A blood cell classification and counting method, electronic device, computer readable storage medium and program product
By using a blood cell counting and classification model, and employing sparse convolution and fully connected layers to process voxel parameters in blood slide preparation, the problem of long detection cycles and errors caused by the step-by-step process of blood cell counting and classification is solved, enabling rapid and accurate classification and counting of blood cells.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE SECOND AFFILIATED HOSPITAL ARMY MEDICAL UNIV
- Filing Date
- 2026-04-21
- Publication Date
- 2026-07-21
AI Technical Summary
Existing blood cell counting and classification methods involve two steps, resulting in long testing cycles and inconsistent results, especially prone to errors when cells overlap.
A blood cell counting and classification model is adopted. By obtaining the voxel parameters and three-dimensional feature parameters of blood slides, sparse convolution operation and feature extraction are performed. The cell classification and counting are integrated by combining sparse convolution with fully connected layers. The three-dimensional feature parameters are used to avoid errors caused by cell overlap.
It integrates blood cell classification and counting, shortens the testing cycle, improves classification accuracy, avoids detection errors caused by cell overlap, and enhances the speed and accuracy of testing.
Smart Images
Figure CN122432765A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of blood cell classification and counting technology, and more specifically, to a blood cell classification and counting method, electronic device, computer-readable storage medium, and program product. Background Technology
[0002] Blood cells are the main formed elements in blood, and their classification and counting are the core content of clinical routine blood tests, playing a vital role in assessing human health and assisting in disease diagnosis and monitoring. Among existing testing methods, fully automated blood analyzers are the mainstream approach. While this method is relatively accurate in counting blood cells, it is less effective in accurately typing cells. To achieve accurate cell typing, two common practices exist: one is manual blood smearing, staining, and classification; the other is automated smearing, staining, and classification by certain machines (such as Mindray workstations), followed by manual review. However, the drawback of both methods is that cell counting and classification require manual review, especially regarding common phenomena like cell overlap, which significantly increases reporting time and may lead to inconsistencies between the counter's classification and the final classification results. How to integrate blood cell counting and classification into a single output, resolve detection errors caused by cell overlap, improve the accuracy of blood cell classification and counting, and make it suitable for rapid clinical testing is a pressing issue that needs to be addressed. Summary of the Invention
[0003] In view of this, the purpose of the embodiments of this application is to provide a blood cell classification and counting method, an electronic device, a computer-readable storage medium and a program product method, which can improve the problems of blood cell classification and counting not being integrated, cell overlap leading to deviation in test results and long clinical testing cycle.
[0004] To achieve the above technical objectives, the technical solution adopted in this application is as follows:
[0005] In a first aspect, embodiments of this application provide a method for blood cell classification and counting, the method comprising:
[0006] Obtain voxel parameters for standard blood slides;
[0007] Based on the voxel parameters of the standard blood slide, the three-dimensional feature parameters of the blood slide are obtained.
[0008] The voxel parameters and the three-dimensional feature parameters are input into the blood cell counting classification model. Based on the voxel parameters, sparse convolution operation is performed through the blood cell counting classification model to extract features from the voxel parameters and obtain voxel data.
[0009] Based on the voxel data and the three-dimensional feature parameters, a first result is obtained, wherein the first result is used to classify and count blood cells.
[0010] According to the first aspect, the blood cell counting classification model includes: a preprocessing module, a 3D sparse feature extraction module, and a task module;
[0011] Based on the voxel parameters, sparse convolution operations are performed using the blood cell counting classification model to extract features from the voxel parameters, resulting in voxel data, including:
[0012] The voxel parameters are input into the preprocessing module. Based on the voxel parameters, the preprocessing module performs normalization and sparse encoding to obtain a sparse tensor.
[0013] The sparse tensor is input into the 3D sparse feature extraction module, and sparse convolution operation is performed through the 3D sparse feature extraction module to obtain the voxel data.
[0014] According to the first aspect, the task module includes: a first branch and a second branch.
[0015] The process of obtaining a first result based on the voxel data and the three-dimensional feature parameters includes:
[0016] The voxel data is input into the first branch, and a first set and a second set are output through the first branch. The first set includes N three-dimensional masks, and the second set includes N independent cells corresponding to the N three-dimensional masks.
[0017] The three-dimensional mask includes the coordinates of all voxels of the corresponding individual cells, where N is a positive integer;
[0018] The voxel data, the first set, and the three-dimensional feature parameters are input into the second branch. Based on the voxel data, the first set, and the three-dimensional feature parameters, the second branch performs a classification operation on all independent cells in the second set to obtain a first result, which includes type labels for N independent cells.
