WO2023046167A1 - 一种细胞识别的方法、装置和系统 - Google Patents

一种细胞识别的方法、装置和系统 Download PDF

Info

Publication number
WO2023046167A1
WO2023046167A1 PCT/CN2022/121340 CN2022121340W WO2023046167A1 WO 2023046167 A1 WO2023046167 A1 WO 2023046167A1 CN 2022121340 W CN2022121340 W CN 2022121340W WO 2023046167 A1 WO2023046167 A1 WO 2023046167A1
Authority
WO
WIPO (PCT)
Prior art keywords
cell
information
traction force
cells
structured
Prior art date
Application number
PCT/CN2022/121340
Other languages
English (en)
French (fr)
Inventor
林哲
Original Assignee
瑞新（福州）科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 瑞新（福州）科技有限公司 filed Critical 瑞新（福州）科技有限公司
Priority to CN202280064594.5A priority Critical patent/CN118076981A/zh
Priority to AU2022350581A priority patent/AU2022350581A1/en
Priority to EP22872210.4A priority patent/EP4407571A1/en
Publication of WO2023046167A1 publication Critical patent/WO2023046167A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Definitions

  • the invention relates to the field of cell analysis and identification, in particular to a method, device and system for cell identification.
  • the current popular single-cell quantitative analysis technology is single-cell sequencing technology (scRNA-seq), which is characterized by quantitative analysis of single-cell transcriptome.
  • scRNA-seq single-cell sequencing technology
  • the disadvantage is that it is equivalent to taking a snapshot of a single cell, which is an intrusion and damage method, and cannot detect the same cell in real time.
  • Other methods, such as immunofluorescence, require staining of cells, which inevitably affects or damages cells, and the process is expensive and complicated.
  • the inventor provides a method for cell identification, comprising the following steps:
  • the cell information includes cell traction force information at a point in the cell acquired based on a cell mechanical sensor, and the cell traction force information includes the size of the cell traction force at the point;
  • the structured cell information includes the number of cells, the number of cell characteristics and characteristic information of each cell characteristic;
  • the cell traction force information also includes the direction of the cell traction force at this point.
  • the cell traction force information also includes the change of the magnitude or direction of the cell traction force at the point within a certain time interval.
  • the cell information also includes cell shape information.
  • the cell information is obtained by performing a cell-limiting operation on the cells.
  • the inventor also provides a device for cell identification, including an information acquisition unit, a preprocessing unit, a learning unit, and an identification unit;
  • the information acquisition unit is used to acquire cell information, the cell information includes the cell traction force information of a certain point in the cell acquired based on the cell mechanical sensor, and the cell traction force information includes the size of the cell traction force at this point;
  • the preprocessing unit is used to preprocess the cell information to form structured cell information;
  • the structured cell information includes the number of cells, the number of cell characteristics and characteristic information of each cell characteristic;
  • the learning unit is used to use structured cell information as input data to establish a cell feature model using supervised, unsupervised or semi-supervised machine learning;
  • the recognition unit is used to apply the cell characteristic model to the classification or clustering of cells of unknown type or state.
  • the cell traction force information also includes the direction of the cell traction force at this point.
  • the cell traction force information also includes the change of the magnitude or direction of the cell traction force at the point within a certain time interval.
  • the cell information also includes cell shape information.
  • the cell information is obtained by performing a cell-limiting operation on the cells.
  • the inventor also provides a cell recognition system, including a cell mechanical sensor and the cell recognition device described in the above technical solution.
  • the cell mechanical sensor includes a nano-pillar array, or a cell traction force detection device with a light reflection layer on the micro-pillar.
  • the cell traction force detection device provided with a light reflection layer on the microcolumn includes:
  • a microcolumn array composed of a plurality of microcolumns that can be deformed by the traction force of the cells is arranged on the base, and the top of the microcolumns or the top of the cylinder surface has a light reflection layer.
  • system for cell identification further includes a cell shape information acquisition device for acquiring cell shape information.
  • the cell shape information acquisition device includes a microscopic camera or a microscopic video camera.
  • system for cell identification further includes a cell definition device, which is used to perform cell definition operations on the cells.
  • the inventor also provides a method for detecting the state of a cell, which includes: using the cell recognition method described in any of the above schemes or the cell recognition device described in any of the above schemes or any of the above schemes Any one of the cell identification systems described in obtains the cell traction force information, and determines the cell state according to the analysis of the cell traction force information;
  • the cell state includes: cell adhesion, cell viability, cell differentiation/activation, cell proliferation and/or cell migration.
  • the cells may be single cells, or multicellular aggregates of any shape formed by two or more cells.
  • the present invention is not limited to the various forms formed by single cells or two or more cells.
  • the present invention uses a cell mechanical sensor to obtain cell mechanical information for cell identification, and this cell identification includes not only the type of the cell, but also the state of the cell;
  • Non-invasive and non-invasive with significant advantages of real-time, high-throughput, and high-resolution, it can be based on the measurement of the traction force of each single cell, and then identify different types of cells, and can be further used to detect the influence of cell force on chemotherapy drugs , even cell sorting, which has great application value in biomedicine and medical treatment; moreover, it can detect multilayer cells, including cell traction such as tumor polymers, so that it can be applied to drug screening, regeneration Scenarios that require the characterization of multicellular aggregates in medicine, gene editing, precision medicine, organ development, and disease modeling
  • Fig. 1 is a schematic diagram of scalarization processing of certain point displacement information in the fourth embodiment of the present invention.
  • Fig. 2 is the result diagram A of using the established cell characteristic model to identify unknown cells or unknown cell phenotypes as described in the extended implementation of the fifth embodiment of the present invention
  • Fig. 3 is the result diagram B of using the established cell characteristic model to identify unknown cells or unknown cell phenotypes as described in the extended implementation of the fifth embodiment of the present invention
  • Fig. 4 is a schematic structural diagram of the device for cell recognition according to the twelfth embodiment of the present invention.
  • FIG. 5 is a schematic structural diagram of the cell recognition system according to the eighteenth embodiment of the present invention.
