Sample analyzer, sample analysis method and sample analysis system

Abstract

The present application relates to a sample analyzer, a sample analysis method and a sample analysis system, including: obtaining the original data corresponding to the sample, the original data including at least one of the original light signal data and the original scatter plot; the original light signal data including at least two paths of light signal data with different angles; inputting the original data into a conversion model, the conversion model including N conversion layers connected in sequence, each conversion layer being used to convert the light signal data according to the corresponding conversion function to obtain converted light signal data; wherein the input of the first conversion layer of the conversion model is the original light signal data, and the input of the Mth conversion layer of the conversion model is the light signal data converted by the M-1th conversion layer; N is greater than or equal to 2, and the maximum value of M is N; generating a converted scatter plot based on the converted light signal data of the Nth conversion layer of the conversion model, and identifying the scatter plot to obtain a particle classification result. This method can improve the accuracy of particle classification.

Description

Technical Field

[0001] The present application relates to the technical field of in vitro diagnosis, and in particular to a sample analyzer, a sample analysis method and a sample analysis system. Background Art

[0002] When performing particle classification detection on samples or analyzing cell particles in samples, the raw light signal data output by the optical detection component usually has a high degree of fusion, and the particle clusters are relatively concentrated and cannot be clearly distinguished. The direct presentation of the raw light signal data does not utilize particle classification, and it is difficult for doctors to analyze and judge when presented.

[0003] In the prior art, a preset logarithmic function is usually used for signal conversion, and the parameter value of the logarithmic function is set manually based on experience according to the scatter plot output by the instrument. However, there are differences in the cell particles contained in various samples. The current conversion method cannot meet the conversion needs of various samples, resulting in unsatisfactory conversion results and low classification accuracy. Summary of the invention

[0004] Based on this, it is necessary to provide a sample analyzer, a sample analysis method and a sample analysis system that can improve the accuracy of particle classification in response to the above technical problems.

[0005] In a first aspect, the present application provides a sample analyzer, comprising:

[0006] A dispensing component, used for dispensing samples and reagents to be tested into the reaction container;

[0007] A reaction component, used for reacting the sample and reagent in the reaction container to obtain a liquid to be tested;

[0008] An optical detection component, used for obtaining raw data of the sample when the liquid to be tested flows through a detection area of ​​the optical detection component;

[0009] A controller, used to obtain raw data corresponding to a sample; the raw data includes at least one of raw light signal data and raw scatter plot; the raw light signal data includes at least two paths of light signal data with different angles; the raw scatter plot is generated based on the raw light signal data; the raw data is input into a conversion model, the conversion model includes N conversion layers connected in sequence, each of the conversion layers corresponds to a different conversion function, and each of the conversion layers is used to convert the light signal data according to the corresponding conversion function to obtain converted light signal data; wherein the input of the first conversion layer of the conversion model is the raw light signal data, and the input of the Mth conversion layer of the conversion model is the light signal data converted by the M-1th conversion layer; N is greater than or equal to 2, and the maximum value of M is N; based on the converted light signal data of the Nth conversion layer of the conversion model, a converted scatter plot is generated, and the scatter plot is identified to obtain a particle classification result;

[0010] A display is used to display the converted scatter plot.

[0011] In a second aspect, the present application provides a sample analysis method, the method comprising:

[0012] Acquire original data corresponding to the sample, the original data including at least one of original light signal data and an original scatter plot; the original light signal data includes at least two paths of light signal data with different angles; the original scatter plot is generated based on the original light signal data;

[0013] The original data is input into a conversion model, wherein the conversion model includes N conversion layers connected in sequence, each of the conversion layers corresponds to a different conversion function, and each of the conversion layers is used to convert the optical signal data according to the corresponding conversion function to obtain converted optical signal data; wherein the input of the first conversion layer of the conversion model is the original optical signal data, and the input of the Mth conversion layer of the conversion model is the optical signal data converted by the M-1th conversion layer; wherein N is greater than or equal to 2, and the maximum value of M is N;

[0014] A converted scatter plot is generated according to the converted light signal data of the Nth conversion layer of the conversion model, and the scatter plot is identified to obtain a particle classification result.

[0015] In one embodiment, the training process of the conversion model includes:

[0016] Constructing an initial conversion model; the initial conversion model includes N conversion layers connected in sequence, and each conversion layer corresponds to a different conversion function;

[0017] The initial conversion model is trained using a sample set. During the training process, the parameters of the conversion function of each conversion layer in the initial conversion model are adjusted based on back propagation. When the initial conversion model meets the iteration end condition, the conversion model is obtained. The sample set includes a sample scatter plot and annotated particle classification results of the sample scatter plot.

[0018] In one embodiment, the initial conversion model is trained using the sample set, and the parameters of the conversion function of each conversion layer in the initial conversion model are adjusted based on back propagation during the training process, and when the initial conversion model meets the iteration end condition, the conversion model is obtained, including:

[0019] Inputting the original optical signal data of the sample scatter plot in the sample set into the initial conversion model to obtain the converted optical signal data of each sample particle output by the Nth conversion layer of the initial conversion model;

[0020] Determine the converted light signal data of the sample particles of each category according to the particle classification result of the sample scatter plot and the converted light signal data of each sample particle;

[0021] Determine the sum of first distances between sample particles of the same category and the sum of second distances between sample particles of different categories based on the converted light signal data of the sample particles of each category;

[0022] Using the difference between the sum of the first distances and the sum of the second distances as the loss value of the initial conversion model;

[0023] During the training process, the loss value is back-propagated to the initial conversion model, and the parameters of the conversion function of each conversion layer in the initial conversion model are adjusted until the initial conversion model meets the iteration end condition to obtain a conversion model.

[0024] In one embodiment, the number of the initial conversion models is multiple; the conversion function corresponding to at least one conversion layer in each of the initial conversion models is different, and / or the number of conversion layers is different;

[0025] When the initial conversion model satisfies the iteration end condition, obtaining the conversion model includes:

[0026] When the number of training times of each of the initial conversion models reaches a preset number of times, and / or the loss value of the initial conversion model tends to be stable, a plurality of candidate conversion models are obtained;

[0027] Determine a loss value corresponding to each of the candidate conversion models;

[0028] The candidate conversion model with the smallest loss value is determined as the conversion model.

[0029] In one embodiment, the initial conversion model is trained using the sample set, and the parameters of the conversion function of each conversion layer in the initial conversion model are adjusted based on back propagation during the training process, including:

[0030] The initial conversion model includes two conversion layers connected in sequence, and the conversion functions corresponding to the two conversion layers connected in sequence are respectively a Hyperlog function and a Logical function;

[0031] Inputting the original light signal data of the sample scatter plot in the sample set into the initial conversion model to obtain the converted light signal data of eosinophils, neutrophils, lymphocytes, and monocytes output by the second conversion layer of the initial conversion model;

[0032] Calculating a first loss value based on the converted optical signal data of the eosinophils and the neutrophils;

[0033] Calculating a second loss value based on the converted optical signal data of the lymphocytes and the monocytes;

[0034] The first loss value is back-propagated to the conversion layer where the Logical function is located in the initial conversion model, and the second loss value is back-propagated to the conversion layer where the Hyperlog function is located in the initial conversion model, and the parameters of the Logical function and the parameters of the Hyperlog function are adjusted until the initial conversion model meets the iteration end condition to obtain the conversion model.

[0035] In one embodiment, when the initial conversion model satisfies the iteration end condition, obtaining the conversion model includes:

[0036] When the initial conversion model satisfies an iteration termination condition and the loss value of the initial conversion model is greater than a first threshold, determining the loss value corresponding to the initial conversion model according to the optical signal data after conversion of sample particles of at least two target categories;

[0037] The corresponding loss value is back-propagated to the target conversion layer corresponding to the target category in the initial conversion model, and the parameters of the conversion function in the target conversion layer are adjusted until the initial conversion model meets the iteration end condition and the loss value of the initial conversion model is less than the first threshold, so as to obtain a conversion model.