[0019] According to the first aspect, when the classification operation is performed on all independent cells in the second set, the classification operation includes:
[0020] Based on the i-th three-dimensional mask corresponding to the i-th independent cell, k feature vectors are extracted from the voxel data, where i = 1, 2, 3…N, and k is the number of voxels contained in the i-th independent cell;
[0021] Based on the k feature vectors and the three-dimensional feature parameters, a fused feature vector is obtained through feature fusion;
[0022] Based on the fused feature vector, it is mapped to M predicted scores through a fully connected layer, where M is the number of blood cell categories;
[0023] Based on the M predicted scores, classification prediction is performed using a preset softmax function to obtain the category confidence of the i-th independent cell, and the type label of the i-th independent cell is obtained based on the category confidence of the i-th independent cell.
[0024] According to the first aspect, the step of inputting the voxel parameters into the preprocessing module, and based on the voxel parameters, performing normalization and sparse encoding through the preprocessing module to obtain a sparse tensor includes:
[0025] Based on the voxel parameters, the Δφ value and three-dimensional coordinate value are extracted, where the Δφ value is the voxel light phase delay value.
[0026] Both the Δφ value and the three-dimensional coordinate value are linearly normalized to obtain the normalized Δφ value and the normalized three-dimensional coordinate value, respectively.
[0027] By iterating through the normalized Δφ value and the voxel parameters, and filtering based on a preset strategy, several effective voxels are obtained.
[0028] Based on the aforementioned effective voxels, corresponding values are extracted from the normalized 3D coordinate values and normalized Δφ values to obtain several effective 3D coordinate values and effective Δφ values.
[0029] The effective voxels, the effective three-dimensional coordinates, and the effective Δφ value are fused to obtain the sparse tensor.
[0030] According to the first aspect, the first result is used for classifying and counting blood cells, including:
[0031] The type labels of the N independent cells in the first result are categorized into M types to obtain M cell groups, where N and M are both positive integers, and M is the number of blood cell categories.
[0032] The number of cells in each of the M cell groups was counted to obtain the number of cells in each cell group.
[0033] According to the first aspect, prior to obtaining the voxel parameters of the standard blood slide, the method further includes:
[0034] Obtain multiple historical voxel parameters, as well as historical 3D feature parameters obtained based on the historical voxel parameters;
[0035] Obtain the historical true results corresponding to the historical voxel parameters and the historical three-dimensional feature parameters;
[0036] The historical voxel parameters, the historical 3D feature parameters, and the historical real results are integrated into training data;
[0037] A blood cell counting classification model is trained and initialized based on the training data, so that after the voxel parameters and the three-dimensional feature parameters are input into the blood cell counting classification model, the blood cell counting classification model outputs the first result.
[0038] Secondly, embodiments of this application also provide an electronic device, the electronic device including a processor and a memory coupled to each other, the memory storing a computer program, and when the computer program is executed by the processor, causing the electronic device to perform the method as described in the first aspect.
[0039] Thirdly, embodiments of this application also provide a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, which, when run on a computer, causes the computer to perform the method described in the first aspect.
[0040] Fourthly, embodiments of this application also provide a program product, characterized in that it includes a computer program, which, when executed by a processor, implements the method described in the first aspect.
[0041] This invention provides a method for blood cell classification and counting. For standard blood slides, voxel parameters containing the three-dimensional spatial coordinates of blood cells and voxel optical phase delay values are obtained. Three-dimensional feature parameters are then calculated based on these voxel parameters. The voxel parameters are then input into the preprocessing module of a blood cell counting and classification model. After linear normalization, effective voxel selection, and data fusion, a sparse tensor is generated. This sparse tensor is then input into a 3D sparse feature extraction module. Voxel data is obtained through 3D sparse convolution and feature fusion. The voxel data is input into the first branch of the model's task module. A three-dimensional mask is generated through sparse 3D convolution, separating overlapping and adhered cell clusters into independent cells. Then, combining the voxel data, the three-dimensional mask, and the three-dimensional feature parameters, single-cell feature extraction and fusion are completed in the second branch of the task module. Finally, after fully connected layer mapping and Softmax classification prediction, the type label of each cell is obtained. Finally, the cell groups are categorized by type label and the counting statistics are completed. This blood cell counting classification model also pre-constructs training data using historical voxel parameters, historical three-dimensional feature parameters, and corresponding real results, and completes training optimization using mean squared error as the loss function to ensure the accuracy of the classification and counting results.
