Cell classification model training method and device and cell classification method and device

Abstract

The invention relates to a cell classification model training method and device and a cell classification method and device. The cell classification model training method comprises the following steps: acquiring a focused single cell quantitative phase microscopic image of a target cell; processing the focused single cell quantitative phase microscopic image to obtain corresponding cell characteristic data, and forming a cell characteristic data set; performing prediction performance screening on the cell characteristic data set to obtain an optimal characteristic data set; and taking the optimal feature data set as sample data, taking the type of the target cell as a label, and training the initial model to obtain a cell classification model. The method and the device provided by the invention have high classification accuracy on various cells, and are suitable for label-free detection and classification of large-scale, multi-category and high-heterogeneity cell populations.

Description

Technical Field

[0001] The present invention belongs to the technical field of cell characterization and classification, and specifically, relates to a cell classification model training method and device, and a cell classification method and device. Background Art

[0002] Single-cell research not only promotes people's understanding of basic biological processes such as cell differentiation, individual development, and disease progression, but also provides new possibilities for drug development and precision diagnosis and treatment. Imaging Flow Cytometry (IFC) can quantitatively analyze the morphology, structure, and multiple biomarkers of cells by capturing single-cell images under laser illumination in the detection area. However, IFC is highly dependent on fluorescent labels, which may lead to problems such as cytotoxicity, nonspecific binding, autofluorescence interference, and complex sample pre-treatment. In addition, this analysis method that relies on specific labels is ineffective for cell populations that lack prior knowledge or whose surface antigens change frequently (such as CTC cells).

[0003] In recent years, research teams have integrated different quantitative phase microscopy (QPM) architectures and IFC for imaging of flow cells, and for cell characterization and classification. However, compared with platforms based on staining, fluorescence and other marker detection, the current QPM-IFC still has gaps in imaging accuracy and classification recognition accuracy, especially when large-scale, highly heterogeneous single-cell populations are detected and analyzed. The main reason is that the rapid flow of microfluidics can cause blurred cell images or distorted shapes due to stretching. Under non-ideal microfluidic focusing, optical defocus of suspended cells is also difficult to avoid. Ultimately, low resolution leads to severe loss of subcellular texture, which in turn makes it impossible to achieve accurate cell characterization and classification. Summary of the invention

[0004] In order to overcome the problems existing in the prior art, the present invention provides a cell classification model training method, comprising:

[0005] Acquire focused single-cell quantitative phase microscopy images of target cells;

[0006] Processing the focused single-cell quantitative phase microscopy image to obtain corresponding cell feature data to form a cell feature data set;

[0007] Performing prediction performance screening on the cell feature data set to obtain an optimal feature data set;

[0008] The optimal feature data set is used as sample data, and the type of target cells is used as a label to train the initial model to obtain the cell classification model.

[0009] Another aspect of the present invention provides a cell classification model training device, comprising:

[0010] An image acquisition module, used to acquire a focused single-cell quantitative phase microscopy image of a target cell;

[0011] An image processing module, used for processing the focused single-cell quantitative phase microscopy image to obtain corresponding cell feature data to form a cell feature data set;

[0012] A feature screening module, used to screen the cell feature data set for predictive performance to obtain an optimal feature data set;

[0013] The model training module is used to train the initial model using the optimal feature data set as sample data and the type of target cells as labels to obtain the cell classification model.

[0014] In another aspect, the present invention provides a cell classification method, comprising:

[0015] Acquire focused single-cell quantitative phase microscopy images of cells to be classified;

[0016] Processing the focused single-cell quantitative phase microscopy image of the cell to be classified to obtain corresponding cell characteristic data;

[0017] Based on the cell feature data, the cells to be classified are classified using a cell classification model, and the cell classification model is trained according to the cell classification model training method.

[0018] Another aspect of the present invention provides a cell classification device, comprising:

[0019] An image acquisition subsystem, used to acquire focused single-cell quantitative phase microscopy images of cells to be classified;

[0020] An image processing subsystem, used for processing the focused single-cell quantitative phase microscopy image of the cell to be classified to obtain corresponding cell characteristic data;

[0021] The cell classification subsystem is used to classify the cells to be classified based on the cell feature data using a cell classification model. The cell classification subsystem also includes the cell classification model training device, which is used to train and obtain the cell classification model.

[0022] The beneficial effects of the present invention are:

[0023] 1. Using focused single-cell quantitative phase microscopy images as initial data for subsequent cell classification model training not only avoids motion blur and image distortion of cells during flow, but also eliminates the need for digital refocusing, greatly shortening the reconstruction time of single-cell quantitative phase microscopy images and saving computing resources.

[0024] 2. By adjusting the flow rate ratio of the sheath fluid to the sample flow in the microfluidic platform, cells of different properties and sizes can be accurately focused on the detection area in the vertical and horizontal directions. By adjusting the recording focal plane of the digital holographic microscope to coincide with the three-dimensional fluid dynamics focusing plane of the sample flow, a single-cell image plane holographic image of three-dimensional fluid dynamics focusing can be directly obtained, and then a single-cell quantitative phase microscopy image in the focused state can be obtained.

[0025] 3. Comprehensively consider the correlation and redundancy of cell features, screen the cell feature data set for predictive performance, and obtain the optimal feature data set. Model training based on this can shorten the training time, and the obtained classification model has high accuracy. In particular, the feature selection algorithm based on mutual information and the incremental feature selection algorithm are used to comprehensively evaluate the importance of each feature in the cell feature data set to cell classification and the redundancy between cell features to achieve predictive performance screening.

[0026] 4. Theoretical simulation and experimental tests confirm that the classification method and device of the present invention have a classification accuracy of 100% for a variety of cells, and are suitable for label-free detection and classification of large-scale, multi-category, and highly heterogeneous cell populations.

[0027] 5. The method and device of the present invention have a wide range of applications and are applicable to any cell or cell-sized particle. They can be used for the classification of normal cells, such as red blood cell counting and morphological research, and can also be used for the confirmation and classification of diseased cells (such as tumor cells). The method and device of the present invention can be used for the detection of cells in or out of human or mammal bodies, and can also be used for the detection of plants, microorganisms, and algae cells. For example, it can be used for the detection and classification of algae cells in environmental monitoring and ecological research, the detection and classification of plankton communities in aquatic ecosystems, and the counting and classification of pollen in ecology and environmental science.

[0028] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present invention.

[0030] Figure 1The figure is a flow chart of a cell classification model training method according to an embodiment of the present invention.

[0031] Figure 2 A top view of a microfluidic chip for a cell classification model training method according to an embodiment of the present invention.

[0032] Figure 3 A schematic diagram of the structure of a digital holographic microscope is shown according to a cell classification model training method according to an embodiment of the present invention.

[0033] Figure 4 (a) to Figure 4 (f) shows the theoretical simulation and experimental verification results of three-dimensional fluid dynamics focusing according to a cell classification model training method according to an embodiment of the present invention. Figure 4 (a) shows the vertical focusing computational fluid dynamics (CFD) simulation results (yz plane), where the sample injection speed is 1 μL / min, R v It represents the ratio of the vertical sheath flow to the sample flow velocity, and the black streamline shows the sample flow height (H s ), scale bar is 30 μm. Figure 4 (b) shows the results of the vertical focusing experiment verification (yz plane). The conditions are consistent with the CFD simulation. The red strip is the sample flow containing the fluorescent dye rhodamine. The dotted lines are the upper and lower boundaries of the sample channel. The scale is 30 μm. H s It is the average of 5 independent repeated experiments. Figure 4 (c) shows the vertical focusing height H s The ratio of the vertical sheath flow to the sample flow velocity R v The change curve of H s (sim) is CFD simulation data, H s (exp) is the average of 5 independent repeated experiments, and the dotted line represents the error. Figure 4 (d) shows the horizontal focusing CFD simulation results (xy plane), where the sample injection speed is 1 μL / min, R v Fixed to 0.5, R h It represents the velocity ratio of the horizontal sheath flow to the sample flow, and the width of the sample flow is shown as a black streamline (W s ), scale bar is 200 μm. Figure 4 (e) shows the experimental verification results of horizontal focusing (xy plane). The conditions are consistent with the CFD simulation. The red strip is the sample flow containing the fluorescent dye rhodamine, the dotted line is the channel boundary, and the scale is 100 μm. W s It is the average of 5 independent repeated experiments. Figure 4 (f) shows the horizontal focus width W s With the ratio of horizontal sheath flow to sample flow rate R h The change curve of W s(sim) is the CFD simulation data, W s (exp) is the average value of 5 independent repeated tests, and the dashed line represents the error.

