A single-cell high-throughput detection system and method based on integrated three-dimensional focusing

By adopting integrated three-dimensional sheath flow focusing channel and neural network integrated learning on the single-cell detection platform, the problem of insufficient electrical detection sensitivity and image detection quality is solved, efficient and stable single-cell detection is achieved, and the accuracy of cell classification is improved. Spend.

CN118688075BActive Publication Date: 2024-11-15AEROSPACE INFORMATION RES INST CAS
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411166056.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-23
Publication Date
2024-11-15
Estimated Expiration
2044-08-23

AI Technical Summary

Technical Problem

The existing single-cell detection platform has shortcomings in electrical detection sensitivity and image detection quality, and there are problems of channel blockage and cell rupture, which affects work efficiency.

Method used

Using a single-cell detection system based on an integrated three-dimensional sheath flow focusing channel, the three-dimensional sheath flow channel is designed through high-precision 3D printing, combining fluid control and neural network integrated learning to achieve high-sensitivity electrical detection and high-quality image detection, and through circulating nerves Networks and convolutional neural networks extract features for cell classification.

Benefits of technology

While achieving high-sensitivity electrical detection and high-quality image detection in single cells, the stability and sustainability of the system are ensured, the accuracy of cell classification is improved, and the problems of channel blockage and cell rupture are avoided.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118688075B_ABST
    Figure CN118688075B_ABST
Patent Text Reader

Abstract

The present invention discloses a single-cell high-throughput detection system and method based on integrated three-dimensional focusing, which belongs to the field of cell detection technology. The system components include: a fluid control module for realizing sample flow and sheath flow channel flow rate control; an integrated three-dimensional sheath flow focusing module for realizing integrated three-dimensional sheath flow focusing; a single-cell bright field image detection module for collecting single-cell bright field images in the focus detection area and processing and analyzing them; a single-cell multi-frequency impedance detection module for collecting single-cell multi-frequency impedance in the focus detection area and processing and analyzing them; a single-cell recognition and classification module based on neural network ensemble learning for training neural networks through detection data to construct ensemble learning. The present invention realizes high-precision three-dimensional sheath flow focusing, improves the single-cell electrical and structural feature detection flux and sensitivity; and realizes high-accuracy cell classification through the data processing method of neural network ensemble learning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of cell detection, and in particular relates to a single-cell high-throughput detection system and method based on integrated three-dimensional focusing. Background Art

[0002] Microscopic examination, as the "gold standard" for white blood cell identification, mainly relies on "Wright's stain" or "Liu's stain" to stain the prepared blood smear, and then uses a microscope to observe cell morphology and classify and count. Although microscopic examination has extremely high classification accuracy in white blood cell identification and classification, it has significant defects such as long detection time and high labor costs. In recent years, with the development of artificial intelligence, automated blood smear detection instruments have gradually been used in clinical practice to improve the classification performance of microscopic examinations and reduce the demand for labor costs, but it still does not solve the problem of low microscopic examination throughput.

[0003] In response to the clinical needs of rapid classification and typing of white blood cells, commercial blood cell analyzers use the principles of electrical impedance and light scattering detection to classify and count various types of cells in peripheral blood, and are currently widely used in clinical hospitals. In the blood cell analyzer, a single cell quickly passes through the detection area, and the changes in scattered light and electrical impedance waveforms caused are collected and used to count and classify cells, with a high detection throughput. In terms of electrical detection, the electrical signals captured by commercial blood cell analyzers at the single-cell level are relatively weak, and the sensitivity of electrical detection is low; in terms of optical detection, light scattering cannot achieve single-cell imaging, the structural information of the measured cells is incomplete, and the accuracy of white blood cell identification and classification is limited. It can only be used as a rapid initial screening tool and cannot replace the "gold standard" of microscopic examination. At the same time, in terms of data processing, the current commercial blood cell analyzers use the method of manually extracting features, the number of extracted features is limited, and the information loss is large, which affects the results of white blood cell classification.

