CN113155713B - Label-free high-content video flow cytometer and method based on transfer learning - Google Patents

Abstract

The invention provides a label-free high content video flow cytometer based on transfer learning and a method thereof, wherein the device comprises: the device comprises an optical excitation module, a sheath flow control module, a data acquisition module and a data processing module; the invention adopts a two-dimensional light scattering technology, and has the sensibility of structural and morphological information; the flow technology is used, and an automatic data processing program based on transfer learning is utilized, so that the operation flow of data acquisition is simplified, and human intervention is reduced; the method without marking is adopted, so that the workload of sample preparation can be greatly reduced, the damage to the sample is reduced, the timeliness is high, the sample requirement is low, and the application field is greatly expanded; the high-speed high-resolution image acquisition component is integrated, so that the space and time resolution of the high-content streaming image can be ensured, and the traceable positioning of the pattern can be performed; the added video trigger mechanism can reduce the redundancy of the acquired data as much as possible.

Description

Label-free high-content video flow cytometry and method based on transfer learning

technical field

The invention relates to the technical field of high-end medical equipment, in particular to a marker-free high-content video flow cytometer and method based on transfer learning.

Background technique

The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

Flow cytometry is one of the most commonly used instruments in the field of single cell analysis. It is both accurate and efficient. It can not only quickly and effectively provide individual cell morphology or component information, but also detect and identify cells in a short time. and count a large number of cells. In general, identification of specific cells using flow cytometry requires the labeling of individual cells with specific fluorescent dyes. The cell labeling process will inevitably introduce various chemical reagents and complicated operating procedures, resulting in additional time and material costs and irreversible changes in the properties of cell samples. In some application fields, the samples need to be tested as soon as possible or the samples need to remain active, and the sample labeling operation required by the existing flow cytometer will have limitations.

The inventors found that imaging flow cytometer is a new type of flow cytometry single-cell detection instrument, which has both the high-throughput characteristics of flow cytometer and the microscopic imaging capability of fluorescence microscope, and can Acquisition of single cell images. Multi-channel imaging is the biggest advantage of imaging flow cytometry, which can realize multiple image acquisition on a single sensor. Therefore, the sensor imaging area occupied by each channel is limited, and the obtained image quality is also relatively limited. When using the existing imaging flow cytometer for single-cell label-free imaging, the number of pixels of the cell image acquired by the bright field channel is low, which can only provide less information about cell morphology and structure, which greatly limits its application. potential. In addition, the data provided by the existing imaging flow cytometer is the image of the sample, and video data cannot be displayed. Static data cannot provide more samples of the original state data, and data retrospective search cannot be performed, which also limits its application fields to a certain extent.

Contents of the invention

In order to solve the deficiencies of the existing technology, the present invention provides a label-free high-content video flow cytometer detection flow cytometer and method based on migration learning, which adopts two-dimensional light scattering technology and has the sensitivity of structural and morphological information ; Use streaming technology and use automatic data processing procedures based on transfer learning to simplify the operation process of data collection and reduce human intervention; use a label-free method, which can greatly reduce the workload of sample preparation and reduce damage to samples. The higher timeliness and lower sample requirements greatly expand the field of use; the integration of high-speed and high-resolution image acquisition components can ensure the spatial and temporal resolution of high-content streaming images, and can perform pattern resiliency. Backtracking positioning; the added video trigger mechanism can minimize the redundancy of collected data.

In order to achieve the above object, the present invention adopts the following technical solutions:

The first aspect of the present invention provides a label-free high-content video flow cytometer based on migration learning.

A label-free high-content video flow cytometer based on migration learning, including: an optical excitation module, a sheath flow control module, a data acquisition module and a data processing module;

The excitation beam emitted by the optical excitation module is transmitted to the sheath flow chamber of the sheath flow control module, the video image is collected through the data acquisition module, and the collected image data is transmitted to the data processing module for cell detection;

Among them, the sheath flow control module can limit the spatial flow of the sample liquid, so that the sample forms a flowing single-cell sequence. At the same time, the excitation beam generated by the optical excitation module is coupled into the sample flow channel after being shaped by the objective lens, so that the sample liquid and the excitation beam Coincident only within the preset area.