  • FIG. 6 is a schematic structural diagram of the cell traction force detection device according to the twentieth embodiment of the present invention.
  • FIG. 7 is a schematic structural diagram of the cell recognition system according to the twentieth embodiment of the present invention.
  • Fig. 8a is a schematic structural diagram of the cell recognition system according to the twentieth embodiment of the present invention.
  • 8b is an image of the light reflection signal of the cell traction force detection device acquired by the information acquisition unit of the cell identification system according to the twentieth embodiment of the present invention.
  • Fig. 8c is a visualization effect diagram of the mechanical size and distribution after processing by the preprocessing unit of the cell recognition system according to the twentieth embodiment of the present invention.
  • Fig. 9a is a physical diagram of a cell traction force detection device using a silicon thin film as a cell confining device
  • Fig. 9b is a fluorescence microscope image of the detection device of the cell traction force using the silicon thin film as the cell confining device under light reflection;
  • Figure 9c is an enlarged view of Figure 9b;
  • Figure 10a is a fluorescence imaging diagram of a mixed system of healthy cells and lung non-small cell cancer cells
  • Fig. 10b is a distribution diagram of the light reflection signal of the cell traction force detection device obtained by the information acquisition unit;
  • Figure 10c is a visualization effect diagram of the mechanical size and distribution after the preprocessing unit
  • Figure 10d is an enlarged view of representative single-cell cell force distributions of healthy cells and lung non-small cell carcinoma cells in Figure 10c;
  • Figure 10e is a comparison of healthy cells and lung non-small cell carcinoma cells in cell morphology
  • Figure 10f is a comparison chart of the reflection signal intensity of healthy cells, lung non-small cell carcinoma cells and the mixture of these two cells in different proportions;
  • Fig. 10g is a cluster analysis diagram obtained based on structured cell information processing after Fig. 10c is structured;
  • Figure 11a is a schematic diagram of the operation flow of the cell viability detection method
  • Figure 11b is a comparison chart of the cell viability measured by the MTT method and the cell viability reflected by the cell traction force after A549 cells were treated with different doses of 5FU for 24 hours;
  • Figure 11c is a comparison chart of the cell viability measured by the MTT method and the cell viability reflected by the cell traction force after A549 cells were treated with different doses of 5FU for different times;
  • Figure 12a is a diagram of the operation process of the cell state detection method
  • Figure 12b is a fluorescence microscope image of M0 macrophages differentiated to M1 state
  • Figure 12c is a fluorescence microscope image of M0 macrophages differentiated to M2 state
  • Figure 12d is a comparison chart of cell adhesion area of M0 macrophages, M1 state and M2 state;
  • Figure 12e is a comparison diagram of cell roundness of M0 macrophages, M1 state and M2 state;
  • Figure 12f is a comparison diagram of the traction force of M0 macrophages, M1 state and M2 state;
  • Figure 13a is the characterization diagram of tumor cell multimers with the first morphology with or without the action of 5-Fu; in the figure, from left to right are the mixed images of cell membrane fluorescence and reflection signals (1), light reflection Signal (2), cell nucleus (3), cell membrane (4) and the visualized image of cell force after processing by optical image analysis software (ImageJ) (5);
  • Figure 13b is a characterization diagram of tumor cell multimers with the second morphology with or without 5-Fu; in the figure, from left to right are the mixed images of cell membrane fluorescence and reflection signals (1), light reflection Signal (2), cell nucleus (3), cell membrane (4) and cell force visualization image (5) processed by optical image analysis software (ImageJ).
  • a method for cell identification (only the size of the cell traction force, no label), comprising the steps of:
  • the cell information is the magnitude of the cell traction force at a certain point in the cell obtained based on the cell mechanical sensor, specifically: using the cell mechanical sensor to collect cell information of multiple cells, including the multiple of each cell point to collect information on the size of the cell traction force, so as to obtain multi-point cell traction force size data in multiple cells;
  • the structured cell information includes the number of cells, the number of cell characteristics, and the characteristic information of each cell characteristic.
  • optimization or improvement can be carried out in the following manner: For a single cell, further information processing is performed on the information on the magnitude of the multi-point cell traction force obtained by it, such as calculating: unit The average value of the area cell traction force; the distribution of the cell traction force in the cell; the information of equal dimensions can be used as a new cell feature and added to the two-dimensional feature matrix described in step S2, that is, to expand the content of P, Then through subsequent machine learning to know which feature can better distinguish cells of different types or states.
  • a method for cell identification (only the size of the cell traction force, with labels), comprising the steps of:
  • the cell information is the magnitude of the cell traction force at a certain point in the cell obtained based on a cell mechanical sensor, specifically including: using a cell mechanical sensor to measure the above-mentioned Information collection of several types of cells, including collecting information on the size of the cell traction force at multiple points of each cell, so as to obtain data on the size of the cell traction force at multiple points in multiple cells;
  • the structured cell information includes the number of cells, the number of cell characteristics, and the characteristic information of each cell characteristic.
  • this embodiment uses Random Forest (Random Forest) Forest, RF) algorithm extracts salient features and estimates model parameters, and then applies it to new cells to estimate the labels corresponding to new cells, that is, to apply the cell feature model to the classification of cells of unknown type or unknown state (i.e. identification,
  • identification The "identification” mentioned in the present invention should be understood as identification in a broad sense, including "classification", that is, to determine the type or state of cells; Cells that may have the same or similar properties, and may be of the same or similar type or state are clustered).
  • machine learning algorithms/ideas such as support vector machine (Support Vector Machine, SVM) or deep learning may also be adopted to complete the corresponding model building and training and learning work.
  • optimization or improvement can be carried out in the following manner: For a single cell, further information processing is performed on the information on the magnitude of the multi-point cell traction force obtained by it, such as calculating: unit The average value of the area cell traction force; the distribution of the cell traction force in the cell; the information of equal dimensions can be used as a new cell feature and added to the two-dimensional feature matrix described in step S2, that is, to expand the content of P, Then through subsequent machine learning to know which feature can better distinguish cells of different types or states.
