Platelet detection method and device based on deep learning and storage medium

Abstract

The invention provides a platelet detection method and device based on deep learning and a storage medium, which can improve the detection efficiency of platelets and reduce the detection cost of the platelets, and the method comprises the following steps: obtaining a whole blood sample blood cell analysis data set, the whole blood sample blood cell analysis data set comprises blood cell analysis data of a plurality of whole blood samples and PLT counting values of PLT-F channels; a PLT-F fusion prediction model is constructed; training the PLT-F fusion prediction model based on the whole blood sample blood cell analysis data set to obtain a trained PLT-F fusion prediction model; acquiring blood cell analysis data of the whole blood sample to be analyzed; the blood cell analysis data of the whole blood sample to be analyzed are input into the trained PLT-F fusion prediction model, a PLT-F prediction result is obtained, and the PLT prediction result comprises the PLT prediction value of the whole blood sample to be analyzed on the PLT-F channel.

Description

Technical Field

[0001] The present application relates to the field of biomedical technology, and in particular to a platelet detection method, device and storage medium based on deep learning. Background Art

[0002] Platelet examination is one of the clinical diagnostic examination items and plays an important role in the diagnosis and treatment of diseases. Clinical laboratories currently mainly use the electrical impedance method (PLT-I) channel to detect platelets and calculate platelet distribution-related parameters by drawing platelet histograms.

[0003] However, factors such as small red blood cells, red blood cell fragments, tiny blood particles and giant platelets often interfere with platelet quantification, causing false increase or decrease in platelets and affecting clinical diagnosis and treatment.

[0004] To address this issue, clinical laboratories typically use additional fluorescent dye reagents to re-examine platelets using the platelet fluorescence detection channel of the detection equipment to eliminate interference, but this requires a large amount of reagents, labor, and detection time, resulting in low platelet detection efficiency and high detection costs. Summary of the invention

[0005] The present application provides a platelet detection method, device and storage medium based on deep learning to solve the problems existing in the related technologies. The technical solution is as follows:

[0006] In a first aspect, the present application provides a platelet detection method based on deep learning, comprising:

[0007] Acquire a blood cell analysis data set of a whole blood sample, wherein the blood cell analysis data set of the whole blood sample includes blood cell analysis data of multiple whole blood samples and PLT count values ​​of a PLT-F channel, each of the blood cell analysis data includes a PLT histogram, an RBC histogram, and a blood cell detection value, the PLT histogram refers to a platelet histogram of a PLT-I channel, the RBC histogram refers to a red blood cell histogram of an RBC channel, the blood cell detection value includes various blood cell count values ​​of a PLT-O channel, and various blood cells include white blood cells WBC, red blood cells RBC, and platelet PLT;

[0008] Constructing a PLT-F fusion prediction model, wherein the input of the PLT-F fusion prediction model is the blood cell analysis data, and the output is the PLT prediction value of the corresponding PLT-F channel;

[0009] Training the PLT-F fusion prediction model based on the whole blood sample blood cell analysis data set to obtain a trained PLT-F fusion prediction model;

[0010] Acquiring the blood cell analysis data of the whole blood sample to be analyzed;

[0011] The blood cell analysis data of the whole blood sample to be analyzed is input into the trained PLT-F fusion prediction model to obtain a PLT-F prediction result, wherein the PLT-F prediction result includes the PLT prediction value of the whole blood sample to be analyzed on the PLT-F channel.

[0012] In one embodiment, the PLT-F fusion prediction model uses one of image stitching, feature stitching and Tensor stitching to perform feature fusion on the blood cell analysis data, where Tensor refers to a tensor.

[0013] In one embodiment, the PLT-F fusion prediction model uses image splicing to perform feature fusion on the blood cell analysis data; constructing the PLT-F fusion prediction model includes:

[0014] Constructing an image stitching module and a first PLT-F prediction module, wherein the image stitching module is used to stitch a blood cell value bar graph, a PLT histogram, and an RBC histogram in order from left to right to obtain a stitching image, the blood cell value bar graph is obtained after image processing of the blood cell detection value, and the first PLT-F prediction module is used to extract a first fusion feature of the stitching image, and predict the PLT value on the PLT-F channel based on the first fusion feature;

[0015] The image stitching module is used as an input module, and the first PLT-F prediction module is used as an output module to construct the PLT-F fusion prediction model.

[0016] In one embodiment, before each of the blood cell analysis data is input into the PLT-F fusion prediction model for training to obtain the trained PLT-F fusion prediction model, and before the blood cell analysis data of the whole blood sample to be analyzed is input into the trained PLT-F fusion prediction model to obtain the PLT-F prediction result, the method also includes: visualizing the blood cell detection values ​​in the blood cell analysis data in the form of a bar chart to obtain a corresponding blood cell value bar chart.

[0017] In one embodiment, the PLT-F fusion prediction model uses a feature splicing method to perform feature fusion on the blood cell analysis data; constructing the PLT-F fusion prediction model includes:

[0018] Constructing a feature extraction module, a feature splicing module and a second PLT-F prediction module, wherein the feature extraction module is used to extract a first feature of the blood cell detection value, a second feature of the PLT histogram and a third feature of the RBC histogram, the feature splicing module is used to splice the first feature, the second feature and the third feature to obtain a spliced ​​feature, and the second PLT-F prediction module is used to extract a second fusion feature of the spliced ​​feature, and predict the PLT value on the PLT-F channel based on the second fusion feature;

[0019] The feature extraction module is used as an input module, the feature splicing module is used as an intermediate module, and the second PLT-F prediction module is used as an output module to construct the PLT-F fusion prediction model.

[0020] In one embodiment, the PLT-F fusion prediction model uses Tensor splicing to perform feature fusion on the blood cell analysis data; constructing the PLT-F fusion prediction model includes:

[0021] Constructing a conversion module, a Tensor splicing module and a third PLT-F prediction module, wherein the conversion module is used to convert the blood cell detection value, the PLT histogram and the RBC histogram into three-channel Tensors respectively to obtain three Tensors in the same dimension, the Tensor splicing module is used to splice the three Tensors along a specified dimension to obtain a spliced ​​Tensor, and the third PLT-F prediction module is used to extract a third fusion feature of the spliced ​​Tensor, and predict the PLT value on the PLT-F channel based on the third fusion feature;

[0022] The PLT-F fusion prediction model is constructed by using the conversion module as the input module, the Tensor splicing module as the intermediate module, and the third PLT-F prediction module as the output module.