[0042] The invention employing the above technical solution has the following advantages:
[0043] The technical solution provided in this application utilizes a blood cell counting classification model to obtain the typing and counting results of blood cells based on the voxel parameters of the blood test sample in blood slides and the three-dimensional feature parameters calculated from the voxel parameters. This achieves integrated classification and counting, shortening the detection cycle compared to the existing "equipment + manual" workflow. By using three-dimensional voxels and 3D sparse convolution, overlapping cells are separated at a three-dimensional level, avoiding duplication or omission in subsequent feature extraction due to cell overlap, thus preventing significant classification and counting errors caused by repetitive or omitted feature extraction. By obtaining three-dimensional feature parameters, this application improves classification accuracy compared to existing two-dimensional image recognition detection for counting, while also avoiding the difficulty in identification caused by overlapping two-dimensional cell images. Attached Figure Description
[0044] This application can be further illustrated by the non-limiting embodiments given in the accompanying drawings. It should be understood that the following drawings only illustrate some embodiments of this application and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained from these drawings without any inventive effort.
[0045] Figure 1 A flowchart illustrating the blood cell classification and counting method provided in the embodiments of this application. Detailed Implementation
[0046] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that similar or identical parts are referred to by the same reference numerals in the drawings or description. Implementations not shown or described in the drawings are forms known to those skilled in the art. In the description of this application, terms such as "first" and "second" are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0047] Please refer to Figure 1 This application provides a method for classifying and counting blood cells, which can be applied to an electronic device and whose steps can be executed or implemented by the electronic device. The electronic device can be, but is not limited to, a personal computer, a smartphone, or other electronic devices. The blood cell classification and counting method may include the following steps:
[0048] S110, Obtain voxel parameters for standard blood slides;
[0049] S120 obtains the three-dimensional feature parameters of the blood slide based on the voxel parameters of the standard blood slide;
[0050] S130 inputs the voxel parameters and the three-dimensional feature parameters into the blood cell counting classification model, performs sparse convolution operation on the blood cell counting classification model based on the voxel parameters, extracts features from the voxel parameters, and obtains voxel data; based on the voxel data and the three-dimensional feature parameters, a first result is obtained, wherein the first result is used to classify and count blood cells.
[0051] In traditional hematology cell differential counting, the common practice is to manually perform blood smears, staining, and classification, or to use equipment such as Mindray workstations to automatically perform smears, staining, and classification followed by manual review. The drawback of these two methods is that separating cell counting and cell classification into two steps increases overall reporting time and may lead to inconsistencies between the counter's classification and the final classification result. This application employs a hematology cell counting and classification model that, by inputting voxel parameters and three-dimensional feature parameters, outputs a first result, completing the task of integrated classification and counting.
[0052] The steps of blood cell classification and counting methods will be explained in detail below:
[0053] In the S110, a holographic phase-comparison microscope can be used to scan the entire cell thickness layer by layer at a set Z-axis step size of 0.5 μm, outputting voxel parameters. The voxel parameters can include X / Y / Z three-dimensional spatial coordinates, with each coordinate point corresponding to a voxel, and each voxel corresponding to a unique Δφ value, which is the voxel optical phase delay value.
[0054] In S120, the three-dimensional optical parameters include the mean, variance, maximum value, gradient mean, and number of phase peaks of the three-dimensional phase; the three-dimensional morphological parameters include cell volume, equivalent sphere volume, three-dimensional aspect ratio, and convex hull volume; and the three-dimensional substructure parameters include cell nuclear volume and nucleocytoplasmic ratio. All of these parameters can be directly obtained through statistical analysis, gradient calculation, and geometric calculation of the three-dimensional voxel clusters of a single cell.
[0055] In S130, the blood cell counting classification model includes: a preprocessing module, a 3D sparse feature extraction module, and a task module. When performing sparse convolution operations based on the voxel parameters through the blood cell counting classification model to process the voxel parameters into voxel data, the specific steps may include:
[0056] S131: Input the voxel parameters into the preprocessing module. Based on the voxel parameters, normalize and sparsely encode them through the preprocessing module to obtain a sparse tensor.
[0057] S132: Input the sparse tensor into the 3D sparse feature extraction module, and perform sparse convolution operation through the 3D sparse feature extraction module to obtain the voxel data.
[0058] Specifically, S131 may include the following steps:
[0059] A1 extracts the Δφ value and three-dimensional coordinate value based on the voxel parameters, where the Δφ value is the voxel light phase delay value.