  • a method for cell identification (only the size of the cell traction force, part of the data has labels, and some of the data has no labels), comprising the following steps:
  • This step specifically includes: using a cell mechanical sensor to collect information on the above-mentioned several types of cells, including collecting information on the size of the cell traction force at multiple points of each cell, so as to obtain data on the size of the cell traction force at multiple points in multiple cells;
  • the structured cell information includes the number of cells, the number of cell characteristics, and the characteristic information of each cell characteristic.
  • optimization or improvement can be carried out in the following manner: For a single cell, further information processing is performed on the information on the magnitude of the multi-point cell traction force obtained by it, such as calculating: unit The average value of the area cell traction force; the distribution of the cell traction force in the cell; the information of equal dimensions can be used as a new cell feature and added to the two-dimensional feature matrix described in step S2, that is, to expand the content of P, Then through subsequent machine learning to know which feature can better distinguish cells of different types or states.
  • a method for cell identification (size and direction of cell traction force, no label), comprising the steps of:
  • the cell information is the magnitude and direction of the cell traction force at a certain point in the cell obtained based on the cell mechanical sensor, specifically including: using the cell mechanical sensor (in this embodiment, a nano-column sensor) to multiple Cell information collection, including collecting information on the magnitude and direction of cell traction force at multiple points of each cell, so as to obtain data on the magnitude and direction of cell traction force at multiple points in multiple cells;
  • the cell mechanical sensor in this embodiment, a nano-column sensor
  • the structured cell information includes the number of cells, the number of cell features, and the feature information of each cell feature.
  • the S2 step in this embodiment generally processes the cell information as follows: Assume that the cell traction force vector data (size and direction) of n points are obtained in a cell, and these points are: (i, i ⁇ 1,2,...n), corresponding to two two-dimensional coordinates as well as Among them, t 0 and t n correspond to the initial and displaced points of the nano-pillars, respectively. Thus, That is, the direction of the force on each coordinate point. In addition, each coordinate point should also have a scalar information, that is, the magnitude of the force d.
  • the direction of the cell axis and the coordinates of the center point can be estimated from the existing data, and each cell can be organized into a vector with the same length based on this.
  • the center point can be calculated as:
  • optimization or improvement can be carried out in the following manner: For a single cell, further information processing is performed on the acquired multi-point cell traction force magnitude or cell traction force direction information, for example Calculate: the average value of the cell traction force per unit area; the distribution of the cell traction force in the cell; the distribution of the cell traction force vector in the cell; the information of equal dimensions can be used as a new cell feature and added to step S2
  • the two-dimensional feature matrix is to expand the content of P, and then learn which feature can better distinguish cells of different types or states through subsequent machine learning.
  • a method for cell identification (size and direction of cell traction force, with labels), comprising the steps of:
  • the cell information is the magnitude and direction of the cell traction force at a certain point in the cell obtained based on the cell mechanical sensor, specifically including: using the cell mechanical sensor Collect information on the above-mentioned several types of cells, including collecting information on the size of the cell traction force at multiple points of each cell, so as to obtain data on the size of the cell traction force at multiple points in multiple cells;
  • the structured cell information includes the number of cells, the number of cell characteristics, and the characteristic information of each cell characteristic.
  • this embodiment uses the random forest (Random Forest, RF) algorithm to extract salient features and estimate model parameters, and then apply it to new cells to estimate the labels corresponding to new cells, that is, to apply the cell feature model to unknown types or Classification of cells of unknown status.
  • machine learning algorithms/ideas such as support vector machine (Support Vector Machine, SVM) or deep learning may also be adopted to complete the corresponding model building and training and learning work.
  • FIG. 2 and Fig. 3 are respectively the result diagram A and the result of using the established cell characteristic model to identify unknown cells or unknown cell phenotypes in the extended implementation of the fifth embodiment of the present invention
  • FIG B different rows in Figure A represent different cell types, and different columns represent different samples.
  • the black dots in the figure are the first 50 significant features extracted by the random forest algorithm, or they can be called significant points.
  • the points here Refers to a certain position in a cell, and the cell traction information obtained from different positions is different.
  • Figure B shows the significant discrimination effect of the top 50 salient features (significant points) on three different cell types.
  • the salient features learned based on labeled data can be used to visualize data dimensionality reduction.
  • Subsequent clustering algorithms can also be used to classify and identify cells based on dimensionality reduction data.
  • optimization or improvement can be carried out in the following manner: For a single cell, further information processing is performed on the acquired multi-point cell traction force magnitude or cell traction force direction information, for example Calculated: the average value of the cell traction force per unit area; the distribution of the cell traction force in the cell; the distribution of the cell traction force vector in the cell and other dimensional information, which can be used as a new cell feature and added to the step S2.
  • the two-dimensional feature matrix described above that is, to expand the content of P, and then learn which feature can better distinguish cells of different types or states through subsequent machine learning.
  • a method for cell identification (cell traction force size and direction, part of the data has labels, and some of the data has no labels), comprising the following steps:
  • the cell information is based on cell mechanics sensors Get the size of the cell traction force at a certain point in the cell. It specifically includes: using cell mechanics sensors to collect information on the above-mentioned several types of cells, including collecting information on the magnitude of cell traction force at multiple points of each cell, so as to obtain data on the magnitude of cell traction force at multiple points in multiple cells;
  • the structured cell information includes the number of cells, the number of cell characteristics, and the characteristic information of each cell characteristic.
  • optimization or improvement can be carried out in the following manner: For a single cell, further information processing is performed on the acquired multi-point cell traction force magnitude or cell traction force direction information, for example Calculate: the average value of the cell traction force per unit area; the distribution of the cell traction force in the cell; the distribution of the cell traction force vector in the cell; the information of equal dimensions can be used as a new cell feature and added to step S2
  • the two-dimensional feature matrix is to expand the content of P, and then learn which feature can better distinguish cells of different types or states through subsequent machine learning.
