Method and system for absolutely counting cells based on micro-fluidic chip, and electronic equipment

Abstract

The invention discloses a method and a system for absolutely counting cells based on a micro-fluidic chip, and electronic equipment, and relates to the technical field of cell counting. The trace cell absolute counting method provided by the invention has the technical advantages of high flux, simplicity in detection and accuracy in detection, and can accurately, simply and conveniently quantify the concentration of a microscopic sample. Therefore, the counting method and system provided by the invention have important significance in clinical detection and biological application.

Description

Technical Field

[0001] The present invention relates to the technical field of cell counting, and in particular to a method and system for absolute cell counting based on a microfluidic chip, and an electronic device. Background Art

[0002] Absolute cell count refers to the specific number of a certain type of cell in a unit volume (such as per microliter or per liter) of body fluid (such as blood, cerebrospinal fluid, etc.), rather than the relative proportion. This method usually does not rely on labeling or other interventions, but is based on direct counting and quantification techniques. In medical testing, absolute cell counts can help doctors accurately understand the number of certain types of cells, thereby providing important information about the patient's health status. By quantifying the abnormal number of specific cells, it can assist in determining the type and severity of the disease, so that the treatment plan can be modified in a targeted manner or treatment monitoring can be carried out flexibly. For example, by performing an absolute count of CD34+T cells in HIV-infected patients (normal value 500-1,600 / μL), if it is <200 / μL, it is defined as the AIDS stage and requires emergency intervention. If the absolute CD4+T cell count of HIV patients returns to >500 / μL, it indicates that immune reconstitution is successful; the diagnosis of chronic lymphocytic leukemia (CLL) requires an absolute count of mature lymphocytes in peripheral blood ≥5×10 9 / L; and the absolute count of circulating tumor cells (CTCs) can be used to assess disease progression and metastasis risk in patients with solid tumors.

[0003] Absolute cell counting involves two steps: counting the number of cells and measuring the volume of a liquid. Currently, numerous methods exist for measuring cell count, including flow cytometry, Coulter counters, cell counting plates, and automated cell counters. However, methods for measuring liquid volume are limited, primarily relying on cell counting plates, MEMS flow sensors, and capillary flowmeters. These methods are effective for larger sample volumes, but many current research areas require absolute cell counts in minute sample volumes. Examples include circulating tumor cell (CTC) liquid biopsies, cerebrospinal fluid (CSF) cell analysis, amniotic fluid / chorionic villus aspirate, single-cell suspensions from biopsied tissue (e.g., lung / liver biopsies), single-cell sequencing, stem cell research, and viral load testing. In these applications, including point-of-care (POCT), traditional methods are ineffective due to the small volume of liquid. For many years, cell biologists have used manual counting with hemocytometers. This method relies on microscopy to identify cells in a fixed volume, which is time-consuming and cumbersome.

[0004] Microfluidic methods have attracted considerable attention due to their precise manipulation of fluids. Through miniaturization and functional integration, microfluidic technology is revolutionizing traditional cell counting methods, demonstrating unique advantages in rare cell detection, bedside diagnostics, and resource-scarce settings. Digital microfluidics uses microchambers to segment single cells and combines them with Poisson distribution statistics to achieve absolute counts (e.g., Bio-Rad's ddPCR technology). Multi-frequency impedance analysis is used to distinguish cell subpopulations (e.g., red blood cells and white blood cells) and achieve differential counting (e.g., microfluidic chips from the Swiss ETH). Combining microscopic imaging with convolutional neural networks (CNNs) allows for the automatic identification of cells in complex samples (e.g., malaria-infected red blood cells). However, these advances still face challenges such as poor anti-interference, high computational latency, low system throughput, cytotoxicity, insufficient sensitivity, and high cost, particularly regarding throughput. A high-throughput, simple, and accurate method for absolute microcell counting is urgently needed.

[0005] Droplet microfluidics is a branch of microfluidics that focuses on generating, manipulating, and analyzing tiny droplets (usually at the pL to nL level) in micron-scale channels. Its core principle is to use fluid mechanics, interfacial tension, and micromachining technology to divide samples into independent micro-reaction units. Droplets act as independent microreactors, with the advantages of high throughput, low sample consumption, and precise control. They show great potential in biomedicine, chemical synthesis, materials science, and other fields. Currently, the field of droplet microfluidics has developed technologies such as single-cell transcriptome and drug screening (CP-seq), which achieves combined perturbations of single cells and drugs through random pairing of droplets, and combines single-cell sequencing to analyze drug response mechanisms. There is also the droplet sequential operation array (SODA) technology, which automatically completes the mixing, transfer, and stimulation of droplets through microfluidic chips to achieve long-term cell culture and drug dose-dependent analysis. The ReSCUE microfluidic platform generates tumor-like and organoid-like cells of different shapes through sliding chip design and supports targeted release for downstream analysis. Droplet technology has the ability to accurately quantify liquids due to precise control of microscale fluids, monodispersity of droplets, real-time monitoring feedback, and high-precision equipment processing.

[0006] In view of this, the present invention is proposed. Summary of the Invention

[0007] The purpose of the present invention is to provide a method and system for absolute cell counting based on a microfluidic chip, and an electronic device to solve the problems of low throughput, complex detection, and inaccurate detection of absolute microcell counting.

[0008] The present invention is achieved in that:

[0009] In a first aspect, the present invention provides a method for absolute cell counting based on a microfluidic chip, comprising the following steps:

[0010] (1) Generating oil-in-water droplets of the cell sample to be tested: generating droplets by a pressure pump, controlling the pressure applied by the pressure pump to the dispersed phase inlet flow channel to be 300 kPa ± 10 kPa, and controlling the pressure applied by the pressure pump to the continuous phase inlet flow channel to be 150 MPa ± 10 MPa; the frequency of droplet generation is 1200 Hz ± 50 Hz; and the cell throughput is at least 1000 cells per second;

[0011] (2) Changing the motion state of the droplets by controlling the pressure conditions at the droplet capture outlet so that the droplets enter the capture array; performing droplet detection in parallel; measuring the static volume of the droplets, establishing a droplet volume calculation model, and calculating the volume of the droplets entering the capture array based on the droplet volume calculation model to obtain the total volume of the droplets;

[0012] The droplet volume calculation model formula is as follows:

[0013] V t =f(S T , w, h, θ);

[0014]

[0015] Among them, S T is the area of ​​the top view, w is the width of a single channel of the microchannel, h is the height of the microchannel, a, b, c are the semi-axis lengths of the ellipsoid along the x-axis, along the y-axis, and along the z-axis, respectively, dz represents the integral, and it is assumed that And c is calculated from the contact angle θ, and the contact angle θ formula is:

[0016]

[0017] The value range of θ is 100°~150°;

[0018] (3) Obtaining bright-field images of droplets on the microfluidic chip using a microscopic imaging platform;

[0019] (4) Processing the bright field image and obtaining cell concentration information: Inputting the bright field image into the trained deep learning neural network, obtaining the time series information and position information of the droplets in the bright field image, and obtaining the number and type of identified cells;

[0020] Acquisition of cell concentration information: The original image of the droplet is processed using a morphological algorithm based on the droplet position information to obtain the droplet contour information. This contour information is combined with the droplet volume model to obtain the volume of each cell type in the sample. The absolute concentration of each cell type is calculated by combining the total cell volume with the number of corresponding cell types.