[0034] Figure 5 (a) to Figure 5 (h) shows the digital holographic microscopy imaging results of microspheres in a three-dimensional hydrodynamic focusing state for a cell classification model training method according to an embodiment of the present invention. Figure 5 (a) and Figure 5 (b) are the flow conditions of 10μm and 20μm monodisperse polystyrene microspheres recorded in the white light mode in the microfluidic chip under three conditions: no focusing, only vertical focusing (two-dimensional focusing), and three-dimensional hydrodynamic focusing, respectively. The scale bar is 50μm. Figure 5 (c) and Figure 5 (d) are the reconstructed amplitude (upper part in the figure) and phase (lower part in the figure) of 10μm and 20μm microspheres in the three-dimensional hydrodynamic focusing state, respectively. The scale bar is 10μm. Figure 5 (e) and Figure 5 (f) are the reconstructed particle sizes of the amplitude and phase of 10μm microspheres under different focusing states (N = 50), respectively. Figure 5 (g) and Figure 5 (h) are the reconstructed particle sizes of the amplitude and phase of 20μm microspheres under different focusing states (N = 50), respectively. Among them, *** indicates p-value < 0.001, ** indicates 0.001 < p-value < 0.01, given based on Student t-Test analysis.

[0035] Figure 6 (a) to Figure 6 (c) shows the training results of different types of cancer cell lines for a cell classification model training method according to an embodiment of the present invention. Figure 6 (a) shows the importance scores of 81 features for classification obtained by the MRMR algorithm based on mutual information, sorted from high to low. Figure 6 (b) shows the incremental feature selection curves of each machine learning model, where the gray shadow is the feature MRMR score value, the blue curve is the accuracy of the ten-fold cross-validation set, and the red curve is the accuracy of the test set. The accuracy marked in each picture is the accuracy of each model on the test set obtained by training with the optimal feature dataset. Figure 6 (c) shows the prediction performance of each machine learning model for different genera of cancer cells, and the indicators include recall rate, precision, F1-score, accuracy, and ROC AUC value.

[0036] Figure 7 (a) to Figure 7(d) shows the quantitative phase microscopy image feature analysis of breast cancer multi-subtype cells in a cell classification model training method according to an embodiment of the present invention. Figure 7 (a) shows single-cell quantitative phase microscopy images of 8 breast cancer-related cell lines, including 2 LA-type cell lines (MCF-7 and T47D), 2 LB-type cells (BT474 and MDA-MB-361), 1 Her2+-type cell (SK-BR-3), 2 TN-type cells (HCC1937 and MDA-MB-231), and 1 normal breast cell (MCF-10A); the color bar represents the scale of the optical path difference (normalized value), and the phase information is represented by the optical path difference. Figure 7 (b) shows the violin plot of the numerical distribution of 6 features of 8 breast cancer-related cell lines (N = 1000), including volume (Bulk), dry weight density (Bulk), kurtosis (1st Order), entropy (GLCM), gray level asymmetry (GGCM), and run length difference (GLRLM). *** indicates p-value < 0.001, ** indicates 0.001 < p-value < 0.01, * indicates 0.01 < p-value < 0.05, ns indicates p-value > 0.05, given based on One-way ANOVA analysis. Figure 7 (c) shows the mean heat map and significance of differences of 81 quantitative phase microscopy image features of breast cancer-related cell lines. The color bar represents the scale of the mean value of each feature of different cell lines (normalized value) (N = 1000). The significance level (α) is taken as 0.05, given based on the post hoc multiple test (Tukey test method) of One-way ANOVA. Figure 7 (d) shows the Pearson correlation coefficient matrix heat map of 81 quantitative phase microscopy image features of breast cancer-related cell lines, where blue represents positive correlation, red represents negative correlation, the size of the correlation is represented by the grid size, and various feature subsets are divided using yellow boxes.

[0037] Figure 8 (a) to Figure 8 (c) shows the training classification results of normal breast cell lines and breast cancer multi-subtype cell lines in a cell classification method according to an embodiment of the present invention. Figure 8 (a) shows the importance ranking of 81 features for classification obtained by the MRMR algorithm based on mutual information. Figure 8 (b) shows the incremental feature selection curves of each model. The gray shadow is the feature MRMR score value, the blue curve is the accuracy of the ten-fold cross-validation set, the red curve is the accuracy of the test set, and the accuracy marked in each picture is the accuracy of each model obtained by training with the optimal feature set on the test set. Figure 8(c) shows the prediction performance of each model for different types of cancer cells, including recall rate, precision rate, F1-score, accuracy rate and ROC AUC value.

[0038] Fig. 9 (a) shows a quantitative phase microscopic image of cells in blood according to an embodiment of the present invention, Fig. 9 (b) and Fig. 9 (c) shows the confusion matrix of the LDA model obtained by training the cell classification model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The preferred embodiments of the present invention will be described in more detail below. Although the preferred embodiments of the present invention are described below, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0040] An embodiment of the present invention provides a cell classification model training method, comprising the following steps.

[0041] Step 1: Obtain a focused single-cell quantitative phase microscopy image of the target cell.

[0042] In the study of combining quantitative phase microscopy with imaging flow cytometry, the main challenge currently faced is to achieve high-precision quantitative phase microscopy acquisition of single cells under fast flow. Since suspended cells moving at high speeds are in natural optical defocus, the collected images are prone to motion blur or elongation distortion. Traditional IFC can focus cells using sheath flow technology, but after integration with quantitative phase microscopy, it is bulky and complex to operate, and requires additional digital refocusing or some image processing algorithms (denoising, etc.) to compensate for the reconstruction accuracy of quantitative phase microscopy, but this is computationally expensive and time-consuming, which does not match the rapid analysis process of IFC.

[0043] The present invention first obtains a focused single-cell quantitative phase microscopy image of a target cell, which refers to an image plane holographic quantitative phase microscopy image of a single cell focused by three-dimensional fluid dynamics. The characteristic of image plane holography is that the object or the image of the object is directly located on the hologram recording plane, and no optical diffraction propagation operation is required in the computer when reproducing it.

[0044] In this embodiment, a three-dimensional fluid dynamics focused single-cell image plane holographic image is acquired by using a microfluidic chip and a digital holographic microscope, and the three-dimensional fluid dynamics focused single-cell image plane holographic image is processed to obtain a focused single-cell quantitative phase microscopy image; wherein, the microfluidic chip is configured to provide a three-dimensional fluid dynamics focused sample flow containing target cells to the digital holographic microscope, and the digital holographic microscope is configured so that its recording focal plane coincides with the focal plane of the sample flow.

[0045] The three-dimensional hydrodynamically focused sample flow refers to a sample flow that achieves hydrodynamic focusing in both the vertical direction (longitudinal direction) and the horizontal direction (transverse direction); the three-dimensional hydrodynamically focused single-cell image plane holographic image refers to the image plane holographic image of a single cell in the three-dimensional hydrodynamically focused sample flow. By making the focal plane of the three-dimensional hydrodynamically focused sample flow coincide with the recording focal plane of the digital holographic microscope, the image plane holographic image can be directly acquired.