  • a method for cell identification (instantaneous value of the cell traction force vector, the variation of the cell traction force vector within a certain time interval, without labels), comprising the following steps:
  • the cell information is the instantaneous value of the vector of the cell traction force at a certain point in the cell obtained by the cell mechanics sensor and the change of the cell traction force vector at this point within a certain time interval; specifically includes: using cell mechanics
  • the sensor collects cell information on multiple cells, including collecting information on the magnitude and direction of cell traction force at multiple points of each cell, so as to obtain data on the magnitude and direction of cell traction force at multiple points in multiple cells;
  • the structured cell information includes the number of cells, the number of cell features, and the feature information of each cell feature.
  • the structured cell information can be It is regarded as a two-dimensional matrix of M N ⁇ P feature matrix (Feature matrix), where N is the number of cells and P is the number of cell features.
  • the acquired mechanical data of a single cell can actually be compared to is an image (instantaneous value) or a video (changes in the time dimension), so that if the dot matrix covered by the current cell is compared to the pixel in the image, and the information recorded at each point (force Multiple cell features such as size and direction) can be compared to the color corresponding to the pixel, and machine learning algorithms in the field of image and video data processing can be used for reference in subsequent machine learning, for example, machine learning widely used in image recognition , more precisely, the Convolutional Neural Network (CNN) in deep learning (deep learning) is used to model and analyze data.
  • CNN Convolutional Neural Network
  • a method for cell identification (the instantaneous value of the cell traction force vector, the variation of the cell traction force vector within a certain time interval, with labels), comprising the following steps:
  • the cell information is the instantaneous value of the vector of the cell traction force at a certain point in the cell obtained based on the cell mechanics sensor and the cell traction force vector of the point at Changes within a certain time interval; specifically include: using cell mechanics sensors to collect cell information on multiple cells, including collecting information on the magnitude and direction of cell traction force at multiple points of each cell, so as to obtain multiple cells in multiple cells. Point cell traction magnitude and direction data;
  • the structured cell information includes the number of cells, the number of cell features, and the feature information of each cell feature.
  • the structured cell information can be It is regarded as a two-dimensional matrix of M N ⁇ P feature matrix (Feature matrix), where N is the number of cells and P is the number of cell features.
  • identification in the present invention should be understood as identification in a broad sense, including “classification”, that is, determining the type of cells or the state of cells; also including “clustering", That is, although the specific cell state or type is unknown, cells that may have the same or similar properties and may be of the same or similar type or state are clustered).
  • this embodiment uses the random forest (Random Forest, RF) algorithm to extract salient features and estimate model parameters, and then apply it to new cells to estimate the labels corresponding to new cells, that is, to apply the cell feature model to unknown types or Classification of cells of unknown status.
  • RF Random Forest
  • machine learning algorithms/ideas such as support vector machine (Support Vector Machine, SVM) or deep learning may also be adopted to complete the corresponding model building and training and learning work.
  • the acquired mechanical data of a single cell can actually be compared to It is an image (instantaneous value) or a video (multiple instantaneous values in a certain time dimension).
  • the dot matrix under the current cell coverage is compared to the pixels in the image, and the information recorded at each point (A variety of cell characteristics such as the magnitude and direction of force) can be compared to the color corresponding to the pixel, and the machine learning algorithm in the field of image and video data processing can be used for reference in subsequent machine learning.
  • image recognition more precisely, Convolutional Neural Network (CNN) in deep learning (deep learning) to model and analyze data.
  • CNN Convolutional Neural Network
  • a method for cell identification (instantaneous value of the cell traction force vector, the variation of the cell traction force vector within a certain time interval, part of the data has labels, and some data has no labels), comprising the following steps:
  • the cell information is the instantaneous value of the vector of the cell traction force at a certain point in the cell obtained based on the cell mechanics sensor and the cell traction force vector of the point at Changes within a certain time interval; specifically include: using cell mechanics sensors to collect cell information on multiple cells, including the magnitude and direction of cell traction force at multiple points of each cell, and continue to collect this information within a certain time range , so as to obtain multi-point cell traction force magnitude and direction data in multiple cells, and the change information of the multi-point cell traction force vector within a certain time interval;
  • the structured cell information includes the number of cells, the number of cell features, and the feature information of each cell feature.
  • the structured cell information can be It is regarded as a two-dimensional matrix of M N ⁇ P feature matrix (Feature matrix), where N is the number of cells and P is the number of cell features.
  • the acquired mechanical data of a single cell can actually be compared to It is an image (instantaneous value) or a video (multiple instantaneous values in a certain time dimension).
  • the dot matrix under the current cell coverage is compared to the pixels in the image, and the information recorded at each point (A variety of cell characteristics such as the magnitude and direction of force) can be compared to the color corresponding to the pixel, and the machine learning algorithm in the field of image and video data processing can be used for reference in subsequent machine learning.
  • image recognition more precisely, Convolutional Neural Network (CNN) in deep learning (deep learning) to model and analyze data.
  • CNN Convolutional Neural Network
  • a method for cell identification which is different from the first to ninth embodiments in that the cell information also includes cell shape information, and the cell shape information is acquired by a microscopic camera or a microscopic camera (when it is necessary to identify When the information within a certain time interval is continuously collected).
  • the cell features in the data preprocessing step also include cell shape information or further information obtained through further processing or analysis of the cell shape information, including: cell size, cell shape, cell nucleus size, cell nucleus shape, cell One or several colors.
  • a cell identification method which is different from the first to tenth embodiments in that the cell information is collected on the premise of restricting cells.
  • the method of restricting the shape of cells includes physical methods, such as setting up a restriction wall that can surround a certain area (area) around several cell mechanics sensors, the height of this restriction wall is higher than the height of the cell mechanics sensors inside, Therefore, it can be regarded as a cell-defining device. Since it sets up a certain space area, only cells of corresponding size can fall into it and contact the cell mechanical sensor, and it can also be considered that there are only a corresponding number (in most cases, The limiting wall can accommodate 1 cell falling into it). In special cases, the restriction wall can also allow the cells entering it to adapt to the shape of the cross-section surrounded by the restriction wall to a certain extent through extrusion, so as to achieve a certain cell shape restriction effect.
