Portable blood cell analysis method based on microfluidic technology

Abstract

The invention discloses a portable blood cell analysis method based on a micro-fluidic technology, which comprises the following steps: preprocessing a collected blood sample, adding the preprocessed blood sample into a micro-fluidic chip, turning on a light source while the sample flows through an image acquisition area, and shooting the sample in real time by using an image sensor to obtain a blood cell image; the method comprises the following steps: carrying out graying, denoising, contrast enhancement and other preprocessing operations on collected images, segmenting cells and backgrounds through a self-adaptive threshold method, and carrying out differential calculation according to motion characteristics of the cells in a system and adjacent three frames of images to finish dynamic capture of target cells, so as to obtain cell image data sets of different postures in a camera view. A spatial pyramid pooling structure SPPF in a YOLOv8 model is replaced by SPPCSPC, a GAM attention mechanism is introduced into a backbone network, sample images are labeled in batches by using a labmeli labeling tool to obtain final sample data sets of different types of cells, and the improved YOLOv8 model is trained to realize classification and counting of the cells.

Description

Technical Field

[0001] The invention belongs to the field of biomedical testing, and in particular relates to a portable blood cell analysis method based on microfluidics technology. Background Art

[0002] White blood cells, also known as leukocytes, are a type of cell in the blood and are a key component of the human immune system. Their main function is to defend the body from pathogens (such as bacteria, viruses and other microorganisms), as well as to participate in immune responses and inflammatory processes. The number of white blood cells in the blood is relatively small, but their important role in the body cannot be ignored. White blood cells are generally divided into five major categories: neutrophils, lymphocytes, monocytes, eosinophils and basophils. In medical examinations, white blood cell counts (WBC) and differential counts (DIFF) are usually performed to assess the status of the immune system and diagnose possible diseases. Abnormalities in white blood cell counts may indicate infection, inflammation, autoimmune diseases or other health problems.

[0003] The current mainstream white blood cell detection methods include electrical impedance method and flow cytometry technology, both of which have complex internal liquid and gas circuits, high cost, many operating steps, and easy failure. Some blood cell analyzers also use imaging methods for counting and classification of white blood cells, but these instruments are not accurate and have a single function. They can only count white blood cells, or they have a complex structure and require a focusing structure and a planar XY two-axis motion mechanism. These methods cannot achieve the portability and accuracy requirements of white blood cell analysis and are difficult to use in home testing scenarios. Summary of the invention

[0004] To solve the above problems, the present invention discloses a portable blood cell analysis method based on microfluidic technology, which has a simple structure, is easy to use, and can accurately classify and count white blood cells in a home setting.

[0005] To achieve the above object, the technical solution of the present invention is as follows: A portable blood cell analysis method based on microfluidic technology includes: after a blood sample is collected by a sampler, it is added to a pretreatment reagent for mixing, the treated sample solution is quantitatively absorbed and added to a microfluidic chip, the sample is moved in a microchannel by capillary action or the driving action of an external pressure source, and while the sample flows through an image acquisition area, a light source is turned on and an image sensor is used to take a real-time photo of the sample to obtain a blood cell image; the collected image is preprocessed by grayscale, denoising, contrast enhancement and other preprocessing operations, and the cells and background are segmented by an adaptive threshold method, and the dynamic capture of the target cell is completed according to the motion characteristics of the cell in the system and the difference calculation of three adjacent frames of images, so as to obtain a cell image dataset of different postures in the camera field of view. By replacing the spatial pyramid pooling structure SPPF in the YOLOv8 model with SPPCSPC (SPP Cross Stage Partial Channel), and introducing the GAM (Global Attention Mechanism) attention mechanism in the backbone network, the labelimg annotation tool is used to batch annotate the sample images to obtain the final sample dataset of different types of cells, and the improved YOLOv8 model is trained to achieve cell classification and counting.