As some possible implementations, the optical excitation module at least includes lasers arranged sequentially along the optical path, a neutral density sheet for light intensity control, a laser diaphragm for collimating and switching the optical path, a mirror for adjusting the direction, and Excitation objective lens used for shaping, the laser beam passes through the excitation objective lens to obtain a compressed excitation beam.

As some possible implementations, the sheath flow control module is used to drive the cell solution to form a single-cell passage, including a sheath flow chamber, a sample fluid injection pump, and a sheath fluid injection pump. The sample fluid flows in from the middle channel of the sheath flow chamber, and the sheath fluid flows from the sheath The channel around the flow chamber flows in, and the velocity of the sheath fluid is greater than the velocity of the sample fluid;

Before the sheath liquid enters the sheath flow chamber, it will pass through a buffer chamber to stabilize the liquid flow and pre-disperse its flow direction, so as to ensure that the sheath flow can simultaneously compress and control the sample liquid in two orthogonal directions.

As a further limitation, the sheath flow control module further includes a waste liquid pool for collecting the sample and the sheath liquid after passing through the observation area.

As some possible implementations, the data acquisition module includes an acquisition optical path and a data path. The acquisition optical path at least includes an acquisition objective lens, a high-speed CMOS camera, and a trigger. The position of the acquisition objective lens is at the front, which is used to control the storage time of the camera;

The data path transmits and stores the high-quality video data acquired by the high-speed CMOS camera to the data processing module.

As some possible implementation methods, the data processing module automatically preprocesses the acquired image data, and then uses the CNN algorithm based on migration learning as a feature extractor to automatically extract features from the pattern, and then uses the SVM algorithm for cell classification. :

preprocessing, including:

Split the image data frame by frame from the video data, then locate the position of the pattern to be recognized from the split frame image and automatically intercept and save it, and filter the obtained image data;

Each image data is processed by morphological algorithm and particle size analysis algorithm to obtain image morphological granularity feature value, judge whether the feature value meets the standard, if it meets the image, keep the image, otherwise the image will be eliminated;

The retained images are further filtered using a trained machine learning model.

As a further limitation, threshold judgment is performed on the morphology and granularity features of the image, and if the image is within the preset ratio interval centered on the mean, the image is retained; otherwise, it is removed as an impurity.

The second aspect of the present invention provides a method for detecting flow cytometry of unmarked high-content video based on migration learning.

A method for detecting label-free high-content video flow cytometry based on migration learning, using the label-free high-content video flow cytometer based on migration learning described in the first aspect of the present invention, comprising the following steps:

Three kinds of cultured cells were processed respectively to form a single cell suspension, prepared into a sample solution with a certain concentration to be tested and placed in a sample syringe, and PBS solution was selected as the sheath fluid and placed in the sheath fluid syringe;

Set the syringe pump parameters and start the syringe pump so that the sheath flow can be formed normally, the sample fluid is 30 μL/h, and the sheath fluid is 800 μL/h;

Start the light source and CMOS camera, observe the formation of sheath flow and imaging effect, and adjust the displacement platform to make the system work in defocus mode;

Start a trigger to collect and record video data of two-dimensional light scattering when a sample flows through;

After the collection, replace the sample fluid and the sheath fluid, and rinse the system with alcohol solution and ultrapure water in sequence;

Automatically process video results using preset classification algorithms, extract desired cell patterns, and use automatic classification algorithms for analysis.

The third aspect of the present invention provides a method for calibrating a label-free high-content video flow cytometer based on migration learning, using the label-free high-content video flow cytometer based on migration learning described in the first aspect of the present invention, including The following steps:

Place each component in the preset position so that each module can work normally, and adjust the position of each module so that the excitation light path, sheath flow and collection light path can be coupled;

Configure Rhodamine 6G solution and place it in the sample liquid syringe, use ultrapure water as the sheath liquid and place it in the sheath liquid syringe, set the syringe pump parameters and start the syringe pump to make the sheath flow system work normally;

Control the translation stage in the acquisition optical path, make the sample liquid flow in the center of the field of view, focus on the sample, and observe the effect of sheath flow;

Start the light source, further calibrate the position of each device in the optical path, observe and further adjust the position of each module, so that the liquid flow can be accurately excited and determine the focal plane of the system;

Start the CMOS camera and trigger to collect and record video data;

After the collection, replace the sample fluid and the sheath fluid, and rinse the system with alcohol solution and ultrapure water in sequence;

Process the acquired data, calculate and verify the effect of the sheath flow.