[0023] In one embodiment, the PLT-F fusion prediction model is trained based on the whole blood sample blood cell analysis data set to obtain a trained PLT-F fusion prediction model including:

[0024] The array consisting of the PLT count values ​​of the PLT-F channel in the whole blood sample blood cell analysis data set is used as a true PLT value group;

[0025] Dividing the whole blood sample blood cell analysis data set into a training set, a test set and a validation set;

[0026] Input each of the blood cell analysis data in the training set into the PLT-F fusion prediction model for training, and during the training process, adjust the network parameters of the PLT-F fusion prediction model according to the prediction results and the true PLT value group to obtain an adjusted PLT-F fusion prediction model;

[0027] Based on the validation set, a ten-fold cross validation method is used to evaluate the performance of the adjusted PLT-F fusion prediction model, and several evaluated PLT-F fusion prediction models are obtained, and the several evaluated PLT-F fusion prediction models correspond to different Pearson correlation coefficients;

[0028] The PLT-F fusion prediction model after target evaluation with the highest Pearson correlation coefficient was selected;

[0029] The PLT-F fusion prediction model after the target evaluation is tested based on the test set. When it is judged that the performance of the PLT-F fusion prediction model after the target evaluation meets the standard according to the test result, the PLT-F fusion prediction model after the target evaluation is used as the trained PLT-F fusion prediction model.

[0030] In a second aspect, the embodiment of the present application further provides a platelet detection device based on deep learning, comprising:

[0031] A first receiving unit is used to obtain a blood cell analysis data set of a whole blood sample, wherein the blood cell analysis data set of the whole blood sample includes blood cell analysis data of multiple whole blood samples and a PLT count value of a PLT-F channel, each of the blood cell analysis data includes a PLT histogram, an RBC histogram, and a blood cell detection value, the PLT histogram refers to a platelet histogram of a PLT-I channel, the RBC histogram refers to a red blood cell histogram of an RBC channel, the blood cell detection value includes various blood cell count values ​​of a PLT-O channel, and various blood cells include white blood cells WBC, red blood cells RBC, and platelet PLT;

[0032] A construction unit, used for constructing a PLT-F fusion prediction model, wherein the input of the PLT-F fusion prediction model is the blood cell analysis data, and the output is the PLT prediction value of the corresponding PLT-F channel;

[0033] A training unit, used for training the PLT-F fusion prediction model based on the whole blood sample blood cell analysis data set to obtain a trained PLT-F fusion prediction model;

[0034] A second receiving unit, used for acquiring the blood cell analysis data of the whole blood sample to be analyzed;

[0035] A prediction unit is used to input the blood cell analysis data of the whole blood sample to be analyzed into the trained PLT-F fusion prediction model to obtain a PLT-F prediction result, wherein the PLT-F prediction result includes the PLT prediction value of the whole blood sample to be analyzed on the PLT-F channel.

[0036] In one embodiment, the PLT-F fusion prediction model uses one of image stitching, feature stitching and Tensor stitching to perform feature fusion on the blood cell analysis data, where Tensor refers to a tensor.

[0037] In one embodiment, the PLT-F fusion prediction model uses image splicing to perform feature fusion on the blood cell analysis data; when the construction unit is used to construct the PLT-F fusion prediction model, it is specifically used to:

[0038] Constructing an image stitching module and a first PLT-F prediction module, wherein the image stitching module is used to stitch a blood cell value bar graph, a PLT histogram, and an RBC histogram in order from left to right to obtain a stitching image, the blood cell value bar graph is obtained after image processing of the blood cell detection value, and the first PLT-F prediction module is used to extract a first fusion feature of the stitching image, and predict the PLT value on the PLT-F channel based on the first fusion feature;

[0039] The image stitching module is used as an input module, and the first PLT-F prediction module is used as an output module to construct the PLT-F fusion prediction model.

[0040] In one embodiment, each of the blood cell analysis data is input into the PLT-F fusion prediction model for training. Before obtaining the trained PLT-F fusion prediction model, the training unit is further used to: image the blood cell detection values ​​in the blood cell analysis data in the form of a bar chart to obtain a corresponding blood cell value bar chart;

[0041] The blood cell analysis data of the whole blood sample to be analyzed is input into the trained PLT-F fusion prediction model. Before obtaining the PLT-F prediction result, the prediction unit is also used to: visualize the blood cell detection values ​​in the blood cell analysis data in the form of a bar chart to obtain a corresponding blood cell value bar chart.

[0042] In one embodiment, the PLT-F fusion prediction model uses a feature splicing method to perform feature fusion on the blood cell analysis data; when the construction unit is used to construct the PLT-F fusion prediction model, it is specifically used to:

[0043] Constructing a feature extraction module, a feature splicing module and a second PLT-F prediction module, wherein the feature extraction module is used to extract a first feature of the blood cell detection value, a second feature of the PLT histogram and a third feature of the RBC histogram, the feature splicing module is used to splice the first feature, the second feature and the third feature to obtain a spliced ​​feature, and the second PLT-F prediction module is used to extract a second fusion feature of the spliced ​​feature, and predict the PLT value on the PLT-F channel based on the second fusion feature;

[0044] The feature extraction module is used as an input module, the feature splicing module is used as an intermediate module, and the second PLT-F prediction module is used as an output module to construct the PLT-F fusion prediction model.