[0006] Technical solution: A portable blood cell analysis method based on microfluidics technology, comprising the following steps: S1. Blood sample collection and processing: First, a quantitative pipette is used to collect a blood sample, and the quantitative pipette is placed in the pretreatment liquid reagent bottle. After manual inversion and mixing and reacting for a specified time, the blood sample completes the dilution, lysis, and staining process and reaches the imaging condition; S2. Image data acquisition: The pretreated blood sample is mixed and added to the sample addition hole of the microfluidic test card, and the test card is placed in the blood cell analyzer. The sample is moved in the microchannel by capillary action or the driving force of an external pressure source. When the sample flows through the image acquisition area, the light source is turned on and the image sensor is used to take a real-time photo of the sample to obtain the image data of the blood cells; S3. Image data preprocessing: The blood cell analyzer performs pre-processing operations such as graying, denoising, and contrast enhancement on the collected images, and segments cells and backgrounds through an adaptive threshold method, and completes dynamic capture of target cells based on the motion characteristics of cells in the system and differential calculation of three adjacent frames of images, and obtains cell image data of different postures in the camera field of view; S4. Image result analysis: The blood cell analyzer replaces the spatial pyramid pooling structure SPPF in the YOLOv8 model with SPPCSPC (SPP Cross Stage Partial Channel), introduces the GAM (Global Attention Mechanism) attention mechanism in the backbone network, uses the labelimg annotation tool to batch annotate sample images to obtain the final sample data set of different types of cells, and trains the improved YOLOv8 model to achieve cell classification and counting.

[0007] Further, in step S2, the microfluidic detection card includes a sample adding area, a detection area, a waste liquid collection area, and a microfluidic channel, the microfluidic channel is connected to the sample adding area, the detection area, and the waste liquid collection area, the sample adding area includes a sealing boss, and the detection area includes a target for camera focusing.

[0008] Further, in steps S2 to S4, the blood cell analyzer includes a stage, a microscopic imaging system, a pressure drive system and a main control system, the stage includes a holder for placing a microfluidic chip and a horizontal movement mechanism, the microscopic imaging system includes a light source, an optical magnification component, an image receiver and a focusing motor, the pressure drive system includes an air pump, a connecting pipeline and a corrugated sealing ring, and the corrugated sealing ring is connected to the sealing boss of the microfluidic chip.

[0009] Furthermore, in step S3, the method for dynamically capturing target cells and acquiring image data of cells in different postures includes the following steps: ① Through the pressure source, the treated cell mixture solution is squeezed into the microchannel at a uniform speed. When the cell solution flows through the camera's field of view, the camera starts to collect images and obtain image data of different cells in different postures; ② Preprocess the real-time cell image in the acquired video image data, and obtain different posture data images of the same target cell in different data frames through sample image processing; ③ In order to accurately obtain the position changes of target cells in different data frames, the inter-frame difference method is performed on three consecutive frames of a single target cell to obtain the motion trajectory of the target cell. The position changes of the target cells in the continuous frame images and the corresponding posture data images are analyzed through the trajectory.

[0010] The above step ① specifically realizes the focusing and image acquisition of the entire image acquisition system by placing a sample card with a focus calibration pattern, wherein the image acquisition device includes an industrial camera with a fixed frame rate, and the specific camera selection needs to be selected according to the flow rate of the flow channel liquid, the flow channel width, the flow channel height, and the dilution concentration of the cell solution in the system.

[0011] The above step ② is mainly divided into the following steps: A: Get a single frame image from video data; B: A binary image containing cell position information after background removal is obtained through adaptive threshold processing and other methods; C: The image positions of different target cells in the original data frame are obtained through the acquired binary image and the motion characteristics of the cells in the system, and the positions of different target cells in adjacent data frames are obtained through differential processing; The method for obtaining the cell movement trajectory in step ③ mainly includes the following processes: A: Select the target cells to be tracked; B: According to the image frame of the selected target cell, two adjacent frames of data images are obtained respectively. In the above step ①, the movement characteristics of the cell in the microfluidic channel have been obtained according to the flow rate of the cell solution and other parameters. According to the movement characteristics of the cell and the movement characteristics of the cell solution, three consecutive frames of the continuous frame images containing the target cell are differentially calculated to calculate the cell movement trajectory of a single target cell. The image area of ​​the same target cell in different frame images is analyzed through the trajectory.

[0012] Furthermore, in step S4, the labelimg annotation tool batch-annotates sample images to obtain a final sample data set of different types of cells, and classifies and counts images of different types of cell data sets through the improved YOLOv8 model. The improvement of the YOLOv8 model mainly includes the optimization of the pyramid pooling structure SPPF and the introduction of the GAM attention mechanism.