The fourth aspect of the present invention provides a method for calibrating a label-free high-content video flow cytometer based on migration learning, using the label-free high-content video flow cytometer based on migration learning described in the first aspect of the present invention, including The following steps:

Prepare the solution of the sample to be tested and place it in the sample syringe, select ultrapure water or phosphate buffered saline solution as the sheath fluid according to the different sample fluids and place it in the sheath fluid syringe;

Set the syringe pump parameters and start the syringe pump so that the sheath flow can be formed normally;

Start the light source and CMOS camera, observe the formation of sheath flow and imaging effect, and adjust the displacement platform to make the system work in defocus mode;

Start the trigger to collect and record video data of two-dimensional light scattering;

After the collection, replace the sample fluid and the sheath fluid, and rinse the system with alcohol solution and ultrapure water in sequence;

Use the preset analysis algorithm to count the number of beads, and use Mie to simulate the pattern of beads under the experimental conditions.

Compared with prior art, the beneficial effect of the present invention is:

1. The flow cytometer and method of the present invention, the collection method is a label-free method, which can greatly reduce the workload of sample preparation and reduce the damage to the sample, has high timeliness and low sample requirements, and can Greatly expand its field of use.

2. The flow cytometer and method of the present invention adopts an automated flow control scheme with simplicity, adjustable parameters with a certain degree of flexibility, and reproducible parameters with high stability after being fixed.

3. The flow cytometer and method described in the present invention realize the multi-directional flow control method through the buffer chamber, and at the same time ensure the stability of the single cell in the excitation direction and the collection direction, and ensure that the excitation, flow and collection positions can be accurate confluence.

4. The flow cytometer and method described in the present invention are based on two-dimensional light scattering technology, have the sensitivity of structural and morphological information, adopt high-magnification acquisition objective lenses, and can acquire rich spatial details.

5. The flow cytometer and method described in the present invention have the ability to acquire large video data, adopt a high-speed CMOS camera, can collect rich time details, and comprehensively record the samples flowing through the collection area within a period of time. Provide high-quality original information.

6. The flow cytometer and method described in the present invention have the ability to obtain high-content information, can effectively use the full resolution of a CMOS camera to obtain information, and improve the spatial details of single-cell information collection.

7. The flow cytometer and method described in the present invention have a trigger collection function, which can reduce redundant data collection without samples and reduce the pressure on data storage.

8. The flow cytometer and method of the present invention adopts an automatic data processing and analysis algorithm, which reduces human subjectivity to the greatest extent.

9. The analysis algorithm used in the flow cytometer and method of the present invention is scalable, and the overall effect of the system is not limited to the current research level.

10. The flow cytometer and method described in the present invention have mobility and scalability, are not limited to a certain type of cells, can be applied to research in different fields very quickly, and have certain universality.

Advantages of additional aspects of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Description of drawings

The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention, and the schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations to the present invention.

FIG. 1 is a schematic diagram of a label-free high-content video flow cytometer based on transfer learning provided in Example 1 of the present invention.

Fig. 2 is the sheath flow effect diagram and intensity scanning diagram under Rhodamine 6G white light illumination provided by Example 2 of the present invention.

Fig. 3 is a graph showing the results of counting the number of small balls in the example provided by Embodiment 3 of the present invention.

Fig. 4 is the experimental pattern and the simulated pattern of two kinds of pellets provided by Example 3 of the present invention.

Fig. 5 is a result graph of cell number statistics provided by Example 4 of the present invention.

FIG. 6 is a graph showing experimental results of three cervical cancer cell lines provided in Example 4 of the present invention.

Among them, 1. Laser; 2. Neutral density film; 3. Laser aperture; 4. Mirror; 5. Excitation objective lens; 6. Sheath flow chamber; 7. Acquisition objective lens; 8. High-speed CMOS camera; and analysis system; 10, trigger; 11, sample liquid; 12, sheath liquid.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the present invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It should be noted that the terminology used here is only for describing specific embodiments, and is not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural, and it should also be understood that when the terms "comprising" and/or "comprising" are used in this specification, they mean There are features, steps, operations, means, components and/or combinations thereof.

In the case of no conflict, the embodiments and the features in the embodiments of the present invention can be combined with each other.