[0004] Due to the high matching between microfluidic technology and biological cell size, the microfluidic detection platform based on micromachining technology has improved the performance of existing single-cell detection methods. In terms of electrical impedance detection, the microfluidic impedance detection platform improves the sensitivity and consistency of single-cell impedance detection through high-precision and high-consistency cell flow channels and microelectrodes made by micromachining technology; in terms of optical detection, the measured cells are restricted to a specific position through micromachining channels, so that they are on the focal plane of optical detection, and the capture of single-cell images is realized, which increases the single-cell structural information. However, in the current microfluidic detection platform, the geometric size of the microchannel is significantly larger than the single cell being measured. More specifically, in terms of electrical detection, the measured cells block fewer electric field lines, and it is impossible to obtain high-sensitivity impedance changes of single cells; in terms of image detection, the measured cells without focus restriction are prone to deviate from the imaging plane during movement, and it is impossible to obtain high-quality image information of single cells. In response to the problems existing in the existing microfluidic single-cell detection platform, a single-cell detection platform based on compression channels is proposed. By using microchannels smaller than the geometric size of cells, the electrical detection sensitivity is increased and the image detection defocus is alleviated. Specifically, during the measurement process, the measured cells are forced to compress and pass through a compression channel that is smaller than their own size, effectively blocking the electric field lines; at the entrance of the smaller compression channel, the cells are confined to a relatively small detection plane, effectively avoiding the occurrence of defocusing. However, this method achieves high-sensitivity electrical signals and high-definition image information through compressed cells. In actual application, there are problems such as channel blockage and cell rupture, which affects work efficiency.

[0005] In signal processing, deep neural networks are widely used for feature extraction and classification of one-dimensional sequence signals (impedance signals) and two-dimensional image signals because they can extract massive features from raw signals and effectively improve classification accuracy. Among them, recurrent neural networks are very effective in processing sequence characteristics because of their high-quality mining and recognition of time series information in sequence signals; convolutional neural networks perform well in image feature extraction and recognition through the key layer structure of convolution and pooling.

[0006] Therefore, it is very meaningful to develop a high-throughput detection system and method for single-cell electrical and image features based on an integrated three-dimensional sheath flow focusing channel, which can achieve high-sensitivity electrical detection and high-quality image detection of single cells while ensuring its stable working ability and improving the accuracy of cell classification through neural network ensemble learning. Summary of the invention

[0007] In order to overcome the technical problems recorded in the above-mentioned background technology, the present invention proposes a single-cell high-throughput detection system and method based on integrated three-dimensional focusing. While realizing high-sensitivity electrical detection and high-quality image detection of single cells, the system and method ensure their stable working ability and improve the accuracy of cell classification through neural network integrated learning.

[0008] The present invention is achieved through the following technical solutions:

[0009] A single-cell high-throughput detection system based on integrated three-dimensional focusing, comprising an integrated three-dimensional sheath flow focusing module, a fluid control module, a single-cell bright-field image detection module and a single-cell multi-frequency impedance detection module;

[0010] The integrated three-dimensional sheath flow focusing module comprises a sample flow injection channel, a sheath flow injection channel and a three-dimensional sheath flow focusing channel. The sample flow injection channel and the sheath flow injection channel are connected to the three-dimensional sheath flow focusing channel after converging. The starting end of the sample flow injection channel is a sample flow inlet, the starting end of the sheath flow injection channel is a sheath flow inlet, and the ending end of the three-dimensional sheath flow focusing channel is a mixed fluid outlet.

[0011] The fluid control module includes a sample flow system and a sheath flow system, wherein the sample flow system includes a sample flow syringe, a sample flow fluid injection pump, a sample flow liquid guiding hose and a sample flow joint, wherein the sample flow syringe is connected to the sample flow inlet via the sample flow liquid guiding hose and the sample flow joint, and the sample flow fluid injection pump is used to control the flow rate of the sample flow; the sheath flow system includes a sheath flow syringe, a sheath flow fluid injection pump, a sheath flow liquid guiding hose and a sheath flow joint, wherein the sheath flow syringe is connected to the sheath flow inlet via the sheath flow liquid guiding hose and the sheath flow joint, and the sheath flow fluid injection pump is used to control the flow rate of the sheath flow;

[0012] The single-cell bright-field image detection module comprises an inverted microscope, a bright-field light source and a high-speed camera, wherein the inverted microscope and the bright-field light source are aligned with the integrated three-dimensional sheath flow focusing module, and the high-speed camera is aligned with the eyepiece of the inverted microscope;

[0013] The single-cell multi-frequency impedance detection module comprises a phase-locked amplifier and a data acquisition card connected thereto for collecting data. The phase-locked amplifier is respectively connected to the sample flow injection channel and the mixed fluid outlet through electrode lines.