  • the contact between cells can be reduced, so that most of the cells are in a single state.
  • Cell state the dimensions of cell size and shape can be reduced, that is, the number of cell features has been reduced in dimensionality, which can reduce the difficulty of the analysis process to a certain extent; moreover, the differentiation state of cells can also be controlled by using specific cell shapes
  • immune cells can be morphologically limited to the M1 state, and more valuable results can be obtained when used in specific scenarios.
  • the cell restriction scheme described in this embodiment has its unique advantages in specific technical scenarios.
  • this embodiment provides a device for cell identification, including an information acquisition unit 1, a preprocessing unit 2, a learning unit 3 and an identification unit 4;
  • the information acquisition unit 1 is used to acquire cell information, the cell information includes the cell traction force information of a certain point in the cell acquired based on the cell mechanical sensor, and the cell traction force information includes the size of the cell traction force at the point;
  • the preprocessing unit 2 is used to preprocess the cell information to form structured cell information;
  • the structured cell information includes the number of cells, the number of cell characteristics and characteristic information of each cell characteristic;
  • the learning unit 3 is used to use structured cell information as input data to establish a cell feature model using supervised, unsupervised or semi-supervised machine learning;
  • the recognition unit 4 is used to apply the cell feature model to the classification or clustering of cells of unknown type or state.
  • the cell identification device described in this embodiment can be used to realize the technical solution of the cell identification method described in the first to third embodiments.
  • a device for cell identification which is different from the twelfth embodiment in that the cell traction force information acquired by the information acquisition unit 1 also includes the direction of the cell traction force at this point.
  • the cell identification device described in this embodiment can be used to realize the technical solution of the cell identification method described in the fourth to sixth embodiments.
  • a device for cell identification which is different from the twelfth and thirteenth embodiments in that the cell traction force information acquired by the information acquisition unit 1 also includes the magnitude or direction of the cell traction force at the point within a certain time interval The change.
  • the cell identification device described in this embodiment can be used to realize the technical solution of the cell identification method described in the seventh to ninth embodiments.
  • a device for cell identification which differs from the twelfth to fourteenth embodiments in that the cell traction force information acquired by the information acquisition unit 1 also includes cell shape information.
  • the cell identification device described in this embodiment can be used to implement the technical solution of the cell identification method described in the tenth embodiment.
  • a device for cell identification which differs from the twelfth to fifteenth embodiments in that the cell information is obtained by performing a cell-limiting operation on the cells.
  • the method of restricting the shape of cells includes physical methods, such as setting up a restriction wall that can surround a certain area (area) around several cell mechanics sensors, the height of this restriction wall is higher than the height of the cell mechanics sensors inside, Therefore, it can be regarded as a cell-defining device. Since it sets up a certain space area, only cells of corresponding size can fall into it and contact the cell mechanical sensor, and it can also be considered that there are only a corresponding number (in most cases, The limiting wall can accommodate 1 cell falling into it). In special cases, the restriction wall can also allow the cells entering it to adapt to the shape of the cross-section surrounded by the restriction wall to a certain extent through extrusion, so as to achieve a certain cell shape restriction effect.
  • the contact between cells can be reduced, so that most of the cells are in a single state.
  • Cell state the dimensions of cell size and shape can be reduced, that is, the number of cell features has been reduced in dimensionality, which can reduce the difficulty of the analysis process to a certain extent; moreover, the differentiation state of cells can also be controlled by using specific cell shapes
  • immune cells can be morphologically limited to the M1 state, and more valuable results can be obtained when used in specific scenarios.
  • the cell restriction scheme described in this embodiment has its unique advantages in specific technical scenarios.
  • the contact between cells can be reduced, so that most of the cells are in a single state.
  • Cell state the dimensions of cell size and shape can be reduced, that is, the number of cell features has been reduced in dimensionality, which can reduce the difficulty of the analysis process to a certain extent; moreover, the differentiation state of cells can also be controlled by using specific cell shapes
  • immune cells can be morphologically limited to the M1 state, and more valuable results can be obtained when used in specific scenarios.
  • the cells will be in a natural state, which can reduce the impact of external intervention on the cells; and, the edge detection of the cells must be performed without morphological restrictions to obtain the cell shape, tropism and other characteristics, thereby assisting the identification and prediction of cells.
  • the cell restriction scheme described in this embodiment has its unique advantages in specific technical scenarios.
  • the cell identification device described in this embodiment can be used to realize the technical solution of the cell identification method described in the eleventh embodiment.
  • a cell recognition system includes a cell mechanical sensor 10 and a cell recognition device 20 .
  • the cell recognition device is the cell recognition device described in the above-mentioned twelfth to sixteenth embodiments, and is used to realize the technical solution described in the cell recognition method described in the above-mentioned first to ninth embodiments.
  • the cell mechanical sensor 10 in this embodiment is a nano-pillar array.
  • any device capable of acquiring cell traction force information can be used as the cell mechanical sensor.
  • this embodiment provides a cell recognition system, which differs from the seventeenth embodiment in that it further includes a cell shape information acquisition device 30 for acquiring cell shape information.
  • the cell morphology information acquisition device 30 described in this embodiment is a microscopic camera. In other embodiments, the cell morphology information acquisition device 30 may also be a microscopic camera or other equipment capable of acquiring cell morphology information.
  • the cell identification system described in this embodiment can be used to implement the technical solution described in the cell identification method described in the tenth embodiment above.
  • a system for cell identification which differs from the seventeenth and eighteenth embodiments in that it further includes a cell defining device 40 for performing a cell defining operation on cells.
  • the cell confining device 40 has the following structure: a confinement wall that can surround a certain area (area) is established around several cell mechanics sensors, and the height of the confinement wall is higher than that of the cell mechanics sensors inside it. , so it can be regarded as a cell-limiting device. Since it sets up a certain space area, only cells of corresponding size can fall into it and contact the cell mechanical sensor, and it can also be considered that only a corresponding number (in most cases , the limit wall can accommodate 1 cell falling into it).