[0013] In the present invention, the spatial pyramid pooling structure SPPF in the YOLOv8 model is replaced by SPPCSPC, and the GAM attention mechanism is introduced into the backbone network. The labelimg annotation tool is used to batch annotate sample images to obtain the final sample data set of different types of cells, and the improved YOLOv8 model is trained to realize the classification and counting of white blood cells.

[0014] Furthermore, in step S4, the image result analysis mainly includes the steps of detection model optimization, sample data set preparation, model training and model reasoning: 1. Change the Backbone network of the YOLOv8 model to introduce SPPCSPC; 2. Introduce the GAM attention mechanism into the backbone network; 3. Collect data of different types of cell samples through the microfluidic control system, and obtain image data of a single type of cell in different postures; 4. Create a sample data set based on the acquired images; 5. Complete the training of the network model based on the sample data set in step 3 and the improved detection network model in steps 1 and 2; 6. Use the trained model file and network model to complete the classification and counting of target cells to be tested.

[0015] In the above step 1, mainly to improve the receptive field of the model and enhance the features of the sample image, the SPP structure and the CSP structure are used to introduce the SPP structure into the YOLOv8 detection model. The introduced SPP structure can improve the model's feature acquisition for cell images of different sizes.

[0016] The SPPCSPC structure can complete the fusion of cell features based on cell images with different postures and enhance the cell features in the network model. Through pooling and convolution operations of cell images of different sizes, the model can obtain more feature information and improve the accuracy of model reasoning.

[0017] In the above step 2, the GAM attention mechanism is introduced into the YOLOv8 backbone network. The GAM attention mechanism can amplify the interactive features of the global dimension of the cell image while reducing the information redundancy of the input samples.

[0018] The GAM attention mechanism is based on the optimization of the CBAM (Channel and Spatial Attention Module) attention mechanism, such as Figure 6 As shown: By redesigning the sub-blocks, when given an input feature map F1 ∈ R CxHxW , the intermediate state F2 and output F3 are defined as: ; ; Where Mc is the channel attention map, Ms is the spatial attention map; In the GAM attention mechanism, the channel attention submodule uses 3D permutation to retain information in three dimensions, and two layers of MLP (multilayer perceptron) are used to amplify the channel and spatial dependencies in the cell image. In the spatial attention submodule, in order to focus on the feature spatial information of the input image, the GAM attention mechanism uses two 7x7 convolutional layers to fuse the spatial information of the cell image. The introduction of the GAM attention mechanism further retains the feature map, and the spatial attention module increases the number of parameters.

[0019] The sample data set obtained in step S3 and the improved detection network model are used to complete the training of the input network model, new sample video images are obtained using step S2, and the trained model is used to complete the input data image reasoning.

[0020] The beneficial effects of the present invention are: 1. The present invention combines microscopic imaging technology with microfluidics technology, without the need for complex and precise mechanical motion mechanisms, to achieve high optical magnification blood cell image acquisition; 2. The present invention analyzes the number of blood cells and different spatial postures based on the optimized network model, which greatly improves the accuracy of blood cell counting and classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of a workflow in an embodiment of the present invention; Figure 2 Schematic diagram of a microfluidic chip in one embodiment of the present invention; Figure 3 is a structural schematic diagram of a blood cell analyzer in one embodiment of the present invention; Figure 4 is a schematic diagram of a differential image in an embodiment of the present invention; Figure 5 is a schematic diagram of blood cell trajectories in one embodiment of the present invention; Figure 6 Schematic diagram of the GAM attention mechanism in one embodiment of the present invention; Figure 7 This is a comparison chart between the test results of an embodiment of the present invention and the test results of a hospital. DETAILED DESCRIPTION