[0014] In the above technical solution, the cross-section of the three-dimensional sheath flow focusing channel is circular and has a diameter of 100 microns.

[0015] In the above technical solution, the cross-section of the sheath flow injection channel is square, with a length of 600 microns and a width of 300 microns.

[0016] In the above technical solution, the sheath flow injection channel includes a first channel and a second channel symmetrically distributed on both sides of the sheath flow injection channel. The first channel and the second channel converge with the sample flow injection channel at the same position and then connect with the three-dimensional sheath flow focusing channel.

[0017] In the above technical solution, the sections of the first channel and the second channel close to the convergence point respectively form an angle of 60° with the sample flow injection channel.

[0018] In the above technical solution, the flow rates of the sample flow fluid injection pump and the sheath flow fluid injection pump are 0.0033 µL / min-205.30 mL / h.

[0019] In the above technical solution, the output frequency of the phase-locked amplifier is 865,800 sampling points / second.

[0020] In the above technical solution, the video image of the high-speed camera has 128 8-1920 1200 resolution and 730-400000 frame rate, with exposure time as low as 1 microsecond.

[0021] A single cell high-throughput detection method based on integrated three-dimensional focusing comprises the following steps:

[0022] Step 1: Use a fluid control module to control the sample flow and sheath flow rates, and pass the high-conductivity sample flow carrying cells and the low-conductivity sheath flow into the microfluidic chip through the sample flow inlet and the sheath flow inlet respectively;

[0023] Step 2: In the three-dimensional sheath flow focusing area, use a single-cell bright-field image detection module to collect image video data of the cell flow process;

[0024] Step 3: Use the single-cell multi-frequency impedance detection module to detect the four-frequency impedance amplitude and phase data when there are no cells passing between the two inserted electrode wires; Step 4: Use the single-cell recognition and classification module based on neural network ensemble learning to build and train neural networks for feature extraction, perform feature fusion based on the extracted features, and build and train support vector machines for cell classification. In the above technical solution, the single-cell bright field image described in step 2 is obtained by processing the image video. First, each frame of the image is subtracted from the background frame, and the size is set to 128 in the detection area. The sensitive area of ​​128 pixels is selected, and the time frame image with the largest pixel change in the sensitive area after subtraction is extracted as the bright field image of each single cell passing through the detection channel, and the frame size is selected as 128 by cropping. 128 pixels in size, so far, the single cell image data set has been obtained. In the above technical solution, the neural network construction and training process described in step 4 is as follows: the four-frequency impedance amplitude and phase data of each single cell are matched one by one with the image video data, 70% of the impedance data set is used as a training set to enter the recurrent neural network for training, and 15% of the data set is used as a test set to verify the training results; 70% of the image data set is used as a training set to enter the convolutional neural network for training, and 15% of the data set is used as a test set to verify the training results; after that, 85% of the data sets participating in the recurrent neural network and the convolutional neural network extract 640 features based on the trained neural network and are further used to train the support vector machine classifier, and 15% of the completely isolated data set is used as an integrated learning test set to verify the support vector machine training results.

[0025] Beneficial effects:

[0026] (1) The present invention designs and manufactures an integrated three-dimensional sheath flow channel based on high-precision 3D printing, realizes integrated focusing and detection, solves the problem of focusing deviation caused by alignment errors, and achieves highly consistent detection results.

[0027] (2) The present invention uses a high-conductivity sample flow solution and a low-conductivity sheath flow solution for three-dimensional focusing. The electric field lines are effectively compressed in the focused sample flow. When the single cell under test passes through the detection area, most of the electric field lines are effectively blocked, effectively improving the impedance change caused by the single cell under test, and realizing high-sensitivity electrical detection. On the other hand, since the electric field line compression is achieved through fluid focusing rather than physical channel size, the stability and sustainability of the system operation are effectively achieved.

[0028] (3) The present invention uses sample flow solution and sheath flow solution for three-dimensional focusing. The movement direction of the cells to be measured in the sample flow is controlled at the center of the channel by the sheath flow solution. By collecting images of the fixed focal plane with a high-speed camera, a high-quality cell image without loss of focus can be obtained. On the other hand, since the cells are fixed on the detection focal plane by fluid focusing rather than physical channel restriction, the stability and sustainability of the system operation are effectively achieved.