[0021] In a second aspect, the present invention provides a system for absolute cell counting based on a microfluidic chip, which comprises: a droplet generation module, a droplet volume calculation module, an image acquisition module, and an image analysis module;

[0022] The droplet generation module is used to generate oil-in-water droplets for the cell sample to be tested. The oil-in-water droplet generation includes: generating droplets by a pressure pump, controlling the pressure applied by the pressure pump to the dispersed phase inlet flow channel to be 300kPa±10kPa, and controlling the pressure applied by the pressure pump to the continuous phase inlet flow channel to be 150kPa±10kPa; the droplet generation frequency is 1200Hz±50Hz;

[0023] The droplet volume calculation module is used to: change the motion state of the droplets by controlling the pressure conditions at the droplet capture outlet, allowing the droplets to enter the capture array; perform droplet detection using droplet parallelism; measure the static volume of the droplets, establish a droplet volume calculation model, and calculate the volume of the droplets entering the capture array based on the droplet volume calculation model to obtain the total volume of the droplets. The droplet volume calculation model formula is as follows:

[0024] V t =f(S T , w, h, θ);

[0025]

[0026] Among them, S T is the area of ​​the top view, w is the width of a single channel of the microchannel, h is the height of the microchannel, a, b, c are the semi-axis lengths of the ellipsoid along the x-axis, along the y-axis, and along the z-axis, respectively, dz represents the integral, and it is assumed that And c is calculated from the contact angle θ, and the contact angle θ formula is:

[0027]

[0028] The value range of θ is 100°~150°;

[0029] The image acquisition module is used to: obtain bright field images of droplets on the microfluidic chip through a microscopic imaging platform;

[0030] The image analysis module is used to process brightfield images and obtain cell concentration information. Processing brightfield images includes inputting the brightfield images into a trained deep learning neural network to obtain time series information and position information of droplets in the brightfield images, and to obtain the number and type of identified cells.

[0031] Acquisition of cell concentration information includes: comparing the original image of the droplet with the droplet position information through morphological algorithm processing to obtain the droplet contour information; combining the contour information with the droplet volume model to obtain the volume of each cell type in the sample, and calculating the concentration of each cell type by the total cell volume and the number of cells of the corresponding type.

[0032] In a third aspect, the present invention further provides an electronic device comprising: a processor and a memory; the processor and the memory are connected, wherein the memory is used to store a computer program, and the processor is used to call the computer program to execute the above-mentioned method for absolute cell counting based on a microfluidic chip.

[0033] In a fourth aspect, the present invention further provides a computer-readable storage medium comprising instructions, which, when executed on a computer, enables the computer to execute the above-mentioned method for absolute blood cell counting based on a microfluidic chip.

[0034] The present invention has the following beneficial effects:

[0035] The present invention proposes a droplet microfluidics-based method for absolute microcell counting. Unlike traditional oil-in-water techniques, this method uses water-in-oil technology to discretize the cell liquid. For each water fragment dispersed by an oil droplet, the liquid volume is calculated by constructing a droplet model. The droplet volume is then measured and calibrated using a microscopic imaging platform. A neural network model is constructed to analyze cell images, identify cell types, and count them, ultimately achieving absolute microcell counting. This method, using water-in-oil technology, significantly improves test throughput, reaching a maximum throughput of 2 μL / minute and 130,000 cells / minute. Validation using clinical samples demonstrated over 95% consistency compared to test results using large-scale hospital equipment. To further verify the accuracy of the absolute microcell counting method provided by the present invention, the present invention further uses confocal microscopy to reconstruct three-dimensional images and calculate the volume of the captured droplets. By combining calculus principles and image processing algorithms, the true volume of the droplets can be accurately calculated. The results of the absolute microcell counting method provided by the present invention were found to be consistent with the microcell content calculated using confocal microscopy.

[0036] Therefore, the absolute microcell counting method provided by the present invention has the technical advantages of high throughput, simple detection, and accurate detection, and can accurately and conveniently quantify the concentration of microscopic samples.

[0037] The absolute cell count provided by the present invention (classifying and counting different types of cells in a quantitative liquid volume) is of great significance in clinical testing and biological applications, such as analysis of blood heterogeneity and drug resistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 This is a flowchart of the absolute counting method;

[0040] Figure 2 Figure 1 shows the image processing results and droplet measurement results;

[0041] Figure 3 Figure 2 shows the cell imaging results and target detection model performance test results.

[0042] Figure 4 It is a curve graph of increasing cell number;

[0043] Figure 5 The figure shows the comparison of clinical results. DETAILED DESCRIPTION

[0044] Reference will now be made in detail to embodiments of the present invention, one or more examples of which are described below. Each example is provided to illustrate, not to limit, the present invention. Indeed, it will be apparent to those skilled in the art that various modifications and variations may be made to the present invention without departing from the scope or spirit of the invention. For example, features illustrated or described as part of one embodiment may be used in another embodiment to produce further embodiments.

[0045] The present invention proposes a method based on droplet microfluidics to achieve absolute counting of trace cells. Unlike traditional oil-in-water technology, this method uses water-in-oil technology to discretize cell liquids. For each water fragment dispersed by the oil droplets, the liquid volume is calculated by constructing a droplet model. After constructing a neural network model for cell image analysis, the cell type is identified and counted, ultimately achieving absolute counting of trace cells. This method uses water-in-oil technology to greatly improve the test throughput, with the highest throughput reaching 2 microliters / minute and 130,000 cells / minute. Finally, 20 clinical samples were used for verification, and compared with the test results of large-scale equipment in the hospital, the result consistency exceeded 95%.

[0046] The absolute cell count provided by the present invention (classifying and counting different types of cells in a quantitative liquid volume) is of great significance in clinical testing and biological applications. For example, analysis of blood heterogeneity and drug resistance analysis are also important. In addition, due to the difficulty of sample collection and the introduction of precision medicine, absolute count studies of small volume samples (usually about 100 microliters) such as bone marrow biopsies and inflammatory synovial fluid have become increasingly necessary. Therefore, the method provided by the present invention is also helpful for absolute counts of small volume samples such as bone marrow biopsies and inflammatory synovial fluid.

[0047] Microfluidics has promoted the development of highly integrated and small-sample detection technologies, providing an excellent detection platform for cell counting. On one hand, microfluidics can complete sample processing, transfer, and detection within centimeter-scale spaces, providing strong support for automated detection. On the other hand, the dimensions of the microchannels are highly compatible with human blood cells, simulating a liquid environment and effectively improving sample utilization. Furthermore, it can also combine multiple detection methods, such as Coulter assays, scattered light detection, and imaging detection. Based on this foundation, microfluidics is easily applicable to cell analysis.

[0048] In order to achieve absolute cell counting, it is necessary to develop sensitive detection methods for micro-samples and accurate volume measurement methods with high feasibility and operability. Existing microfluidic devices can be divided into thermal or non-thermal types according to their working principles. Thermal flow sensors are widely used in trace sample measurements. However, the complex manufacturing process of embedded electrodes and thermal elements makes them difficult to integrate into PDMS devices. Non-thermal sensors mainly rely on the mechanical deformation of moving parts, including measuring the deflection of PDMS cantilever beams, evaluating the elongation of stretchable spring-like structures, and evaluating the rotation of teardrop-shaped indicators. Thanks to these highly integrated design methods, on-chip flow measurements can be performed in traditional PDMS microfluidic systems, but they are often limited by manufacturing difficulties and difficult to match with complex detection equipment.

[0049] Glossary

[0050] Time series information refers to physical, chemical or biological parameter data collected continuously over time (such as flow rate, pressure, temperature, concentration, fluorescence signal, etc.).

[0051] Cell throughput is the number of cells passing through the detector per unit time.