[0046] Specifically, Figure 2 As shown, the microfluidic chip is provided with a sample channel 1, a first sheath flow channel 2, and a second sheath flow channel 3 arranged along a horizontal plane. The first sheath flow channel 2 and the second sheath flow channel 3 are symmetrically arranged on both sides of the sample channel 1 and connected to the sample channel through a confluence port 4.

[0047] Two first sheath flows are respectively injected into the sample channel from above and below the sample channel at the second flow rate in the vertical direction at the symmetrical positions before and after the sample flow injection point 5, the sample flow is injected into the sample channel at the first flow rate in the vertical direction, and two second sheath flows are injected into the sample channel 1 through the first sheath flow channel 2 and the second sheath flow channel 3 at the third flow rate. The ratio of the second flow rate to the first flow rate is used as the first flow rate ratio, and the ratio of the third flow rate to the sum of the first flow rate and the second flow rate is used as the second flow rate ratio. The first flow rate ratio and / or the second flow rate ratio are adjusted so that the sample flow flows in the sample channel along a straight line located on the recording focal plane of the digital holographic microscope. That is, by adjusting the first flow rate ratio and the second flow rate ratio, the longitudinal focusing and lateral focusing of the sample flow can be achieved, that is, three-dimensional fluid dynamic focusing. Among them, the first sheath flow and the second sheath flow are the same sheath flow, and the first sheath flow and the second sheath flow are not mixed with the sample flow.

[0048] The first sheath flow channel and the second sheath flow channel are both zigzag-shaped, and at the confluence, the first sheath flow channel and the second sheath flow channel have the same angle with the sample channel. The first sheath flow channel and the second sheath flow channel can also be designed in other appropriate forms, such as streamlined.

[0049] This embodiment sets up a microfluidic chip based on three-dimensional fluid dynamics focusing to achieve strict and precise single-cell manipulation, and builds an image plane pre-amplification off-axis digital holographic microscope based on a Mach-Zehnder interferometer structure to capture the complex wavefront of single-cell samples.

[0050] Specifically, a microinjection pump is used as the power source, and the first sheath flow with the same flow rate continuously and symmetrically squeezes the sample flow in the longitudinal direction, so that the sample flow is focused longitudinally. After the three immiscible laminar flows advance along the channel to the confluence, they are symmetrically clamped by the two second sheath flows that converge at equal angles, thereby achieving transverse focusing of the sample flow in the sample channel, and the cells will flow in an orderly manner along the central axis of the sample channel. On the one hand, the sheath flow wrapping limits the motion blur caused by cell floating and stacking, and on the other hand, it reduces the risk of sample contamination and channel blockage caused by contact between the sample flow and the channel wall.

[0051] In this embodiment, at the confluence, the angle between the first sheath flow channel and the second sheath flow channel and the sample channel is 30° to 60° (preferably 45°), and the first flow rate ratio R v Set to 0-1.5 (preferably 0.25), the second flow rate ratio R h Set to 0-5 (preferably 2), and measure the changes in the width and height of the focused sample flow by changing the flow rate of the sample flow. When the flow rate of the sample flow jumps from 1-20 μL / min (1.46 m / s), the focus width and height fluctuate only in a small range, proving that the hydrodynamic focusing state is still stable at high speed.

[0052] Afterwards, the three-dimensional fluid dynamics focused sample flow is introduced into the digital holographic microscope, the structure of which is as follows: Figure 2 As shown. The image plane pre-magnification off-axis digital holographic microscope based on the Mach-Zehnder interferometer structure built in this embodiment, the single-frequency solid laser emits a laser with a wavelength of 632.8nm. The laser is collimated by the optical path calibration reflector group and then incident along the center of the optical path. It is then filtered by a spatial light filter composed of an aperture and a collimating lens to produce a uniform parallel light spot, which is divided into an object light wave and a reference light wave after passing through a polarization beam splitter. The object light path is equipped with a 40x objective lens and a tube lens with NA 0.75. After irradiating the sample, the microscope objective lens performs imaging and magnification; the reference light path is provided with a microscope objective lens with the same parameters as the object light path to balance the surface error as much as possible and reduce the amount of calculation for subsequent digital compensation processing in the computer. In addition, a filter is added to balance the light intensity attenuation caused by the sample in the object light path. Finally, the two beams of light interfere at a small angle on the imaging plane, and the single-cell holographic image of the sample flow is automatically and continuously acquired by a fast high-resolution camera (sCMOS) at a resolution of 496×496 at a speed of 30fps, with a pixel size of 2.5μm. Those skilled in the art can also choose other existing appropriate digital holographic microscopes as long as their recording focal plane is adjustable.

[0053] By precisely adjusting the recording focal plane of the digital holographic microscope to coincide with the focal plane of the three-dimensional fluid dynamics focused sample flow provided by the microfluidic chip, a three-dimensional fluid dynamics focused single-cell image plane holographic image can be directly obtained, which can quickly reconstruct accurate quantitative phase microscopy images without the need for digital refocusing, a step that usually accounts for about 60% of the time required to reconstruct quantitative phase microscopy images and consumes a large amount of computing resources.

[0054] The three-dimensional fluid dynamics focused single-cell image plane holographic image is processed to obtain the focused single-cell quantitative phase microscopy image. The specific process includes:

[0055] (1) Spectral filtering: The frequency spectrum of the single-cell holographic image and the background holographic image is obtained by Fourier transform. Then, according to the principle of holographic reconstruction, the original image (+1 level item) is filtered out separately and frequency-shifted to the center of the spectrum.

[0056] (2) Numerical reconstruction: The frequency spectrum is transformed into the spatial domain using the inverse Fourier transform to obtain the object light wave amplitude, and then the object's wrapping phase can be extracted through the inverse tangent operation;

[0057] (3) Phase processing: First, double-exposure phase subtraction is used to perform phase filtering (i.e., cell wrapping phase minus background wrapping phase). Then, the least squares unwrapping algorithm is used to obtain the continuous phase of the single cell. Finally, the distortion is eliminated by quadratic surface fitting to obtain a single-cell quantitative phase microscopy image.

[0058] <Fabrication and testing of microfluidic chips>

[0059] In this embodiment, a microfluidic chip can be made according to the following process. First, use AutoCAD to draw the chip channel pattern, print the pattern into a mask, and then follow the standard process of soft lithography to prepare the chip. Each channel (including the sample channel, the first sheath flow channel, and the second sheath flow channel) on the microfluidic chip of this embodiment is 100 μm wide and 35 μm high. The first sheath flow channel and the second sheath flow channel are both broken line-shaped, and the angles between the first sheath flow channel and the second sheath flow channel and the sample channel at the confluence are both 45°. The size of each channel can be adjusted according to the type of target cells or the actual application scenario, which is easily understood by those skilled in the art.

[0060] After obtaining the microfluidic chip, you can first perform a simulation test on it to ensure the feasibility of its function. Figure 4 (a) to Figure 4(f), In this embodiment, a simplified channel model is established in the finite element simulation software COMSOL Multiphysics, and the single-phase flow module is used to simulate the three-dimensional fluid dynamics focusing process to explore the influence of different flow rates on the focusing effect of the sample flow. Specifically, the channel in the focusing area is three-dimensionally modeled, and the channel cross-section (xz) is consistent with the actual chip size (100μm×35μm); the laminar flow physical field is configured, assuming that the laminar flow is incompressible, immiscible, and all side walls have no slip; the sample flow boundary is displayed with velocity streamlines. First, the flow rate ratio (R v ) varies from 0 to 1.5, increasing by 0.25. The height (H) of the focused sample flow in the longitudinal section (yz) of the measuring channel is s ), the simulation shows that it decreases from 35μm to 6μm, which is consistent with the first-order exponential decay relationship. After that, since the focusing process is "vertical first and then horizontal", the fixed R v The flow rate ratio (R h ) varies from 0 to 5, increasing by 0.5. The width (W) of the focused sample flow in the cross section (xy) of the measurement channel is s ), the simulation shows a decrease from 100 μm to 6 μm, which is also consistent with the first-order exponential decay relationship.