[0029] (4) The present invention uses multiple deep neural network integrated learning methods to process impedance and image signals. In more detail, the long short-term memory neural network in the recurrent neural network is used to process the multi-frequency impedance signal of a single cell, and hundreds of electrical features are extracted; the residual neural network in the convolutional neural network is used to process the bright field image signal of a single cell, and hundreds of structural features are extracted; the two parts of features are integrated and learned by the support vector machine, that is, by the feature-level fusion method, the high-precision classification of the measured cells is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a single-cell high-throughput detection system device and module diagram based on integrated three-dimensional focusing;

[0031] Figure 2 Schematic diagram of a three-dimensional sheath flow focusing module of a single-cell high-throughput detection system based on integrated three-dimensional focusing; (I) is the overall structure of the microfluidic chip, (II) is a top view of the overall structure of the microfluidic chip, and (III) is the cross-sectional structure of the three-dimensional sheath flow focusing channel;

[0032] Figure 3 Schematic diagram of cells passing through the three-dimensional sheath flow focusing area; (Ⅰ) indicates that the cell enters the channel, (Ⅱ) indicates that the cell reaches the three-dimensional sheath flow focusing area, and (Ⅲ) indicates that the cell reaches the focusing detection area;

[0033] Figure 4 This is a schematic diagram of the bright field image and impedance signal when cells pass through;

[0034] Figure 5 Schematic diagram of the neural network construction and training process;

[0035] Figure 6 The physical picture of the integrated three-dimensional sheath flow focusing module; (I) is the physical picture of the integrated three-dimensional sheath flow focusing channel, (II) is the physical picture of the sensitive structure under a 4x microscope, and (III) is the physical picture of the sensitive structure under a 40x microscope;

[0036] Figure 7 This is the working diagram of the integrated three-dimensional sheath flow focusing module; (I) sample flow rate 5 μL / min, sheath flow rate 10 μL / min, (II) sample flow rate 5 μL / min, sheath flow rate 25 μL / min, (III) sample flow rate 5 μL / min, sheath flow rate 50 μL / min.

[0037] Among them, 1 is an integrated three-dimensional sheath flow focusing module, 1-1 is a sample flow inlet, 1-2 is a detection electrode inlet, 1-3 is a mixed fluid outlet, 1-4 is a sheath flow inlet, 1-5 is a sheath flow injection channel, 1-6 is a sample flow injection channel, 1-7 is a three-dimensional sheath flow focusing area, 1-8 is a focusing detection area, 2 is a fluid control module, 3 is a single-cell bright field image detection module, 3-1 is an inverted microscope, 3-2 is a high-speed camera, and 4 is a single-cell multi-frequency impedance detection module. DETAILED DESCRIPTION

[0038] The specific implementation mode of the present invention is described in detail below. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation mode and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

[0039] Example 1: A single-cell high-throughput detection system based on integrated three-dimensional focusing

[0040] The hardware devices required for this embodiment are as follows Figure 1 As shown, the system consists of: a fluid control module 2, which is used to realize the flow rate control of the sample flow and sheath flow channel; an integrated three-dimensional sheath flow focusing module 1, which is used to realize the integrated three-dimensional sheath flow focusing; a single-cell bright field image detection module 3, which is used to collect the single-cell bright field image in the focus detection area and process and analyze it; a single-cell multi-frequency impedance detection module 4, which is used to collect the single-cell multi-frequency impedance in the focus detection area and process and analyze it.