[0052] IOU (Intersection over Union): IOU measures the degree of overlap between two bounding boxes. A bounding box usually represents an object location predicted by the model or the actual object location in the labeled data (ground truth).

[0053] In a first aspect, the present invention provides a method for absolute cell counting based on a microfluidic chip, comprising the following steps:

[0054] (1) Generating oil-in-water droplets of the cell sample to be tested: generating droplets by a pressure pump, controlling the pressure applied by the pressure pump to the dispersed phase inlet flow channel to be 300 kPa ± 10 kPa, and controlling the pressure applied by the pressure pump to the continuous phase inlet flow channel to be 150 MPa ± 10 MPa; the frequency of droplet generation is 1200 Hz ± 50 Hz; and the cell throughput is at least 1000 cells per second;

[0055] (2) Changing the motion state of the droplets by controlling the pressure conditions at the droplet capture outlet so that the droplets enter the capture array; performing droplet detection in parallel; measuring the static volume of the droplets, establishing a droplet volume calculation model, and calculating the volume of the droplets entering the capture array based on the droplet volume calculation model to obtain the total volume of the droplets;

[0056] The droplet volume calculation model formula is as follows:

[0057] V t =f(S T , w, h, θ);

[0058]

[0059] Among them, S T is the area of ​​the top view, w is the width of a single channel of the microchannel, h is the height of the microchannel, a, b, c are the semi-axis lengths of the ellipsoid along the x-axis, along the y-axis, and along the z-axis, respectively, dz represents the integral, and it is assumed that And c is calculated from the contact angle θ, and the contact angle θ formula is:

[0060]

[0061] The value range of θ is 100°~150°;

[0062] (3) Obtaining bright-field images of droplets on the microfluidic chip using a microscopic imaging platform;

[0063] (4) Processing the bright field image and obtaining cell concentration information: Inputting the bright field image into the trained deep learning neural network, obtaining the time series information and position information of the droplets in the bright field image, and obtaining the number and type of identified cells;

[0064] Acquisition of cell concentration information: The original image of the droplet is processed using a morphological algorithm based on the droplet's position information to obtain the droplet's contour information. This contour information is combined with the droplet volume model to obtain the total volume of the sample liquid passing through the microfluidic system. By calculating the total volume of the sample liquid and the number of cells corresponding to it, the absolute concentration of each cell type is obtained.

[0065] A peripheral precision pump delivers the cell suspension into the microfluidic chip, forming monodisperse droplets in the confocal area (i.e., the microscopic imaging area).

[0066] Droplet generation depends on the dimensionless capillary number (Ca = ηv / γ), and the droplet size can be changed by controlling the flow rate of the two-phase flow.

[0067] Although droplet generation can reach thousands of times per second, there are specific time delays in the image acquisition and transmission process, resulting in slow detection speeds. This invention utilizes a droplet parallelization strategy to increase system detection throughput. Droplet parallelization utilizes flow field forces to convert a single channel (80 microns) into eight sub-channels (30 microns wide) without the need for droplet splitting.

[0068] Under existing experimental conditions, the detection system provided by the present invention achieves a droplet detection rate of 1200 times per second and a liquid throughput of up to 2 microliters per minute.

[0069] In step (1), by setting the pressure conditions of the dispersed phase and the continuous phase to 300 MPa and 150 MPa respectively, the inventor unexpectedly found that there was no overlap between red blood cells and white blood cells in the image, which significantly improved the accuracy of cell classification and counting.

[0070] In an alternative embodiment, the image acquisition equipment in step (3) includes an inverted microscope (Nikon Ti2, 40x), a high-speed camera (Phantom), a computer (Windows 10), and an external light source to obtain high-resolution brightfield images of the droplets for subsequent analysis. The microfluidic chip is fixed in the field of view, and the microchannel is kept in the focal plane by adjusting the microscope adjustment knob, and the exposure time is set to 20 microseconds to eliminate motion blur.

[0071] In a preferred embodiment of the present invention, the trained deep learning neural network is obtained by training in the following manner: using a training set containing classified target cell images as input data, and performing unsupervised training on the deep learning neural network corresponding to the training set;

[0072] In a preferred embodiment of the present invention, the deep learning neural network is a Yolo series deep learning neural network;

[0073] In a preferred embodiment of the present invention, the Yolo series deep learning neural network is selected from Yolo-V6, Yolo-V8 or Yolo-V11;

[0074] In a preferred embodiment of the present invention, the Yolo series deep learning neural network is selected from Yolo-V11. The inventors have found that the Yolo-V11 deep learning neural network has higher classification and counting accuracy than other Yolo-V6 and Yolo-V8 deep learning neural networks.

[0075] In a preferred embodiment of the present invention, the model training parameters are as follows: the momentum is set to 0.937, the initial learning rate is 0.0005, and the weight decay rate is 0.005; under the above model training parameters, the trained model has good cell type classification and counting effects.

[0076] In a preferred embodiment of the present invention, a batch size of 32 and an input image resolution of 1024 pixels were used. 330 iterations were performed to better analyze the training process. During training, the neural network was gradually fine-tuned to approximate the true value, and the trained model was then evaluated on a test set.

[0077] In a preferred embodiment of the present invention, the training data further includes: image preprocessing before training before inputting the training data, and the image preprocessing before training includes grayscale mapping and image sharpening.

[0078] In a preferred embodiment of the present invention, step (4) further includes: image preprocessing before cell counting before inputting the bright field image into the trained deep learning neural network; the image preprocessing before cell counting is selected from: background phase reduction, high-pass filtering and grayscale conversion.

[0079] In a preferred embodiment of the present invention, before the morphological algorithm performs morphological processing in step (4), the method further includes: first using a droplet algorithm to identify droplet trajectories and a single-frame droplet image; then, performing a preprocessing operation to remove the influence of small objects on droplet identification;

[0080] In a preferred embodiment of the present invention, the preprocessing is selected from at least one of the following: threshold processing, connected domain identification and image sharpening;

[0081] In a preferred embodiment of the present invention, the morphological algorithm is a morphological algorithm in OpenCV.

[0082] In a preferred embodiment of the present invention, the volume of a single droplet is obtained by combining the contour information with the droplet volume model, thereby obtaining the total volume of the sample liquid passing through the microfluidic system.

[0083] In step (1), the continuous phase is fluorinated oil, silicone oil, vegetable oil or mineral oil, and the dispersed phase includes a whole blood solution and a diluent; or, the continuous phase includes a whole blood solution and a diluent, and the dispersed phase is isopropyl palmitate.

[0084] In the initial stage of model establishment, fluorinated oil, silicone oil, vegetable oil or mineral oil is used as the continuous phase, and whole blood solution is used as the dispersed phase to collect image features and fit the volume model calculation formula. In one embodiment, a person skilled in the art can verify the formula of the aforementioned droplet volume calculation model as needed to evaluate the fit of the droplet volume calculation model to the droplet volume. In the droplet volume model, a confocal microscope is used to measure the droplet volume, and the droplet volume is detected with the help of a fluorescence field. During the detection, a mixture of an analytical diluent and fluorescein (the mixing mass ratio is 95-105:1) is used as the dispersed phase, and fluorinated oil is used as the continuous phase to obtain fluorescent droplets that are bright in the confocal microscope field of view to measure the volume, thereby completing the volume model verification.

[0085] The inventors found in actual experiments that using isopropyl palmitate as the dispersed phase and a whole blood solution and a diluent as the continuous phase can greatly improve the cell flux.