[0038] In one of the embodiments, the original light signal data includes front scattered pulse signal data, side scattered pulse signal data and fluorescence pulse signal data;

[0039] Before the step of inputting the original data into the conversion model, the method further comprises:

[0040] Acquire a second threshold corresponding to the front scattered pulse signal data, a third threshold corresponding to the side scattered pulse signal data, and a fourth threshold corresponding to the fluorescent pulse signal data; when the front scattered pulse signal data is less than or equal to the second threshold, adjust the front scattered pulse signal data to a first preset value;

[0041] When the side scattered pulse signal data is less than or equal to the third threshold, adjusting the front scattered pulse signal data to a second preset value;

[0042] When the fluorescence pulse signal data is less than or equal to the fourth threshold, the pre-scattered pulse signal data is adjusted to a third preset value.

[0043] In one embodiment, before the step of inputting the original data into the conversion model, the method further comprises:

[0044] The original optical signal data at least includes pre-scattered pulse signal data;

[0045] Obtaining a fifth threshold corresponding to the forward scattered pulse signal data;

[0046] When the preceding scattered pulse signal data is greater than the fifth threshold, a random value is superimposed on the fifth threshold to obtain adjusted preceding scattered pulse signal data.

[0047] In a third aspect, the present application provides a sample analysis system, including a sample analyzer and a cloud / server; the sample analyzer is used to obtain raw data of the sample and send it to the cloud / server;

[0048] The cloud / server stores a computer program, which implements the steps of the above method when executed by the cloud / server, and sends the converted scatter plot to the sample analyzer for display.

[0049] The above-mentioned sample analyzer, sample analysis method and sample analysis system obtain the original data corresponding to the sample, wherein the original data includes at least one of the original light signal data and the original scatter plot, and the original light signal data includes at least two paths of light signal data with different angles; the original scatter plot is generated based on the original light signal data, and the original data is input into the conversion model for light signal conversion, wherein the conversion model includes N conversion layers connected in sequence, each conversion layer corresponds to a different conversion function, each conversion layer is used to convert the light signal data according to the corresponding conversion function to obtain converted light signal data, the input of the first conversion layer of the conversion model is the original light signal data, the input of the Mth conversion layer of the conversion model is the light signal data converted by the M-1th conversion layer, and then the converted scatter plot is obtained according to the converted light signal data of the Nth conversion layer of the conversion model, and the scatter plot is identified to obtain the particle classification result, so that different types of particles in the scatter plot can be distinguished, the fusion degree between particles can be reduced, and the boundary between particles can be made more obvious, so that the final particle classification result can be more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A diagram showing an application environment of a sample analysis method in an embodiment;

[0051] Figure 2 A schematic diagram of a sample analysis method according to an embodiment;

[0052] Figure 3 is a schematic flow chart of a sample analysis method in another embodiment;

[0053] Figure 4 is a structural block diagram of a sample analyzer in one embodiment;

[0054] Figure 5 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] The sample analysis method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the sample analyzer 104 is connected to the cloud / server 102, and the sample analyzer 104 includes an optical detection component, which can be used to perform optical detection on the liquid to be tested. The sample analyzer 104 obtains the original data corresponding to the sample, and the original data includes at least one of the original light signal data and the original scatter plot; the original light signal data includes at least two paths of light signal data with different angles; the original scatter plot is generated based on the original light signal data; the sample analyzer 104 inputs the original data into the conversion model, and the conversion model includes N conversion layers connected in sequence, and each conversion layer corresponds to a different conversion function, and each conversion layer is used to convert the light signal data according to the corresponding conversion function to obtain the converted light signal data; the input of the first conversion layer of the conversion model is the original light signal data, and the input of the Mth conversion layer of the conversion model is the light signal data converted by the M-1th conversion layer; N is greater than or equal to 2, and the maximum value of M is N; the sample analyzer 104 generates a converted scatter plot according to the converted light signal data of the Nth conversion layer of the conversion model, and identifies the scatter plot to obtain the particle classification result. The sample analyzer 104 can also send the original data of the sample to the cloud / server 102, and the cloud / server 102 inputs the original data into the conversion model to obtain a converted scatter plot and sends it to the sample analyzer 104. The sample analyzer 104 is used to display the converted scatter plot and the particle classification results obtained by its identification.

[0057] In some embodiments, the sample analyzer includes a dispensing component, a reaction component, an optical detection component, a controller, and a display, wherein the dispensing component is used to dispense samples and reagents to be detected into a reaction container; the reaction component is used to react the samples and reagents in the reaction container to obtain a liquid to be tested; the optical detection component is used to obtain raw data of the sample when the liquid to be tested flows through the detection area of ​​the optical detection component; the controller is used to output the particle classification results in the sample; and the display is used to display the converted scatter plot. In one embodiment, as Figure 2 As shown, a sample analysis method is provided, which is applied to Figure 1 The sample analyzer 104 in the embodiment is used as an example to illustrate, and the following steps are included:

[0058] S202, obtaining original data corresponding to the sample, the original data including at least one of original optical signal data and original scatter plot. The original optical signal data includes at least two optical signal data with different angles. The original scatter plot is generated based on the original optical signal data.

[0059] Among them, after the sample and the reagent react, a test liquid can be obtained. The test liquid refers to a sample liquid containing various types of cells. The test liquid includes but is not limited to blood samples and body fluid samples. Body fluid samples include but are not limited to body cavity fluid, cerebrospinal fluid, pleural effusion, ascites, sac of the heart or joint fluid, etc. The original data is the input data of the conversion model. The original data is obtained based on the optical detection component when the test liquid flows through the detection area of ​​the optical detection component. When the laser irradiates the test liquid, the various light signals generated by each particle in the test liquid due to light irradiation are the original light signal data. Optionally, the original light signal data can be the pulse signal data directly output by the light detector, or it can be the data after filtering, amplification and other processing.

[0060] Different angles refer to different angles in three-dimensional space, that is, two paths of optical signal data are in two directions of X, Y, and Z respectively. At least two paths of optical signal data with different angles refer to at least two types of optical signal data among forward scattered pulse signal data, side scattered pulse signal data, and fluorescent pulse signal data. The forward scattered pulse signal data, the side scattered pulse signal data, and the fluorescent pulse signal data are all commonly used detection parameters, wherein the forward scattered pulse signal data is a scattered light signal generated along the direction of laser propagation when the laser beam irradiates the cell or particle, the side scattered pulse signal data is a scattered light signal generated along the direction perpendicular to the direction of laser propagation when the laser beam irradiates the cell or particle, and the fluorescent pulse signal data is a fluorescent signal emitted after the fluorescent marker on the cell or particle is excited when the laser beam irradiates the cell or particle, and these fluorescent signals are usually captured by the detector laterally.

[0061] The original scatter plot is a graph / matrix drawn from at least two optical signal data with different angles included in the original optical signal data. For example, when the original optical signal data includes two optical signal data with different angles, namely, front scattered pulse signal data and side scattered pulse signal data, the generated original scatter plot is a graph drawn from the two optical signal data, namely, front scattered pulse signal data and side scattered pulse signal data.

[0062] S204, inputting the original data into a conversion model, the conversion model includes N conversion layers connected in sequence, each conversion layer corresponds to a different conversion function, and each conversion layer is used to convert the optical signal data according to the corresponding conversion function to obtain converted optical signal data. The input of the first conversion layer of the conversion model is the original optical signal data, and the input of the Mth conversion layer of the conversion model is the optical signal data converted by the M-1th conversion layer. N is greater than or equal to 2, and the maximum value of M is N.