Example 1

As shown in FIG. 1 , Embodiment 1 of the present invention provides a label-free high-content video flow cytometer based on transfer learning. It mainly includes: optical excitation module, sheath flow control module, data acquisition and data processing module.

The optical excitation module includes: laser 1 (532nm), neutral density sheet 2 (50%, 32%, 10%, 1%, etc. optional), laser aperture 3, mirror 4 and excitation objective lens 5 (4X); laser 1 Generate an excitation beam, adjust the intensity through a neutral density film, and then focus it to the observation area by the objective lens, and the laser diaphragm acts as an optical path switch.

In this embodiment, the purpose of the optical excitation module is to spatially shape the laser light generated by the laser, so that it forms an approximately cylindrical excitation area in the observation area. Combined with the limitation of the sheath flow module on the spatial flow of the sample liquid, the sample and excitation The beams only overlap in a limited small area, thus greatly improving the efficiency of collection.

The sheath flow control module includes: a sheath flow chamber 6, a syringe pump for driving the sample fluid 11 and the sheath fluid 12, and a waste fluid pool. The syringe pump is the power source for the fluid flow and consists of a sample syringe pump and a sheath fluid syringe pump. The syringe pump presses the liquid into the sheath flow chamber, the sample liquid flows in from the middle channel of the sheath flow chamber, and the sheath liquid flows in from the surrounding channels; the velocity of the sheath fluid must be much greater than the velocity of the sample fluid, so that the sheath flow can maintain a sufficient length in the flow direction The distance is convenient for observation; the waste liquid pool is used to collect the sample and sheath fluid after flowing through the observation area.

The data acquisition and processing module includes: acquisition objective lens 7 (40X), high-speed CMOS camera 8, data processing and analysis system 9, trigger 10, and a set of precise displacement devices. Other lenses need to be increased or decreased according to the experimental design. The module needs to be able to precisely focus and defocus to meet the needs of data acquisition; the focus of the trigger is slightly ahead of the high-speed CMOS camera, so that the camera can be controlled in time when the sample passes; The video data is transmitted and stored for subsequent processing steps; during data processing, the system can automatically locate and intercept patterns in the video data, and analyze them through machine learning and deep learning methods to realize cell identification and classification.

In this embodiment, the data acquisition and processing module collects high-content pattern videos as quickly as possible to form video big data, which can be used for subsequent classification analysis and traceable pattern positioning.

The specific process is as follows: the semiconductor pump laser generates a 532nm laser beam with a diameter of 1.052mm, which enters the excitation objective lens after adjusting the light intensity and the direction of the mirror through the neutral density sheet; after being reshaped by the excitation objective lens 4 times, it is coupled into the sheath flow chamber ;

At the same time, the syringe pump drives the sample liquid and the sheath liquid into the sheath flow chamber to form a sheath flow, and the sample liquid is restricted to flow in the area close to the excitation beam; the collection objective lens is also focused on the overlapping part of the sample liquid and the excitation beam, and its Efficiency has also been greatly improved; the collected data is transferred to a computer for storage, processing and analysis.

Example 2

In order to test the working effect and stability of each module in Example 1, Rhodamine 6G solution was used in this example for test verification and calibration. In the experiment, rhodamine 6G solution was selected as the sample liquid, and ultrapure water was selected as the sheath liquid. Use the device described in the embodiment to acquire video data under white light illumination, intercept one frame for analysis, scan the intensity of the 10 rows of pixels in the middle of the frame, make an average intensity curve, and obtain the actual range of sheath flow.

Specific steps:

(1) According to the scheme design described in embodiment 1, each component is placed in a preset position, so that each module can work normally;

(2) Weigh 4mg of rhodamine 6G solute and dissolve it in 4mL of ultrapure water, stir until fully dissolved and place it in the sample syringe, extract ultrapure water as the sheath fluid and place it in the sheath fluid syringe; set the parameters of the syringe pump and start Syringe pump, sample fluid 960μL/h, sheath fluid 9600μL/h.

(3) Use white light for auxiliary lighting, control the displacement stage of the collection optical path, so that the sample is located in the center of the field of view for observation;

(4) Start the laser light source, further calibrate the position of each device in the optical path, so that the excitation beam, sheath flow and collection optical path can be accurately coupled;

(5) Turn off the laser light source, start the high-speed CMOS camera and trigger, and collect and record high-quality video data;

(6) After the collection, replace the sample solution and the sheath fluid, and use 75% alcohol solution and ultrapure water to flush the system in sequence;

(7) Scan the pixel intensity in the middle row of the image obtained by scanning one frame, calculate and verify the effect of the sheath flow.