[0022] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0023] like Figure 1 As shown, the portable blood cell analysis method based on microfluidic technology described in the present invention comprises the following steps: S1. Blood sample collection and processing: First, a quantitative pipette is used to collect a blood sample, and the quantitative pipette is placed in the pretreatment liquid reagent bottle. After manual inversion and mixing and reacting for a specified time, the blood sample completes the dilution, lysis, and staining process and reaches the imaging condition; S2. Image data acquisition: The pretreated blood sample is mixed and added to the sample addition hole of the microfluidic test card, and the test card is placed in the blood cell analyzer. The sample is moved in the microchannel by capillary action or the driving force of an external pressure source. When the sample flows through the image acquisition area, the light source is turned on and the image sensor is used to take a real-time photo of the sample to obtain the image data of the blood cells; S3. Image data preprocessing: The blood cell analyzer performs pre-processing operations such as graying, denoising, and contrast enhancement on the collected images, and segments cells and backgrounds through an adaptive threshold method, obtains a video data stream of cell images through an image acquisition system, and completes dynamic capture of target cells based on the movement characteristics of cells in the system and differential calculation of three adjacent frames of images, and obtains cell image data of different postures in the camera field of view; S4. Image result analysis: The blood cell analyzer replaces the spatial pyramid pooling structure SPPF in the YOLOv8 model with SPPCSPC (SPP Cross Stage Partial Channel), introduces the GAM (Global Attention Mechanism) attention mechanism in the backbone network, uses the labelimg annotation tool to batch annotate sample images to obtain the final sample data set of different types of cells, and trains the improved YOLOv8 model to achieve cell classification and counting.

[0024] like Figure 2 As shown, the microfluidic detection card of the present invention includes a sample adding area 23, a detection area 24, a waste liquid collection area 26, and a microfluidic channel 25. The microfluidic channel is connected to the sample adding area, the detection area and the waste liquid collection area. The sample adding area includes a sealing boss 22, and the detection area includes a target 27 for camera focusing.

[0025] The schematic diagram of the system structure of the present invention is as follows Figure 3 The blood cell analyzer comprises a stage, a microscopic imaging system, a pressure driving system and a main control system. The stage comprises a holder for placing a microfluidic chip and a horizontal moving mechanism. The microscopic imaging system comprises a light source, an optical magnification component, an image receiver and a focusing motor. The pressure driving system comprises an air pump, a connecting pipeline and a corrugated sealing ring. The corrugated sealing ring is connected to the sealing boss of the microfluidic chip.

[0026] The method for dynamically capturing target cells and acquiring image data of cells in different postures of the present invention comprises the following steps: ① Through the pressure source, the treated cell mixture solution is squeezed into the microchannel at a uniform speed. When the cell solution flows through the camera's field of view, the camera starts to collect images and obtain image data of different cells in different postures; ② Preprocess the real-time cell image in the acquired video image data, and obtain different posture data images of the same target cell in different data frames through sample image processing; ③ In order to accurately obtain the position changes of target cells in different data frames, the inter-frame difference method is performed on three consecutive frames of a single target cell to obtain the motion trajectory of the target cell. The position changes of the target cells in the continuous frame images and the corresponding posture data images are analyzed through the trajectory.

[0027] The above step ① specifically realizes the focusing and image acquisition of the entire image acquisition system by placing a sample card with a focus calibration pattern, wherein the image acquisition device includes an industrial camera with a fixed frame rate, and the specific camera selection needs to be selected according to the flow rate of the flow channel liquid, the flow channel width, the flow channel height, and the dilution concentration of the cell solution in the system.

[0028] The above step ② is mainly divided into the following steps: A: Get a single frame image from video data; B: A binary image containing cell position information after background removal is obtained through adaptive threshold processing and other methods; C: The image positions of different target cells in the original data frame are obtained through the acquired binary image and the motion characteristics of the cells in the system, and the positions of different target cells in adjacent data frames are obtained through differential processing; The specific implementation process of the method for obtaining the cell movement trajectory in step ③ is as follows Figure 4 , 5 As shown: A: Select a data frame image to obtain the target cell, and extract the previous frame p0 and the next frame image P2 adjacent to the current frame P1 in the video stream; B: As shown in the figure, the difference image F0 between data frames P1 and P0, and the difference image F1 between data frames P2 and P1 are calculated respectively; C: By performing an AND operation on the two differential images F0 and F1, the movement position of the cell is obtained through the differential image F0 and F1 and the operation graph; D: Repeat the difference operation of the above three adjacent frames. For example, in the M frames of the video stream, each frame is recorded as P1, P2...PM. After selecting the target cell, calculate the three adjacent frames of P2...PM-1 respectively, and perform an AND operation on the differential images obtained from the three frames to obtain the movement trajectory of the cells in the three frames. Iterate the operation until the PM-1 frame data image in the data stream is calculated, and finally obtain the movement trajectory of the cell through the above execution operation.