[0060] A2 performs linear normalization on both the Δφ value and the three-dimensional coordinate value to obtain the normalized Δφ value and the normalized three-dimensional coordinate value, respectively.
[0061] A3 iterates through the normalized Δφ value and the voxel parameters, and filters them based on a preset strategy to obtain several effective voxels;
[0062] A4 extracts corresponding values from the normalized 3D coordinate values and normalized Δφ values based on the aforementioned effective voxels, thereby obtaining several effective 3D coordinate values and effective Δφ values.
[0063] A5 fuses the effective voxels, the effective three-dimensional coordinates, and the effective Δφ value to obtain the sparse tensor.
[0064] In A2, the value of Δφ is linearly mapped to the interval [0,1], as shown in the formula:
[0065]
[0066] Δφ is the voxel phase delay value, Δφmax is the maximum voxel phase delay value, and Δφmin is the minimum voxel phase delay value. The three-dimensional coordinates are linearly mapped to the interval [-1, 1] using the following formula:
[0067]
[0068] In A3, a number of valid voxels are obtained through screening based on a preset strategy. The screening rule can be: if the Δφ value is > 0.05, the voxel is determined to be a valid voxel; otherwise, it is an invalid voxel.
[0069] In S132, a 4×4×4 3D convolutional kernel can be used for the sparse tensor. Generalized sparse convolution is performed on each valid voxel in 3D space. A spatial hash table is used to quickly find other valid voxels within a 4×4×4 neighborhood of the given voxel. The features of these neighboring voxels are then weighted and summed to generate a new 32-dimensional feature map. ReLU nonlinear activation is applied to zero out negative features to filter out invalid responses, retaining discriminative positive features. This allows the network to automatically learn and extract low-level 3D structural features such as voxel edges, brightness variations, and textures. A 4×4×4 3D deconvolutional kernel is used on the low-level 3D structural features to restore the feature map to its original input size, outputting a 32-channel high-resolution sparse feature map. A 1×1×1 sparse convolution is then applied to the high-resolution sparse feature map, fusing the 32 channels to generate a 1-channel attention weight map. The attention weights are multiplied by the 32-dimensional features voxel-by-voxel to obtain the voxel data.
[0070] By analyzing overlapping cell clusters in three dimensions, this method reduces issues such as missed counts and typing errors caused by cell overlap, thus lowering the error rate. This results in test results that more closely reflect the actual distribution of blood cells, offering higher confidence compared to traditional methods.
[0071] In 130, the blood cell count classification model further includes a task module, which comprises a first branch and a second branch. When obtaining the first result based on the voxel data and the three-dimensional feature parameters, the specific steps include:
[0072] B1: Input the voxel data into the first branch, and output a first set and a second set through the first branch. The first set includes N three-dimensional masks, and the second set includes N independent cells corresponding to the N three-dimensional masks. The three-dimensional mask includes all voxel coordinates of the corresponding independent cell, where N is a positive integer.
[0073] B2: Input the voxel data, the first set, and the three-dimensional feature parameters into the second branch. Based on the voxel data, the first set, and the three-dimensional feature parameters, the second branch performs a classification operation on all independent cells in the second set to obtain a first result. The first result includes type labels for N independent cells.
[0074] In B1, the overlapping cell clusters are divided into N independent single-cell 3D masks through the first branch. Sparse 3D convolution is used to extract the distinguishing features between cells and background layer by layer, and Softmax activation is performed to obtain the probability that each effective voxel belongs to a cell or background. The maximum probability is taken to generate a binary 3D mask (1=cell, 0=background), resulting in the first set including N 3D masks.
[0075] In B2, a classification operation is performed on all independent cells in the second set, and the first result specifically includes:
[0076] B21: Based on the i-th three-dimensional mask corresponding to the i-th independent cell, extract k feature vectors from the voxel data, where i=1,2,3…N, and k is the number of voxels contained in the i-th independent cell;
[0077] B22: Based on the k feature vectors and the three-dimensional feature parameters, a fused feature vector is obtained through feature fusion;
[0078] B23: Based on the fused feature vector, it is mapped to M predicted scores through a fully connected layer, where M is the number of blood cell categories;
[0079] B24: Based on the M predicted scores, classification prediction is performed using a preset softmax function to obtain the category confidence of the i-th independent cell, and the type label of the i-th independent cell is obtained based on the category confidence of the i-th independent cell.