[0045] In one embodiment, the PLT-F fusion prediction model uses Tensor splicing to perform feature fusion on the blood cell analysis data; when the construction unit is used to construct the PLT-F fusion prediction model, it is specifically used to:

[0046] Constructing a conversion module, a Tensor splicing module and a third PLT-F prediction module, wherein the conversion module is used to convert the blood cell detection value, the PLT histogram and the RBC histogram into three-channel Tensors respectively to obtain three Tensors in the same dimension, the Tensor splicing module is used to splice the three Tensors along a specified dimension to obtain a spliced ​​Tensor, and the third PLT-F prediction module is used to extract a third fusion feature of the spliced ​​Tensor, and predict the PLT value on the PLT-F channel based on the third fusion feature;

[0047] The PLT-F fusion prediction model is constructed by using the conversion module as the input module, the Tensor splicing module as the intermediate module, and the third PLT-F prediction module as the output module.

[0048] In one embodiment, the training unit is used to train the PLT-F fusion prediction model based on the whole blood sample blood cell analysis data set to obtain the trained PLT-F fusion prediction model, and is specifically used to:

[0049] The array consisting of the PLT count values ​​of the PLT-F channel in the whole blood sample blood cell analysis data set is used as a true PLT value group;

[0050] Dividing the whole blood sample blood cell analysis data set into a training set, a test set and a validation set;

[0051] Input each of the blood cell analysis data in the training set into the PLT-F fusion prediction model for training, and during the training process, adjust the network parameters of the PLT-F fusion prediction model according to the prediction results and the true PLT value group to obtain an adjusted PLT-F fusion prediction model;

[0052] Based on the validation set, a ten-fold cross validation method is used to evaluate the performance of the adjusted PLT-F fusion prediction model, and several evaluated PLT-F fusion prediction models are obtained, and the several evaluated PLT-F fusion prediction models correspond to different Pearson correlation coefficients;

[0053] The PLT-F fusion prediction model after target evaluation with the highest Pearson correlation coefficient was selected;

[0054] The PLT-F fusion prediction model after the target evaluation is tested based on the test set. When it is judged that the performance of the PLT-F fusion prediction model after the target evaluation meets the standard according to the test result, the PLT-F fusion prediction model after the target evaluation is used as the trained PLT-F fusion prediction model.

[0055] In a third aspect, an embodiment of the present application further provides a computer device, comprising: a memory and a processor, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement a method in any one of the above-mentioned aspects, wherein the memory and the processor communicate with each other through an internal connection path.

[0056] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. When the computer program runs on a computer, the method in any one of the above-mentioned embodiments is implemented.

[0057] The advantages or beneficial effects of the above technical solution include at least:

[0058] The present application applies artificial intelligence technology (such as deep learning technology) to count platelets, which can integrate multimodal feature data, effectively correct poor platelet histograms, and quickly obtain true platelet values. It provides a screening method for specimens with poor platelet histograms, which is beneficial to shorten the blood cell analysis and detection time, thereby improving the detection efficiency of platelets and reducing the detection cost of platelets.

[0059] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present application will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in the present application and should not be regarded as limiting the scope of the present application.

[0061] Figure 1 A flowchart of a platelet detection method based on deep learning provided in an embodiment of the present application;

[0062] Figure 2 An example of a PLT histogram provided in an embodiment of the present application;

[0063] Figure 3 An example of an RBC histogram provided in an embodiment of the present application;

[0064] Figure 4 An example of a blood cell value bar graph provided in an embodiment of the present application;

[0065] Figure 5 An example of a mosaic image provided in an embodiment of the present application;

[0066] Figure 6 An example diagram of a PLT-F fusion prediction model provided in an embodiment of the present application;

[0067] Figure 7 An example diagram of another PLT-F fusion prediction model provided in an embodiment of the present application;

[0068] Figure 8 An example diagram of another PLT-F fusion prediction model provided in an embodiment of the present application;

[0069] Fig. 9 An example diagram of an evaluation result corresponding to an evaluated PLT-F fusion prediction model provided in an embodiment of the present application;

[0070] Fig.10 An example diagram of excluding small-volume red blood cells and red blood cell fragments by combining platelet and red blood cell features provided in an embodiment of the present application;

[0071] Fig.11 A structural block diagram of a platelet detection device based on deep learning provided in an embodiment of the present application;

[0072] Fig.12 A structural block diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0073] In the following, only some exemplary embodiments are briefly described. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the spirit or scope of the present application. Therefore, the drawings and descriptions are considered to be exemplary and non-restrictive in nature.

[0074] In the related technology, due to the complex composition of human blood, when counting platelets in routine blood tests, platelet histograms often show poor graphics, which is mainly due to the interference of small red blood cells, cell debris and other factors. In order to solve this problem, the current industry guidelines recommend the use of platelet fluorescence channels (PLT-F channels) to distinguish platelets from other blood components through nucleic acid dyes, but this method is time-consuming to review and the cost of nucleic acid dyes is high, resulting in low platelet detection efficiency and high detection costs.

[0075] Based on this, the embodiments of the present application provide a corresponding platelet detection solution based on deep learning. By applying artificial intelligence technology (such as deep learning technology) to count platelets, it is possible to effectively correct poor platelet histogram graphics, quickly obtain true platelet values, and provide a screening method for specimens with poor platelet histogram graphics, thereby improving the detection efficiency of platelets and reducing the detection cost of platelets.

[0076] The technical solution provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0077] Figure 1 FIG. 1 is a flow chart of a platelet detection method based on deep learning according to an embodiment of the present application. Figure 1 As shown, the method may include the following steps:

[0078] S110, obtaining a blood cell analysis data set of a whole blood sample.

[0079] In one embodiment, the whole blood sample blood cell analysis data set may include but is not limited to: blood cell analysis data of multiple whole blood samples and PLT count values ​​of PLT-F channel, and each blood cell analysis data may include but is not limited to: PLT histogram, RBC histogram and blood cell detection value. Among them, PLT histogram refers to the platelet histogram of PLT-I channel, RBC histogram refers to the red blood cell histogram of RBC channel, and blood cell detection value may include various blood cell count values ​​of PLT-O channel, and various blood cells include white blood cells (White blood cell, WBC), red blood cells (Red blood cell, RBC) and platelets (Platelet, PLT).

[0080] Exemplarily, the PLT histogram can be as follows Figure 2As shown, the RBC histogram can be Figure 3 shown.