[0061] Then, the fluorescent dye Rhodamine B was used to visualize the sample flow and detect the three-dimensional fluid dynamic focusing effect in the actual microfluidic chip. A 1mL plastic syringe was used to absorb the sheath flow and sample flow, and the syringe was fixed on the syringe pump. A silicone hose (inner diameter 0.4mm, outer diameter 0.9mm) was connected to the syringe needle, and a 23G steel needle (inner diameter 0.3mm, outer diameter 0.6mm) was connected to the other end of the hose. The steel needle was directly inserted into the reserved injection port (aperture 0.6mm) on the PDMS chip to connect the syringe and the microfluidic chip. Three micro-injection pumps were used independently to control the injection flow rate of each inlet, and the sheath flow and sample flow were injected into the channel at different flow rates. The fluid focusing situation can be observed under an inverted fluorescence microscope, and the fluorescence photos are taken using software. The results show that the experimental measurement values ​​are basically consistent with the simulation values, proving that by adjusting the flow rate ratio, the three-dimensional focusing effect can be finely and accurately controlled to meet the focusing requirements of cells of different sizes.

[0062] <Reliability test of focused single-cell quantitative phase microscopy images>

[0063] This example uses polystyrene microsphere standards to test the accuracy of the digital holographic microscope for the reconstruction of particle amplitude and phase under different flow focusing states. First, the particle size of the microsphere is measured from the holographically reconstructed amplitude reconstruction image to verify the accuracy of the amplitude reconstruction. Then, the particle size of the microsphere is measured from the holographically reconstructed quantitative phase microscopy image to verify the accuracy of the phase reconstruction. The basis for this is that there is a linear relationship between the quantitative phase microscopy image and the actual thickness of the object, as shown in the following formula (1):

[0064]

[0065] Among them, λ refers to the wavelength of the incident light, n1 refers to the refractive index of the sample, n0 refers to the refractive index of the ambient medium, d refers to the actual thickness of the object, φ specifies the phase, and OPD refers to the optical path difference. Therefore, when the refractive index of the medium and the refractive index of the object are known, the true height of the object can be calculated according to this relationship.

[0066] Specifically, 10 μm and 20 μm microspheres were diluted to about 10 μm using 1X PBS buffer. 6 The prepared microsphere solution was used as the sample flow, and the 1X PBS buffer containing 0.1% Tween 20 was used as the sheath flow. The flow rate of the sample flow injection pump was fixed at 1

[0067] μL / min, and the flow rate ratio of sheath flow to sample flow is adjusted to the corresponding focusing state. The CMOS camera is set to continuous acquisition mode to capture and record the holographic image of polystyrene microspheres, and the quantitative phase microscopy image is reconstructed. The particle size of the microspheres is calculated through the above relationship, and compared with the known particle size parameters, the deviation is calculated, and the reliability of the phase reconstruction results is proved.

[0068] This example performs label-free digital holographic microscopy detection of three-dimensional hydrodynamically focused cells. Based on standardized physical properties and sizes consistent with a variety of cells to be analyzed, this example uses 10μm and 20μm polystyrene microspheres to simulate cells and verify the detection effect of the present invention on three-dimensional hydrodynamically focused microparticles. First, block the holographic light path and qualitatively observe the microsphere focusing process under an external white light source ( Figure 5 (a) Figure 5 (b)). The concentration of microsphere suspension is ~10 6 / mL, the flow rate slowed down to 1μL / min. Without any focusing, the microspheres stacked vertically and floated horizontally through the channel, with different degrees of defocus. After applying the first sheath flow constraint in the vertical direction, the microspheres were compressed to the same plane in the longitudinal direction, but still floated and swayed in the horizontal direction. After continuing to apply the second sheath flow constraint in the horizontal direction, the microspheres were focused to the central axis of the channel. The focusing condition of the 10μm microsphere is R v=0.75(H s =9.7μm), R h =3(W s =9.6μm), the focusing condition of 20μm microsphere is R v =0.125(H s =21.3μm), R h =1.25(W s =19.5μm). By removing the shielding of white light and holographic optical path, the holographic interference image of a single microsphere can be collected within the preset detection window. After frequency domain filtering and numerical reconstruction, the amplitude reconstruction image of a single microsphere is obtained ( Figure 5 (c) Figure 5 (d)); after phase extraction, unwrapping and distortion compensation, the quantitative phase microscopy image of a single microsphere is finally reconstructed ( Figure 5 (c) Figure 5 (d)).

[0069] The reconstruction effects of the amplitude and quantitative phase microscopy images of the microparticles under different focusing states were quantitatively compared. First, the length and width of the microspheres in the amplitude reconstruction image were measured, and the average value was taken as the amplitude particle size value of the microspheres; then, the difference between the highest and lowest phases of the microspheres in the quantitative phase microscopy image was extracted, and the quantitative phase microscopy image particle size value of the microspheres was converted according to formula (1). 50 microspheres in different focusing states were taken for statistical analysis ( Figure 5 (e) to Figure 5 (h)), the results show that three-dimensional hydrodynamic focusing is a prerequisite for accurately reconstructing the amplitude and quantitative phase microscopy images of microspheres. Especially for quantitative phase microscopy images, the particle size values ​​of 10μm and 20μm microspheres in the non-focused state are 7.13μm and 16.5μm, which are very different from the actual particle size of the microspheres; after three-dimensional hydrodynamic focusing, the particle size values ​​of the quantitative phase microscopy images are 10.4μm and 19.8μm, respectively ( Figure 5 f), Figure 5 (h)), which is basically consistent with the true value. In addition, three-dimensional fluid dynamic focusing further improves the measurement consistency. Although the measurement accuracy of microsphere amplitude and quantitative phase microscopy image particle size seems to have improved when only vertical focusing is performed, the degree of discreteness between the measurement values ​​is high. The three-dimensional fluid dynamic focusing strictly limits the motion trajectory of the particles by focusing in the horizontal and vertical directions, thereby ensuring the stability of the measurement results. The CV value (that is, the ratio of the standard deviation to the mean) comprehensively reflects the accuracy and stability of the liquid flow focusing and measurement system in the IFC. According to calculations, the CV values ​​of the amplitude measurement of 10μm and 20μm microspheres in this embodiment are 3.3% and 1.9%, and the CV values ​​of the phase measurement are 4.3% and 4.7%, which are very close to the 2-3% level of mature commercial flow cytometers.

[0070] Step 2: Process the focused single-cell quantitative phase microscopy image to obtain the corresponding cell feature data to form a cell feature data set.

[0071] After obtaining the focused single-cell quantitative phase microscopy image of the target cell, it is processed to obtain the corresponding cell feature data, including:

[0072] Step S21: pre-processing the focused single-cell quantitative phase microscopy image, including filling small holes and breaks in the image, smoothing the boundaries, etc.

[0073] Step S22: Convert the preprocessed single-cell quantitative phase microscopy image into a binary image.

[0074] Step S23: using the maximum connected domain in the binary image as a mask to segment the binary image to obtain a cell ROI region.

[0075] Step S24: extracting cell feature data based on the cell ROI region. Specifically, the cell feature data includes: 1. Morphological feature data: describing the geometric features of ROI / cell, including volume, surface area, sphericity, etc.; the dry weight of the cell can also be calculated based on the phase information; 2. First-order grayscale feature data (i.e., first-order histogram feature data): describing the relevant statistical features of different grayscales (voxel intensities) and probability distributions of ROI / cells, but without considering the spatial interaction between them, only calculated based on the global grayscale histogram, including maximum value, minimum value, mean, standard deviation, variance, skewness, kurtosis, entropy, energy, etc. These features reflect the symmetry, uniformity and local intensity distribution changes of the measured voxels; 3. High-order texture feature data: describing the spatial distribution relationship of ROI / cell grayscale values, including grayscale co-occurrence matrix (GLCM), grayscale gradient co-occurrence matrix (GGCM), grayscale run length matrix (GLRLM), grayscale size region matrix (GLSZM), neighborhood grayscale difference matrix (NGTDM), grayscale dependence matrix (GLDM), etc. In this embodiment, the relevant features of the first three matrices are selected. Finally, 81 features are extracted for each target cell to construct the cell feature data of each target cell.