[0041] The fluid control module 2 is used to realize the flow rate control of the sample flow and sheath flow channels, and includes two fluid injection pumps, two sets of liquid guide hoses, disposable syringes and connectors. The fluid injection pump can output the required flow rate by manual adjustment. Using a disposable syringe with a capacity of 1 mL, the fluid injection pump can provide an adjustable flow rate in the range of 0.0033 µL / min-205.30 mL / h. It is connected to the integrated three-dimensional sheath flow focusing module through the liquid guide tube and the connector to provide the integrated three-dimensional sheath flow focusing module 1 with adjustable injection flow rates of sample flow and sheath flow. The integrated three-dimensional sheath flow focusing module 1 is used to realize integrated three-dimensional sheath flow focusing. The main part is an insulating resin material microfluidic chip realized by high-precision 3D printing technology, with a minimum layer thickness of 10 microns. The overall structure of the microfluidic chip is as shown in the figure. Figure 2 As shown in (Ⅰ), the integrated three-dimensional sheath flow focusing channel is mainly composed of a sample flow injection channel 1-6, a sheath flow injection channel 1-5, a three-dimensional sheath flow focusing region 1-7, a detection electrode inlet 1-2, and a mixed fluid outlet 1-3; in addition, 1-1 represents a sample flow inlet, and 1-4 represents a sheath flow inlet. Figure 2 (II) is a top view of the overall structure of the microfluidic chip, wherein the sample flow injection channel 1-6 is used to introduce a high conductivity sample flow solution whose flow rate is controlled by the fluid control module 2. The cells to be tested are suspended in the sample flow solution, and its cross-sectional shape is circular with a cross-sectional diameter of 600 microns, and gradually changes to 100 microns when approaching the three-dimensional sheath flow focusing area, which is much larger than the diameter of most peripheral blood cells (10-25 microns); the sheath flow injection channel 1-5 is composed of two symmetrically distributed sample flow injection circular channels, and the two channels have the same inlet, which is convenient for achieving a three-dimensional centrally symmetrical focusing effect. The channel is used to introduce a low conductivity sheath flow solution whose flow rate is controlled by the fluid control module, and its cross-sectional shape is circular, and changes into a square with a length of 600 microns and a width of 300 microns when entering the three-dimensional sheath flow focusing channel; the cross-sectional structure of the three-dimensional sheath flow focusing channel is as shown in FIG. Figure 2As shown in (III), the cross section of the main channel is a circle with a diameter of 100 microns, which is used for sample flow injection and detection. The sheath flow channel enters the focusing area through a square channel with a cross section of 600 microns in length and 300 microns in width, and forms a three-dimensional focus through a conical area with a cross-sectional angle of 60°. By adjusting the sample flow, sheath flow velocity and flow rate ratio, the focusing effect of different focusing area diameters from 0 microns to 100 microns can be achieved. The schematic diagram of the cell passing through the three-dimensional sheath flow focusing area is shown in Figure 3 As shown, (I) indicates that the cell enters the channel, (II) indicates that the cell reaches the three-dimensional sheath flow focusing area 1-7, and (III) indicates that the cell reaches the focusing detection area 1-8. The detection electrode inlet is located on the sample flow channel and is a circular hole with a diameter of 1 mm. The electrode is inserted into the sample flow solution with higher conductivity during impedance detection. It is sealed with raw tape through a flat joint to avoid affecting the flow rate of the sample flow. The cross-section of the mixed fluid outlet is a circle with a diameter of 600 microns. It is connected to the atmosphere at the outlet through a flat joint. At the same time, the other electrode insertion port for impedance detection shares a physical channel with the mixed fluid outlet.

[0042] The single cell bright field image detection module 3 is used to collect the single cell bright field image in the focused detection area and process and analyze it, and includes an inverted microscope 3-1, a bright field light source and a high-speed camera 3-2 (such as Figure 1 The bright field light source provides the incident light required for imaging, and the high-speed camera 3-2 can record 128 8-1920 The video image has a resolution of 1200 and a frame rate of 730-400000. The exposure time can be as low as 1 microsecond, and a high-power objective lens is used to perform high-speed imaging of the detection area.

[0043] The single cell multi-frequency impedance detection module 4 is used to collect the single cell multi-frequency impedance in the focus detection area and process and analyze it. It includes a phase-locked amplifier and a data acquisition card. According to the needs of the embodiment, it can accurately detect the presence or absence of impedance changes, with an output frequency of 865,800 sampling points / second, and is connected to the two electrode inlets of the aforementioned three-dimensional sheath flow focusing module by inserting an electrode line.

[0044] Example 2: A single-cell high-throughput detection method based on integrated three-dimensional focusing

[0045] A single-cell high-throughput detection method based on integrated three-dimensional focusing in this embodiment is based on a single-cell high-throughput detection system based on integrated three-dimensional focusing in the above-mentioned embodiment 1. In this embodiment, the system also includes a single-cell recognition and classification module based on neural network ensemble learning. The single-cell recognition and classification module based on neural network ensemble learning is used to train the neural network through detection data to construct ensemble learning, so as to realize single-cell electrical and impedance feature extraction and classification, including a recurrent neural network for processing impedance sequence signals, a convolutional neural network for processing bright field image signals and a support vector machine for feature fusion.