[0081] As an example, a re-examination sample database for the fluorescence method (PLT-F) channel, a re-examination sample database for the optical method (PLT-O) channel, a re-examination sample database for the electrical impedance method (PLT-I) channel, and a re-examination sample database for the RBC channel can be established. Afterwards, the blood cell analysis results of each whole blood sample that has been re-examined can be collected, and then the corresponding blood cell analysis result data of each whole blood sample can be correspondingly saved in these re-examination sample databases, so that the whole blood sample blood cell analysis data set can be directly obtained later.

[0082] In a large number of blood routine test specimens, platelet graphs often show irregularities. Based on this, through this example, the corrected platelet count results can be directly obtained.

[0083] Furthermore, since the current industry standards recommend platelet re-examination and blood cell smear microscopy. Based on this, in the process of saving the corresponding blood cell analysis result data of each whole blood sample in these re-examination specimen databases in this example, the re-examination platelet count result can be compared with the historical analysis result of the blood transfusion recipient; if it is found that the re-examination platelet count result is significantly different from the historical analysis result of the blood transfusion recipient, the same whole blood sample is subjected to blood smear microscopy, and then the re-examination platelet count result is corrected based on the blood smear microscopy result, and the corrected platelet count result is saved in these re-examination specimen databases; or, if it is found that the re-examination platelet count result is slightly different from the historical analysis result of the blood transfusion recipient, it is judged that the review is passed, and the re-examination platelet count result is directly saved in these re-examination specimen databases.

[0084] That is, through this example, the embodiment of the present application can provide a better review strategy to cope with the detection of clinical abnormal platelet specimens.

[0085] S120. Construct a PLT-F fusion prediction model.

[0086] In one embodiment, the input of the PLT-F fusion prediction model is blood cell analysis data, and the output is the PLT prediction value corresponding to the PLT-F channel.

[0087] In one embodiment, the PLT-F fusion prediction model can use but is not limited to one of image stitching, feature stitching and Tensor stitching to perform feature fusion on blood cell analysis data, wherein Tensor refers to tensor.

[0088] In one embodiment, when the PLT-F fusion prediction model uses image splicing to perform feature fusion on blood cell analysis data, the implementation process of step S120 may include the following steps:

[0089] S121, constructing an image splicing module and a first PLT-F prediction module.

[0090] In a specific implementation, the image stitching module is used to stitch the blood cell value bar graph, PLT histogram and RBC histogram in order from left to right to obtain a stitching image. The blood cell value bar graph is obtained by image processing of the blood cell detection value. The first PLT-F prediction module is used to extract the first fusion feature of the stitching image, and predict the PLT value on the PLT-F channel based on the first fusion feature.

[0091] As an example, if the PLT count value of the PLT-O channel in the blood cell analysis data is 500, and the maximum value of the platelet count test results of other channels in the blood cell analysis data is 1500, when the image height is set to 200px, the PLT count value of the PLT-O channel can be plotted as a bar histogram with a pixel height of (500 / 1500)*200px, which is similar to Figure 4 The left bar histogram in .

[0092] That is, in this example, it can be understood that the blood cell value bar graph is a graphical presentation of the blood cell count values ​​of each type of PLT-O channel in the form of a bar graph, that is, the count value of each type of blood cell is presented as a bar histogram, and the height of the bar histogram represents the specific value of the count value of each type of blood cell.

[0093] In specific implementation, Figure 2-Figure 4 Taking the PLT histogram, RBC histogram and blood cell value bar graph as an example, the image stitching module can sequentially stitch the Figure 4 , Figure 2 and Figure 3 Splice and get Figure 5 The mosaic shown.

[0094] As an example, the first PLT-F prediction module can be implemented using a first deep neural network. The first deep neural network can be implemented using a convolutional neural network, a Transformer network, etc. For example, it can also be implemented using a VGG network, a ResNet network, or an Incepition network. The Transformer network can include a Vision Transformer network, a Win Transformer network, etc.

[0095] Exemplarily, the first deep neural network may include but is not limited to: a first convolution layer, a residual module, a pooling layer and a linear layer, the residual module includes a plurality of residual blocks, and each residual block includes a plurality of second convolution layers. Among them, the first convolution layer is used to perform a first convolution on the first fusion feature to obtain a first convolution result. The residual module is used to perform a second convolution on the first convolution result to obtain a second convolution result. The pooling layer is used to perform global average pooling on the second convolution result to obtain a global average pooling result. The linear layer is used to map the global average pooling result to obtain a prediction result.

[0096] S122. Use the image stitching module as an input module and the first PLT-F prediction module as an output module to construct the PLT-F fusion prediction model.

[0097] As an example, in this embodiment, the framework of the PLT-F fusion prediction model can be as follows Figure 6 shown.

[0098] In another embodiment, when the PLT-F fusion prediction model uses feature splicing to perform feature fusion on blood cell analysis data, the implementation process of step S120 may include the following steps:

[0099] S121a, construct a feature extraction module, a feature splicing module and a second PLT-F prediction module.

[0100] In a specific implementation, the feature extraction module is used to extract the first feature of the blood cell detection value, the second feature of the PLT histogram, and the third feature of the RBC histogram. The feature splicing module is used to splice the first feature, the second feature, and the third feature to obtain a spliced ​​feature. The second PLT-F prediction module is used to extract the second fusion feature of the spliced ​​feature, and predict the PLT value on the PLT-F channel based on the second fusion feature.

[0101] As an example, the feature extraction module may be composed of, but not limited to, a second deep neural network, a third deep neural network, and a first multilayer perceptron (MLP). The first MLP is used to extract the first feature of the blood cell detection value, the second deep neural network is used to extract the second feature of the PLT histogram, and the third deep neural network is used to extract the third feature of the RBC histogram.

[0102] Exemplarily, the second deep neural network and the third deep neural network can both be implemented as existing convolutional neural networks, Transformer networks, etc. In specific implementation, the second deep neural network and the third deep neural network can also be implemented using existing vgg networks, resnet networks, or incepition networks, which are not described in detail in the embodiments of the present application. Among them, the second deep neural network and the third deep neural network can be set the same or different, which is not limited in the embodiments of the present application.