[0029] The image result analysis of the present invention mainly includes detection model optimization, sample data set preparation, model training and model reasoning. The specific steps are: 1. Change the Backbone network of the YOLOv8 model to introduce SPPCSPC; 2. Introduce the GAM attention mechanism into the backbone network; 3. Collect data of different types of cell samples through the microfluidic control system, and obtain image data of a single type of cell in different postures; 4. Complete the training of the network model based on the sample data set in step 3 and the improved detection network model in steps 1 and 2; 5. Use the sample data set to complete the model training after the optimization in step 4 and implement the classification and counting of target cells in the data stream.

[0030] The improvements to the YOLOv8 model mainly include the optimization of the pyramid pooling structure SPPF and the introduction of the GAM attention mechanism.

[0031] The Spatial Pyramid Pooling Network (SPPNet) is a deep learning network structure for processing input images with different sizes and scales. It introduces a spatial pyramid pooling layer to allow the network to model different scales and scales of input images, thereby improving the robustness and generalization ability of the model.

[0032] SPPF is a technology that simplifies and optimizes SPP. SPP divides the input image into sub-regions of different sizes by constructing multiple pyramid structures of different scales, and performs pooling operations on each sub-region. SPPF improves computational efficiency by reducing the number of pyramid levels and sub-regions.

[0033] The main idea of ​​SPPF is to reduce the computational complexity by simplifying the pyramid structure while maintaining the idea of ​​spatial pyramid pooling. Specifically, SPPF only retains some levels and sub-regions in the pyramid, thereby reducing the amount of computation. At the same time, SPPF also adopts some optimization strategies, such as using more efficient pooling functions and parallel computing, to further improve computational efficiency.

[0034] The attention mechanism has been widely used in the field of deep learning research. In the algorithm model using the attention mechanism, the built model can selectively focus on different parts of the input sequence. When building a model, in order to highlight the model's ability to understand and analyze different sample data, users often use the attention mechanism to assign different weights to different parts of the data input in the data set to highlight more important information.

[0035] In the field of computer vision, attention mechanisms have been widely used in scenarios such as image classification and target detection, enabling the model to focus on important areas or features in the image.

[0036] The main advantage of the GAM attention mechanism is that it retains information in both channel and space while enhancing the model's cross-dimensional interactions. The GAM attention mechanism improves the performance of deep neural networks by reducing information loss and amplifying global interaction representations. The GAM attention mechanism also introduces a 3D arrangement with multi-layer perceptrons for channel attention enhancement and combines it with a convolutional spatial attention submodule to improve the overall detection performance of the model.

[0037] In the above step 1, mainly to improve the receptive field of the model and enhance the features of the sample image, the SPP structure and the CSP structure are used to introduce the SPP structure into the YOLOv8 detection model. The introduced SPP structure can improve the model's feature acquisition for cell images of different sizes.

[0038] The SPPCSPC structure can complete the fusion of cell features based on cell images with different postures and enhance the cell features in the network model. Through pooling and convolution operations of cell images of different sizes, the model can obtain more feature information and improve the accuracy of model reasoning.

[0039] In the above step 2, the GAM attention mechanism is introduced into the YOLOv8 backbone network. The GAM attention mechanism can amplify the interactive features of the global dimension of the cell image while reducing the information redundancy of the input samples.

[0040] The GAM attention mechanism is based on the optimization of the CBAM (Channel and Spatial Attention Module) attention mechanism, such as Figure 6 As shown: By redesigning the sub-blocks, when given an input feature map F1 ∈ R CxHxW , the intermediate state F2 and output F3 are defined as: F2=Mc(F1)⊗F1 F3=Ms(F1)⊗F2 Where Mc is the channel attention map, Ms is the spatial attention map; In the GAM attention mechanism, the channel attention submodule uses 3D permutation to retain information in three dimensions, and two layers of MLP (multilayer perceptron) are used to amplify the channel and spatial dependencies in the cell image. In the spatial attention submodule, in order to focus on the feature spatial information of the input image, the GAM attention mechanism uses two 7x7 convolutional layers to fuse the spatial information of the cell image. The introduction of the GAM attention mechanism further retains the feature map, and the spatial attention module increases the number of parameters.