[0086] Although silicone oil, vegetable oil, mineral oil, and electronic fluorinated liquid can be used as the continuous phase, their cell flux effects are far inferior to isopropyl palmitate.

[0087] The addition of fluorescein facilitates three-dimensional imaging of the droplets according to the principle of fluorescence imaging. Fluorescent agents include, for example, sodium fluorescein, carboxyfluorescein, etc. Carboxyfluorescein may be, for example, 5-carboxyfluorescein, 6-carboxyfluorescein, or 5(6)-carboxyfluorescein. 5(6)-carboxyfluorescein is a mixture of 5-carboxyfluorescein and 6-carboxyfluorescein in any ratio.

[0088] In a preferred embodiment of the present invention, the analytical diluent is selected from COULTER ISOTON III;

[0089] In a preferred embodiment of the present invention, the sample to be tested is one or more of a blood sample, a bone marrow biopsy sample, an inflammatory synovial fluid, a pleural effusion, an ascites, a saliva, a urine, a tissue or a fetal early detection sample.

[0090] In a second aspect, the present invention provides a system for absolute cell counting based on a microfluidic chip, which comprises: a droplet generation module, a droplet volume calculation module, an image acquisition module, and an image analysis module;

[0091] The droplet generation module is used to generate oil-in-water droplets for the cell sample to be tested. The oil-in-water droplet generation includes: generating droplets by a pressure pump, controlling the pressure applied by the pressure pump to the dispersed phase inlet flow channel to be 300MPa±10MPa, and controlling the pressure applied by the pressure pump to the continuous phase inlet flow channel to be 150MPa±10MPa; the droplet generation frequency is 1200Hz±50Hz;

[0092] The droplet volume calculation module is used to: change the motion state of the droplets by controlling the pressure conditions at the droplet capture outlet to allow the droplets to enter the capture array; perform droplet detection using droplet parallelism; measure the static volume of the droplets, establish a droplet volume calculation model, and calculate the volume of the droplets entering the capture array based on the droplet volume calculation model to obtain the total volume of the droplets;

[0093] The droplet volume calculation model formula is:

[0094] V t =f(S T , w, h, θ);

[0095]

[0096] Among them, S T is the area of ​​the top view, w is the width of a single channel of the microchannel, h is the height of the microchannel, a, b, c are the semi-axis lengths of the ellipsoid along the x-axis, along the y-axis, and along the z-axis, respectively, dz represents the integral, and it is assumed that And c is calculated from the contact angle, and the contact angle formula is:

[0097] The value range of θ is 100°~150°;

[0098] The image acquisition module is used to: obtain bright field images of droplets on the microfluidic chip through a microscopic imaging platform;

[0099] The image analysis module is used to process brightfield images and obtain cell concentration information. Processing brightfield images includes inputting the brightfield images into a trained deep learning neural network to obtain time series information and position information of droplets in the brightfield images, and to obtain the number and type of identified cells.

[0100] Acquisition of cell concentration information includes: comparing the original image of the droplet with the droplet position information through morphological algorithm processing to obtain the droplet contour information; combining the contour information with the droplet volume model to obtain the volume of each cell type in the sample, and calculating the concentration of each cell type by the total cell volume and the number of cells of the corresponding type.

[0101] The image acquisition module, consisting of an inverted microscope (Nikon Ti2, 40x), a high-speed camera (Phantom), a computer (Windows 10), and an external light source, acquired high-resolution brightfield images of the droplets for subsequent analysis. The microfluidic chip was fixed in the field of view, and the microchannels were kept in focus by adjusting the microscope knobs. The exposure time was set to 20 microseconds to eliminate motion blur.

[0102] The system also includes a droplet capture module, which uses a chip with a capture structure to capture droplets. Dynamic images of the droplets are acquired using a confocal microscope (e.g., a high-speed camera). Three-dimensional image reconstruction and volume calculation of the captured droplets are performed to evaluate the accuracy of the droplet volume calculation model. The droplet capture module serves only as a validation of the droplet volume calculation model and assesses its accuracy. During actual absolute cell counting, this module can be enabled or disabled as needed.

[0103] In a preferred embodiment of the present invention, the trained deep learning neural network is obtained by training in the following manner: using a training set containing images of classified target cells (such as red blood cells, white blood cells, and platelets) as input data, and performing unsupervised training on the deep learning neural network corresponding to the training set;

[0104] In a preferred embodiment of the present invention, the deep learning neural network is a Yolo series deep learning neural network;

[0105] In a preferred embodiment of the present invention, the Yolo series deep learning neural network is selected from Yolo-V6, Yolo-V8 or Yolo-V11;

[0106] In a preferred embodiment of the present invention, the Yolo series deep learning neural network is selected from Yolo-V11;

[0107] In a preferred embodiment of the present invention, the model training parameters are as follows: the momentum is set to 0.937, the initial learning rate is 0.0005, and the weight decay rate is 0.005;

[0108] In a preferred embodiment of the present invention, the batch size is set to 32 and the input image resolution is 1024.

[0109] In a third aspect, the present invention further provides an electronic device comprising: a processor and a memory; the processor and the memory are connected, wherein the memory is used to store a computer program, and the processor is used to call the computer program to execute the above-mentioned method for absolute cell counting based on a microfluidic chip.

[0110] In a fourth aspect, the present invention further provides a computer-readable storage medium comprising instructions, which, when executed on a computer, enables the computer to execute the above-mentioned method for absolute blood cell counting based on a microfluidic chip.

[0111] Specifically, the electronic device may include a memory, a processor, a bus, and a communication interface, wherein the memory, processor, and communication interface are electrically connected to each other directly or indirectly to enable data transmission or interaction. For example, these components may be electrically connected to each other via one or more buses or signal lines. The processor may process information and / or data related to target identification to perform one or more functions described in this application.

[0112] The memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0113] A processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including a central processing unit (CPU) or a network processor (NP). It can also be 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.

[0114] To make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely below. Where specific conditions are not specified in the embodiments, conventional conditions or conditions recommended by the manufacturer are used. Where the manufacturer of the reagents or instruments is not specified, all are conventional products that can be purchased commercially.

[0115] The features and performance of the present invention are further described in detail below with reference to the embodiments.

[0116] Example 1

[0117] This embodiment provides a microfluidic chip-based absolute cell counting system for accurate quantification of cell concentration. The system comprises a droplet generation module, a droplet volume calculation module, an image acquisition module, and an image analysis module. The first two modules work together to control droplet motion, while the second two modules are responsible for image acquisition and information processing.

[0118] The droplet generation module includes a pressure pump, a reservoir, and piping. The droplet generation state and droplet encapsulation effect are controlled by adjusting the flow rates of the dispersed and continuous phases. The pressure applied by the pressure pump to the dispersed phase inlet channel is controlled to be 300 MPa, and the pressure applied to the continuous phase inlet channel is controlled to be 150 MPa.

[0119] The droplet volume calculation module, primarily composed of a precision pump, controls the pressure at the droplet capture outlet to change the motion of the droplets, allowing them to enter the capture array. The captured droplets can be used for 3D imaging (which can be used to verify the volume model) and volume model construction. By combining dynamic droplet images with static volume measurements, a droplet volume calculation model is established. Based on this droplet volume calculation model, the volume of the droplets entering the capture array is calculated to obtain the total volume of the droplets.

[0120] The droplet volume calculation model formula is:

[0121] V t =f(S T , w, h, θ);

[0122]

[0123] Among them, S T is the area of ​​the top view, w is the width of a single channel of the microchannel, h is the height of the microchannel, a, b, c are the semi-axis lengths of the ellipsoid along the x-axis, along the y-axis, and along the z-axis, respectively, dz represents the integral, and it is assumed that And c is calculated from the contact angle θ, and the contact angle θ formula is:

[0124]

[0125] The value range of θ is 100°~150°.