[0063] Among them, the conversion model refers to a trained model. The conversion model includes N conversion layers connected in sequence, each conversion layer corresponds to a different conversion function, and each conversion layer is used to convert the input optical signal data according to the corresponding conversion function to obtain converted optical signal data. The input of the first conversion layer of the conversion model is the original optical signal data, and the input of the Mth conversion layer of the conversion model is the optical signal data converted by the M-1th conversion layer. For example, the conversion model includes a first conversion layer and a second conversion layer connected in sequence, the input of the first conversion layer is the original optical signal data, the output data of the first conversion layer is optical signal data A converted from the original optical signal data, the input of the second conversion layer is optical signal data A, and the output of the second conversion layer is optical signal data B converted from optical signal data A. By using a conversion model including N conversion layers connected in sequence to convert optical signal data, the accuracy of the conversion can be improved, so that the final particle classification result is more accurate.

[0064] After the original data is input into the conversion model, if the original data is an original scatter plot, the conversion model will also obtain the original optical signal data through the original scatter plot, so that the conversion layer of the conversion model can convert the original optical signal data.

[0065] The conversion function refers to a function that can convert optical signal data. Conversion functions include but are not limited to Hyperlog function, Logical function, and Linlog function. Furthermore, there are differences between the conversion properties of each conversion function. For example, the Hyperlog function can distinguish lymphocytes and monocytes in the sample more clearly, and the Logical function can distinguish eosinophils and neutrophils in the sample more clearly. When the conversion functions corresponding to the first conversion layer and the second conversion layer connected in sequence in the conversion model are respectively the Hyperlog function and the Logical function, the lymphocytes and monocytes in the sample can be distinguished, and the eosinophils and neutrophils in the sample can be distinguished more clearly, so that the boundaries of various types of particles in the generated scatter plot can be made more obvious.

[0066] Optionally, the sample analyzer inputs the acquired raw data into the trained conversion model. If the raw data is an original scatter plot, the conversion model first obtains the raw optical signal data through the original scatter plot, and then uses each conversion layer of the conversion model to convert the raw optical signal data to obtain the optical signal data converted by the Nth conversion layer of the conversion model. The conversion model includes N conversion layers connected in sequence, each conversion layer corresponds to a different conversion function, and each conversion layer is used to convert the optical signal data according to the corresponding conversion function to obtain converted optical signal data. The input of the first conversion layer of the conversion model is the raw optical signal data, and the input of the Mth conversion layer of the conversion model is the optical signal data converted by the M-1th conversion layer. N is greater than or equal to 2, and the maximum value of M is N.

[0067] S206, obtaining a converted scatter plot according to the converted optical signal data of the Nth conversion layer of the conversion model, and identifying the scatter plot to generate a particle classification result.

[0068] Among them, before obtaining the scatter plot, it is necessary to preset the parameters represented by each coordinate axis of the scatter plot, and then draw the scatter plot according to the light signal data after each particle is converted. For example, the front scatter pulse signal is used as the X-axis of the scatter plot, the side scatter pulse signal is used as the Y-axis of the scatter plot, and the fluorescence pulse signal is used as the Z-axis of the scatter plot. Then, according to the front scatter pulse signal, side scatter pulse signal and fluorescence pulse signal after each particle is converted, the position of each particle in the scatter plot is determined respectively, so as to obtain the scatter plot corresponding to the sample. The parameters represented by each coordinate axis of the scatter plot can be preset or set in real time.

[0069] Identification refers to classifying and identifying each particle in the scatter plot to distinguish different types of particles in the scatter plot. Classification and identification can be achieved through clustering models, classification models, or annotation software.

[0070] In some embodiments, after obtaining the particle classification results, the sample analyzer can also mark each type of particle in the scatter plot according to the particle classification results, and display the marked scatter plot in the display interface of the sample analyzer. Marking methods include but are not limited to symbol marking, text marking, and color marking. Use red to mark particles of type A, and use blue to mark particles of type B; or use asterisks to mark particles of type A, and use circles to mark particles of type B; or mark "A" near particles of type A, and mark "B" near particles of type B. By marking each type of particle in the scatter plot, users can distinguish each particle more intuitively.

[0071] In some embodiments, the original optical signal data includes three optical signal data at different angles: front scattered pulse signal data, side scattered pulse signal data, and fluorescent pulse signal data. The converted scatter plot is a three-dimensional scatter plot, which can be displayed on the display of the sample analyzer.

[0072] Optionally, the sample analyzer obtains the parameters represented by each coordinate axis in the preset scatter plot, and generates a scatter plot corresponding to the sample according to the converted light signal data of each particle output by the Nth conversion layer of the conversion model. The sample analyzer uses a clustering model to classify and identify each particle in the scatter plot to obtain a particle classification result.

[0073] In the above-mentioned sample analysis method, by obtaining the original data corresponding to the sample, wherein the original data includes at least one of the original light signal data and the original scatter plot, and the original light signal data includes at least two paths of light signal data with different angles; the original scatter plot is generated based on the original light signal data, and the original data is input into the conversion model for light signal conversion, wherein the conversion model includes N conversion layers connected in sequence, each conversion layer corresponds to a different conversion function, and each conversion layer is used to convert the light signal data according to the corresponding conversion function to obtain converted light signal data, the input of the first conversion layer of the conversion model is the original light signal data, the input of the Mth conversion layer of the conversion model is the light signal data obtained by conversion by the M-1th conversion layer, and then the converted scatter plot is obtained according to the converted light signal data of the Nth conversion layer of the conversion model, and the scatter plot is identified to obtain the particle classification result, so that different types of particles in the scatter plot can be distinguished, the fusion degree between particles can be reduced, and the boundary between particles can be made more obvious, so that the final particle classification result can be more accurate.

[0074] In one embodiment, the training process of the conversion model includes:

[0075] Construct an initial conversion model. The initial conversion model includes N conversion layers connected in sequence, and each conversion layer corresponds to a different conversion function.

[0076] The sample set is used to train the initial conversion model. During the training process, the parameters of the conversion function of each conversion layer in the initial conversion model are adjusted based on back propagation. When the initial conversion model meets the iteration end condition, the conversion model is obtained. The sample set includes sample scatter plots and the particle classification results of the labeled sample scatter plots.

[0077] The initial conversion model is constructed by N conversion functions, where N is greater than or equal to 2. The conversion function refers to a function that can convert optical signal data, and there are differences in the conversion properties of each conversion function.

[0078] The iteration end condition refers to that the number of training times of the initial conversion model reaches the preset number of times, and / or the loss value of the initial conversion model tends to be stable. For example, the preset number of training times is 100 times, and the initial conversion model meets the iteration end condition after 100 training times, and / or the loss value of the initial conversion model is always maintained within a certain range during the n training processes.

[0079] The parameters of the conversion function of each conversion layer are the conversion parameters in the conversion function. For example, the formula of the HyperLog function is , where d, r, b are conversion parameters in the HyperLog conversion function, x is the optical signal data to be converted, and y is the converted optical signal data.

[0080] After the initial conversion model satisfies the iteration end condition and the conversion model is obtained, the conversion model and its parameters can be applied to other instruments of the same type to convert optical signal data. The other instruments of the same type can be instruments with the same liquid path structure, reagents, etc.

[0081] The number of sample scatter plots included in the sample set is at least one. The particle classification result is the classification result of each sample particle in the sample scatter plot, that is, the category to which each sample particle in the sample scatter plot belongs. The particle classification result of the labeled sample scatter plot can be realized by labeling software.

[0082] Optionally, the sample analyzer constructs an initial conversion model based on N conversion functions, wherein the initial conversion model includes N sequentially connected conversion layers, and each conversion layer corresponds to a different conversion function. The sample analyzer continuously trains the constructed initial conversion model, and adjusts the parameters of the conversion functions of each conversion layer in the initial conversion model through back propagation during the training process until the initial conversion model meets the iteration end condition, thereby obtaining a conversion model.

[0083] In this embodiment, an initial conversion model is constructed and trained using a sample set. During the training process, the parameters of the conversion function of each conversion layer in the initial conversion model are adjusted based on back propagation. When the initial conversion model meets the iteration end condition, the conversion model is obtained. This can prevent overfitting during the model training process, improve the generalization ability of the model on new data, save computer resources, improve training efficiency, and improve the efficiency of determining the parameters of each conversion function.