  • the restriction wall can also allow the cells entering it to adapt to the shape of the cross-section surrounded by the restriction wall to a certain extent through extrusion, so as to achieve a certain cell shape restriction effect.
  • the cell identification system described in this embodiment can be used to realize the technical solution described in the cell identification method described in the eleventh embodiment above.
  • a cell recognition system which is different from the seventeenth, eighteenth, and nineteenth embodiments in that the cell mechanical sensor 10 in this embodiment is a cell traction force detection device with a light reflection layer on a microcolumn , in other embodiments, any device capable of acquiring cell traction force information may also be used as a cell mechanical sensor.
  • FIG. 6 is a schematic structural diagram of a cell pulling force detection device.
  • the cell traction force detection device in this embodiment includes a light-transmitting base 101 and a microcolumn 103 arranged on the base 101 that can be deformed by the cell traction force.
  • the top of the microcolumn 103 is Coated with a light reflection layer 1031, the thickness of the light reflection layer 1031 is 5nm (in other embodiments, the thickness of the light reflection layer 1031 can be between 5nm-20nm-the thickness of the coating is related to the coating material, in Under the premise of applying the same coating material, the choice of coating thickness should be limited to ensure the light transmission effect, the stability of the micro-pillar, and the connection with the micro-pillar not to fall off).
  • the cylinders of the micropillars 103 can transmit light, and the clusters of arrows in opposite directions in the figure represent incident light and reflected light. (Note: The term "coating" is used in this embodiment, which only means that the light reflection layer 1031 in this embodiment can be prepared by a coating process, and does not limit that the light reflection layer 1031 must be prepared by a coating process)
  • FIG. 7 is a schematic structural diagram of a cell recognition system related to the twenty-first embodiment of the present invention.
  • the system shown in Figure 7 also involves: under the base 101 An optical signal generating device 102 with a light source is provided, and the light emitted by the light source is irradiated from the light-transmitting base 101 of the cell traction force detection device 10 to the light reflection layer of the microcolumn 103 through the incident optical path; the information acquisition unit 1 uses After detecting the light reflected from the light reflection layer 1031 on the top of the microcolumn 103 , the light reflected by the light reflection layer 1031 enters the cell identification device 20 after passing through the reflection optical path and the beam splitter 104 .
  • the information acquisition unit 1 (optical signal detection device) of the cell identification device 20 obtains the reflected light signal
  • the information is preprocessed by the preprocessing unit 2 to form structured cell information;
  • the structured cell information includes cell number, The number of cell features and the feature information of each cell feature.
  • the structured cell information can be regarded as a two-dimensional feature matrix (Feature matrix), where N is the number of cells, and P is the number of cell features.
  • the cell feature is the size of the cell traction force and/or the distribution of the cell traction force in the cell; then, using the above-mentioned structured cell information as input data, use supervised, unsupervised or semi-supervised learning unit 3 to learn to establish a cell feature model , and use the structured cell information of a large number of cells to train the cell feature model; finally, the cell feature model is applied to the classification or clustering of cells of unknown type or unknown state through the recognition unit 4 .
  • the microcolumn 12 When the microcolumn 12 is not stressed, the microcolumn should remain upright, so as to reflect the detection light to the greatest extent; and when the microcolumn 103 is in contact with the cell, the microcolumn 103 bends under the action of the cell traction force, Resulting in a reduced level of light reflection. Therefore, when the cell traction force is greater, the obtained light reflection signal should be smaller, so that the size of the cell traction force at this point can be easily deduced by observing the intensity of the light reflection signal.
  • the optical splitter 104 may be a transflective or other equivalent optical element, the main purpose of which is to simplify the design of the optical path.
  • the optical signal generating device 102 may be LED, halogen lamp, laser (eg, infrared laser), or other light sources, or other devices with these light sources, which are not specifically limited in the present invention.
  • the information acquisition unit 1 in the cell recognition device can be a microscope, a charge-coupled device CCD, a complementary metal oxide semiconductor CMOS, a photomultiplier tube PMT and a photoelectric converter PT, a film, or other
  • the optical signal detection elements with the same function are not specifically limited in the present invention.
  • the preprocessing unit, learning unit, and identification unit in the device for cell identification are the image/data processing device 201 that integrates these units, for example it can be optical image analysis software ImageJ, Matlab, Fluoview , Python, or other optical image/data analysis components with the same function, or the combined use of these analysis software, the present invention is not specifically limited.
  • Fig. 8a is a schematic structural diagram of the cell recognition system
  • Fig. 8b is an image of the light reflection signal of the cell traction force detection device obtained by the information acquisition unit
  • Fig. 8c is a visualization effect diagram of the mechanical size and distribution.
  • each microcolumn is provided with a metal reflective layer, and the side is provided with an antireflection layer; when there are no cells, the light irradiates the microcolumn from below, and it will be completely reflected.
  • the information acquisition unit (such as a CCD camera) in the device recognized by the cell is fully received; but when the cell is attached to the microcolumn, the cell force generated by the cell movement will make the microcolumn tilt, thereby reducing the reflected signal. After analyzing the reflected signal, the strength of the cell force can be calculated.
  • the image of the light reflection signal of the cell traction force detection device and the enlarged image of the local cell adhesion area are collected by an information acquisition unit (such as a CCD camera) (as shown in FIG. 8b ). Then, the image in Fig. 8b is further processed by the preprocessing unit to convert it into a more intuitive visualization effect of mechanical size and distribution in Fig. 8c.
  • an information acquisition unit such as a CCD camera
  • the specific processing process is as follows: First, based on Figure 8b, obtain the bright field reflection signal map (I, focusing on the cell); then, filter the high-frequency signal after performing Fourier transform on the image, and perform the inverse Fourier transform operation, thereby calculating and obtaining Then, the I and I 0 images are further processed to convert the reflection signal map into a more intuitive cell mechanics map (I 0 signal value minus I Signal value) after normalization to obtain a more intuitive cell traction strength map j.