[0103] Exemplarily, the first MLP may be implemented by an existing MLP, and the embodiments of the present application will not be described in detail herein.

[0104] As an example, the feature concatenation module can be used to concatenate the first feature, the second feature, and the third feature using the torch.cat function.

[0105] As an example, the second PLT-F prediction module is implemented by using a second MLP, wherein the second MLP can be implemented by using an existing MLP, which will not be described in detail in the embodiment of the present application.

[0106] In a specific implementation, the second MLP and the first MLP may be configured to be the same or different, and this embodiment of the present application does not limit this.

[0107] S122a, using the feature extraction module as an input module, the feature splicing module as an intermediate module, and the second PLT-F prediction module as an output module to construct the PLT-F fusion prediction model.

[0108] As an example, in this embodiment, the PLT-F fusion prediction model can be as follows Figure 7 shown.

[0109] In another embodiment, when the PLT-F fusion prediction model uses Tensor splicing to perform feature fusion on blood cell analysis data, the implementation process of step S120 may include the following steps:

[0110] S121A, construct conversion module, Tensor splicing module and the third PLT-F prediction module.

[0111] In specific implementation, the conversion module is used to convert the blood cell detection value, PLT histogram and RBC histogram into three-channel Tensors respectively to obtain three Tensors in the same dimension. The Tensor splicing module is used to splice the three Tensors along the specified dimension to obtain a spliced ​​Tensor. The third PLT-F prediction module is used to extract the third fusion feature of the spliced ​​Tensor and predict the PLT value on the PLT-F channel based on the third fusion feature.

[0112] As an example, the conversion module can use the Tensor function to convert the PLT histogram and the RBC histogram into three-channel (3, width, height) Tensors to obtain two Tensors (these two Tensors can be called image Tensors), and the blood cell detection values ​​can be expanded into a Tensor with the same height or width as the first two Tensors through Tensor.expand (this Tensor can be called a data Tensor), that is, the Tensor can be (3, width, Num) or (3, Num, height), so that the three converted Tensors are in the same dimension.

[0113] As an example, the third PLT-F prediction module can be implemented using a fourth deep neural network. The fourth deep neural network can be implemented in the same or similar manner as the first deep neural network, and the embodiments of the present application will not be repeated here.

[0114] S122A, using the conversion module as the input module, the Tensor splicing module as the intermediate module, and the third PLT-F prediction module as the output module to construct the PLT-F fusion prediction model.

[0115] As an example, in this embodiment, the PLT-F fusion prediction model can be as follows Figure 8 shown.

[0116] In an embodiment of the present application, by executing step S120, it is possible to extract the fusion features of the PLT and RBC-related results of each whole blood sample through the PLT-F fusion prediction model, and predict the PLT value of each whole blood sample on the PLT-F channel based on the fusion features, so that the PLT-related data on the PLT-O channel, PLT-I channel and RBC channel can be combined to predict the corresponding PLT count value on the PLT-I channel.

[0117] S130. Train the PLT-F fusion prediction model based on the blood cell analysis data set of the whole blood sample to obtain a trained PLT-F fusion prediction model.

[0118] In one implementation, the implementation process of step 130 may include the following steps:

[0119] S131. Use the array consisting of the PLT count values ​​of the PLT-F channel in the whole blood sample blood cell analysis data set as the real PLT value group.

[0120] S132. Divide the whole blood sample blood cell analysis data set into a training set, a test set, and a validation set.

[0121] In specific implementation, the whole blood sample blood cell analysis data set can be divided into a training set, a test set, and a validation set in a corresponding ratio according to actual needs. For example, the whole blood sample blood cell analysis data set can be divided into a training set, a test set, and a validation set in a ratio of 70%, 15%, and 15%.

[0122] S133. Input each blood cell analysis data in the training set into the PLT-F fusion prediction model for training. During the training process, adjust the network parameters of the PLT-F fusion prediction model according to the prediction results and the real PLT value group to obtain the adjusted PLT-F fusion prediction model.

[0123] For example, each blood cell analysis data in the training set is input into the PLT-F fusion prediction model to obtain the corresponding prediction result, and then the corresponding prediction result is compared with the real PLT value in the real PLT value group corresponding to the training set. If the difference is large, the network parameters of the PLT-F fusion prediction model are adjusted.

[0124] S134. Based on the validation set, a ten-fold cross-validation method is used to evaluate the performance of the adjusted PLT-F fusion prediction model to obtain several evaluated PLT-F fusion prediction models.

[0125] In specific implementation, several evaluated PLT-F fusion prediction models correspond to different Pearson correlation coefficients.

[0126] As an example, the evaluation results of a PLT-F fusion prediction model after evaluation can be as follows: Fig. 9 shown.

[0127] By executing step S134, it is convenient to select a PLT-F fusion prediction model with better performance.

[0128] S135. Select the PLT-F fusion prediction model after target evaluation with the highest Pearson correlation coefficient.

[0129] That is, the evaluated PLT-F fusion prediction model with the highest Pearson correlation coefficient can be selected from several evaluated PLT-F fusion prediction models as the target evaluated PLT-F fusion prediction model.

[0130] S136. Based on the test set, the PLT-F fusion prediction model after target evaluation is tested. When the performance of the PLT-F fusion prediction model after target evaluation is judged to be up to standard according to the test result, the PLT-F fusion prediction model after target evaluation is used as the trained PLT-F fusion prediction model.

[0131] As an example, if the accuracy of the PLT-F fusion prediction model after the target evaluation is judged to be above the preset value according to the test results, it can be judged that its performance meets the standard. The preset value can be set according to actual needs, such as 96%, which is not limited in the embodiment of the present application.

[0132] In one embodiment, in step S130, if the PLT-F fusion prediction model uses image stitching to perform feature fusion on the blood cell analysis data, then before each blood cell analysis data is input into the PLT-F fusion prediction model, the blood cell detection values ​​in each blood cell analysis data can be imaged in the form of a bar chart to obtain a corresponding blood cell value bar chart.