[0084] In one embodiment, the initial conversion model is trained using a sample set. During the training process, the parameters of the conversion functions of each conversion layer in the initial conversion model are adjusted based on back propagation. When the initial conversion model meets the iteration end condition, the conversion model is obtained, including:

[0085] The original optical signal data of the sample scatter diagram in the sample set is input into the initial conversion model to obtain the converted optical signal data of each sample particle output by the Nth conversion layer of the initial conversion model.

[0086] According to the particle classification result of the sample scatter plot and the converted light signal data of each sample particle, the converted light signal data of each category of sample particles are determined.

[0087] Based on the converted optical signal data of each category of sample particles, the sum of the first distances between sample particles of the same category and the sum of the second distances between sample particles of different categories are determined.

[0088] The difference between the sum of the first distance and the sum of the second distance is used as the loss value of the initial conversion model.

[0089] During the training process, the loss value is back-propagated to the initial conversion model, and the parameters of the conversion function of each conversion layer in the initial conversion model are adjusted until the initial conversion model meets the iteration end condition to obtain the conversion model.

[0090] The converted light signal data of sample particles of each category refers to the converted light signal data of each sample particle of each category. For example, the particle category of sample particle A is M1, and the particle category of sample particle B is M2. After the conversion of the initial conversion model, the converted light signal data of sample particle A and sample particle B are obtained, thereby determining the converted light signal data of sample particle A of particle category M1 and the converted light signal data of sample particle B of particle category M2.

[0091] The sum of the first distances is the result of adding the distances between the optical signal data of each sample particle belonging to the same category in the converted optical signal data output by the Nth conversion layer of the initial conversion model, and the sum of the second distances is the result of adding the distances between the optical signal data of each sample particle of different categories in the converted optical signal data output by the Nth conversion layer of the initial conversion model. For example, sample particle 1, sample particle 2, and sample particle 3 are category A, and sample particle 4 is category B. The sum of the first distances is the result of adding the distance between sample particle 1 and sample particle 2, the distance between sample particle 1 and sample particle 3, and the distance between sample particle 2 and sample particle 3. The sum of the second distances is the result of adding the distance between sample particle 1 and sample particle 4, the distance between sample particle 2 and sample particle 4, and the distance between sample particle 3 and sample particle 4.

[0092] The difference refers to the sum of the first distance minus the sum of the second distance. By determining the loss value of the initial conversion model based on the difference between the sum of the first distance and the sum of the second distance, the distance between sample particles of the same category can be reduced and the distance between sample particles of different categories can be increased, thereby better capturing the similarities and differences between sample particles and improving the learning ability of the initial conversion model.

[0093] In some embodiments, when determining the loss value of the initial conversion model, it can also be calculated in a weighted manner, that is, different weight information is assigned to the sum of the first distance and the sum of the second distance. For example, the loss value of the initial conversion model loss=a*sum0-b*sum1, sum0 is the sum of the first distance, sum1 is the sum of the second distance, a is the weight information of the sum of the first distance, b is the weight information of the sum of the second distance, and a and b are different. The weight information can be set by the user.

[0094] Optionally, the sample analyzer inputs the original optical signal data of the sample scatter plot in the sample set into the initial conversion model for conversion, and obtains the converted optical signal data of each sample particle output by the Nth conversion layer of the initial conversion model. The sample analyzer determines the converted optical signal data of each category of sample particles based on the converted optical signal data output by the Nth conversion layer of the initial conversion model and the particle classification result of the sample scatter plot. Based on the converted optical signal data of each sample particle in each category, the sample analyzer adds the distances between the converted optical signal data of each sample particle of the same category to obtain the sum of the first distances, and adds the distances between the converted optical signal data of each sample particle of different categories to obtain the sum of the second distances. The sample analyzer determines the loss value of the initial conversion model based on the difference obtained by subtracting the sum of the second distances from the sum of the first distances. During the training process, the sample analyzer back-propagates the loss value to the initial conversion model, and adjusts the parameters of the conversion function of each conversion layer in the initial conversion model according to the back-propagated loss value, until the initial conversion model is trained for a preset number of times and / or the loss value of the initial conversion model tends to be stable, thereby obtaining a conversion model.

[0095] In this embodiment, the loss value of the initial conversion model is determined based on the difference between the sum of the first distances and the sum of the second distances, which can reduce the distance between sample particles of the same category and increase the distance between sample particles of different categories, thereby better capturing the similarities and differences between sample particles and improving the learning ability of the initial conversion model.

[0096] In one embodiment, there are multiple initial conversion models, and the conversion functions corresponding to at least one conversion layer in each initial conversion model are different, and / or the number of conversion layers is different.

[0097] When the initial conversion model meets the iteration end condition, the conversion model is obtained, including:

[0098] When the number of training times of each initial conversion model reaches a preset number of times, and / or the loss value of the initial conversion model tends to be stable, a plurality of candidate conversion models are obtained.

[0099] Determine the loss value corresponding to each candidate transformation model.

[0100] The candidate conversion model with the smallest loss value is determined as the conversion model.

[0101] Among them, different conversion functions have different conversion properties. For example, Hyperlog function and Logical function are two conversion functions with different conversion properties. Hyperlog function is very useful when dealing with logarithmic models with non-zero intercepts, and can effectively reveal details and trends in the data, especially when the data is unevenly distributed or there are extreme values. By using HyperLog function, the distribution characteristics of the data can be better displayed, making data visualization more intuitive and easy to understand. Logical function combines the characteristics of linear and logarithmic scaling, and can effectively display the distribution characteristics of the data.

[0102] The iteration end condition refers to the initial conversion model being trained for a preset number of times and / or the loss value of the initial conversion model being stable. After each initial conversion model is trained for a preset number of times and / or the loss value of the initial conversion model is stable, each initial conversion model can be determined as a candidate conversion model, and the loss value corresponding to each candidate conversion model is the loss value of each initial conversion model when it meets the iteration end condition.

[0103] The training of each initial conversion model can be performed simultaneously or sequentially. When multiple initial conversion models are trained simultaneously, the waiting time for obtaining the conversion model can be reduced, and the utilization rate of computer resources can be improved. When each initial conversion model is trained sequentially, complex task scheduling and resource management mechanisms are not required, which reduces the complexity of the system.

[0104] Since, when performing optical signal data conversion, using a model with a smaller loss value for data conversion can make the conversion result more accurate, the candidate conversion model with the smallest loss value is determined as the conversion model.

[0105] Optionally, when the number of training times of each initial conversion model reaches a preset number and / or the loss value of the initial conversion model tends to be stable, the sample analyzer determines each initial conversion model as a candidate conversion model, and determines the loss value corresponding to each initial conversion model when the iteration stops as the loss value corresponding to the candidate conversion model. The sample analyzer selects the candidate conversion model with the smallest loss value as the conversion model.

[0106] In this embodiment, by determining the candidate conversion model with the smallest loss value as the conversion model, when performing optical signal data conversion, the converted optical signal data can be made more accurate.

[0107] In one embodiment, the initial conversion model is trained using the sample set, and the parameters of the conversion function of each conversion layer in the initial conversion model are adjusted based on back propagation during the training process, including:

[0108] The initial conversion model includes two conversion layers connected in sequence, and the conversion functions corresponding to the two conversion layers connected in sequence are Hyperlog function and Logical function respectively.

[0109] The original optical signal data of the sample scatter plot in the sample set is input into the initial conversion model to obtain the converted optical signal data of eosinophils, neutrophils, lymphocytes, and monocytes output by the second conversion layer of the initial conversion model.

[0110] Based on the optical signal data after conversion of eosinophils and neutrophils, the first loss value is calculated.

[0111] The second loss value is calculated based on the light signal data converted from the lymphocytes and monocytes.