[0046] Specifically, the long short-term memory network (LSTM) in the recurrent neural network is used to realize the feature extraction and recognition classification of the 8-dimensional impedance sequence signal, and 128 high-dimensional features for classification are extracted; the residual neural network (ResNet18) in the convolutional neural network is used to realize the recognition and classification of the pixel size of 128 Feature extraction and recognition classification of 128 bright field images were performed to extract 512 high-dimensional features for classification. The 128 features based on impedance signals and the 512 features based on image signals as mentioned above were fused in parallel to obtain 640 high-dimensional features, and support vector machines were used to implement cell classification based on the 640 high-dimensional features.

[0047] The specific implementation method of this embodiment includes experimental operations for obtaining raw data and a method for processing raw data:

[0048] In the experimental operation of this embodiment, the fluid control module, the integrated three-dimensional sheath flow focusing module, the single-cell bright field image detection module, and the single-cell multi-frequency impedance detection module are first connected. The fluid control module connects two injection pumps to the sample flow inlet and the sheath flow inlet in the integrated three-dimensional sheath flow focusing module through a liquid guide hose and a flat connector; the single-cell multi-frequency impedance detection module is connected to the detection electrode inlet and the mixed fluid outlet in the integrated three-dimensional sheath flow focusing module through an electrode line; the optical path of the single-cell bright field image detection module is placed in the fine three-dimensional sheath flow focusing detection area in the integrated three-dimensional sheath flow focusing module, and imaging is performed through a high-power objective lens. Before the sample is introduced, all channels in the microfluidic chip are filled with isopropanol. The purpose is to first use an organic solvent to infiltrate the hydrophobic inner wall of the channel to ensure that no bubbles are generated when the sample flow is introduced. Next, a certain concentration of cell suspension is connected to the sample flow inlet using a fluid injection pump, and a low conductivity sheath flow solution is connected to the sheath flow inlet. The osmotic pressure equal to that of the cells is maintained by dissolving sucrose in the sheath flow to avoid cell rupture or shrinkage caused by osmotic pressure mismatch. The fluid control module is used to control the flow rate and flow rate ratio of the sample flow and the sheath flow. Under the action of three-dimensional focusing, the sample flow is focused by the sheath flow into a circular cross-section that can be controlled by the flow rate and flow rate ratio. At the same time, most of the electric field lines generated by the AC voltage are confined to the low-conductivity sample flow. The size of the focused sample flow is controlled by adjusting the flow rate ratio of the sample flow and the sheath flow. The cells suspended in the sample flow pass through the three-dimensional focusing channel and the mixing fluid channel driven by the fluid; the single-cell bright field image detection module is used to collect image video data of the cell flow process, such as Figure 4 As shown in (I), the impedance data when the single-cell multi-frequency impedance detection module is used to detect whether there are cells passing between the two inserted electrode wires is as follows Figure 4 As shown in (II), impedance series data and image video data are used as the original data of the experiment.

[0049] According to the above description of this embodiment, a single cell high-throughput detection method based on integrated three-dimensional focusing is proposed, comprising the following steps:

[0050] Step 1: Use a fluid control module to control the sample flow and sheath flow rates, and pass the high-conductivity sample flow carrying cells and the low-conductivity sheath flow into the microfluidic chip through the sample flow inlet and the sheath flow inlet respectively;

[0051] Step 2: In the three-dimensional sheath flow focusing area, use a single-cell bright-field image detection module to collect image video data of the cell flow process;

[0052] Step 3, using a single-cell multi-frequency impedance detection module to detect the four-frequency impedance amplitude and phase data when there is a cell passing between two inserted electrode wires;

[0053] Step 4: Use the single-cell recognition and classification module based on neural network ensemble learning to build and train neural networks for impedance data and image data for feature extraction, perform feature fusion based on the extracted features, and build and train a support vector machine for cell classification.

[0054] The single cell bright field image described in step 2 is obtained by processing the image video. First, each frame of the image is subtracted from the background frame, and the size is set to 128 in the detection area. The sensitive area of ​​128 pixels is selected, and the time frame image with the largest pixel change in the sensitive area after subtraction is extracted as the bright field image of each single cell passing through the detection channel, and the frame size is selected as 128 by cropping. The size is 128 pixels. So far, the single-cell image dataset has been obtained.