[0126] The image acquisition module, consisting of an inverted microscope (Nikon Ti2, 40x), a high-speed camera (Phantom), a computer (Windows 10), and an external light source, was used to acquire high-resolution brightfield images of the droplets for subsequent analysis. The microfluidic chip was fixed in the field of view, and the microfluidic channel (or microchannel) of the microfluidic chip was kept in the focal plane by adjusting the microscope knobs. The exposure time was set to 20 microseconds to eliminate motion blur.

[0127] The image analysis module is used to process video images and obtain cell concentration information for the sample being measured. Processing brightfield images involves: inputting the brightfield image into a trained deep learning neural network to obtain the time series and position information of the droplets in the brightfield image, and determining the number and type of cells identified;

[0128] Acquisition of cell concentration information includes: comparing the original image of the droplet with the position information of the droplet through morphological algorithm processing to obtain the contour information of the droplet; combining the contour information with the droplet volume model to obtain the total volume of the sample liquid passing through the microfluidic system, and calculating the concentration of each cell type by the total volume of the cells and the number of cells of the corresponding type.

[0129] Example 2

[0130] This example provides a method for absolute cell counting based on the system of Example 1. The working principle is as follows:

[0131] In principle, the digital droplet method utilizes the surface tension of a liquid to transform a continuous liquid into thousands of incompatible droplets. The total sample concentration is determined by accumulating the detection results of each droplet. The volume flow rate (Q = Q1 + Q2 + ...) is obtained by adding the volumes of each droplet, and the number of cells in the sample is obtained by summing the number of encapsulated cells. This linear superposition relationship ensures that the sample detection accuracy is consistent with the droplet detection accuracy.

[0132] This paper constructs an optical imaging platform and an automated recognition algorithm for droplet detection. A peripheral precision pump delivers a cell suspension into a microfluidic chip, where monodisperse droplets are formed in the confocal region. Although droplet generation can reach thousands of times per second, the image acquisition and transmission process exhibits a specific time delay, resulting in slow detection. This paper employs a parallel droplet detection strategy to increase system throughput.

[0133] The microfluidic chip consists of three components: droplet generation, parallelization, and droplet detection. See the schematic diagram. Figure 1A in Figure 1. Droplet generation relies on the dimensionless capillary number (Ca = ηv / γ), and droplet size is varied by controlling the two-phase flow rate. Droplet parallelization utilizes flow field forces to transform a single channel (80 microns) into eight subchannels (30 microns wide) without droplet fragmentation. A contour recognition algorithm can be used to calculate droplet volume information. Under existing experimental conditions, the system achieved a droplet detection rate of 1200 times per second and a liquid throughput of up to 2 microliters per minute.

[0134] By using the principle of optical imaging, we can obtain high-resolution cell video images. By measuring the volume of multiple droplets, we can obtain a three-dimensional model of the droplets, such as Figure 1 As shown in Figure B. The volume of the droplet can be accurately calculated using its top view area, surface contact angle, and channel height. The three types of blood cells differ significantly in morphology and size, which will help image recognition algorithms achieve classification, recognition, and counting. Deep learning algorithms are used to classify and count cells in droplets. We selected the Yolo-v11 algorithm as the network model. The process of cell counting using deep learning is shown in the figure below. Figure 1 As shown in Figure C. The Yolo algorithm uses an end-to-end network structure and is designed to obtain target location and category information through sequential convolution, which has efficient detection capabilities and accurate detection accuracy.

[0135] The specific steps of sample processing, microfluidic chip fabrication, and absolute counting method are as follows:

[0136] 1. Sample Preparation and Processing

[0137] The cell diluent constituting the continuous phase is prepared by mixing whole blood solution with analytical diluent (COULTER ISOTON III) in a ratio of 1:80. Before being added to the reservoir, it needs to be thoroughly mixed using a vortex mixer. Among them, blood samples were provided by healthy subjects from the First Affiliated Hospital of Zhengzhou University (Zhengzhou, China) and collected during physical examinations. It should be noted that the present invention has been approved by the Ethics Review Committee of the First Affiliated Hospital of Zhengzhou University. A laboratory-specific blood cell counting plate (MARIENFELD, Germany) is used to determine the concentration value of the blood cell solution. The dispersed phase uses isopropyl palmitate (Isopropyl Palmitate, China) with a density and viscosity of 0.86 g / mL and 6.2 mPa-S, respectively. In the process of modeling the capture droplet volume, it is necessary to use the principle of fluorescence imaging to perform three-dimensional imaging of the droplets. Sodium fluorescein (sodium fluorescein) is mixed with the above-mentioned analytical diluent in a mass ratio of 1:100 as the discrete phase for droplet generation.

[0138] 2. Fabrication of Microfluidic Devices (Chips)

[0139] A PDMS (polydimethylsiloxane) microfluidic chip was designed and fabricated. The geometry was designed using L-edit software, and the chip was fabricated using soft lithography and rapid prototyping techniques. The fabrication process is roughly divided into three steps:

[0140] First, the channel geometry was printed on a transparent chrome mask, and then the SU-8 wafer was fabricated on a silicon wafer using conventional photolithography. Next, the PDMS base, PDMS curing agent, and active agent (Triton X-100) were mixed in a mass ratio of 1000:100:1 and poured into the master mold. After curing in an 80°C oven for 2 hours, a PDMS sheet with a channel network was formed. When the PDMS sheet cooled, it was cut and punched. Finally, the PDMS cover and glass slide surfaces were oxidized using the principle of oxygen plasma bonding to complete the assembly of the microfluidic chip.

[0141] 3. Absolute cell counting steps:

[0142] (1) Droplet generation and cell encapsulation

[0143] A diluted blood cell solution is used for the dispersed phase, while a fluorinated oil is used for the continuous phase. Before the experiment, the continuous phase is passed into the flow channel (also called the channel) of the microfluidic chip for a period of time to check the tightness and connectivity of the path. Droplets are generated in the microfluidic chip by controlling the two-phase flow pressure using a pneumatic pump. The pressure applied by the pressure pump to the dispersed phase inlet flow channel is controlled to be 300MPa, and the pressure applied by the pressure pump to the continuous phase inlet flow channel is controlled to be 150MPa. By setting the pressure conditions of the dispersed phase and the continuous phase to 300MPa and 150MPa respectively, the inventors unexpectedly found that there was no overlap between red blood cells and white blood cells in the image, which significantly improved the accuracy of cell classification and counting.

[0144] The droplet generation frequency is approximately 1200 Hz, and the cell throughput is approximately 2000 cells per second. The image acquisition frame rate can be further increased to achieve even higher cell throughput.

[0145] (2) Droplet capture

[0146] By combining dynamic imaging with static volume measurements, a droplet volume calculation model was established. The microfluidic chip features two outlets: a capture outlet and a waste outlet. Due to the high flow resistance at the capture outlet, droplets typically flow out through the waste outlet. When the actual droplet volume needs to be measured, a precision pump is used to sort the droplets, directing them into the capture array structure. A preset pressure signal ensures the stability of the sorting process and prevents droplet breakup. Rectangular pulses are often used in experiments.