[0133] In the embodiment of the present application, by executing step S130, a PLT-F fusion prediction model with better performance can be obtained, which is convenient for improving the detection efficiency of platelets and reducing the detection cost of platelets.

[0134] S140, obtaining blood cell analysis data of the whole blood sample to be analyzed.

[0135] In a specific implementation, the blood cell analysis data of the whole blood sample to be analyzed is the same as or similar to the blood cell analysis data in the above-mentioned whole blood sample blood cell analysis data set, and the embodiments of the present application will not be described in detail here.

[0136] S150, inputting the blood cell analysis data of the whole blood sample to be analyzed into the trained PLT-F fusion prediction model to obtain the PLT-F prediction result.

[0137] In one embodiment, the PLT-F prediction result may include the PLT prediction value of the whole blood sample to be analyzed on the PLT-F channel.

[0138] In one embodiment, in step S150, if the PLT-F fusion prediction model uses image stitching to perform feature fusion on the blood cell analysis data, then before the blood cell analysis data of the whole blood sample to be analyzed is input into the trained PLT-F fusion prediction model, the blood cell detection values ​​in the blood cell analysis data of the whole blood sample to be analyzed can be imaged in the form of a bar chart to obtain a corresponding blood cell value bar chart.

[0139] In the embodiment of the present application, by executing step S150, the platelet count test result of the whole blood sample to be analyzed on the PLT-O channel, the platelet histogram on the PLT-I channel, and the features of the red blood cell histogram on the RBC channel can be extracted, and then feature fusion can be performed to obtain the fusion feature of the PLT and RBC related results of the whole blood sample to be analyzed. In this way, the PLT prediction value of the whole blood sample to be analyzed on the PLT-F channel is predicted by the fusion feature, such as Fig.10 As shown, small-volume red blood cells and red blood cell fragments can be excluded by combining the platelet and red blood cell characteristics, and the true PLT value of the sample with a bad PLT-I histogram can be predicted, which is beneficial to improving detection efficiency, reducing costs and shortening blood cell analysis detection time.

[0140] In summary, the platelet detection method based on deep learning provided in the embodiment of the present application, by applying artificial intelligence technology (such as deep learning technology) to count platelets, can fuse multimodal feature data, can effectively correct poor platelet histograms, and realize rapid acquisition of true platelet values. It provides a screening method for specimens with poor platelet histograms, which is beneficial to shortening the blood cell analysis and detection time, thereby improving the detection efficiency of platelets and reducing the detection cost of platelets.

[0141] In practical applications, the prediction results of this deep learning-based platelet detection method are highly consistent with the detection results of the PLT-F instrument method, and can replace the PLT-F instrument method.

[0142] Fig.11 FIG. 1 is a block diagram of a platelet detection device based on deep learning according to an embodiment of the present application. Fig.11 As shown, the device may include:

[0143] The first receiving unit 210 is used to obtain a blood cell analysis data set of a whole blood sample, wherein the blood cell analysis data set of a whole blood sample includes PLT count values ​​of a PLT-F channel of blood cell analysis data of a plurality of whole blood samples, each blood cell analysis data includes a PLT histogram, an RBC histogram and a blood cell detection value, the PLT histogram refers to a platelet histogram of a PLT-I channel, the RBC histogram refers to a red blood cell histogram of an RBC channel, the blood cell detection value includes various blood cell count values ​​of a PLT-O channel, and various blood cells include white blood cells WBC, red blood cells RBC and platelet PLT;

[0144] A construction unit 220 is used to construct a PLT-F fusion prediction model, the input of the PLT-F fusion prediction model is the blood cell analysis data, and the output is the PLT prediction value of the corresponding PLT-F channel;

[0145] A training unit 230, for training the PLT-F fusion prediction model based on the whole blood sample blood cell analysis data set to obtain a trained PLT-F fusion prediction model;

[0146] The second receiving unit 240 is used to obtain blood cell analysis data of the whole blood sample to be analyzed;

[0147] The prediction unit 250 is used to input the blood cell analysis data of the whole blood sample to be analyzed into the trained PLT-F fusion prediction model to obtain a PLT-F prediction result, which includes the PLT prediction value of the whole blood sample to be analyzed on the PLT-F channel.

[0148] In one embodiment, the PLT-F fusion prediction model uses one of the image stitching method, feature stitching method and Tensor stitching method to perform feature fusion on the blood cell analysis data, where Tensor refers to a tensor.

[0149] In one embodiment, the PLT-F fusion prediction model uses image splicing to perform feature fusion on blood cell analysis data; when the construction unit 220 is used to construct the PLT-F fusion prediction model, it is specifically used to:

[0150] Constructing an image stitching module and a first PLT-F prediction module, wherein the image stitching module is used to stitch the blood cell value histogram, the PLT histogram and the RBC histogram in order from left to right to obtain a stitching image, the blood cell value histogram is obtained after the blood cell detection value is image-processed, and the first PLT-F prediction module is used to extract a first fusion feature of the stitching image, and predict the PLT value on the PLT-F channel based on the first fusion feature;

[0151] The image stitching module is used as an input module, and the first PLT-F prediction module is used as an output module to construct a PLT-F fusion prediction model.

[0152] In one embodiment, each blood cell analysis data is input into the PLT-F fusion prediction model for training. Before obtaining the trained PLT-F fusion prediction model, the training unit 230 is further used to: image the blood cell detection values ​​in the blood cell analysis data in the form of a bar chart to obtain a corresponding blood cell value bar chart;

[0153] The blood cell analysis data of the whole blood sample to be analyzed is input into the trained PLT-F fusion prediction model. Before obtaining the PLT-F prediction result, the prediction unit 250 is also used to: visualize the blood cell detection values ​​in the blood cell analysis data in the form of a bar chart to obtain a corresponding blood cell value bar chart.