[0041] In a specific embodiment, the present invention is a blood cell analyzer for a portable blood cell analysis method based on microfluidic technology, which greatly reduces the complexity of the instrument and is easy to use and operate. The blood sample detection is performed by the microfluidic chip and the improved YOLOv8 detection model, which greatly improves the detection accuracy. The present invention uses hospital samples with known test results for comparative detection, showing a high result correlation (r>0.99) ( Figure 7 ).

[0042] It should be noted that the above content only illustrates the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.

Claims

1. A portable blood cell analysis method based on microfluidics technology, characterized in that: The steps include: S1. Blood sample collection and processing: First, a quantitative pipette is used to collect a blood sample, and the quantitative pipette is placed in the pretreatment liquid reagent bottle. After manual inversion and mixing and reacting for a specified time, the blood sample completes the dilution, lysis, and staining process and reaches the imaging condition; S2. Image data acquisition: The pretreated blood sample is mixed and added to the sample addition hole of the microfluidic test card, and the test card is placed in the blood cell analyzer. The sample is moved in the microchannel by capillary action or the driving force of an external pressure source. When the sample flows through the image acquisition area, the light source is turned on and the image sensor is used to take a real-time photo of the sample to obtain the image data of the blood cells; S3. Image data preprocessing: The blood cell analyzer performs preprocessing operations on the collected images, and segments cells and backgrounds through an adaptive threshold method, completes dynamic capture of target cells based on the motion characteristics of cells in the system and differential calculation of three adjacent frames of images, and obtains cell image data of different postures in the camera field of view; S4. Image result analysis: The blood cell analyzer replaces the spatial pyramid pooling structure SPPF in the YOLOv8 model with SPPCSPC, introduces the GAM attention mechanism in the backbone network, uses the labelimg annotation tool to batch annotate sample images to obtain the final sample data set of different types of cells, and trains the improved YOLOv8 model to achieve cell classification and counting.

2. According to claim 1, a portable blood cell analysis method based on microfluidics technology is characterized in that: In step S2, the microfluidic detection card includes a sample adding area, a detection area, a waste liquid collection area, and a microchannel, the microchannel is connected to the sample adding area, the detection area, and the waste liquid collection area, the sample adding area includes a sealing boss, and the detection area includes a target for camera focusing.

3. According to claim 1, a portable blood cell analysis method based on microfluidics technology is characterized in that: In steps S2 to S4, the blood cell analyzer includes a stage, a microscopic imaging system, a pressure drive system and a main control system. The stage includes a holder for placing a microfluidic chip and a horizontal movement mechanism. The microscopic imaging system includes a light source, an optical magnification component, an image receiver and a focusing motor. The pressure drive system includes an air pump, a connecting pipeline and a corrugated sealing ring. The corrugated sealing ring is connected to the sealing boss of the microfluidic chip.

4. According to claim 1, a portable blood cell analysis method based on microfluidics technology is characterized in that: In step S3, the method for dynamically capturing the target cells and acquiring image data of cells in different postures includes: The treated cell mixture solution is squeezed into the microchannel at a uniform speed through a pressure source. When the cell solution flows through the camera's field of view, the camera starts to collect images and obtain image data of different cells in different postures. Preprocessing the real-time cell image in the acquired video image data, and acquiring different posture data images of the same target cell in different data frames through sample image processing; In order to accurately obtain the position changes of target cells in different data frames, the inter-frame difference method is performed on three consecutive frames of a single target cell to obtain the motion trajectory of the target cell. The position changes of the target cells in the consecutive frame images and the corresponding posture data images are analyzed through the trajectory.

5. According to claim 4, a portable blood cell analysis method based on microfluidics technology is characterized in that: The method for obtaining image data of different cells in different postures includes: placing a sample card with a focus calibration pattern and an image acquisition device to achieve focusing and image acquisition of the entire image acquisition system, wherein the image acquisition device includes an industrial camera with a fixed high frame rate, and the high frame rate camera is selected according to the flow rate of the flow channel liquid, the flow channel width, the flow channel height, and the dilution concentration of the cell solution in the system.