In this example, the obtained results are shown in (a) in Figure 2 and (b) in Figure 2. The sheath flow effect is obvious and stable. After calculation, the FWHM of the average intensity of the 10 lines in the middle of the image is 26 μm, and the maximum width is about 40 μm , which proves that the device described in Example 1 has better sheath flow effect and stability.

Example 3

In order to further verify the sensitivity of the detection in the micron scale and the stability in the time dimension of Example 1, this embodiment uses two standard polystyrene microspheres for verification and calibration. The sizes of the two microspheres were chosen to be 3.87 μm and 4.19 μm. The experiment counted the change of the number of 3.87μm spheres in about 20s, and compared with the results of the two kinds of microspheres simulated by Mie. The results are shown in Figure 3 and Figure 4.

Specific steps:

(1) Dissolve 2 μL of the standard microsphere stock solution in 4 mL of ultrapure water, prepare the solution of the sample to be tested and place it in the sample syringe, select ultrapure water as the sheath fluid and place it in the sheath fluid syringe;

(2) Set the parameters of the syringe pump and start the syringe pump so that the sheath flow can be formed normally, the sample solution: 30 μL/h, the sheath fluid 800 μL/h;

(3) Start the light source and high-speed CMOS camera, observe the formation of the sheath flow and the imaging effect, and adjust the displacement platform to make the system work in the defocusing mode;

(4) Start the trigger to collect and record high-content video data of two-dimensional light scattering;

(5) After the collection, replace the sample solution and the sheath fluid, and use 75% alcohol solution and ultrapure water to rinse the system in sequence;

(6) Use the analysis algorithm to count the number of microspheres, and use Mie to simulate the pattern of microspheres under the experimental conditions.

In this example, about 20 seconds of collected 3.87 μm microspheres were counted, and about 352 microspheres were obtained, which proves that the device described in Example 1 has high time stability. The results of the microspheres simulated by Mie were compared with the pattern results obtained by the experiment, which proves that the device described in Example 1 has a relatively high micron-scale resolution.

Example 4

In this example, three cervical cancer cell lines were detected and analyzed using a high-quality video flow cytometry device based on two-dimensional light scattering. The three cells used were: Caski, HeLa and C33-A. The algorithm automatically processed the acquired video data, selected 5,000 patterns from each category for model training, and selected 600 patterns for result verification, realizing the automatic classification of three types of cells.

Specific steps:

(1) Treat the three kinds of cultured cells to form a single cell suspension, prepare a sample solution to be tested with a concentration of about 500,000 per milliliter and place it in a sample syringe, select PBS solution as the sheath fluid and placed in the sheath fluid syringe;

(2) Set the parameters of the syringe pump and start the syringe pump so that the sheath flow can be formed normally, the sample solution: 30 μL/h, the sheath fluid 800 μL/h;

(3) Start the light source and high-speed CMOS camera, observe the formation of the sheath flow and the imaging effect, and adjust the displacement platform to make the system work in the defocusing mode;

(4) Start the trigger to collect and record high-content video data of two-dimensional light scattering;

(5) After the collection, replace the sample solution and the sheath fluid, and use 75% alcohol solution and ultrapure water to rinse the system in sequence;

(6) Use the analysis algorithm to automatically process the video results, extract the required cell pattern, and use the automatic classification algorithm to analyze, the results are shown in Figure 5 and Figure 6.

In this example, the automatic classification algorithm of CNN-SVM is used, and the CNN algorithm based on transfer learning is used as the feature extractor to automatically extract features from the pattern, and then the SVM algorithm is used for classification. The 600 patterns of each class verified as the final result in the analysis do not overlap with the 5000 patterns of each class used for training, and they are all randomly selected. The classification results are shown in Table 1.

Table 1: Cell classification results of three cervical cancer cell lines

Specifically, digital cell filtering is performed before cell classification, mainly using morphological particle size analysis methods and machine learning algorithms to filter two-dimensional light scattering video data frame by frame.