[0147] Specifically, the pressure conditions at the droplet capture outlet are controlled to change the motion state of the droplets so that the droplets enter the capture array; droplet detection is performed in parallel using droplets; the static volume of the droplets is measured, a droplet volume calculation model is established, and the volume of the droplets entering the capture array is calculated based on the droplet volume calculation model to obtain the total volume of the droplets;

[0148] The droplet volume calculation model formula is as follows:

[0149] V t =f(S T , w, h, θ);

[0150]

[0151] Among them, S T is the area of ​​the top view, w is the width of a single channel of the microchannel, h is the height of the microchannel, a, b, c are the semi-axis lengths of the ellipsoid along the x-axis, along the y-axis, and along the z-axis, respectively, dz represents the integral, and it is assumed that And c is calculated from the contact angle θ, and the contact angle θ formula is:

[0152]

[0153] The value range of θ is 100°~150°.

[0154] The above formula of the droplet volume calculation model is an optimized formula. When performing droplet volume calculation, those skilled in the art can directly apply the above formula to calculate the droplet volume.

[0155] (3) Image collection

[0156] The bright field image of the droplets on the microfluidic chip was obtained using the microscopic imaging platform of Example 1.

[0157] (4) Process bright field images and obtain cell concentration information:

[0158] i. Dataset and neural network training

[0159] In this example, a deep learning algorithm was used to classify and count cells within droplets, with the Yolo-v11 algorithm selected as the network model. The Yolo family of deep neural networks was first proposed at the 2016 CVPR conference and has seen rapid development in recent years. The Yolo algorithm utilizes an end-to-end network architecture and uses sequential convolution to acquire target location and category information, resulting in efficient detection capabilities and accurate accuracy.

[0160] The dataset annotated in this embodiment contains a total of 621 red blood cell images, 572 white blood cell images, and 929 platelet images (wherein the red blood cell images are diluted with whole blood, and the white blood cell and platelet datasets are centrifuged). It is divided into a training set (498 red blood cells, 451 white blood cells, and 778 platelets), a validation set (68 red blood cells, 57 white blood cells, and 88 platelets), and a test set (55 red blood cells, 64 white blood cells, and 94 platelets). There is no overlap between the three groups of data to ensure the independence of each group of data. The size of all images in the dataset is unified to 832×800 pixels. All original images were processed by image processing programs during the initial training phase, including grayscale mapping and image sharpening.

[0161] The Yolo-V11 model training parameters are as follows: momentum is set to 0.937, initial learning rate is 0.0005, and weight decay rate is 0.005. The batch size is set to 32, and the input image resolution is 1024. 330 iterations were performed to better analyze the training process. During training, the neural network is gradually fine-tuned to approach the true value, and the trained model is then evaluated on the test set. The weight file is saved after each iteration, and the mean average performance (mAP) on the validation set is calculated every three epochs.

[0162] ii. Blood cell counting using a trained deep learning neural network

[0163] The trained deep learning neural network can classify and count all intact droplets and cells in the entire image. By combining the image's time series and location information with the original image, effective information such as droplet size and cell type can be determined.

[0164] Specifically: First, the acquired droplet bright field images need to be preprocessed to eliminate background noise and improve the accuracy of droplet and cell identification, such as background phase reduction, high-pass filtering, and grayscale conversion.

[0165] The processed image is then input into the trained deep learning neural network to obtain the time series information and position information of the droplets in the bright field image to determine the number of the three types of blood cells in each droplet.

[0166] Finally, the deep learning neural network output file with time series and location information is processed to obtain the accurate number and type of the three types of cells identified, and the number of the three types of blood cells in the sample is counted.

[0167] iii. Sample volume measurement and cell concentration quantification

[0168] Image processing is used to obtain the position and size information of the droplets.

[0169] First, a droplet algorithm is used to identify droplet trajectories and a single-frame droplet image. Preprocessing operations, such as thresholding, connected domain identification, and image sharpening, are then used to remove the influence of small objects on droplet identification. Next, a morphological algorithm is used to process the droplet's original image based on its positional information to obtain its contour information. Specifically, contour information such as area and aspect ratio is obtained. The entire image algorithm is implemented using OpenCV's built-in algorithm modules, making the process relatively simple.

[0170] Finally, the contour information is combined with the droplet volume calculation model to obtain the total volume of the sample liquid. The absolute concentration of red blood cells (red blood cells / L) = the total number of red blood cells in the measured droplet / the total volume of the sample liquid.

[0171] Experimental Example 1

[0172] This experimental example provides a specific optimization process for the droplet volume calculation model:

[0173] Droplet volume measurement and calibration

[0174] Due to the effect of surface tension, the squeezed droplets will not completely fill the microchannel, but will form a specific gap at the edge, which is bullet-shaped. Traditional image analysis methods treat droplets as cylinders, and there are significant errors in volume quantification. For a characteristic droplet with a length of 80 microns, the total number of pixels is about 9600 (20*80*2*2), which is approximately 440 ((20+80)*2*2), and the total volume error is about 4.5%. For smaller droplets, the edge occupies a greater proportion of the droplet area, resulting in greater errors. The minimum detected droplet (the droplet diameter matches the width of the microchannel) has a volume of about 0.03 nanoliters and an error rate of 13%. In order to more accurately measure the droplet volume, the existing model needs to be optimized.

[0175] In this experimental example, the detection structure of the microfluidic chip contains multiple 30-micron subchannels (much smaller than the diameter of the generated droplets). This design allows the channel sidewalls to provide uniform extrusion pressure on the droplets. After extrusion, the droplet sidewalls will be more regular, and the correlation between droplet length and volume will be more significant. Multiple droplet volumes with length information were tested to model this consistent relationship. The statistical results were also fitted to a linear model.

[0176] It's important to note that existing volume measurement methods, such as confocal microscopy and optical vector methods, are only applicable to stationary and slow-moving droplets. Therefore, this experimental example employed confocal microscopy to reconstruct 3D images and calculate the volume of captured droplets, further validating the accuracy of the droplet volume calculation model constructed in Example 2.

[0177] The confocal microscope measurement method is as follows:

[0178] A confocal microscope was used to reconstruct three-dimensional images and calculate the volume of the captured droplets. The selected imaging platform was a Leica series Stellaris confocal microscope with an objective lens magnification of 60 times, and excitation and emission wavelengths of 488 nm and 510 nm, respectively. The general operating steps are as follows: First, adjust the stage, place the droplet to be measured in the center of the field of view, and turn the focus knob to obtain a clear droplet image. Then, set the optical path conditions, shooting conditions, image gain and other parameters on the software console to adjust the signal-to-noise ratio of the fluorescence image. Finally, a 1-micron resolution tomography scan was performed within the imaging area to obtain a two-dimensional slice image of the droplet.

[0179] By combining calculus principles with image processing algorithms, the true volume of a droplet can be accurately calculated. In this experimental example, the slice image is thresholded and connected domains are identified to determine the droplet's slice size. The geometric volume of the droplet is then determined by integrating the areas of multiple slice images. Finally, the true volume of the droplet is restored based on the image resolution, completing the volume measurement.

[0180] According to the above confocal microscope measurement method, this experimental example will use a high-speed camera to obtain the droplet plane motion image, such as Figure 2 As shown in Figure A, the original image is differentiated and contours are recognized. The confocal microscope measures the stereo image of the stationary droplet and captures the stationary droplet through a capture operation. Figure 2 Figure B shows the 3D measurement process of the confocal microscope. The length and volume fitting diagram is shown in Figure 2 As shown in Figure C, the corrected volume calculation model (the droplet volume calculation model provided in Example 2) achieves a measurement accuracy exceeding 92.1%. This means that the droplet volume calculation model of Example 2 is highly consistent or well-matched with the results measured using a confocal microscope. The droplet volume calculation model provided in Example 2 can be used to accurately calculate the volume of droplets generated by microfluidic chips.