[0076] Based on the cell characteristic data of each cell, a cell characteristic data set can be established. The cell characteristic data set established in the present embodiment covers immune cells, tumor cells, red blood cells, white blood cells, stem cells, germ cells, algae cells, microbial cells and planktonic cells. Wherein, immune cells include but are not limited to T cells, B cells, natural killer cells (NK cells), macrophages, granulocytes, dendritic cells (DC cells), mast cells, monocytes and platelets. Tumor cells include but are not limited to lung cancer cells, colorectal cancer cells, cervical cancer cells, breast cancer cells, human monocytic leukemia cells, prostate cancer cells, skin cancer cells, gastric cancer cells, liver cancer cells, cervical cancer cells, esophageal cancer cells, thyroid cancer cells, bladder cancer cells, lymphoma cells, pancreatic cancer cells, renal cancer cells, uterine cancer cells, oral cancer cells, skin melanoma cells, ovarian cancer cells, brain cancer cells, multiple myeloma cells, nasopharyngeal cancer cells, gallbladder cancer cells, glioblastoma cells and testicular cancer cells. In one embodiment, tumor cells include lung cancer cells A549, colorectal cancer cells HCT116, cervical cancer cells Hela, normal breast cells MCF-10A, breast cancer cells T47D, breast cancer cells MCF-7, breast cancer cells BT474, breast cancer cells MDA-MB-361, breast cancer cells SK-BR-3, breast cancer cells HCC1937, breast cancer cells MDA-MB-231, human monocytic leukemia cells THP-1. Leukocytes include but are not limited to neutrophils, eosinophils and basophils. Stem cells include but are not limited to embryonic stem cells, adult stem cells, induced pluripotent stem cells, mesenchymal stem cells, hematopoietic stem cells, neural stem cells, epithelial stem cells, muscle stem cells, liver stem cells, skin stem cells, pancreatic stem cells and corneal stem cells. Germ cells include but are not limited to sperm cells, spore cells, pollen cells. Microbial cells include but are not limited to bacterial cells and fungal cells. Algae cells include but are not limited to green algae, diatoms, and cyanobacteria.

[0077] Step 3: Screen the cell feature data set for prediction performance to obtain the optimal feature data set.

[0078] When there are too many input features, the classification model has a high risk of overfitting and a high total computational cost. Therefore, it is hoped that the feature dimension can be further reduced while ensuring high accuracy. For some classification models, the higher the feature order, the higher the accuracy of the classification model, but for other classification models, the result is just the opposite. Therefore, the importance of features to the classification model cannot be generalized based on the feature category alone.

[0079] To this end, this embodiment performs prediction performance screening on the cell feature data set to obtain the optimal feature data set, specifically including:

[0080] The importance of all cell features involved in the cell feature dataset is ranked using a feature selection algorithm based on mutual information;

[0081] Based on the ranking, the cell features used for training are determined by an incremental feature selection algorithm, and a subset of the corresponding cell feature data set is selected as the optimal feature data set according to the determined cell features.

[0082] In particular, this embodiment uses the MRMR (Max-Relevance and Min-Redundancy) algorithm based on mutual information to evaluate the classification performance of each cell feature one by one and filter out low-value features. Compared with some methods that process features discretely, the MRMR algorithm based on mutual information simultaneously weighs redundancy (between feature variables) and correlation (feature variables and target variables) to give a feature importance ranking. The ranking results are shown in Figure 6 (a).

[0083] In addition, combined with the incremental feature selection algorithm, according to the feature importance ranking obtained above, the features are added one by one in the order of importance from high to low, and the incremental feature selection (IFS) curve ( Figure 6 (b)). As can be seen from the figure, when the number of features reaches a certain level, the classification accuracy will not be increased or even decrease if more features are added, indicating that redundancy or overfitting does occur. In addition, the more features there are, the longer the model training takes and the slower the prediction speed. After comprehensively weighing the accuracy and efficiency, the cell features used for training are finally determined, and a subset of the corresponding cell feature data set is selected as the optimal feature data set based on the determined cell features for subsequent training.

[0084] Step 4: Use the optimal feature data set as sample data and the type of target cells as labels to train the initial model to obtain a cell classification model.

[0085] After determining the optimal feature data set, the optimal feature data set is used as sample data, and the type of target cell is used as a label to train the initial model to obtain a cell classification model. The initial model can be a linear discriminant analysis (LDA) model, a support vector machine (SVM) model, a neural network (NN) model, a logistic regression (LR) model, a nearest neighbor (KNN) model, a naive Bayes model, a tree model, etc. In particular, the initial model can be a subspace KNN or a Bagging Tree model. The initial model can also be other appropriate machine learning models.

[0086] In this embodiment, the overall classification accuracy of the LDA and SVM models can reach 100% under the optimal feature data set. In view of the fact that the LDA model requires fewer features, shorter training time, and faster prediction speed, LDA is determined to be the best classification model. Further, the recall rate, precision rate, F1 score, accuracy rate, and ROC AUC value can be used to evaluate the classification effect of each classification model on different types of cells (see Figure 6 (c)). In general, after sufficient training, the prediction accuracy of each classification model for various cells is above 90%.

[0087] The present invention also relates to a cell classification method, comprising:

[0088] Acquire focused single-cell quantitative phase microscopy images of cells to be classified;

[0089] Processing the focused single-cell quantitative phase microscopy images of the cells to be classified to obtain corresponding cell characteristic data;

[0090] Based on the cell feature data, the cells to be classified are classified using a cell classification model, and the cell classification model is trained according to the cell classification model training method.

[0091] Among them, the steps of obtaining the focused single-cell quantitative phase microscopy image of the cell to be classified and processing the focused single-cell quantitative phase microscopy image of the cell to be classified to obtain the corresponding cell feature data are similar to those in the aforementioned training method and will not be repeated here.

[0092] The embodiment of the present invention also provides a cell classification model training device, comprising:

[0093] An image acquisition module, used to acquire a focused single-cell quantitative phase microscopy image of a target cell;

[0094] An image processing module is used to process the focused single-cell quantitative phase microscopy image to obtain corresponding cell feature data to form a cell feature data set;

[0095] A feature screening module is used to screen the cell feature data set for predictive performance and obtain the optimal feature data set;

[0096] The model training module is used to train the initial model using the optimal feature data set as sample data and the type of target cells as labels to obtain a cell classification model.

[0097] Preferably, a three-dimensional fluid dynamics focused single-cell image plane holographic image is acquired by a microfluidic chip and a digital holographic microscope, and the three-dimensional fluid dynamics focused single-cell image plane holographic image is processed to obtain a focused single-cell quantitative phase microscopy image; wherein, the microfluidic chip is configured to provide a three-dimensional fluid dynamics focused sample flow containing target cells to the digital holographic microscope, and the digital holographic microscope is configured so that its recording focal plane coincides with the focal plane of the sample flow.