The morphological particle size analysis method mainly quickly removes simple impurities such as cell debris and air bubbles in the video. Machine learning algorithms are mainly used to remove impurities with more complex morphological changes, such as complex impurities in clinical samples.

The analysis algorithm of morphology and granularity can extract the intensity and gradient related information of the speckle in the pattern, and the threshold of discrimination is based on the data within the 60% range (±30%) centered on the mean value according to the statistical data. In the step of machine learning, the pre-trained machine learning model is used for discrimination. The training set is a priori pattern and impurity data set, and the training network model is CNN.

The CNN-SVM classification part includes CNN feature extractor and SVM classifier. The CNN feature extractor is composed of a series of convolutional pooling layers of neural networks, which input two-dimensional light scattering pattern training data and output feature vectors; the CNN feature extractor used in the present invention uses a migration learning method from the natural atlas image classification framework Migrated from it, it can reduce the time of network construction and the number of images used for training. According to the input feature vector, the SVM classifier automatically optimizes the classification function by finding the optimal parameters, and realizes the automatic classification of samples. The CNN network used is an improved Inception v3 network. The network first has an alternating structure of 5 convolutional layers and 2 pooling layers, and then is formed by combining three sub-network modules, and finally the output results are integrated by the average pooling layer.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (7)

1. The utility model provides a no mark high connotation video flow cytometer based on migration study which characterized in that:

comprising the following steps: the device comprises an optical excitation module, a sheath flow control module, a data acquisition module and a data processing module;

the optical excitation module at least comprises a laser, a neutral density sheet, a laser diaphragm, a reflecting mirror and an excitation objective lens, wherein the laser, the neutral density sheet, the laser diaphragm, the reflecting mirror and the excitation objective lens are sequentially arranged along an optical path, the laser diaphragm is used for collimating and switching the optical path, the reflecting mirror is used for adjusting the direction, the excitation objective lens is used for shaping, and compressed excitation light beams are obtained after laser passes through the excitation objective lens;

the excitation light beam emitted by the optical excitation module is transmitted to the sheath flow chamber of the sheath flow control module, video image acquisition is carried out through the data acquisition module, and acquired image data is transmitted to the data processing module for cell detection;

the sheath flow control module can limit the space flow of the sample liquid to enable the sample to form a flowing unmarked single cell sequence, and meanwhile, the excitation light beam generated by the optical excitation module is coupled into the sample flow channel after being shaped by the objective lens, so that the sample liquid and the excitation light beam are overlapped only in a preset area;

the sheath flow control module is used for driving the cell solution to form a single cell passage and comprises a sheath flow chamber, a sample liquid injection pump and a sheath liquid injection pump, wherein the sample liquid flows in from a middle channel of the sheath flow chamber, the sheath liquid flows in from a surrounding channel of the sheath flow chamber, and the speed of the sheath liquid is higher than that of the sample liquid;

before the sheath fluid enters the sheath flow chamber, the sheath fluid passes through a buffer chamber, so that the fluid flow is stable, the flowing direction of the sheath fluid is pre-dispersed, and the sheath fluid is ensured to simultaneously perform compression control on the sample fluid in two orthogonal directions;

the data acquisition module is provided with a trigger, and the focus of the trigger is slightly advanced than that of the high-speed CMOS camera;

the laser beam enters the excitation objective lens after the light intensity is adjusted by the neutral density plate and the direction is adjusted by the reflecting mirror; shaping by 4 times of excitation objective lens and coupling into sheath flow chamber; simultaneously, the injection pump drives the sample liquid and the sheath liquid to flow into the sheath flow chamber to form sheath flow, and the sample liquid is limited to flow in a region close to the excitation light beam; the acquisition objective lens focuses on the superposition part of the sample liquid and the excitation light beam;

the data processing module carries out automatic preprocessing on the acquired image data, realizes digital cell filtration, then adopts a CNN algorithm based on transfer learning as a feature extractor, automatically extracts features from patterns, and then carries out cell classification by utilizing an SVM algorithm, wherein:

pretreatment, comprising:

splitting image data frame by frame for video data, then positioning the pattern position to be identified from the split frame image, automatically intercepting and storing, and filtering the obtained image data;

processing each image data by a morphological algorithm and a granularity analysis algorithm to obtain an image morphological granularity characteristic value, judging whether the characteristic value accords with a standard, if so, reserving the image, otherwise, rejecting the image;

the retained image is further filtered for digital cells using a trained machine learning model.