[0181] Experimental Example 2

[0182] This experimental example provides a classification experiment after cell imaging and a deep learning neural network test after training according to the counting method provided in Example 2.

[0183] 1. Cell Imaging

[0184] The images of white blood cells and red blood cells were collected in the form of single samples, and the samples were processed using the discrete phase cell preparation method described in Example 1.

[0185] Since platelets are smaller and more distinct from the other two cell types, images were collected manually. Figure 3A. During cell movement, there is a certain degree of deflection. During imaging, red blood cells exhibit different morphological characteristics, such as double rings, dumbbells, etc. ( Figure 3 Left image of A). White blood cells are more regular spherical ( Figure 3 Platelets appear as bright dots ( Figure 3 (right image of center A).

[0186] 2. Train and test using Yolo-v11 network

[0187] After obtaining the dataset of three types of blood cells, the training process was carried out on the Yolo-v11 model. Compared with its predecessor, for example, the Yolo-v8 training model, the average recognition accuracy of the three types of blood cells was only 88%, and a larger dataset (more than 8,000 images) was required to achieve a better improvement. Yolo-v11 no longer requires too many trainings and too large datasets to achieve a better effect. The loss (Avgloss) and mAP of each iteration during the training process are as follows: Figure 3 As shown in D. Figure 3 As shown in D in Figure 3, the model learns more efficiently and converges faster in the early stages of training. As training progresses, the convergence rate of the curve gradually slows. Ultimately, when the number of iterations reaches approximately 330, the model's learning ability gradually saturates, and the final loss value stabilizes at around 0.4. As the loss value decreases and stabilizes, the model gradually converges. The mAP curve also shows a similar trend. Here, the IoU threshold is set to 0.5 when calculating mAP, meaning that if the calculated IoU is greater than 0.5, the detection is considered correct.

[0188] In this model, single-frame cell detection achieves high accuracy. This is due to the shear force generated by the channel design and the high pump pressure, which causes slight morphological changes in red blood cells while the morphological changes in white blood cells are not obvious. The shear force has a positive effect on distinguishing red blood cells from white blood cells. Platelets are much smaller than red blood cells and white blood cells, and the morphological distinction is clear. However, because the pixels occupied by a single platelet are too small, the initial training results for platelets were not very ideal. By adding Gaussian noise and motion blur, and adjusting the classification loss weight, the training results for platelets reached a satisfactory level.

[0189] The closer the indicated PR curve is to the (1,1) point, the better the model performance is. The corresponding PR curve is shown in Figure 3 Middle B. Here, all curves for the three cell lines and the oil droplet are very close to the upper right corner. After training, a well-trained cell detection model can make classification decisions and determine location coordinates. The average accuracy of these three blood cell (AP) detections was 99.11%, 99.10%, and 98.45%, respectively.

[0190] from Figure 3 It can be seen from the C confusion matrix that most of the recognition errors can be divided into two categories: a very small number of white blood cells are mistaken for red blood cells, and platelets are omitted by the system of the present invention, and the frequency of platelet omission dominates these error events.

[0191] Experimental Example 3

[0192] This experimental example tests the effectiveness and stability of the system provided in Example 1.

[0193] To evaluate the effectiveness and stability of the detection system, this experimental example used the system to conduct blood cell count and concentration comparison experiments, achieving differential count and absolute quantification of blood cells. The continuous phase of the cell diluent consisted of a proportional mixture of whole blood solution and an analytical diluent (COULTER ISOTON III). The dispersed phase employed isopropyl palmitate (China), with a density and viscosity of 0.86 g / mL and 6.2 mPa-s, respectively. This procedure significantly improved overall throughput without compromising accuracy and stability.

[0194] Figure 4 Center A shows the results of using the system to test a 30-fold diluted blood sample from a normal person. Figure 4 The two curves shown in A represent the ratio of the number of red blood cells to the number of white blood cells (R 2 =0.9956) and the ratio of red blood cells to platelets (R 2 =0.9972). As the number of red blood cells increases, the increase in the number of white blood cells and platelets is in line with expectations, and the two fitting curves are good.

[0195] In the experiment, whole blood samples were diluted into cell solutions of different ratios (30 times, 50 times, 70 times, 90 times, 110 times, 130 times, and 150 times), and the number of cells in the droplets was accurately counted. Figure 4 As shown in Figure B, the cell count at different dilution ratios is proportional to time, and the higher the dilution ratio, the smaller the slope of the count curve. This means that the microfluidic system can maintain normal operation at high throughput (approximately 1200 drops / second) and can stably test blood samples of varying concentrations. Currently, the system has an image acquisition frame rate of 200 fps, which can maintain a stable throughput level for 30 minutes.

[0196] Based on the above results, it can be concluded that the system can maintain good stability within a large concentration range and can accurately detect the relative concentration ratios among RBC, PLT, and WBC.

[0197] Experimental Example 4

[0198] Clinical blood sample testing.

[0199] To verify the quantitative accuracy of the system, this experiment compared the detection results of the LH750 (Beckman Coulter, USA) and the microfluidic system. In this experiment, an 80-fold diluted blood sample was selected for detection.

[0200] This experiment used 20 samples. Each concentration sample was repeated three times under the same experimental test conditions and processing, and the average was taken to obtain the microfluidic system concentration results. The image acquisition time for each experiment was 10 minutes.

[0201] Figure 5 A in the figure shows the three-dimensional diagram of the error range between the microfluidic system and clinical samples, where E R Represents the error range of red blood cells, E W Represents the error range of white blood cells, E P The error range of platelets is shown in Figure 2. It can be seen that the error range is less than 10%. Figure 5 Figure B shows a violin plot of 60 experiments using the microfluidic system on 20 clinical samples. The smaller mean values ​​(0.31, 0.68, -0.76) and standard deviations (6.36%, 6.32%, 7.33%) demonstrate the sensitivity and specificity of this system. Figure 5 C in the middle represents the comparison of the red blood cell concentration test results obtained using the microfluidic system and the LH750, R 2 =0.9813. Figure 5 D in the middle represents the comparison of the leukocyte concentration test results obtained using the microfluidic system and LH750, R 2 =0.9829. Figure 5 E in the middle represents the comparison of the red blood cell concentration test results obtained using the microfluidic system and the LH750

[0202] R2 = 0.9812. These three figures demonstrate that there is good consistency between the microfluidic system measurement results and clinical methods.

[0203] In summary, by combining target detection technology and volume quantification technology, the microfluidic system of the present invention achieves absolute counting of blood cells under laboratory conditions. According to the observation conditions of cell imaging, a droplet microfluidic chip with multiple parallel channels was designed and manufactured. Subsequently, a microscopic imaging system was constructed around the microfluidic chip, and image acquisition and analysis were completed; target detection and data analysis were achieved using the YOLO v11 network algorithm. Finally, a concentration control experiment was set up to verify the effectiveness and stability of the detection system. This method has the advantages of no labeling and high precision. It can achieve rapid cell counting through morphological methods, and the real-time throughput can reach 2000 cells per second.