6. According to claim 4, a portable blood cell analysis method based on microfluidics technology is characterized in that: The acquired video image data is used to pre-process the target cell image, and the method for acquiring the data frame containing the target cell image includes: Get a single frame image from video data; A binary image containing cell position information is obtained through an adaptive threshold processing method; The cell image position in the original data frame is obtained through the binary image after background removal and the movement characteristics of the cells in the system, and the position of different target cells in adjacent data frames is obtained through differential processing.

7. According to claim 4, a portable blood cell analysis method based on microfluidics technology is characterized in that: The method for obtaining the cell movement trajectory comprises: Select the target cells to be tracked; According to the image frame of the selected target cell, two adjacent frames of data images are obtained respectively, and the movement position characteristics of the cell in the microfluidic channel are obtained according to the flow rate of the cell solution. According to the movement characteristics of the cell and the movement characteristics of the cell solution, the frame image of the single target cell is subjected to the inter-frame difference method for three consecutive frames to calculate the movement trajectory of the target cell in the image of adjacent frames, and the image area of ​​the target cell in the single frame image is analyzed by trajectory.

8. According to claim 1, a portable blood cell analysis method based on microfluidics technology is characterized in that: In step S4, the spatial pyramid pooling structure SPPF in the YOLOv8 model is replaced with SPPCSPC, and the GAM attention mechanism is introduced into the backbone network. The sample images are labeled in batches using the labelimg annotation tool to obtain the final sample data set of different types of cells, and the improved YOLOv8 model is trained to achieve the classification and counting of different cells, including: Change the Backbone network of the YOLOv8 model to introduce SPPCSPC; Introduce the GAM attention mechanism into the backbone network; The sample data is collected through the microfluidic control system, and the image data of a single target cell in different postures is acquired; Create a sample data set based on the acquired images; The improved detection network model completes the inference model training through different types of cell sample data sets; Use the trained model file and network model to complete the classification and counting of target cells to be tested.

9. The portable blood cell analysis method based on microfluidics technology according to claim 8, characterized in that: The method of introducing the GAM attention mechanism into the Backbone network of the YOLOv8 model includes: By redesigning the sub-blocks, given an input feature map F1 ∈ R CxHxW , the intermediate state F2 and output F3 are defined as: ； ； Where Mc is the channel attention map, Ms is the spatial attention map; In the GAM attention mechanism, the channel attention submodule uses 3D permutation to retain information in three dimensions, and two layers of multilayer perceptrons are used to amplify the channel and spatial dependencies in the cell image; In the spatial attention submodule, in order to focus on the feature spatial information of the input image, the GAM attention mechanism uses two 7X7 convolutional layers to fuse the spatial information of the cell image; The feature map is further preserved by introducing the GAM attention mechanism, while the spatial attention module increases the number of parameters.

10. The portable blood cell analysis method based on microfluidics technology according to claim 8, characterized in that: The method for obtaining a sample data set using the acquired images includes: Select the target cell and obtain a data frame image, extract the previous frame p0 and the next frame image P2 adjacent to the current frame P1 in the video stream; Calculate the difference image F0 between data frames P1 and P0, and the difference image F1 between data frames P1 and P2 respectively; By performing an AND operation on the two differential images F0 and F1, the movement position of the cell is obtained through the differential images F0 and F1 and the operation graph; Repeat the difference operation of the above three adjacent frames. In the M frames of the video stream, each frame is recorded as P1, P2, ... M , after selecting the target cell, calculate P2…P M-1 The three adjacent frames of differential images are obtained, and the differential images obtained from the three frames of images are ANDed to obtain the movement trajectory of the cells in the three frames of images; the P in the data stream is calculated through iterative operations. M-1 Frame data image, through the above operations, the movement trajectory of the cell is finally obtained; In the actual detection process, different target cell images are used to iterate the above process in the entire video frame data stream, and finally the complete trajectory of different cells in the entire camera field of view and data images of different postures in different frame images are obtained; Use the labelimg annotation tool to batch annotate sample images to obtain the final sample data set of different types of cells.

Images