. The I / O interface 745 is connected to an input device such as a keyboard, mouse, touch screen, or other control buttons, allowing the user to directly input data into the control device 740 using the input device. Furthermore, the I / O interface 745 may also be connected to a display device 740 having a display function, such as an LCD screen, a touch screen, or an LED display. The control device 740 can output processed data as image display data to the display device 740 for display, such as classification result data, prompt information, etc. The communication interface 744 is an interface that can use any currently known communication protocol. The communication interface 744 communicates with the outside world via a network. The control device 740 can transmit data to any device connected to the network using a specific communication protocol via the communication interface 744.

[0093] The control device 740 includes a processor and a storage medium storing a computer program. The control device 740 is configured to perform the following steps when the computer program is executed by the processor: obtaining a pre-trained target neural network classification model, wherein the target neural network classification model is obtained by training multiple groups of blood training samples using a machine learning algorithm, and each group of blood training samples includes an impedance pulse signal of blood cells and label information for representing a classification result of the impedance pulse signal; and inputting the target impedance pulse signal into the target neural network classification model to obtain a classification result of the target impedance pulse signal output by the target neural network classification model, wherein the classification result includes a blood cell type or interference.

[0094] In some embodiments, the target neural network classification model obtains a classification result of the target impedance pulse signal by analyzing pulse morphology features of the target impedance pulse signal. Pulse morphology features include any combination of the following: pulse position, pulse peak value, pulse width, pulse peak-to-width ratio, pulse rise time, pulse fall time, pulse line slope, pulse baseline information, total pulse area information, area difference before and after the pulse peak, and noise information.

[0095] In some embodiments, the control device 740 is configured so that when the computer program is executed by the processor, the control device 740 can further perform the following steps: if the classification result of the target impedance pulse signal includes interference, then statistically analyzing relevant information of the target impedance pulse signal of the interference type; the relevant information includes distribution information and / or morphological information; and based on the relevant information, generating prompt information corresponding to the relevant information. In some embodiments, the control device 740 is configured so that when the computer program is executed by the processor, the control device 740 can further perform the following steps: if the classification result of the target impedance pulse signal includes blood cell type, then statistically analyzing the number of target impedance pulse signals of the same blood cell type; and determining the number of target impedance pulse signals as the number of blood cells of the same blood cell type.

[0096] In some embodiments, the control device 740 is configured to, when the computer program is executed by the processor, further perform the following steps: obtain multiple groups of blood training samples, each group of the blood training samples including an impedance pulse signal and label information for representing the classification result of the impedance pulse signal, the classification result including blood cell type or interference; obtain an initial neural network classification model; input the multiple groups of the blood training samples into the initial neural network classification model to obtain label information output by the initial neural network classification model, and when the relationship between the output label information and the label information of the blood training samples meets the convergence condition, stop training to obtain the target neural network classification model.

[0097] In the above embodiment, the impedance pulse signal of the blood training sample includes: an impedance pulse signal generated by the blood sample without interference factors, an impedance pulse signal generated by interference factors, and an impedance pulse signal generated by the blood sample with interference factors.

[0098] The display device 750 is used to display the classification results obtained by the control device 740. For example, the display device 750 is configured as a user interface. In some embodiments, the display device 750 is also used to display prompt information obtained by the control device 740.

[0099] This document is described with reference to various exemplary embodiments. However, those skilled in the art will recognize that changes and modifications may be made to the exemplary embodiments without departing from the scope of this document. For example, the various operational steps and components used to perform the operational steps may be implemented in different ways (e.g., one or more steps may be deleted, modified, or incorporated into other steps) depending on the specific application or considering any number of cost functions associated with the operation of the system.

[0100] The terms "first," "second," and so on, used in the specification and claims herein and in the accompanying drawings are used to distinguish between different items, not to describe a particular order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, or apparatus.

[0101] Additionally, as will be appreciated by those skilled in the art, the principles of this disclosure may be embodied in a computer program product on a computer-readable storage medium pre-loaded with computer-readable program code. Any tangible, non-transitory computer-readable storage medium may be used, including magnetic storage devices (hard disks, floppy disks, etc.), optical storage devices (CD-ROMs, DVDs, Blu-ray discs, etc.), flash memory, and / or the like. These computer program instructions may be loaded onto a general-purpose computer, a special-purpose computer, or other programmable data processing device to form a machine, such that the instructions executed on the computer or other programmable data processing device can generate a device that implements a specified function. These computer program instructions may also be stored in a computer-readable memory, which can instruct the computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory can form an article of manufacture that includes an implementation device that implements the specified function. The computer program instructions may also be loaded onto a computer or other programmable data processing device, causing the computer or other programmable device to execute a series of operational steps to produce a computer-implemented process, such that the instructions executed on the computer or other programmable device provide the steps for implementing the specified function.