[0080] In B21, based on all coordinates of the independent cell's 3D mask, all effective voxel features corresponding to that cell are extracted from the voxel data output by the backbone. Simultaneously, global average pooling is performed, averaging the 32 channels to obtain a 1×32 vector, preserving the overall average features of the cell. Then, global max pooling is performed, taking the maximum value for each of the 32 channels to obtain a 1×32 vector, preserving the most salient features of the cell. The two 32-dimensional vectors are concatenated to obtain a 1×64 feature vector.
[0081] In B22, the 64-dimensional feature vector and the three-dimensional feature parameters are fused to obtain the fused feature vector.
[0082] Compared to traditional two-dimensional images that only capture cell surface features, three-dimensional feature parameters include three categories of features: three-dimensional optical, three-dimensional morphological, and three-dimensional substructure features. These are used to characterize the intrinsic properties of blood cells, such as their three-dimensional phase, geometric morphology, and nucleocytoplasmic relationship. This provides multiple three-dimensional criteria for cell classification, improving classification accuracy compared to two-dimensional image recognition methods. It also avoids the problems of duplicate or missed identifications caused by overlapping two-dimensional images, resulting in more accurate typing results.
[0083] In B23, a three-layer fully connected classification head is used to fuse feature vectors into M dimensions, resulting in M predicted scores, corresponding to M types of cells.
[0084] In B24, a preset softmax function outputs M class probabilities, and the class with the highest probability is taken as the cell's type label. The softmax function... (k=1,2,3…M), all p kThe maximum value p max As a type label for the cell.
[0085] In S130, the first result is used to classify and count blood cells. The type labels of the N independent cells in the first result are summarized into M types to obtain M cell groups, where N and M are both positive integers, and M is the number of blood cell categories.
[0086] The number of cells in each of the M cell groups was counted to obtain the total number of cells in each group. Finally, the cells were classified, and the number of cells in each type was determined.
[0087] The blood cell count classification model mentioned in S130 can be trained in the following way:
[0088] Multiple historical voxel parameters and historical 3D feature parameters obtained based on the historical voxel parameters are acquired; historical true results corresponding to the historical voxel parameters and the historical 3D feature parameters are obtained; the historical voxel parameters, the historical 3D feature parameters, and the historical true results are integrated into training data; an initialized blood cell counting classification model is trained based on the training data, so that after the voxel parameters and the 3D feature parameters are input into the blood cell counting classification model, the blood cell counting classification model outputs the first result. In this embodiment, the collected training dataset is divided into a training set and a test set in an 8:2 ratio, with 80% of the data used for model training and the remaining 20% used for testing.
[0089] The loss function during training can be expressed as:
[0090]
[0091] This represents the total loss value;
[0092] This represents the first result of the blood cell count classification model. Corresponding historical results The mean square error.
[0093] Historical voxel parameters and historical 3D feature parameters are used together as input features, and historical real results are used as supervision signals to construct a training dataset. A loss function L is used for training. After training, only the voxel parameters need to be input during the testing phase, and the model can directly output the first result.
[0094] This application provides an electronic device that may include a processing module and a memory. The memory stores a computer program, which, when executed by the processor, enables the electronic device to perform the corresponding steps in the aforementioned blood cell classification and counting method.
[0095] In this embodiment, the processor can be an integrated circuit chip with signal processing capabilities. For example, the processor can be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.
[0096] The memory can be, but is not limited to, random access memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, etc.
[0097] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the electronic device described above can be referred to the corresponding steps in the aforementioned method, and will not be elaborated further here.
[0098] This application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program that, when run on a computer, causes the computer to perform the blood cell classification and counting method as described in the above embodiments.
[0099] Computer-readable storage media may be magnetic disks, optical disks, read-only memory, random access memory, flash memory, USB flash drives, hard disks, or solid-state drives, etc., and may also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implement the methods shown in the above embodiments.
[0100] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the aforementioned blood cell classification and counting method. The computer program product may exist in a computer-readable storage medium in forms including, but not limited to, source files, executable files, and installation package files.
[0101] Based on the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by hardware or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, electronic device, or network device, etc.) to execute the methods described in the various implementation scenarios of this application.
[0102] In the embodiments provided in this application, it should be understood that the disclosed methods can also be implemented in other ways. The method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or part of code, which includes one or more executable instructions for implementing a specified logical function. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.