2. The transfer learning-based label-free high content video flow cytometer of claim 1, wherein:

the sheath flow control module also includes a reservoir of waste fluid for collecting the sample and sheath fluid after flowing through the observation region.

3. The transfer learning-based label-free high content video flow cytometer of claim 1, wherein:

the data acquisition module comprises an acquisition optical path and a data path, the acquisition optical path at least comprises an acquisition objective lens, a high-speed CMOS camera and a trigger, the field of view of the acquisition objective lens is positioned at the overlapping part of the sample liquid and the excitation light beam, and the trigger is positioned in front of the acquisition objective lens and is used for controlling the storage time of the camera;

the data path transmits and stores high-quality video data acquired by the high-speed CMOS camera to the data processing module.

4. The transfer learning-based label-free high content video flow cytometer of claim 3, wherein:

and judging the form and granularity characteristics of the image by a threshold, if the image is in a section with a preset proportion and the mean value is the center, reserving the image, otherwise, removing the image as impurities.

5. A label-free high-content video stream cell detection method based on transfer learning is characterized by comprising the following steps of: use of the label-free high content video flow cytometer based on transfer learning of any of claims 1-4, comprising the steps of:

respectively processing the three cultured cells to form single cell suspension, preparing a sample solution to be tested with certain concentration, placing the sample solution into a sample injector, and selecting a PBS solution as sheath fluid and placing the PBS solution into the sheath fluid injector;

setting parameters of a syringe pump and starting the syringe pump to enable sheath flow to be formed normally, wherein the sample liquid is 30 mu L/h, and the sheath liquid is 800 mu L/h;

starting a light source and a CMOS camera, observing the sheath flow forming condition and the imaging effect, and adjusting a displacement platform to enable the system to work in a decoking mode;

starting a trigger, and collecting and recording video data scattered by two-dimensional light;

after the collection is finished, replacing the sample liquid and the sheath liquid, and flushing the system by using alcohol solution and ultrapure water in sequence;

and automatically processing the video result by using a preset classification algorithm, extracting the required cell pattern, and analyzing by using the automatic classification algorithm.

6. A calibration method of a label-free high-content video flow cytometer based on transfer learning is characterized by comprising the following steps of: use of the label-free high content video flow cytometer based on transfer learning of any of claims 1-4, comprising the steps of:

placing all the components at preset positions to enable all the modules to work normally, and adjusting the positions of all the modules to enable the excitation light path, the sheath flow and the acquisition light path to be coupled;

preparing rhodamine 6G solution, placing the rhodamine 6G solution in a sample liquid injector, placing ultrapure water serving as sheath liquid in the sheath liquid injector, setting injection pump parameters, and starting an injection pump to enable a sheath flow system to work normally;

controlling a displacement table in an acquisition light path to enable sample liquid flow to be positioned in the center of a visual field, focusing a sample, and observing a sheath flow effect;

starting a light source, further calibrating the positions of devices in the light path, observing and further adjusting the positions of the modules, so that liquid flow can be accurately excited and a system focusing plane is determined;

starting a CMOS camera and a trigger, and collecting and recording video data;

after the collection is finished, replacing the sample liquid and the sheath liquid, and flushing the system by using alcohol solution and ultrapure water in sequence;

the acquired data is processed to calculate and verify the effect of the sheath flow.

7. A calibration method of a label-free high-content video flow cytometer based on transfer learning is characterized by comprising the following steps of: use of the label-free high content video flow cytometer based on transfer learning of any of claims 1-4, comprising the steps of:

preparing a solution of a sample to be tested, placing the solution into a sample injector, selecting ultrapure water or phosphate buffer salt solution as sheath solution according to different sample solutions, and placing the sheath solution into the sheath solution injector;

setting parameters of the injection pump and starting the injection pump to enable sheath flow to be formed normally;

starting a light source and a CMOS camera, observing the sheath flow forming condition and the imaging effect, and adjusting a displacement platform to enable the system to work in a decoking mode;

starting a trigger, and collecting and recording video data scattered by two-dimensional light;

after the collection is finished, replacing the sample liquid and the sheath liquid, and flushing the system by using alcohol solution and ultrapure water in sequence;

the number of pellets was counted using a preset analytical algorithm and the microsphere pattern was simulated under experimental conditions using Mie.

Images