[0112] The first loss value is back-propagated to the conversion layer where the Logical function is located in the initial conversion model, and the second loss value is back-propagated to the conversion layer where the Hyperlog function is located in the initial conversion model, and the parameters of the Logical function and the parameters of the Hyperlog function are adjusted until the initial conversion model meets the iteration end condition to obtain the conversion model.

[0113] Among them, the first loss value can be determined according to the sum of the distances between the converted light signal data of each eosinophil, the sum of the distances between the converted light signal data of each centriole, the sum of the distances between each centriole and each eosinophil, and the sum of the distances between each eosinophil and each centriole. Specifically, the sum of the distances between the converted light signal data of each eosinophil is added to the sum of the distances between the converted light signal data of each centriole to obtain the first result after addition, the sum of the distances between each centriole and each eosinophil is added to the sum of the distances between each eosinophil and each centriole to obtain the second result after addition, and the difference between the first result and the second result is determined as the first loss value.

[0114] The second loss value can be determined based on the sum of the distances between the light signal data converted by each lymphocyte, the sum of the distances between the light signal data converted by each monocyte, the sum of the distances between each monocyte and each lymphocyte, and the sum of the distances between each lymphocyte and each monocyte. Specifically, the sum of the distances between the light signal data converted by each lymphocyte and the sum of the distances between the light signal data converted by each monocyte are added to obtain a third result after addition, the sum of the distances between each monocyte and each lymphocyte and the sum of the distances between each lymphocyte and each monocyte are added to obtain a fourth result after addition, and the difference between the third result and the fourth result is determined as the second loss value.

[0115] In some embodiments, when determining the first loss and the second loss value, they can also be calculated in a weighted manner. For example, the first loss value loss1 of the initial conversion model = a1*sum2-b1*sum3, where sum2 is the first result after adding the sum of the distances between the converted optical signal data of each eosinophil and the sum of the distances between the converted optical signal data of each centrocyte, and sum3 is the second result after adding the sum of the distances between each centrocyte and each eosinophil and the sum of the distances between each eosinophil and each centrocyte, a1 is the weight information of the first result, and b1 is the weight information of the second result. a1 and b1 are different, and the weight information can be set by the user.

[0116] The iteration end condition refers to that the number of training times of the initial conversion model reaches a preset number and / or the loss value of the initial conversion model tends to be stable.

[0117] Optionally, the initial conversion model includes two sequentially connected conversion layers, and the conversion functions corresponding to the two sequentially connected conversion layers are Hyperlog function and Logical function respectively. The sample analyzer inputs the original light signal data of the sample scatter plot in the sample set into the initial conversion model for conversion, and obtains the converted light signal data of eosinophils, neutrophils, lymphocytes, and monocytes output by the second conversion layer of the initial conversion model. The sample analyzer determines the first loss value according to the sum of the distances between the converted light signal data of each eosinophil, the sum of the distances between the converted light signal data of each neutrophil, the sum of the distances between each neutrophil and each eosinophil, and the sum of the distances between each eosinophil and each neutrophil, and determines the second loss value according to the sum of the distances between the converted light signal data of each lymphocyte, the sum of the distances between the converted light signal data of each monocyte, the sum of the distances between each monocyte and each lymphocyte, and the sum of the distances between each lymphocyte and each monocyte. During the training process, the sample analyzer back-propagates the first loss value to the conversion layer where the Logical function is located in the initial conversion model, and back-propagates the second loss value to the conversion layer where the Hyperlog function is located in the initial conversion model, and adjusts the parameters of the Logical function and the parameters of the Hyperlog function until the training times of the initial conversion model reach the preset times, and / or the loss value of the initial conversion model tends to be stable, so as to obtain the conversion model.

[0118] In one embodiment, when the initial conversion model meets the iteration end condition, the conversion model is obtained, including:

[0119] When the initial conversion model meets the iteration end condition and the loss value of the initial conversion model is greater than the first threshold, the loss value corresponding to the initial conversion model is determined according to the optical signal data after conversion of sample particles of at least two target categories.

[0120] The corresponding loss value is back-propagated to the target conversion layer corresponding to the target category in the initial conversion model, and the parameters of the conversion function in the target conversion layer are adjusted until the initial conversion model meets the iteration end condition and the loss value of the initial conversion model is less than the first threshold, thereby obtaining the conversion model.

[0121] The iteration end condition refers to that the number of training times of the initial conversion model reaches a preset number of times and / or the loss value of the initial conversion model tends to be stable.

[0122] The loss value corresponding to the initial conversion model is greater than the first threshold value, which means that when the number of training times of the initial conversion model reaches the preset number of times and / or the loss value of the initial conversion model tends to be stable, the loss value corresponding to the initial conversion model is still very high. If the initial conversion model with a high loss value is directly used as the conversion model in the conversion scenario of the optical signal data, the optical signal data cannot be accurately converted, and thus a more accurate particle classification result cannot be obtained. Therefore, the parameters of the initial conversion model need to be adjusted again to reduce the loss value of the initial conversion model. The first threshold value is usually pre-set in the sample analyzer by the user based on experience.

[0123] The target conversion layer is determined by the conversion properties of the conversion function and the target category. For example, the Hyperlog function can distinguish lymphocytes and monocytes more clearly, and the Logical function can distinguish eosinophils and neutrophils more clearly. When at least two target categories of sample particles are lymphocytes and monocytes, the parameters of the conversion layer where the Hyperlog function is located can be adjusted according to the loss values ​​obtained from the optical signal data of lymphocytes and monocytes. At this time, the conversion layer where the Hyperlog function is located is the target conversion layer. When at least two categories of sample particles are eosinophils and neutrophils, the parameters of the conversion layer where the Logical function is located can be adjusted according to the loss values ​​obtained from the optical signal data of eosinophils and neutrophils. At this time, the conversion layer where the Logical function is located is the target conversion layer.

[0124] Back-propagating the corresponding loss value to the target conversion layer corresponding to the target category in the initial conversion model, and adjusting the parameters of the conversion function in the target conversion layer can be understood as first refining the loss value of the initial conversion model, and then back-propagating to the target conversion layer in the initial conversion model after refinement to adjust the parameters of the conversion function in the target conversion layer.

[0125] Optionally, when the number of training times of the initial conversion model reaches a preset number of times and / or the loss value of the initial conversion model tends to be stable and the loss value corresponding to the initial conversion model is greater than a first threshold, the sample analyzer calculates the loss value corresponding to the initial conversion model based on the optical signal data after conversion of sample particles of at least two target categories, and then back-propagates the corresponding loss value to the target conversion layer corresponding to the target category of the initial conversion model, adjusts the parameters of the conversion function in the target conversion layer, and obtains the conversion model until the number of training times of the initial conversion model reaches a preset number of times and / or the loss value of the initial conversion model tends to be stable and the loss value of the initial conversion model is less than the first threshold.

[0126] In this embodiment, when the initial conversion model meets the iteration end condition and the loss value of the initial conversion model is greater than the first threshold, the loss value corresponding to the initial conversion model is determined according to the optical signal data after the conversion of sample particles of at least two target categories, and the corresponding loss value is back-propagated to the target conversion layer corresponding to the target category in the initial conversion model, and the parameters of the conversion function in the target conversion layer are adjusted. In this way, the loss value of the model can be further refined, and the distance for distinguishing sample particles of each target category can be increased, making it easier to distinguish sample particles of the target category.

[0127] In one embodiment, the raw light signal data includes front scattered pulse signal data, side scattered pulse signal data and fluorescence pulse signal data; before the step of inputting the raw data into the conversion model, the sample analysis method further includes:

[0128] Acquire a second threshold corresponding to the front scattered pulse signal data, a third threshold corresponding to the side scattered pulse signal data, and a fourth threshold corresponding to the fluorescent pulse signal data;

[0129] When the front scattered pulse signal data is less than or equal to the second threshold, adjusting the front scattered pulse signal data to a first preset value;

[0130] When the side scattered pulse signal data is less than or equal to the third threshold, adjusting the front scattered pulse signal data to a second preset value;

[0131] When the fluorescence pulse signal data is less than or equal to the fourth threshold, the forward scattered pulse signal data is adjusted to a third preset value.