[0154] In one embodiment, the PLT-F fusion prediction model uses a feature splicing method to perform feature fusion on blood cell analysis data; when the construction unit 220 is used to construct the PLT-F fusion prediction model, it is specifically used to:

[0155] Constructing a feature extraction module, a feature splicing module and a second PLT-F prediction module, wherein the feature extraction module is used to extract a first feature of a blood cell detection value, a second feature of a PLT histogram and a third feature of a RBC histogram, the feature splicing module is used to splice the first feature, the second feature and the third feature to obtain a spliced ​​feature, and the second PLT-F prediction module is used to extract a second fusion feature of the spliced ​​feature, and predict the PLT value on the PLT-F channel based on the second fusion feature;

[0156] The feature extraction module is used as the input module, the feature splicing module is used as the intermediate module, and the second PLT-F prediction module is used as the output module to construct a PLT-F fusion prediction model.

[0157] In one embodiment, the PLT-F fusion prediction model uses Tensor splicing to perform feature fusion on blood cell analysis data; when the construction unit 220 is used to construct the PLT-F fusion prediction model, it is specifically used to:

[0158] Constructing a conversion module, a Tensor splicing module and a third PLT-F prediction module, wherein the conversion module is used to convert the blood cell detection value, the PLT histogram and the RBC histogram into three-channel Tensors respectively to obtain three Tensors in the same dimension, the Tensor splicing module is used to splice the three Tensors along the specified dimension to obtain a spliced ​​Tensor, and the third PLT-F prediction module is used to extract the third fusion feature of the spliced ​​Tensor, and predict the PLT value on the PLT-F channel based on the third fusion feature;

[0159] The conversion module is used as the input module, the Tensor splicing module is used as the intermediate module, and the third PLT-F prediction module is used as the output module to construct a PLT-F fusion prediction model.

[0160] In one embodiment, the training unit 230 is used to train the PLT-F fusion prediction model based on the whole blood sample blood cell analysis data set to obtain the trained PLT-F fusion prediction model, and is specifically used to:

[0161] The array consisting of the PLT count values ​​of the PLT-F channel in the whole blood sample blood cell analysis data set is used as the real PLT value group;

[0162] The whole blood sample blood cell analysis dataset is divided into a training set, a test set, and a validation set;

[0163] Each blood cell analysis data in the training set is input into the PLT-F fusion prediction model for training. During the training process, the network parameters of the PLT-F fusion prediction model are adjusted according to the prediction results and the real PLT value group to obtain the adjusted PLT-F fusion prediction model;

[0164] Based on the validation set, the performance of the adjusted PLT-F fusion prediction model was evaluated using the ten-fold cross-validation method, and several evaluated PLT-F fusion prediction models were obtained. The several evaluated PLT-F fusion prediction models corresponded to different Pearson correlation coefficients.

[0165] The PLT-F fusion prediction model after target evaluation with the highest Pearson correlation coefficient was selected;

[0166] The PLT-F fusion prediction model after target evaluation is tested based on the test set. When the performance of the PLT-F fusion prediction model after target evaluation is judged to be up to standard according to the test results, the PLT-F fusion prediction model after target evaluation is used as the trained PLT-F fusion prediction model.

[0167] The functions of each unit in the deep learning-based platelet detection device of the embodiment of the present application can be found in the corresponding description in the above method and will not be repeated here.

[0168] Fig.12 FIG. 2 shows a block diagram of a computer device according to an embodiment of the present application. Fig.12 As shown, the computer device includes: a memory 310 and a processor 320. The memory 310 stores instructions, and the instructions are loaded and executed by the processor 320 to implement the platelet detection method based on deep learning in the above embodiment. The number of the memory 310 and the processor 320 can be one or more.

[0169] The computer device also includes:

[0170] The communication interface 330 is used to communicate with external devices and perform data exchange transmission.

[0171] If the memory 310, the processor 320 and the communication interface 330 are implemented independently, the memory 310, the processor 320 and the communication interface 330 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.12 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0172] Optionally, in a specific implementation, if the memory 310, the processor 320 and the communication interface 330 are integrated on a chip, the memory 310, the processor 320 and the communication interface 330 can communicate with each other through an internal interface.

[0173] An embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program runs on a computer, the method provided in the embodiment of the present application is implemented.

[0174] An embodiment of the present application also provides a chip, which includes a processor for calling and executing instructions stored in a memory, so that a communication device equipped with the chip executes the method provided by the embodiment of the present application.

[0175] An embodiment of the present application also provides a chip, including: an input interface, an output interface, a processor and a memory, wherein the input interface, the output interface, the processor and the memory are connected via an internal connection path, and the processor is used to execute the code in the memory. When the code is executed, the processor is used to execute the method provided in the embodiment of the application.

[0176] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. It is worth noting that the processor may be a processor supporting the Advanced RISC Machines (ARM) architecture.

[0177] Further, optionally, the above-mentioned memory may include a read-only memory and a random access memory, and may also include a non-volatile random access memory. The memory may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may include a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may include a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available. For example, static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct rambus RAM (DR RAM).

[0178] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.

[0179] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0180] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0181] Any process or method description in the flow chart or otherwise described herein can be understood to represent a module, fragment or portion of a code including one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0182] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0183] It should be understood that the various parts of the present application can be implemented with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium, and when the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0184] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk or an optical disk, etc.

[0185] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A platelet detection method based on deep learning, characterized in that: include: Acquire a blood cell analysis data set of a whole blood sample, wherein the blood cell analysis data set of the whole blood sample includes blood cell analysis data of multiple whole blood samples and PLT count values ​​of a PLT-F channel, each of the blood cell analysis data includes a PLT histogram, an RBC histogram, and a blood cell detection value, the PLT histogram refers to a platelet histogram of a PLT-I channel, the RBC histogram refers to a red blood cell histogram of an RBC channel, the blood cell detection value includes various blood cell count values ​​of a PLT-O channel, and various blood cells include white blood cells WBC, red blood cells RBC, and platelet PLT; Constructing a PLT-F fusion prediction model, wherein the input of the PLT-F fusion prediction model is the blood cell analysis data, and the output is the PLT prediction value of the corresponding PLT-F channel; Training the PLT-F fusion prediction model based on the whole blood sample blood cell analysis data set to obtain a trained PLT-F fusion prediction model; Acquiring the blood cell analysis data of the whole blood sample to be analyzed; The blood cell analysis data of the whole blood sample to be analyzed is input into the trained PLT-F fusion prediction model to obtain a PLT-F prediction result, wherein the PLT-F prediction result includes the PLT prediction value of the whole blood sample to be analyzed on the PLT-F channel.