[0055] The neural network construction and training process described in step 4 is as follows Figure 5 As shown, each single cell impedance data corresponds to the image data one by one. 70% of the impedance data set was used as a training set to enter the recurrent neural network for training, and 15% of the data set was used as a test set to verify the training results; 70% of the image data set was used as a training set to enter the convolutional neural network for training, and 15% of the data set was used as a test set to verify the training results; then, 85% of the data sets participating in the recurrent neural network and convolutional neural network extracted 640 features based on the trained neural network and were further used to train the support vector machine classifier, and 15% of the completely isolated data set was used as an integrated learning test set to verify the support vector machine training results.

[0056] like Figure 6 As shown, Figure 6 (I) is a physical picture of the integrated three-dimensional sheath flow focusing channel. Figure 6 (II) is a real picture of the sensitive structure under a 4x microscope. Figure 6 (III) is a real picture of the sensitive structure under a 40x microscope.

[0057] Figure 7 The integrated three-dimensional sheath flow focusing structure works under different flow rate ratios, where the dark part is the sample flow and the light part is the sheath flow. The system working conditions are (Ⅰ): sample flow rate 5μL / min, sheath flow rate 10μL / min; (Ⅱ): sample flow rate 5μL / min, sheath flow rate 25μL / min; (Ⅲ): sample flow rate 5μL / min, sheath flow rate 50μL / min; the focusing interfaces of different widths shown in the figure prove that the method can adjust the focusing width by adjusting the flow rate ratio.

[0058] In this embodiment, as a preferred implementation method, the sheath flow is a deionized water insulating solution in which sucrose is dissolved, and a low-conductivity solution that is miscible with the water system does not affect the realization of basic functions; the sample flow is a phosphate buffered saline (PBS for short), and a conductive solution such as iodixanol prepared with physiological saline, cell culture medium, and PBS does not affect the realization of basic functions.

[0059] In this embodiment, as a preferred implementation method, the three-dimensional sheath flow focusing module is formed by 3D printing using photosensitive resin materials. The above-mentioned three-dimensional fluid focusing module can also be formed using different 3D printers and printing materials, and femtosecond laser quartz processing technology.

[0060] In this embodiment, as a preferred implementation method, the shape and size of the interface between the sample flow channel inlet, sheath flow channel inlet, detection electrode inlet, mixed solution outlet and the fluid control module in the three-dimensional sheath flow focusing module can be changed without affecting the basic function.

[0061] In this embodiment, as a preferred implementation method, three-dimensional sheath flow focusing is the basic unit of this method, which can be easily arranged in parallel and series in the direction of cell migration, and even the combination of certain structures will bring different effects.

[0062] In this embodiment, as a preferred implementation method, the channel cross-section in the three-dimensional fluid focusing module is circular, and can also be replaced by a rectangular or semicircular shape.

[0063] In this embodiment, as a preferred implementation method, fluid is used to drive the cell solution through the channel, but other methods can also be used, such as applying negative pressure to the sheath flow outlet channel and the sample flow outlet channel.