[0098] Preferably, the microfluidic chip is provided with a sample channel, a first sheath flow channel, and a second sheath flow channel arranged along a horizontal plane, and the first sheath flow channel and the second sheath flow channel are symmetrically arranged on both sides of the sample channel and are connected to the sample channel through a confluence port;

[0099] Preferably, the sample flow is injected into the sample channel at a first flow velocity along the vertical direction, two first sheath flows are injected into the sample channel from above and below the sample channel at a second flow velocity at positions symmetrical before and after the sample flow injection point, respectively, and two second sheath flows are injected into the sample channel through the first sheath flow channel and the second sheath flow channel at a third flow velocity; the ratio of the second flow velocity to the first flow velocity is used as the first flow velocity ratio, and the ratio of the third flow velocity to the sum of the first flow velocity and the second flow velocity is used as the second flow velocity ratio, and the first flow velocity ratio and / or the second flow velocity ratio are adjusted to achieve three-dimensional fluid dynamic focusing of the sample flow.

[0100] Preferably, the first sheath flow channel and the second sheath flow channel are both zigzag-line shaped, and at the confluence, the angles between the first sheath flow channel and the second sheath flow channel and the sample channel are equal.

[0101] Preferably, in terms of composition, the first sheath flow and the second sheath flow are the same sheath flow. In microfluidics, the first sheath flow and the second sheath flow exhibit a laminar flow mode, so the two do not mix.

[0102] Preferably, the cell feature data includes at least one of morphological feature data, first-order grayscale feature data, and high-order texture feature data of the cell.

[0103] Preferably, the initial model is one of a linear discriminant analysis (LDA) model, a support vector machine (SVM) model, a neural network (NN) model, a logistic regression (LR) model, a nearest neighbor (KNN) model, a naive Bayes model, and a tree model.

[0104] Preferably, the image processing module processes the focused single-cell quantitative phase microscopy image to obtain cell feature data according to the following steps:

[0105] Preprocessing of focused single-cell quantitative phase microscopy images;

[0106] The preprocessed single-cell quantitative phase microscopy images were converted into binary images;

[0107] The maximum connected domain in the binary image is used as a mask to segment the binary image to obtain the cell ROI area;

[0108] Based on the cell ROI area, cell feature data is extracted.

[0109] Preferably, the feature screening module performs prediction performance screening on the cell feature dataset according to the following steps to obtain the optimal feature dataset:

[0110] The importance of all cell features involved in the cell feature dataset is ranked using a feature selection algorithm based on mutual information;

[0111] Based on the sorting, the cell features used for training are determined by an incremental feature selection algorithm, and a subset of the corresponding cell feature data set is selected as the optimal feature data set according to the determined cell features.

[0112] Preferably, the target cells are selected from at least one of immune cells, tumor cells, red blood cells, white blood cells, stem cells, germ cells, algae cells, microbial cells and planktonic cells, preferably at least two;

[0113] The immune cells are preferably selected from at least one of T cells, B cells, natural killer cells, macrophages, granulocytes, dendritic cells, mast cells, monocytes and platelets;

[0114] The tumor cells are preferably selected from at least one of lung cancer cells, colorectal cancer cells, cervical cancer cells, breast cancer cells, human monocytic leukemia cells, prostate cancer cells, skin cancer cells, gastric cancer cells, liver cancer cells, cervical cancer cells, esophageal cancer cells, thyroid cancer cells, bladder cancer cells, lymphoma cells, pancreatic cancer cells, renal cancer cells, uterine cancer cells, oral cancer cells, skin melanoma cells, ovarian cancer cells, brain cancer cells, multiple myeloma cells, nasopharyngeal cancer cells, gallbladder cancer cells, glioblastoma cells and testicular cancer cells;

[0115] The leukocytes are preferably at least one selected from neutrophils, eosinophils and basophils;

[0116] The stem cells are preferably selected from at least one of embryonic stem cells, adult stem cells, induced pluripotent stem cells, mesenchymal stem cells, hematopoietic stem cells, neural stem cells, epithelial stem cells, muscle stem cells, liver stem cells, skin stem cells, pancreatic islet stem cells and corneal stem cells;

[0117] The reproductive cell is preferably selected from at least one of a sperm cell, a spore cell, and a pollen cell;

[0118] The microbial cells are preferably at least one selected from bacterial cells and fungal cells.

[0119] The present invention also provides a cell classification device, comprising:

[0120] An image acquisition subsystem, used to acquire focused single-cell quantitative phase microscopy images of cells to be classified;

[0121] An image processing subsystem, used for processing the focused single-cell quantitative phase microscopy image of the cells to be classified to obtain corresponding cell characteristic data;

[0122] The cell classification subsystem is used to classify cells to be classified based on cell feature data using a cell classification model. The cell classification subsystem also includes a cell classification model training device for training to obtain a cell classification model.

[0123] Preferably, the image acquisition subsystem comprises a microfluidic chip and a digital holographic microscope, the microfluidic chip and the digital holographic microscope are used to acquire a three-dimensional fluid dynamics focused single-cell image plane holographic image to obtain a focused single-cell quantitative phase microscopic image, wherein the microfluidic chip is configured to provide the digital holographic microscope with a three-dimensional fluid dynamics focused sample flow containing cells to be classified, and the digital holographic microscope is configured so that its recording focal plane coincides with the focal plane of the sample flow;

[0124] Preferably, the microfluidic chip is provided with a sample channel, a first sheath flow channel, and a second sheath flow channel arranged along a horizontal plane. The first sheath flow channel and the second sheath flow channel are arranged on both sides of the sample channel and are connected to the sample channel through a confluence port. A sample flow injection point is provided on the sample channel.

[0125] Preferably, two first sheath flows are injected into the sample channel from above and below the sample channel at a second flow rate along the vertical direction before and after the sample flow injection point, respectively; the sample flow is injected into the sample channel through the sample flow injection point at a first flow rate along the vertical direction; and two second sheath flows are injected into the sample channel through the first sheath flow channel and the second sheath flow channel at a third flow rate; the ratio of the second flow rate to the first flow rate is used as the first flow rate ratio, and the ratio of the third flow rate to the sum of the first flow rate and the second flow rate is used as the second flow rate ratio; the first flow rate ratio and / or the second flow rate ratio are adjusted to achieve three-dimensional fluid dynamic focusing of the sample flow.

[0126] Preferably, the first sheath flow channel and the second sheath flow channel are both zigzag-shaped and symmetrically arranged on both sides of the sample channel. At the confluence, the angles between the first sheath flow channel and the second sheath flow channel and the sample channel are equal. Two first sheath flows are injected into the sample channel from above and below the sample channel at a second flow rate along the vertical direction at positions symmetrical before and after the sample flow injection point.

[0127] Preferably, the first sheath flow and the second sheath flow are the same sheath flow.

[0128] Example 1

[0129] In this embodiment, the cell classification model training method and classification method of the present invention are applied to heterogeneous subtype cells of breast cancer. This embodiment uses cell lines with the same organ origin and more similar morphological characteristics as target cells, including 1 normal breast cell line (N=1000), 7 breast cancer cell lines of different subtypes (N=1000×7) and blank background (N=3000). This group of cell lines was selected because they cover all major breast cancer subtypes (Luminal A / Luminal B / Her2-enriched / TN). In clinical practice, different subtypes guide the choice of clinical treatment options and affect the patient's sensitivity to treatment strategies and prognostic effects. Therefore, it is of great practical value to finely distinguish them, which can benefit patients better.

[0130] See also Figure 7 (a) Based on the results of quantitative phase microscopy image reconstruction of 8 types of cells, TN-type cell lines have certain characteristic similarities with normal breast cells. The average particle sizes of the three cell populations of T47D, BT474 and SK-BR-3 are close and smaller than those of other cell populations. The distribution of some features is visualized as a violin plot ( Figure 7 (b) ), reflecting the differences between different subtypes of breast cancer cells and normal breast cells. Figure 7 (c) summarizes the feature mean distribution of 8 types of cells, and also performs variance analysis and multiple comparison tests. Compared with the feature datasets of different cancer cell lines, the feature dataset of breast cancer multi-subtype cell lines is obviously more complex, which is reflected in the similarity of a single feature between different subtypes (the mean difference is not significant). According to statistics, only one feature among 81 cell features has significant differences in pairwise comparisons, revealing that the classification task of breast cancer multi-subtypes is more difficult. Figure 7 (d) is the Pearson correlation coefficient matrix between features. High correlation still tends to appear in similar features and high-order features, and the frequency of high correlation of features is higher than when distinguishing different types of cancer cells, indicating that the degree of feature redundancy increases and the training difficulty of the classification model becomes higher.