[0204] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for absolute cell counting based on a microfluidic chip, characterized in that: It includes the following steps: (1) Generating oil-in-water droplets of the cell sample to be tested: generating droplets by a pressure pump, controlling the pressure applied by the pressure pump to the dispersed phase inlet flow channel to be 300 kPa ± 10 kPa, and controlling the pressure applied by the pressure pump to the continuous phase inlet flow channel to be 150 kPa ± 10 kPa; the frequency of droplet generation is 1200 Hz ± 50 Hz; and the cell throughput is at least 1000 cells per second; (2) changing the motion state of the droplets by controlling the pressure conditions at the droplet capture outlet, so that the droplets enter the capture array; Performing droplet detection in parallel using droplets; measuring the static volume of the droplets, establishing a droplet volume calculation model, and calculating the volume of the droplets entering the capture array based on the droplet volume calculation model to obtain the total volume of the droplets; The droplet volume calculation model formula is: V t =f(S T ,w,h,θ); Among them, S T is the area of ​​the top view, w is the width of a single channel of the microchannel, h is the height of the microchannel, a, b, c are the semi-axis lengths of the ellipsoid along the x-axis, along the y-axis, and along the z-axis, respectively, dz represents the integral, and it is assumed that And c is calculated from the contact angle θ, and the contact angle θ formula is: The value range of θ is 100°~150°; (3) Obtaining bright-field images of droplets on the microfluidic chip using a microscopic imaging platform; (4) Processing the bright field image and obtaining cell concentration information: Inputting the bright field image into the trained deep learning neural network, obtaining the time series information and position information of the droplets in the bright field image, and obtaining the number and type of identified cells; Acquiring cell concentration information: Based on the droplet position information, the original image of the droplet is compared with the droplet through morphological algorithm processing to obtain the droplet contour information; combining this contour information with the droplet volume model, the total volume of the sample liquid passing through the microfluidic system is obtained. By calculating the total volume of the sample liquid and the number of cell types corresponding to it, the absolute concentration of each cell type is obtained.

2. The method for absolute cell counting based on a microfluidic chip according to claim 1, characterized in that: The trained deep learning neural network is obtained by training in the following manner: using a training set containing classified target cell images as input data, and performing unsupervised training on the deep learning neural network corresponding to the training set; Preferably, the deep learning neural network is a Yolo series deep learning neural network; Preferably, the Yolo series deep learning neural network is selected from Yolo-V6, Yolo-V8 or Yolo-V11; Preferably, the Yolo series deep learning neural network is selected from Yolo-V11; Preferably, the model training parameters are as follows: momentum is set to 0.937, initial learning rate is 0.0005, and weight decay rate is 0.005; Preferably, the batch size is set to 32 and the input image resolution is 1024; Preferably, the training data further includes: image preprocessing before training before inputting the training data, and the image preprocessing before training includes grayscale mapping and image sharpening.

3. The method for absolute cell counting based on a microfluidic chip according to claim 1, characterized in that: Step (4) before inputting the bright field image into the trained deep learning neural network, also includes: image preprocessing before cell counting; the image preprocessing before cell counting is selected from: background phase reduction, high-pass filtering and grayscale conversion.

4. The method for absolute cell counting based on a microfluidic chip according to claim 3, characterized in that: Before the morphological algorithm performs morphological processing in step (4), the method further includes: first using a droplet algorithm to identify droplet trajectories and a single-frame droplet image; then, performing a preprocessing operation to remove the influence of small objects on droplet identification; Preferably, the preprocessing is selected from at least one of the following: threshold processing, connected domain identification and image sharpening; Preferably, the morphological algorithm is a morphological algorithm within OpenCV.

5. The method for absolute cell counting based on a microfluidic chip according to claim 4, characterized in that: The contour information is combined with the droplet volume model to obtain the volume of a single droplet, and then the total volume of the sample liquid passing through the microfluidic system is obtained.

6. The method for absolute cell counting based on a microfluidic chip according to claim 1, characterized in that: In step (1), the continuous phase is fluorinated oil, silicone oil, vegetable oil or mineral oil, and the dispersed phase comprises a whole blood solution and a diluent; or, the continuous phase comprises a whole blood solution and a diluent, and the dispersed phase is isopropyl palmitate; Preferably, the analytical diluent is selected from COULTER ISOTON III; Preferably, the sample to be tested is one or more of a blood sample, a bone marrow biopsy sample, an inflammatory synovial fluid, a pleural effusion, an ascites, a saliva, a urine, a tissue or a fetal early detection sample.

7. A system for absolute cell counting based on a microfluidic chip, characterized in that: It includes: Droplet generation module, droplet volume calculation module, image acquisition module and image analysis module; The droplet generation module is used to generate oil-in-water droplets on the cell sample to be tested, wherein the oil-in-water droplet generation includes: generating droplets by a pressure pump, controlling the pressure applied by the pressure pump to the dispersed phase inlet flow channel to be 300kPa±10kPa, and controlling the pressure applied by the pressure pump to the continuous phase inlet flow channel to be 150kPa±10kPa; the droplet generation frequency is 1200Hz±50Hz; The droplet volume calculation module is used to: change the motion state of the droplets by controlling the pressure conditions of the droplet capture outlet to allow the droplets to enter the capture array; perform droplet detection using droplet parallelism; measure the static volume of the droplets, establish a droplet volume calculation model, and calculate the volume of the droplets entering the capture array based on the droplet volume calculation model to obtain the total volume of the droplets. The formula of the droplet volume calculation model is as follows: The droplet volume calculation model formula is: V t =f(S T ,w,h,θ); Among them, S T is the area of ​​the top view, w is the width of a single channel of the microchannel, h is the height of the microchannel, a, b, c are the semi-axis lengths of the ellipsoid along the x-axis, along the y-axis, and along the z-axis, respectively, dz represents the integral, and it is assumed that And c is calculated from the contact angle, and the contact angle formula is: The value range of θ is 100°~150°; The image acquisition module is used to: obtain a bright field image of the droplets on the microfluidic chip through a microscopic imaging platform; The image analysis module is used to process bright field images and obtain cell concentration information; processing bright field images includes: inputting the bright field images into a trained deep learning neural network, obtaining time series information and position information of droplets in the bright field images, and obtaining the number and type of identified cells; The acquisition of the cell concentration information includes: comparing the original image of the droplet with the position information of the droplet through morphological algorithm processing to obtain the contour information of the droplet; combining the contour information with the droplet volume model to obtain the volume of each type of cell in the sample, and calculating the concentration of each type of cell by the total volume of the cells and the number of cells of the corresponding type.

8. The system for absolute cell counting based on a microfluidic chip according to claim 7, characterized in that: The trained deep learning neural network is obtained by training in the following manner: using a training set containing classified target cell images as input data, and performing unsupervised training on the deep learning neural network corresponding to the training set; Preferably, the deep learning neural network is a Yolo series deep learning neural network; Preferably, the Yolo series deep learning neural network is selected from Yolo-V6, Yolo-V8 or Yolo-V11; Preferably, the Yolo series deep learning neural network is selected from Yolo-V11; Preferably, the model training parameters are as follows: momentum is set to 0.937, initial learning rate is 0.0005, and weight decay rate is 0.005; Preferably, the batch size is set to 32 and the input image resolution is 1024; Preferably, the system also includes a droplet capture module, which captures droplets through a chip with a capture structure, obtains dynamic images of the droplets through a confocal microscope, and evaluates the fitting accuracy of the droplet volume calculation model by performing three-dimensional image reconstruction and volume calculation on the captured droplets.

9. An electronic device, characterized in that: It includes: processor and memory; The processor is connected to the memory, wherein: The memory is used to store a computer program, and the processor is used to call the computer program to execute the method for absolute cell counting based on a microfluidic chip according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that The method comprises instructions which, when run on a computer, enable the computer to execute the method for absolute blood cell counting based on a microfluidic chip as claimed in any one of claims 1 to 6.

Images