2. The method according to claim 1, characterized in that The PLT-F fusion prediction model uses one of an image splicing method, a feature splicing method and a Tensor splicing method to perform feature fusion on the blood cell analysis data, where Tensor refers to a tensor.

3. The method according to claim 2, characterized in that The PLT-F fusion prediction model uses image splicing to perform feature fusion on the blood cell analysis data; constructing the PLT-F fusion prediction model includes: Constructing an image stitching module and a first PLT-F prediction module, wherein the image stitching module is used to stitch a blood cell value bar graph, a PLT histogram, and an RBC histogram in order from left to right to obtain a stitching image, the blood cell value bar graph is obtained after image processing of the blood cell detection value, and the first PLT-F prediction module is used to extract a first fusion feature of the stitching image, and predict the PLT value on the PLT-F channel based on the first fusion feature; The image stitching module is used as an input module, and the first PLT-F prediction module is used as an output module to construct the PLT-F fusion prediction model.

4. The method according to claim 3, characterized in that Before inputting each of the blood cell analysis data into the PLT-F fusion prediction model for training to obtain the trained PLT-F fusion prediction model, and before inputting the blood cell analysis data of the whole blood sample to be analyzed into the trained PLT-F fusion prediction model to obtain the PLT-F prediction result, the method also includes: visualizing the blood cell detection values ​​in the blood cell analysis data in the form of a bar chart to obtain a corresponding blood cell value bar chart.

5. The method according to claim 2, characterized in that: The PLT-F fusion prediction model uses a feature splicing method to perform feature fusion on the blood cell analysis data; constructing the PLT-F fusion prediction model includes: Constructing a feature extraction module, a feature splicing module and a second PLT-F prediction module, wherein the feature extraction module is used to extract a first feature of the blood cell detection value, a second feature of the PLT histogram and a third feature of the RBC histogram, the feature splicing module is used to splice the first feature, the second feature and the third feature to obtain a spliced ​​feature, and the second PLT-F prediction module is used to extract a second fusion feature of the spliced ​​feature, and predict the PLT value on the PLT-F channel based on the second fusion feature; The feature extraction module is used as an input module, the feature splicing module is used as an intermediate module, and the second PLT-F prediction module is used as an output module to construct the PLT-F fusion prediction model.

6. The method according to claim 2, characterized in that The PLT-F fusion prediction model uses Tensor splicing to perform feature fusion on the blood cell analysis data; constructing the PLT-F fusion prediction model includes: Constructing a conversion module, a Tensor splicing module and a third PLT-F prediction module, wherein the conversion module is used to convert the blood cell detection value, the PLT histogram and the RBC histogram into three-channel Tensors respectively to obtain three Tensors in the same dimension, the Tensor splicing module is used to splice the three Tensors along a specified dimension to obtain a spliced ​​Tensor, and the third PLT-F prediction module is used to extract a third fusion feature of the spliced ​​Tensor, and predict the PLT value on the PLT-F channel based on the third fusion feature; The PLT-F fusion prediction model is constructed by using the conversion module as the input module, the Tensor splicing module as the intermediate module, and the third PLT-F prediction module as the output module.

7. The method according to any one of claims 1 to 6, characterized in that The PLT-F fusion prediction model is trained based on the whole blood sample blood cell analysis data set to obtain a trained PLT-F fusion prediction model including: The array consisting of the PLT count values ​​of the PLT-F channel in the whole blood sample blood cell analysis data set is used as a true PLT value group; Dividing the whole blood sample blood cell analysis data set into a training set, a test set and a validation set; Input each of the blood cell analysis data in the training set into the PLT-F fusion prediction model for training, and during the training process, adjust the network parameters of the PLT-F fusion prediction model according to the prediction results and the true PLT value group to obtain an adjusted PLT-F fusion prediction model; Based on the validation set, a ten-fold cross validation method is used to evaluate the performance of the adjusted PLT-F fusion prediction model, and several evaluated PLT-F fusion prediction models are obtained, and the several evaluated PLT-F fusion prediction models correspond to different Pearson correlation coefficients; The PLT-F fusion prediction model after target evaluation with the highest Pearson correlation coefficient was selected; The PLT-F fusion prediction model after the target evaluation is tested based on the test set. When it is judged that the performance of the PLT-F fusion prediction model after the target evaluation meets the standard according to the test result, the PLT-F fusion prediction model after the target evaluation is used as the trained PLT-F fusion prediction model.

8. A platelet detection device based on deep learning, characterized in that: include: A first receiving unit is used to obtain a blood cell analysis data set of a whole blood sample, wherein the blood cell analysis data set of the whole blood sample includes blood cell analysis data of multiple whole blood samples and PLT count values ​​of a PLT-F channel, each of the blood cell analysis data includes a PLT histogram, an RBC histogram and a blood cell detection value, the PLT histogram refers to a platelet histogram of a PLT-I channel, the RBC histogram refers to a red blood cell histogram of an RBC channel, the blood cell detection value includes various blood cell count values ​​of a PLT-O channel, and various blood cells include white blood cells WBC, red blood cells RBC and platelet PLT; A construction unit, used for constructing a PLT-F fusion prediction model, wherein the input of the PLT-F fusion prediction model is the blood cell analysis data, and the output is the PLT prediction value of the corresponding PLT-F channel; A training unit, used for training the PLT-F fusion prediction model based on the whole blood sample blood cell analysis data set to obtain a trained PLT-F fusion prediction model; A second receiving unit, used for acquiring the blood cell analysis data of the whole blood sample to be analyzed; A prediction unit is used to input the blood cell analysis data of the whole blood sample to be analyzed into the trained PLT-F fusion prediction model to obtain a PLT-F prediction result, wherein the PLT-F prediction result includes the PLT prediction value of the whole blood sample to be analyzed on the PLT-F channel.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed on a computer, the method according to any one of claims 1 to 7 is implemented.

Images