[0131] The combined strategy of MRMR and IFS is still used for feature screening. Figure 8 (a) Sort the MRMR scores of each feature. The results show that although more features are generally needed and it takes longer, after sufficient training, the LDA model still gives a 100% classification accuracy ( Figure 8 (b)). The classification performance of the neural network model and the SVM model also remained stable at a high level. Compared with the basic KNN and Tree models, the classification accuracy of the subspace KNN and Bagging Tree models increased by 3.4% and 13.2% respectively.

[0132] Evaluate the prediction performance of each classification model for different types of cells ( Figure 8 (c)), the prediction accuracy of breast cancer subtypes is generally not as good as that of different types of cancer cell lines, which is in line with common sense because cell groups are more similar to each other. In summary, the strategy training based on quantitative phase microscopy image hierarchical features and MRMR combined with IFS shows universal applicability on various machine learning models. The above research results also further prove the ability of this method to perform multi-class cell / particle fine classification tasks.

[0133] Example 2

[0134] This example applies the cell classification model training method and classification method of the present invention to simulated label-free detection of circulating tumor cells in blood to demonstrate the practical application potential of the present invention. Blood samples were obtained from healthy volunteers, see Fig. 9 (a), respectively, quantitative phase microscopic images of red blood cells (RBC) and peripheral blood mononuclear cells (PBMC) in the blood were obtained. MDA-MB-231 cells were used as a model of breast cancer CTC, and Hela cells were used as a model of cervical cancer CTC. The LDA model was trained according to the feature extraction, model training and other methods of the present invention, and its confusion matrix ( Fig. 9 (b), (c)) show the ability to identify and classify these cancer cells and background blood cells with an accuracy rate close to 100%.

[0135] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A cell classification model training method, characterized in that: include: Acquire focused single-cell quantitative phase microscopy images of target cells; Processing the focused single-cell quantitative phase microscopy image to obtain corresponding cell feature data to form a cell feature data set; Performing prediction performance screening on the cell feature data set to obtain an optimal feature data set; The optimal feature data set is used as sample data, and the type of target cells is used as a label to train the initial model to obtain the cell classification model.

2. The cell classification model training method according to claim 1, characterized in that: A three-dimensional fluid dynamically focused single-cell image plane holographic image is acquired through a microfluidic chip and a digital holographic microscope, and the three-dimensional fluid dynamically focused single-cell image plane holographic image is processed to obtain the focused state single-cell quantitative phase microscopy image; wherein, the microfluidic chip is configured to provide the digital holographic microscope with a three-dimensional fluid dynamically focused sample flow containing target cells, and the digital holographic microscope is configured so that its recording focal plane coincides with the focal plane of the sample flow.

3. The cell classification model training method according to claim 2, characterized in that: The microfluidic chip is provided with a sample channel, a first sheath flow channel, and a second sheath flow channel arranged along a horizontal plane, wherein the first sheath flow channel and the second sheath flow channel are arranged on both sides of the sample channel and are connected to the sample channel through a confluence port, and a sample flow injection point is provided on the sample channel; Preferably, two first sheath flows are injected into the sample channel from above and below the sample channel at a second flow rate along the vertical direction before and after the sample flow injection point, respectively; the sample flow is injected into the sample channel through the sample flow injection point at a first flow rate along the vertical direction; and two second sheath flows are injected into the sample channel through the first sheath flow channel and the second sheath flow channel at a third flow rate; the ratio of the second flow rate to the first flow rate is used as the first flow rate ratio, and the ratio of the third flow rate to the sum of the first flow rate and the second flow rate is used as the second flow rate ratio; the first flow rate ratio and / or the second flow rate ratio are adjusted to enable the sample flow to achieve three-dimensional fluid dynamic focusing.

4. The cell classification model training method according to claim 3, characterized in that: The first sheath flow channel and the second sheath flow channel are both zigzag-shaped and symmetrically arranged on both sides of the sample channel. At the confluence, the first sheath flow channel and the second sheath flow channel have an equal included angle with the sample channel; two first sheath flows are respectively injected into the sample channel from above and below the sample channel at a second flow rate along a vertical direction at positions symmetrical before and after the sample flow injection point; and / or The first sheath flow and the second sheath flow are the same sheath flow.

5. The cell classification model training method according to claim 1, characterized in that: The cell feature data includes at least one of cell morphological feature data, first-order grayscale feature data, and high-order texture feature data; and / or The initial model is one of a linear discriminant analysis (LDA) model, a support vector machine (SVM) model, a neural network (NN) model, a logistic regression (LR) model, a nearest neighbor (KNN) model, a naive Bayes model, and a tree model.

6. The cell classification model training method according to claim 1, characterized in that: The focused single-cell quantitative phase microscopy image is processed to obtain cell feature data according to the following steps: Preprocessing the focused single-cell quantitative phase microscopy image; The preprocessed single-cell quantitative phase microscopy images were converted into binary images; Using the maximum connected domain in the binary image as a mask to segment the binary image to obtain a cell ROI region; Based on the cell ROI area, cell feature data is extracted.

7. The cell classification model training method according to claim 1, characterized in that: The cell feature data set is screened for predictive performance, and the optimal feature data set obtained includes: sorting the importance of all cell features involved in the cell feature dataset by a feature selection algorithm based on mutual information; Based on the ranking, the cell features used for training are determined by an incremental feature selection algorithm, and a subset of the corresponding cell feature data set is selected as the optimal feature data set according to the determined cell features.

8. The cell classification model training method according to claim 1, characterized in that: The target cells are selected from at least one of immune cells, tumor cells, red blood cells, white blood cells, stem cells, reproductive cells, algae cells, microbial cells and plankton cells, preferably at least two; The immune cells are preferably selected from at least one of T cells, B cells, natural killer cells, macrophages, granulocytes, dendritic cells, mast cells, monocytes and platelets; The tumor cells are preferably selected from at least one of lung cancer cells, colorectal cancer cells, cervical cancer cells, breast cancer cells, human monocytic leukemia cells, prostate cancer cells, skin cancer cells, gastric cancer cells, liver cancer cells, cervical cancer cells, esophageal cancer cells, thyroid cancer cells, bladder cancer cells, lymphoma cells, pancreatic cancer cells, kidney cancer cells, uterine cancer cells, oral cancer cells, skin melanoma cells, ovarian cancer cells, brain cancer cells, multiple myeloma cells, nasopharyngeal cancer cells, gallbladder cancer cells, glioblastoma cells and testicular cancer cells; The leukocytes are preferably selected from at least one of neutrophils, eosinophils and basophils; The stem cells are preferably selected from at least one of embryonic stem cells, adult stem cells, induced pluripotent stem cells, mesenchymal stem cells, hematopoietic stem cells, neural stem cells, epithelial stem cells, muscle stem cells, liver stem cells, skin stem cells, pancreatic islet stem cells and corneal stem cells; The reproductive cell is preferably selected from at least one of a sperm cell, a spore cell, and a pollen cell; The microbial cells are preferably selected from at least one of bacterial cells and fungal cells.

9. A cell classification model training device, characterized in that: include: An image acquisition module, used to acquire a focused single-cell quantitative phase microscopy image of a target cell; An image processing module, used for processing the focused single-cell quantitative phase microscopy image to obtain corresponding cell feature data to form a cell feature data set; A feature screening module, used to screen the cell feature data set for predictive performance to obtain an optimal feature data set; The model training module is used to train the initial model using the optimal feature data set as sample data and the type of target cells as labels to obtain the cell classification model.