[0103] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for blood cell classification and counting, characterized in that, The method includes: Obtain voxel parameters for standard blood slides; Based on the voxel parameters of the standard blood slide, the three-dimensional feature parameters of the blood slide are obtained. The voxel parameters and the three-dimensional feature parameters are input into the blood cell counting classification model. Based on the voxel parameters, sparse convolution operation is performed through the blood cell counting classification model to extract features from the voxel parameters and obtain voxel data. Based on the voxel data and the three-dimensional feature parameters, a first result is obtained, wherein the first result is used to classify and count blood cells.
2. The method according to claim 1, characterized in that, The blood cell counting classification model includes: a preprocessing module, a 3D sparse feature extraction module, and a task module; Based on the voxel parameters, sparse convolution operations are performed using the blood cell counting classification model to extract features from the voxel parameters, resulting in voxel data, including: The voxel parameters are input into the preprocessing module. Based on the voxel parameters, the preprocessing module performs normalization and sparse encoding to obtain a sparse tensor. The sparse tensor is input into the 3D sparse feature extraction module, and sparse convolution operation is performed through the 3D sparse feature extraction module to obtain the voxel data.
3. The method according to claim 2, characterized in that, The task module includes: a first branch and a second branch. The process of obtaining a first result based on the voxel data and the three-dimensional feature parameters includes: The voxel data is input into the first branch, and a first set and a second set are output through the first branch. The first set includes N three-dimensional masks, and the second set includes N independent cells corresponding to the N three-dimensional masks. The three-dimensional mask includes the coordinates of all voxels of the corresponding individual cells, where N is a positive integer; The voxel data, the first set, and the three-dimensional feature parameters are input into the second branch. Based on the voxel data, the first set, and the three-dimensional feature parameters, the second branch performs a classification operation on all independent cells in the second set to obtain a first result, which includes type labels for N independent cells.
4. The method according to claim 3, characterized in that, When performing a classification operation on all independent cells in the second set, the classification operation includes: Based on the i-th three-dimensional mask corresponding to the i-th independent cell, k feature vectors are extracted from the voxel data, where i = 1, 2, 3…N, and k is the number of voxels contained in the i-th independent cell; Based on the k feature vectors and the three-dimensional feature parameters, a fused feature vector is obtained through feature fusion; Based on the fused feature vector, it is mapped to M predicted scores through a fully connected layer, where M is the number of blood cell categories; Based on the M predicted scores, classification prediction is performed using a preset softmax function to obtain the category confidence of the i-th independent cell, and the type label of the i-th independent cell is obtained based on the category confidence of the i-th independent cell.
5. The method according to claim 2, characterized in that, The step of inputting the voxel parameters into the preprocessing module, and performing normalization and sparse encoding based on the voxel parameters to obtain a sparse tensor includes: Based on the voxel parameters, the Δφ value and three-dimensional coordinate value are extracted, where the Δφ value is the voxel light phase delay value. Both the Δφ value and the three-dimensional coordinate value are linearly normalized to obtain the normalized Δφ value and the normalized three-dimensional coordinate value, respectively. By iterating through the normalized Δφ value and the voxel parameters, and filtering based on a preset strategy, several effective voxels are obtained. Based on the aforementioned effective voxels, corresponding values are extracted from the normalized 3D coordinate values and normalized Δφ values to obtain several effective 3D coordinate values and effective Δφ values. The effective voxels, the effective three-dimensional coordinates, and the effective Δφ value are fused to obtain the sparse tensor.
6. The method according to claim 3, characterized in that, The first result is used to classify and count blood cells, including: The type labels of the N independent cells in the first result are categorized into M types to obtain M cell groups, where N and M are both positive integers, and M is the number of blood cell categories. The number of cells in each of the M cell groups was counted to obtain the number of cells in each cell group.
7. The method according to claim 1, characterized in that, Prior to obtaining the voxel parameters of the standard blood slide, the method further includes: Obtain multiple historical voxel parameters, as well as historical 3D feature parameters obtained based on the historical voxel parameters; Obtain the historical true results corresponding to the historical voxel parameters and the historical three-dimensional feature parameters; The historical voxel parameters, the historical 3D feature parameters, and the historical real results are integrated into training data; A blood cell counting classification model is trained and initialized based on the training data, so that after the voxel parameters and the three-dimensional feature parameters are input into the blood cell counting classification model, the blood cell counting classification model outputs the first result.
8. An electronic device, characterized in that, The electronic device includes a processor and a memory coupled together, the memory storing a computer program that, when executed by the processor, causes the electronic device to perform the method as described in any one of claims 1 to 7.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when run on a computer, causes the computer to perform the method as described in any one of claims 1 to 7.
10. A program product, characterized in that, It includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.