[0064] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A single-cell high-throughput detection method based on integrated three-dimensional focusing, characterized in that: The detection system used in the method includes an integrated three-dimensional sheath flow focusing module, a fluid control module, a single-cell bright field image detection module and a single-cell multi-frequency impedance detection module; The integrated three-dimensional sheath flow focusing module comprises a sample flow injection channel, a sheath flow injection channel and a three-dimensional sheath flow focusing channel. The sample flow injection channel and the sheath flow injection channel are connected to the three-dimensional sheath flow focusing channel after being converged. The starting end of the sample flow injection channel is a sample flow inlet, the starting end of the sheath flow injection channel is a sheath flow inlet, and the terminating end of the three-dimensional sheath flow focusing channel is a mixed fluid outlet; the sheath flow injection channel comprises a first channel and a second channel symmetrically distributed on both sides of the sheath flow injection channel, the first channel and the second channel are connected to the three-dimensional sheath flow focusing channel after being converged with the sample flow injection channel at the same position; the sections of the first channel and the second channel close to the convergence point are respectively at an angle of 60° with the sample flow injection channel; The fluid control module includes a sample flow system and a sheath flow system, wherein the sample flow system includes a sample flow syringe, a sample flow fluid injection pump, a sample flow liquid guiding hose and a sample flow joint, wherein the sample flow syringe is connected to the sample flow inlet via the sample flow liquid guiding hose and the sample flow joint, and the sample flow fluid injection pump is used to control the flow rate of the sample flow; the sheath flow system includes a sheath flow syringe, a sheath flow fluid injection pump, a sheath flow liquid guiding hose and a sheath flow joint, wherein the sheath flow syringe is connected to the sheath flow inlet via the sheath flow liquid guiding hose and the sheath flow joint, and the sheath flow fluid injection pump is used to control the flow rate of the sheath flow; The single-cell bright-field image detection module comprises an inverted microscope, a bright-field light source and a high-speed camera, wherein the inverted microscope and the bright-field light source are aligned with the integrated three-dimensional sheath flow focusing module, and the high-speed camera is aligned with the eyepiece of the inverted microscope; The single-cell multi-frequency impedance detection module includes a phase-locked amplifier and a data acquisition card connected thereto to collect data, and the phase-locked amplifier is respectively connected to the sample flow injection channel and the mixed fluid outlet through electrode lines; The method comprises the following steps: Step 1: Use a fluid control module to control the sample flow and sheath flow rates, and pass the high-conductivity sample flow carrying cells and the low-conductivity sheath flow through the sample flow inlet and the sheath flow inlet, respectively, to enter the microfluidic chip; Step 2: In the three-dimensional sheath flow focusing area, use a single-cell bright-field image detection module to collect image video data of the cell flow process; Step 3, using a single-cell multi-frequency impedance detection module to detect four-frequency impedance amplitude and phase data between two inserted electrode lines when cells pass through and when no cells pass through; Step 4: Use a single cell recognition and classification module based on neural network ensemble learning to construct and train a neural network for feature extraction based on the four-frequency impedance amplitude and phase data obtained in step 3 and the image video data obtained in step 2, perform feature fusion based on the extracted features, and construct and train a support vector machine for cell classification; The construction and training process of the neural network in step 4 is as follows: the four-frequency impedance amplitude and phase data of each single cell are matched one by one with the image video data, 70% of the impedance data set is used as the training set to enter the recurrent neural network for training, and 15% of the data set is used as the test set to verify the training results; 70% of the image data set is used as the training set to enter the convolutional neural network for training, and 15% of the data set is used as the test set to verify the training results; then, 640 features are extracted based on the trained neural network, 85% of the data sets participating in the recurrent neural network and the convolutional neural network are used to train the support vector machine classifier, and 15% of the completely isolated data sets are used as the integrated learning test set to verify the support vector machine training results.

2. The single-cell high-throughput detection method based on integrated three-dimensional focusing according to claim 1, characterized in that: The cross section of the three-dimensional sheath flow focusing channel is circular, and the diameter is 100 microns.

3. The single-cell high-throughput detection method based on integrated three-dimensional focusing according to claim 1, characterized in that: The sheath flow injection channel has a square cross section with a length of 600 microns and a width of 300 microns.

4. The single-cell high-throughput detection method based on integrated three-dimensional focusing according to claim 1, characterized in that: The flow rates of the sample flow fluid injection pump and the sheath flow fluid injection pump are 0.0033 µL / min-205.30 mL / h.

5. The single-cell high-throughput detection method based on integrated three-dimensional focusing according to claim 1, characterized in that: The output frequency of the lock-in amplifier is 865,800 samples per second.

6. The single-cell high-throughput detection method based on integrated three-dimensional focusing according to claim 1, characterized in that: The video image of the high-speed camera has 128 8-1920 1200 resolution and 730-400000 frame rate, with exposure time as low as 1 microsecond.

7. The single-cell high-throughput detection method based on integrated three-dimensional focusing according to claim 1, characterized in that: In step 2, the single cell bright field image is obtained by processing the image video. First, each frame of the image is subtracted from the background frame, and the size is set to 128 in the detection area. The sensitive area of ​​128 pixels is selected, and the time frame image with the largest pixel change in the sensitive area after subtraction is extracted as the bright field image of each single cell passing through the detection channel, and the frame size is selected as 128 by cropping. 128 pixels size.

Citation Information

Patent Citations

  • Device for rapidly detecting number of bacteria in beverage

    CN112697679A

  • Three-dimensional focusing high-flux micro-fluidic chip and application and manufacturing method thereof

    CN116673079A

  • Impedance flow type detection chip and preparation method and application thereof

    CN116786181A

  • Label-free cell classification and screening system based on hybrid transfer learning

    US20220383629A1