10. The cell classification model training device according to claim 9, characterized in that: A three-dimensional fluid dynamically focused single-cell image plane holographic image is acquired through a microfluidic chip and a digital holographic microscope, and the three-dimensional fluid dynamically focused single-cell image plane holographic image is processed to obtain the focused state single-cell quantitative phase microscopy image; wherein, the microfluidic chip is configured to provide the digital holographic microscope with a three-dimensional fluid dynamically focused sample flow containing target cells, and the digital holographic microscope is configured so that its recording focal plane coincides with the focal plane of the sample flow.

11. The cell classification model training device according to claim 10, characterized in that: The microfluidic chip is provided with a sample channel, a first sheath flow channel, and a second sheath flow channel arranged along a horizontal plane, wherein the first sheath flow channel and the second sheath flow channel are arranged on both sides of the sample channel and are connected to the sample channel through a confluence port, and a sample flow injection point is provided on the sample channel; Preferably, two first sheath flows are injected into the sample channel from above and below the sample channel at a second flow rate along the vertical direction before and after the sample flow injection point, respectively; the sample flow is injected into the sample channel through the sample flow injection point at a first flow rate along the vertical direction; and two second sheath flows are injected into the sample channel through the first sheath flow channel and the second sheath flow channel at a third flow rate; the ratio of the second flow rate to the first flow rate is used as the first flow rate ratio, and the ratio of the third flow rate to the sum of the first flow rate and the second flow rate is used as the second flow rate ratio; the first flow rate ratio and / or the second flow rate ratio are adjusted to enable the sample flow to achieve three-dimensional fluid dynamic focusing.

12. The cell classification model training device according to claim 11, characterized in that: The first sheath flow channel and the second sheath flow channel are both zigzag-shaped and symmetrically arranged on both sides of the sample channel. At the confluence, the first sheath flow channel and the second sheath flow channel have an equal included angle with the sample channel; two first sheath flows are respectively injected into the sample channel from above and below the sample channel at a second flow rate along a vertical direction at positions symmetrical before and after the sample flow injection point; and / or The first sheath flow and the second sheath flow are the same sheath flow.

13. The cell classification model training device according to claim 9, characterized in that: The cell feature data includes at least one of cell morphological feature data, first-order grayscale feature data, and high-order texture feature data; and / or The initial model is one of a linear discriminant analysis (LDA) model, a support vector machine (SVM) model, a neural network (NN) model, a logistic regression (LR) model, a nearest neighbor (KNN) model, a naive Bayes model, and a tree model.

14. The cell classification model training device according to claim 9, characterized in that: The image processing module processes the focused single-cell quantitative phase microscopy image to obtain cell feature data according to the following steps: Preprocessing the focused single-cell quantitative phase microscopy image; The preprocessed single-cell quantitative phase microscopy images were converted into binary images; Using the maximum connected domain in the binary image as a mask to segment the binary image to obtain a cell ROI region; Based on the cell ROI area, cell feature data is extracted.

15. The cell classification model training device according to claim 9, characterized in that: The feature screening module performs prediction performance screening on the cell feature data set according to the following steps to obtain the optimal feature data set: sorting the importance of all cell features involved in the cell feature dataset by a feature selection algorithm based on mutual information; Based on the ranking, the cell features used for training are determined by an incremental feature selection algorithm, and a subset of the corresponding cell feature data set is selected as the optimal feature data set according to the determined cell features.

16. The cell classification model training device according to claim 9, characterized in that: The target cells are selected from at least one of immune cells, tumor cells, red blood cells, white blood cells, stem cells, reproductive cells, algae cells, microbial cells and plankton cells, preferably at least two; The immune cells are preferably selected from at least one of T cells, B cells, natural killer cells, macrophages, granulocytes, dendritic cells, mast cells, monocytes and platelets; The tumor cells are preferably selected from at least one of lung cancer cells, colorectal cancer cells, cervical cancer cells, breast cancer cells, human monocytic leukemia cells, prostate cancer cells, skin cancer cells, gastric cancer cells, liver cancer cells, cervical cancer cells, esophageal cancer cells, thyroid cancer cells, bladder cancer cells, lymphoma cells, pancreatic cancer cells, kidney cancer cells, uterine cancer cells, oral cancer cells, skin melanoma cells, ovarian cancer cells, brain cancer cells, multiple myeloma cells, nasopharyngeal cancer cells, gallbladder cancer cells, glioblastoma cells and testicular cancer cells; The leukocytes are preferably selected from at least one of neutrophils, eosinophils and basophils; The stem cells are preferably selected from at least one of embryonic stem cells, adult stem cells, induced pluripotent stem cells, mesenchymal stem cells, hematopoietic stem cells, neural stem cells, epithelial stem cells, muscle stem cells, liver stem cells, skin stem cells, pancreatic islet stem cells and corneal stem cells; The reproductive cell is preferably selected from at least one of a sperm cell, a spore cell, and a pollen cell; The microbial cells are preferably selected from at least one of bacterial cells and fungal cells.

17. A cell classification method, characterized in that: include: Acquire focused single-cell quantitative phase microscopy images of cells to be classified; Processing the focused single-cell quantitative phase microscopy image of the cell to be classified to obtain corresponding cell characteristic data; Based on the cell feature data, the cells to be classified are classified using a cell classification model, wherein the cell classification model is obtained by training according to the cell classification model training method according to any one of claims 1-8.

18. A cell classification device, characterized in that: include: An image acquisition subsystem, used to acquire focused single-cell quantitative phase microscopy images of cells to be classified; An image processing subsystem, used for processing the focused single-cell quantitative phase microscopy image of the cell to be classified to obtain corresponding cell characteristic data; A cell classification subsystem is used to classify the cells to be classified based on the cell feature data using a cell classification model. The cell classification subsystem also includes a cell classification model training device according to any one of claims 9 to 16, which is used to train and obtain the cell classification model.

19. The cell classification device according to claim 18, characterized in that: The image acquisition subsystem comprises a microfluidic chip and a digital holographic microscope, wherein the microfluidic chip and the digital holographic microscope are used to acquire a three-dimensional fluid dynamics focused single-cell image plane holographic image to obtain the focused single-cell quantitative phase microscopic image, wherein the microfluidic chip is configured to provide the digital holographic microscope with a three-dimensional fluid dynamics focused sample flow containing cells to be classified, and the digital holographic microscope is configured so that its recording focal plane coincides with the focal plane of the sample flow; Preferably, the microfluidic chip is provided with a sample channel, a first sheath flow channel, and a second sheath flow channel arranged along a horizontal plane, the first sheath flow channel and the second sheath flow channel are arranged on both sides of the sample channel and are connected to the sample channel through a confluence port, and a sample flow injection point is provided on the sample channel; Preferably, two first sheath flows are respectively injected into the sample channel from above and below the sample channel at a second flow rate along a vertical direction before and after a sample flow injection point, the sample flow is injected into the sample channel through the sample flow injection point along a vertical direction at a first flow rate, and two second sheath flows are injected into the sample channel through a first sheath flow channel and a second sheath flow channel at a third flow rate; the ratio of the second flow rate to the first flow rate is taken as a first flow rate ratio, and the ratio of the third flow rate to the sum of the first flow rate and the second flow rate is taken as a second flow rate ratio, and the first flow rate ratio and / or the second flow rate ratio are adjusted to achieve three-dimensional fluid dynamic focusing of the sample flow; Preferably, the first sheath flow channel and the second sheath flow channel are both zigzag-shaped and symmetrically arranged on both sides of the sample channel. At the confluence, the first sheath flow channel, the second sheath flow channel and the sample channel have an equal included angle; two first sheath flows are respectively injected into the sample channel from above and below the sample channel at a second flow rate along a vertical direction at positions symmetrical before and after the sample flow injection point; Preferably, the first sheath flow and the second sheath flow are the same sheath flow.

Images