  • the difference between this embodiment and the twentieth embodiment lies in that the cell limiting device 40 in this embodiment is a silicon thin film.
  • Fig. 9a is a physical diagram of a cell traction force detection device using a silicon thin film as a cell limiting device. Each well is equipped with micropillars, which restrict the shape and migration of cells through the silicon film, and at the same time control the contact or adhesion between cells; Fluorescence microscope image of the device, Figure 9c is a magnified view of Figure 9b.
  • the size of each hole of the silicon membrane can be set to fit the size of a single cell, suitable for single cell attachment, thereby limiting cell contact, cell shape and its migration range.
  • This embodiment specifically provides a method for applying the cell traction information obtained by the cell identification system described in the twentieth embodiment to identifying cells.
  • Figure 10a is the fluorescence imaging diagram of the mixed system of healthy cells and lung non-small cell cancer cells
  • Figure 10b is the light reflection signal distribution diagram of the cell traction force detection device obtained by the information acquisition unit
  • Figure 10c is the mechanical Visualization effect diagram of size and distribution
  • Figure 10d is an enlarged view of the representative single-cell cell force distribution of healthy cells and lung non-small cell cancer cells in Figure 10c
  • Figure 10e is a healthy cell and lung non-small cell cancer cells in the cell Morphological comparison chart
  • Figure 10f is a comparison chart of the reflection signal intensity of healthy cells, lung non-small cell carcinoma cells, and the two cells mixed in different proportions
  • Figure 10g is a structured processing of Figure 10c, based on Cluster analysis graph obtained by structured cell information processing.
  • this embodiment takes healthy cells (Normal) and lung non-small cell carcinoma cell lines (Cancer) as detection objects, uses two different fluorescent dyes (Dil&DIO) to pre-stain the cell membranes of healthy cells and lung cancer cells, and uses a certain After mixing in proportion, add to the same cell traction force detection device (in other embodiments, it may be added to different independent cell traction force detection devices).
  • healthy cells Normal
  • lung non-small cell carcinoma cell lines Cancer
  • the image of the light reflection signal of the cell traction force detection device (as shown in Figure 10b) is collected through the information acquisition unit (a microscope is used in this embodiment), and the high-resolution force field distributions in the two types of cells are directly obtained by the information acquisition unit.
  • the rendering can be converted into a readable light intensity attenuation signal (reflecting the strength of the cell force) and displayed in the picture (as shown in Figure 10c). According to the difference in the degree of light attenuation of the two types of cells shown in the picture in FIG. 10 c , the two types of cells can be visually distinguished (qualitative analysis) by visual observation.
  • the optical reflection signal in Fig. 10c is further processed by an optical signal analysis device.
  • this embodiment uses ImageJ and Python analysis software (in other embodiments, other image/data analysis software can also be used) to collect information on the obtained cell force field in Figure 10c, including multi-point Carry out cell traction size information collection, thereby obtaining multi-point cell traction size data in multiple cells; preprocess the acquired cell traction size information to form structured cell information; and obtain healthy cells and lungs based on structured cell information analysis
  • the comparison results of non-small cell cancer cells in terms of cell morphology (as shown in FIG. 10e ).
  • Feature Matrix feature matrix
  • the learning unit of the supervised image/data processing device learns to establish cell characteristics model, and use the structured cell information of a large number of cells to train the cell feature model to obtain the cluster analysis diagram shown in Figure 10g, and then apply the obtained cell feature model to the cell of unknown type or unknown state classification recognition.
  • the recognition unit of the image/data processing device can be used to cluster and classify normal healthy cells and cancer cells. type, so as to realize the identification of unknown cell types.
  • Figure 10e shows that there is no statistically significant difference in the morphology (including cell adhesion area and cell roundness) of different cells;
  • Figure 10f shows the obvious difference in the reflected signal intensity (reflecting cell force) between normal cells and tumor cells , and after mixing normal cells and tumor cells in a certain proportion, the reflected signal intensity and the mixing proportion have a certain linear relationship. It can be seen that compared with other characteristics of cells (such as cell adhesion area, cell roundness and other morphological information in Fig. Perform more intuitive and accurate identification (quantitative and qualitative analysis).
  • tumor cells showed higher traction force magnitudes than normal cells, and the distribution was more uneven. It can be seen that after the cell traction force is visualized in the form of an image, the force field characteristics of different cells can be seen intuitively by the naked eye; and further, the force field size of each point of different cells is structured by image analysis software , the comprehensive analysis obtained the cell morphology information in Fig. 10e, the reflection signal intensity in Fig. 10f (reflecting the cell force) and the cluster analysis diagram in Fig. 10g.
  • the present invention can cluster, type and quantitatively analyze different cells (such as healthy cells and non-small cell lung cancer cells in this embodiment) through comprehensive analysis of the force field structured information at each point of the cell, thereby realizing accurate cell types identify.
  • the cell identification system including the cell traction force detection device of this embodiment, it can not only be visually distinguished by the naked eye for qualitative analysis, but also more intuitive and accurate for the state and type of cells based on the measured cell mechanical characteristics. Identification (quantitative and qualitative analysis), and confirmed that the cell force field as a marker can better distinguish cell types.
  • This embodiment specifically provides the cell traction force information obtained by the cell identification system described in the twentieth embodiment, which is applied to monitor the viability of the cells.
  • Figure 11a is a schematic diagram of the operation flow of the cell viability detection method
  • Figure 11b is the cell viability measured by the MTT method after A549 cells were treated with different doses of 5FU for 24 hours and the cell identification system based on this embodiment The comparison chart of the obtained cell traction force
  • Figure 11c is a comparison chart of the cell viability measured by the MTT method and the cell traction force obtained by the cell recognition system of this embodiment after A549 cells were treated with different doses of 5FU for different periods of time.