[0102] The foregoing detailed description has been described with reference to various embodiments. However, those skilled in the art will recognize that various modifications and changes can be made without departing from the scope of this disclosure. Therefore, the present disclosure will be considered in an illustrative rather than a restrictive sense, and all such modifications will be included within its scope. Similarly, the advantages, other advantages and solutions to the problems of the various embodiments have been described above. However, the benefits, advantages, solutions to the problems and any elements that can produce these, or make them more specific, should not be interpreted as critical, required or necessary. The term "comprising" and any other variants used in this article are all non-exclusive inclusions, so that a process, method, article or device that includes a list of elements includes not only these elements, but also other elements that are not explicitly listed or do not belong to the process, method, system, article or device. In addition, the term "coupled" and any other variants used in this article refer to physical connections, electrical connections, magnetic connections, optical connections, communication connections, functional connections and / or any other connections.

[0103] The above embodiments merely illustrate several implementation methods, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the spirit of the present invention, and these modifications and improvements fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A blood cell classification method, characterized in that: include: obtaining a plurality of target impedance pulse signals of a blood sample to be classified; Obtaining a pre-trained target neural network classification model, wherein the target neural network classification model is obtained by training multiple groups of blood training samples using a machine learning algorithm, and each group of blood training samples includes an impedance pulse signal and label information for representing a classification result of the impedance pulse signal, wherein the label information of the classification result includes label information corresponding to a blood cell type and label information corresponding to interference; The target impedance pulse signal is input into the target neural network classification model to obtain a classification result for each target impedance pulse signal output by the target neural network classification model, wherein the classification result includes blood cell type or interference.

2. The blood cell classification method according to claim 1, characterized in that: The target neural network classification model obtains the classification result of the target impedance pulse signal by analyzing the pulse morphology characteristics of the target impedance pulse signal.

3. The blood cell classification method according to claim 2, characterized in that: The pulse morphology features include any combination of the following: pulse position, pulse peak, pulse width, pulse peak width ratio, pulse rise time, pulse fall time, pulse line slope, pulse baseline information, pulse total area information, area difference before and after the pulse peak, and noise information.

4. The blood cell classification method according to claim 3, characterized in that: Inputting the target impedance pulse signal into the target neural network classification model to obtain a classification result of the target impedance pulse signal output by the target neural network classification model includes: The target impedance pulse signal is input into the target neural network classification model so that the target neural network classification model performs the following classification steps: extracting a plurality of pulse morphology features from the target impedance pulse signal; Performing weighted calculation on each pulse morphology feature and the weight corresponding to the pulse morphology feature to obtain a eigenvalue weighted result; Comparing the weighted result of the feature value with a preset classification threshold to determine the corresponding target classification threshold; A classification result of the target classification threshold is identified as a classification result of the target impedance pulse signal.

5. The blood cell classification method according to any one of claims 1 to 4, characterized in that: Also includes: If the classification result of the target impedance pulse signal includes interference, collecting relevant information of the target impedance pulse signal of the interference type; The relevant information includes distribution information and / or morphological information; According to the relevant information, prompt information corresponding to the relevant information is output.

6. The blood cell classification method according to any one of claims 1 to 4, characterized in that: Also includes: If the classification result of the target impedance pulse signal includes the blood cell type, counting the number of target impedance pulse signals of the same blood cell type; The number of the target impedance pulse signals is determined as the number of blood cells of the same blood cell type.

7. The blood cell classification method according to claim 1, characterized in that: The blood cell types include: red blood cells or platelets.

8. The blood cell classification method according to claim 1, wherein: The training steps of the target neural network classification model include: Obtaining multiple groups of blood training samples, each group of blood training samples includes an impedance pulse signal and label information for representing a classification result of the impedance pulse signal, wherein the classification result includes a blood cell type or interference; Obtain an initial neural network classification model; Multiple groups of blood training samples are input into the initial neural network classification model to obtain label information output by the initial neural network classification model. When the relationship between the output label information and the label information of the blood training samples meets the convergence condition, the training is stopped to obtain the target neural network classification model.

9. The blood cell classification method according to claim 8, characterized in that: The impedance pulse signal of the blood training sample includes: an impedance pulse signal generated by the blood sample without interference factors, an impedance pulse signal generated by interference factors, and an impedance pulse signal generated by the blood sample with interference factors.

10. A blood analyzer, characterized in that: include: a sample collection module, for obtaining a blood sample; a detection module, configured to obtain a plurality of target impedance pulse signals of the blood sample; A processor configured to at least obtain a pre-trained target neural network classification model, wherein the target neural network classification model is obtained by training multiple groups of blood training samples using a machine learning algorithm, and each group of blood training samples includes an impedance pulse signal of blood cells and label information for representing a classification result of the impedance pulse signal, wherein the label information of the classification result includes label information corresponding to the blood cell type and label information corresponding to interference; and input the target impedance pulse signal into the target neural network classification model to obtain a classification result for each target impedance pulse signal output by the target neural network classification model, wherein the classification result includes a blood cell type or interference; A human-computer interaction module is configured to output the classification result.

11. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is loaded and executed by a processor, the blood cell classification method according to any one of claims 1 to 9 is implemented.

Images