[0132] Among them, the second threshold is a threshold used to cut off the extremely low value in the front scattered pulse signal data. By cutting off the extremely low value by the second threshold, the noise in the front scattered pulse signal data can be eliminated, thereby improving the quality of the front scattered pulse signal data.

[0133] The third threshold is a threshold used to cut off extremely low values ​​in the side scattered pulse signal data. By cutting off extremely low values ​​through the third threshold, noise in the side scattered pulse signal data can be eliminated, thereby improving the quality of the side scattered pulse signal data.

[0134] The fourth threshold is a threshold used to cut off extremely low values ​​in the fluorescence pulse signal data. By cutting off extremely low values ​​through the fourth threshold, noise in the fluorescence pulse signal data can be eliminated, thereby improving the quality of the fluorescence pulse signal data.

[0135] The second threshold corresponding to the front scattered pulse signal data, the third threshold corresponding to the preset side scattered pulse signal data, and the fourth threshold corresponding to the fluorescence pulse signal data can be preset by the user based on experience. The second threshold, the third threshold, and the fourth threshold can be pre-stored in the storage component of the sample analyzer, and read from the storage component when comparison is required.

[0136] The first preset value, the second preset value, and the third preset value can be set by the user based on experience. For example, the first preset value can be 0 or other values. The original light signal data of each particle includes front scatter pulse signal data, side scatter pulse signal data, and fluorescent pulse signal data, so it is necessary to compare each type of original light signal data of each particle with the corresponding threshold value, so as to determine the light signal data that needs to be adjusted. For example, the front scatter pulse signal data of particle A is 1, the front scatter pulse signal data in the original light signal data of particle B is 3, and the second threshold corresponding to the front scatter pulse signal data is 2, then the front scatter pulse signal data of particle A needs to be adjusted to the first preset value, and the front scatter pulse signal data of particle B does not need to be adjusted to the first preset value. For another example, the side scatter pulse signal data of particle A is 1, the side scatter pulse signal data of particle B is 3, and the third threshold value corresponding to the side scatter pulse signal data is 1, then the side scatter pulse signal data of particle A needs to be adjusted to the second preset value, and the side scatter pulse signal data of particle B needs to be adjusted to the second preset value, the fluorescence pulse signal data of particle A is 2, the fluorescence pulse signal data of particle B is 4, and the fourth threshold value corresponding to the side scatter pulse signal data is 3, then the fluorescence pulse signal data of particle A needs to be adjusted to the third preset value, and the fluorescence pulse signal data of particle B does not need to be adjusted to the third preset value.

[0137] In some embodiments, a threshold value corresponding to each type of raw light signal data in the sample scatter plot is obtained. The raw light signal data in the sample scatter plot is the light signal data obtained by optical detection of the sample particles. From each type of raw light signal data, a class to be adjusted that is less than or equal to the corresponding threshold value is determined. The raw light signal data of the class to be adjusted is adjusted to a preset value, and the light signal data adjusted to the preset value and the raw light signal data of the remaining classes are input into the first conversion layer of the initial conversion model for conversion. In this way, the generalization ability of the initial conversion model can be improved.

[0138] In some embodiments, after adjusting the optical signal data, the sample analyzer inputs the adjusted optical signal data and the unadjusted original optical signal data into the first conversion layer of the conversion model for conversion.

[0139] Optionally, the sample analyzer obtains a second threshold corresponding to the front scatter pulse signal data, a third threshold corresponding to the side scatter pulse signal data, and a fourth threshold corresponding to the fluorescence pulse signal data, and compares the front scatter pulse signal data of each particle with the corresponding second threshold, compares the side scatter pulse signal data of each particle with the corresponding third threshold, compares the fluorescence pulse signal data of each particle with the corresponding fourth threshold, and when the front scatter pulse signal data is less than or equal to the second threshold, adjusts the front scatter pulse signal data to a first preset value, when the side scatter pulse signal data is less than or equal to the third threshold, adjusts the front scatter pulse signal data to a second preset value, and when the fluorescence pulse signal data is less than or equal to the fourth threshold, adjusts the front scatter pulse signal data to the third preset value.

[0140] In this embodiment, by obtaining the threshold values ​​corresponding to the front scattered pulse signal data, the side scattered pulse signal data, and the fluorescent pulse signal data, and when the front scattered pulse signal data is less than or equal to the second threshold, the front scattered pulse signal data is adjusted to the first preset value, when the side scattered pulse signal data is less than or equal to the third threshold, the front scattered pulse signal data is adjusted to the second preset value, and when the fluorescent pulse signal data is less than or equal to the fourth threshold, the front scattered pulse signal data is adjusted to the third preset value. In this way, it can be ensured that the adjusted light signal data is within a certain range, the consistency of the data is improved, and the influence of extreme values ​​can be reduced, so that the adjusted data is more concise and easy for subsequent processing and analysis.

[0141] In one embodiment, before the step of inputting the original data into the conversion model, the sample analysis method further comprises:

[0142] The original optical signal data at least includes pre-scattered pulse signal data;

[0143] A fifth threshold corresponding to the pre-scattered pulse signal data is obtained.

[0144] When the pre-scattered pulse signal data is greater than the fifth threshold, a random value is superimposed on the fifth threshold to obtain adjusted pre-scattered pulse signal data.

[0145] The fifth threshold is a threshold used to compensate for the extremely high value in the pre-scattered pulse signal data. By compensating the extremely high value with the fifth threshold, the influence of the extremely high value in the pre-scattered pulse signal data on the conversion model can be reduced, thereby avoiding distortion or overfitting of the conversion model. The fifth threshold corresponding to the pre-scattered pulse signal data can be stored in a storage component of the sample analyzer, and can be directly obtained from the storage component when comparison is required.

[0146] The original light signal data of each particle includes the pre-scattered pulse signal data, so it is necessary to compare the pre-scattered pulse signal data of each particle with the corresponding fifth threshold value, so as to determine the pre-scattered pulse signal data that needs to be adjusted. The random value superimposed on the basis of the fifth threshold value can be determined within a preset range. For example, if the pre-scattered pulse signal data 5 of particle A is greater than the fifth threshold value 4, then a random value 3 is taken within the range of 1 to 95 and superimposed on the fifth threshold value 4, so that the adjusted pre-scattered pulse signal data 7 can be obtained.

[0147] In some embodiments, after adjusting the pre-scattered pulse signal data, the sample analyzer inputs the adjusted pre-scattered pulse signal data and the unadjusted optical signal data into the first conversion layer of the conversion model for conversion.

[0148] Optionally, the sample analyzer obtains the fifth threshold value corresponding to the pre-scattered pulse signal data, and compares each pre-scattered pulse signal data of each particle with the corresponding fifth threshold value to obtain multiple comparison results. The sample analyzer determines the pre-scattered pulse signal data that needs to be adjusted based on the comparison result of the pre-scattered pulse signal data of each particle with the corresponding fifth threshold value. The sample analyzer determines the result after superimposing the random value on the fifth threshold value as the adjusted pre-scattered pulse signal data. Among them, the random value corresponding to each pre-scattered pulse signal data that needs to be adjusted can be the same or different.

[0149] In this embodiment, when the front scattered pulse signal data is greater than the fifth threshold, a random value is superimposed on the fifth threshold to adjust the front scattered pulse signal data. This ensures that the adjusted front scattered pulse signal data is within a certain range, thereby improving data consistency.

[0150] The present application also provides an application scenario, which applies the above-mentioned sample analysis method. Specifically, the application of the sample analysis method in this application scenario is as follows: the sample analyzer obtains a sample set, the sample set includes a sample scatter plot, and the particle classification results of the labeled sample scatter plot, and the sample scatter plot has original light signal data, wherein the original light signal data in the sample scatter plot includes sample pre-scattered pulse signal data, sample side-scattered pulse signal data, and sample fluorescence pulse signal data. The sample analyzer obtains the threshold corresponding to each type of original light signal data in the sample scatter plot. The sample analyzer determines the class to be adjusted that is less than or equal to the corresponding threshold from each type of original light signal data. The sample analyzer adjusts the original light signal data of the class to be adjusted to a preset value. The sample analyzer obtains the threshold corresponding to the pre-scattered pulse signal data, and compares each sample pre-scattered pulse signal data of each particle with the corresponding threshold to obtain multiple comparison results. The sample analyzer determines the sample pre-scattered pulse signal data that needs to be adjusted based on the comparison result of the sample pre-scattered pulse signal data of each particle with the corresponding threshold. The sample analyzer superimposes a random value on the sample pre-scattered pulse signal data that needs to be adjusted to obtain adjusted sample pre-scattered pulse signal data, so as to achieve data preprocessing by adjusting to a preset value and / or superimposing a random value.

[0151] The sample analyzer constructs multiple initial conversion models based on multiple conversion functions, wherein each initial conversion model includes N conversion layers connected in sequence, each conversion layer corresponds to a different conversion function, at least one conversion layer in each initial conversion model corresponds to a different conversion function, and / or the number of conversion layers is different.

[0152] The sample analyzer converts the preprocessed optical signal data based on each initial conversion model to obtain the converted optical signal data of each sample particle output by the Nth conversion layer of each initial conversion model. The sample analyzer determines the converted optical signal data of each category of sample particles based on the converted optical signal data output by the Nth conversion layer of each initial conversion model and the particle classification result of the sample scatter plot. Based on the converted optical signal data of each sample particle of each category, the sample analyzer adds the distances between the converted optical signal data of each sample particle of the same category to obtain the sum of the first distances, and adds the distances between the converted optical signal data of each sample particle of different categories to obtain the sum of the second distances. The sample analyzer determines the loss value of each initial conversion model based on the difference obtained by subtracting the sum of the second distances from the sum of the first distances.

[0153] During the training process, the sample analyzer back-propagates the loss value to the initial conversion model, and adjusts the parameters of the conversion function of each conversion layer in the initial conversion model according to the back-propagated loss value, until the number of training times of each initial conversion model reaches a preset number, and / or the loss value of the initial conversion model tends to be stable. The sample analyzer determines each initial conversion model as a candidate conversion model, and determines the loss value corresponding to each initial conversion model when the iteration stops as the loss value corresponding to the candidate conversion model. The sample analyzer uses the candidate conversion model with the smallest loss value as the conversion model.

[0154] When the number of training times of the initial conversion model reaches a preset number of times and / or the loss value of the initial conversion model tends to be stable and the loss value corresponding to the initial conversion model is greater than a first threshold, the sample analyzer calculates the loss value corresponding to the initial conversion model based on the optical signal data after conversion of sample particles of at least two target categories, and then back-propagates the corresponding loss value to the target conversion layer corresponding to the target category of the initial conversion model, and adjusts the parameters of the conversion function in the target conversion layer, until the number of training times of the initial conversion model reaches a preset number of times and / or the loss value of the initial conversion model tends to be stable and the loss value of the initial conversion model is less than the first threshold, thereby obtaining a conversion model.

[0155] The optical detection component in the sample analyzer irradiates the liquid to be tested flowing through the detection area with light, and collects various optical signal data generated by each particle in the liquid to be tested due to the light irradiation. The sample analyzer obtains raw data, wherein the raw data includes at least one of raw optical signal data and raw scatter plot. The raw optical signal data includes at least two paths of optical signal data with different angles. The raw scatter plot is generated based on the raw optical signal data. After the raw data is input into the conversion model, if the raw data is a raw scatter plot, the conversion model will also obtain the raw optical signal data through the raw scatter plot, so that the conversion layer of the conversion model can convert the raw optical signal data. The sample analyzer obtains the second threshold corresponding to the front scattered pulse signal data, the third threshold corresponding to the side scattered pulse signal data, and the fourth threshold corresponding to the fluorescence pulse signal data, and compares the front scattered pulse signal data of each particle with the corresponding second threshold, compares the side scattered pulse signal data of each particle with the corresponding third threshold, and compares the fluorescence pulse signal data of each particle with the corresponding fourth threshold. When the front scattered pulse signal data is less than or equal to the second threshold, the front scattered pulse signal data is adjusted to the first preset value, when the side scattered pulse signal data is less than or equal to the third threshold, the front scattered pulse signal data is adjusted to the second preset value, and when the fluorescence pulse signal data is less than or equal to the fourth threshold, the front scattered pulse signal data is adjusted to the third preset value. The sample analyzer obtains the fifth threshold corresponding to the front scattered pulse signal data, and compares each front scattered pulse signal data of each particle with the corresponding fifth threshold to obtain multiple comparison results. The sample analyzer determines the front scattered pulse signal data that needs to be adjusted according to the comparison result of the front scattered pulse signal data of each particle with the corresponding fifth threshold. The sample analyzer determines the result after superimposing the random value on the fifth threshold as the adjusted front scattered pulse signal data. The random values ​​corresponding to each pre-scattered pulse signal data that needs to be adjusted may be the same or different.

[0156] The sample analyzer uses the trained conversion model to convert the preprocessed raw optical signal data to obtain the optical signal data converted by the Nth conversion layer of the conversion model. The conversion model includes N conversion layers connected in sequence, each conversion layer corresponds to a different conversion function, and each conversion layer is used to convert the input optical signal data according to the corresponding conversion function to obtain converted optical signal data. The input of the first conversion layer of the conversion model is the original optical signal data, and the input of the Mth conversion layer of the conversion model is the optical signal data converted by the M-1th conversion layer. The sample analyzer obtains the parameters represented by each coordinate axis in the pre-set scatter plot, and generates a converted scatter plot based on the converted optical signal data of each particle output by the Nth conversion layer of the conversion model. The sample analyzer uses a clustering model to classify and identify each particle in the scatter plot, obtains the particle classification result, and displays the converted scatter plot on the display. The specific process is as follows: Figure 3 shown.

[0157] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0158] Based on the same inventive concept, the embodiment of the present application also provides a sample analyzer for implementing the sample analysis method involved above. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above method, so the specific limitations in one or more sample analyzer embodiments provided below can refer to the limitations of the sample analysis method above, and will not be repeated here.

[0159] In one embodiment, Figure 4 As shown, a sample analyzer is provided, comprising:

[0160] The data acquisition module 402 is used to acquire the original data corresponding to the sample, wherein the original data includes at least one of the original optical signal data and the original scatter plot. The original optical signal data includes at least two optical signal data with different angles. The original scatter plot is generated based on the original optical signal data.

[0161] The data conversion module 404 is used to input the original data into the conversion model. The conversion model includes N conversion layers connected in sequence. Each conversion layer corresponds to a different conversion function. Each conversion layer is used to convert the optical signal data according to the corresponding conversion function to obtain converted optical signal data. Among them, the input of the first conversion layer of the conversion model is the original optical signal data, and the input of the Mth conversion layer of the conversion model is the optical signal data converted by the M-1th conversion layer. N is greater than or equal to 2, and the maximum value of M is N.

[0162] The particle classification module 406 is used to generate a converted scatter plot according to the converted optical signal data of the Nth conversion layer of the conversion model, and identify the scatter plot to obtain a particle classification result.

[0163] Each module in the above sample analyzer can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0164] In one embodiment, a computer device is provided. The computer device may be a sample analyzer, and its internal structure diagram may be as shown in FIG. Figure 5 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store original optical signal data, conversion functions, optical signal data converted by each conversion layer, scatter plots, and particle classification results. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a sample analysis method is implemented.

[0165] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0166] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0167] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0168] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0169] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0170] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0171] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0172] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A sample analyzer, characterized in that: include: A dispensing component, used for dispensing samples and reagents to be tested into the reaction container; A reaction component, used for reacting the sample and reagent in the reaction container to obtain a liquid to be tested; An optical detection component, used for obtaining raw data of the sample when the liquid to be tested flows through a detection area of ​​the optical detection component; A controller is used to obtain raw data corresponding to a sample; the raw data includes at least one of raw optical signal data and raw scatter plot; the raw optical signal data includes at least two optical signal data with different angles; the raw scatter plot is generated based on the raw optical signal data; the raw data is input into a conversion model, the conversion model includes N conversion layers connected in sequence, each conversion layer corresponds to a different conversion function, each conversion layer is used to convert the optical signal data according to the corresponding conversion function to obtain converted optical signal data; wherein the input of the first conversion layer of the conversion model is the raw optical signal data, and the input of the Mth conversion layer of the conversion model is the M-1th conversion layer. layer conversion; N is greater than or equal to 2, and the maximum value of M is N; according to the converted light signal data of the Nth conversion layer of the conversion model, a converted scatter plot is generated, and the scatter plot is identified to obtain a particle classification result; the conversion model is obtained by training the initial conversion model, and during the training process of the initial conversion model, the loss value corresponding to the initial conversion model is back-propagated to the target conversion layer in the initial conversion model to obtain the conversion model; the loss value corresponding to the initial conversion model is determined according to the converted light signal data of sample particles of at least two target categories; the target conversion layer is determined by the conversion properties of the conversion function and the target category; A display is used to display the converted scatter plot.

2. A sample analysis method, characterized in that: The method comprises: Acquire original data corresponding to the sample, the original data including at least one of original light signal data and an original scatter plot; the original light signal data includes at least two paths of light signal data with different angles; the original scatter plot is generated based on the original light signal data; The original data is input into a conversion model, wherein the conversion model includes N conversion layers connected in sequence, each of the conversion layers corresponds to a different conversion function, and each of the conversion layers is used to convert the optical signal data according to the corresponding conversion function to obtain converted optical signal data; wherein the input of the first conversion layer of the conversion model is the original optical signal data, and the input of the Mth conversion layer of the conversion model is the optical signal data converted by the M-1th conversion layer; wherein N is greater than or equal to 2, and the maximum value of M is N; According to the converted optical signal data of the Nth conversion layer of the conversion model, a scatter plot after conversion is generated, and the scatter plot is identified to obtain a particle classification result; the conversion model is obtained by training an initial conversion model, and during the training process of the initial conversion model, the loss value corresponding to the initial conversion model is back-propagated to the target conversion layer in the initial conversion model to obtain the conversion model; the loss value corresponding to the initial conversion model is determined according to the optical signal data after conversion of sample particles of at least two target categories; the target conversion layer is determined by the conversion properties of the conversion function and the target category.

3. The method according to claim 2, characterized in that The training process of the conversion model includes: Constructing an initial conversion model; the initial conversion model includes N conversion layers connected in sequence, and each conversion layer corresponds to a different conversion function; The initial conversion model is trained using a sample set. During the training process, the parameters of the conversion function of the target conversion layer in the initial conversion model are adjusted based on back propagation. When the initial conversion model meets the iteration end condition, the conversion model is obtained. The sample set includes a sample scatter plot and annotated particle classification results of the sample scatter plot.

4. The method according to claim 3, characterized in that The initial conversion model is trained using the sample set, and the parameters of the conversion function of the target conversion layer in the initial conversion model are adjusted based on back propagation during the training process, and the conversion model is obtained when the initial conversion model meets the iteration end condition, including: Inputting the original optical signal data of the sample scatter plot in the sample set into the initial conversion model to obtain the converted optical signal data of each sample particle output by the Nth conversion layer of the initial conversion model; Determine the converted light signal data of the sample particles of each category according to the particle classification result of the sample scatter plot and the converted light signal data of each sample particle; Determine the sum of first distances between sample particles of the same category and the sum of second distances between sample particles of different categories based on the converted light signal data of the sample particles of each category; Taking the difference between the sum of the first distances and the sum of the second distances as the loss value of the initial conversion model; During the training process, the loss value is back-propagated to the initial conversion model, and the parameters of the conversion function of the target conversion layer in the initial conversion model are adjusted until the initial conversion model meets the iteration end condition to obtain the conversion model.

5. The method according to claim 3, characterized in that: There are multiple initial conversion models; the conversion function corresponding to at least one conversion layer in each of the initial conversion models is different, and / or the number of conversion layers is different; When the initial conversion model satisfies the iteration end condition, obtaining the conversion model includes: When the number of training times of each of the initial conversion models reaches a preset number of times, and / or the loss value of the initial conversion model tends to be stable, a plurality of candidate conversion models are obtained; Determine a loss value corresponding to each of the candidate conversion models; The candidate conversion model with the smallest loss value is determined as the conversion model.

6. The method according to claim 3, characterized in that The method of training the initial conversion model using the sample set and adjusting the parameters of the conversion function of the target conversion layer in the initial conversion model based on back propagation during the training process includes: The initial conversion model includes two conversion layers connected in sequence, and the conversion functions corresponding to the two conversion layers connected in sequence are respectively a Hyperlog function and a Logical function; Inputting the original light signal data of the sample scatter plot in the sample set into the initial conversion model to obtain the converted light signal data of eosinophils, neutrophils, lymphocytes, and monocytes output by the second conversion layer of the initial conversion model; Calculating a first loss value based on the converted optical signal data of the eosinophils and the neutrophils; Calculating a second loss value based on the converted optical signal data of the lymphocytes and the monocytes; The first loss value is back-propagated to the conversion layer where the Logical function is located in the initial conversion model, and the second loss value is back-propagated to the conversion layer where the Hyperlog function is located in the initial conversion model, and the parameters of the Logical function and the parameters of the Hyperlog function are adjusted until the initial conversion model meets the iteration end condition to obtain the conversion model.

7. The method according to claim 4, characterized in that When the initial conversion model satisfies the iteration end condition, obtaining the conversion model includes: When the initial conversion model satisfies an iteration termination condition and the loss value of the initial conversion model is greater than a first threshold, determining the loss value corresponding to the initial conversion model according to the optical signal data after conversion of sample particles of at least two target categories; The corresponding loss value is back-propagated to the target conversion layer corresponding to the target category in the initial conversion model, and the parameters of the conversion function in the target conversion layer are adjusted until the initial conversion model meets the iteration end condition and the loss value of the initial conversion model is less than the first threshold, so as to obtain a conversion model.

8. The method according to claim 2, characterized in that: The original light signal data includes front scattered pulse signal data, side scattered pulse signal data and fluorescence pulse signal data; Before the step of inputting the original data into the conversion model, the method further comprises: Acquire a second threshold corresponding to the front scattered pulse signal data, a third threshold corresponding to the side scattered pulse signal data, and a fourth threshold corresponding to the fluorescent pulse signal data; when the front scattered pulse signal data is less than or equal to the second threshold, adjust the front scattered pulse signal data to a first preset value; When the side scattered pulse signal data is less than or equal to the third threshold, adjusting the front scattered pulse signal data to a second preset value; When the fluorescence pulse signal data is less than or equal to the fourth threshold, the pre-scattered pulse signal data is adjusted to a third preset value.

9. The method according to claim 2, characterized in that: Before the step of inputting the original data into the conversion model, the method further comprises: The original optical signal data at least includes pre-scattered pulse signal data; Obtaining a fifth threshold corresponding to the forward scattered pulse signal data; When the preceding scattered pulse signal data is greater than the fifth threshold, a random value is superimposed on the fifth threshold to obtain adjusted preceding scattered pulse signal data.

10. A sample analysis system, characterized in that: Including sample analyzer, cloud / server; The sample analyzer is used to obtain raw data of the sample and send it to the cloud / server; A computer program is stored on the cloud / server, and when the computer program is executed by the cloud / server, the steps of the method described in any one of claims 2 to 9 are implemented, and the converted scatter plot is sent to the sample analyzer for display.

Images