  • non-small cell lung cancer cells A549 were cultured on multiple cell traction force detection devices, and treated with different doses of 5-fluorouracil (5-FU), a drug that inhibits cell proliferation, and passed the twentieth implementation
  • a cell recognition system described in the example monitors the cell traction force at different time points, and monitors the cell proliferation and cytotoxicity at different time points through the CCK-8 kit.
  • the cell viability determined by the MTT assay is used as the control group, and the obtained Data for Figure 11b and Figure 11c.
  • the cell viability measured by the MTT assay method and the cell viability reflected by the cell traction force gradually decrease in a dose-dependent manner. , that is, the cell traction force is positively correlated with cell viability.
  • the cell viability determined by the MTT method did not change significantly; however, by measuring the traction force, the cell traction force could be observed at an earlier time point before the decrease in cell metabolic activity was detected by the MTT method Specifically, at a treatment dose of 0.5 ⁇ M, an obvious decreasing trend can appear at 6 h, and at a treatment dose of 1 ⁇ M, an obvious decreasing trend can appear at 3 h, so that the reduction of cell viability can be more sensitively characterized.
  • the direct detection of cell traction force by the cell identification system including the cell traction force detection device of this embodiment is a highly sensitive and effective method for evaluating the activity of cells responding to drugs.
  • This embodiment specifically provides the cell traction force information acquired by the cell identification system described in the twentieth embodiment, and analyzes and determines the cell state according to the cell traction force information.
  • Figure 12a is a diagram of the operation process of the cell state detection method
  • Figure 12b is a fluorescence microscope image of M0 macrophages differentiated to M1 state
  • Figure 12c is a fluorescence microscope image of M0 macrophages differentiated to M2 state
  • Fig. 12d is a comparison diagram of cell adhesion area of M0 macrophages, M1 state and M2 state
  • Fig. 12e is a comparison diagram of cell roundness of M0 macrophages, M1 state and M2 state
  • Fig. 12f is a comparison diagram of M0 macrophages, Traction comparison chart of M1 state and M2 state.
  • macrophages are used as detection objects, which are respectively added to microcolumns of different independent cell traction force detection devices, and endotoxin LPS and interleukin IL4 are used to guide macrophages to differentiate from M0 to M1 and M2 state, with M0 state as the control group.
  • endotoxin LPS and interleukin IL4 are used to guide macrophages to differentiate from M0 to M1 and M2 state, with M0 state as the control group.
  • collect Figure 12b and Figure 12c through the information acquisition unit (a microscope is used in this embodiment), and perform further image processing and data analysis on Figure 12b and Figure 12c through the image/data processing device (ImageJ and Python), This was transformed into structured information and analyzed to generate the data of Figures 12d-12f. From the data in Figures 12a to 12f, it can be seen that M0 macrophages and their differentiated M1 and M2 states are There are obvious differences between them.
  • This embodiment specifically provides the cell traction force information obtained by the cell identification system described in the twentieth embodiment, and the cell state is determined according to the analysis of the cell traction force information.
  • the multicellular aggregates provided in this example are combined on the cell traction force detection device in various ways, and this example provides two specific ways of combination:
  • the first combination method set the culture medium on the microcolumn of the cell traction force detection device, and transplant the cells to the culture medium on the microcolumn to obtain multicellular aggregates; in other embodiments, this combination method, in The cell traction force information is output in a visualized form, and the cell culture process can be monitored in real time, so as to be applied to the influence of chemical, biological and physical external stimuli such as culture medium and drugs on cell growth;
  • the second combination method directly adhering the cultured multicellular aggregate to the microcolumn of the cell traction force detection device for detection.
  • this embodiment provides a tumor cell multimer cultured on a cell traction force detection device, which is applied to a drug sensitivity test, including the following steps:
  • tumor cell polymers apply FN 50 ⁇ g/mL on the top of the microcolumn of the cell traction force detection device, and sterilize with ultraviolet light for 30 minutes; breast cancer cells MCF-7 (about 1 ⁇ 10 5 ⁇ 9 ⁇ 10 5 1) planted on the top surface of the microcolumn of the cell traction force detection device (the number of microcolumns is not limited), and then the cell traction force monitoring device is immersed in 3dGRO TM Spheroid Medium (S3077) culture medium and cultivated for more than 3 days to guide the tumor cellular multimer production;
  • 3dGRO TM Spheroid Medium S3077
  • Figure 13a is a characterization diagram of tumor cell multimers with the first morphology with or without 5-Fu;
  • Figure 13b is a tumor cell multimer with the second morphology in Characterization diagrams with/without 5-Fu effect; from left to right are the mixed image of cell membrane fluorescence and reflection signal (1), light reflection signal (2), cell nucleus (3), cell membrane (4) and optical image Visualization image of cell force processed by analysis software (Image J) (5).
  • Image J analysis software
  • two representative forms are selected for cell morphology detection.
  • the first form refers to the cell form in which two larger cells stick together;
  • the second form refers to a group of small cells sticking together. cell morphology together.
  • the cell traction force detection device of the present invention can measure the cell traction force of multicellular aggregates (such as tumor polymers), and can be used to monitor the viability of cell polymers through cell traction force, and can distinguish different cells form.
  • cellular multimer refers to:
  • Cells are the basic structural and functional units of organisms. Cells usually multiply or differentiate to form two or more cells that aggregate together to form a cell population, namely: multicellular aggregates; multicellular aggregates include tumor polymers and other in vitro or in vivo cultured cells. Obtained cell groups.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Medical Informatics (AREA)
  • Biophysics (AREA)
  • Multimedia (AREA)
  • Databases & Information Systems (AREA)
  • Image Analysis (AREA)
  • Investigating Or Analysing Biological Materials (AREA)
  • Image Processing (AREA)

Abstract

为解决细胞识别的问题，发明人提供了一种细胞识别的方法，包括如下步骤：获取细胞信息，所述细胞信息包括基于细胞力学传感器获取的细胞中某点的细胞牵引力信息，所述细胞牵引力信息包括该点细胞牵引力的大小；对细胞信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息；以结构化细胞信息作为输入数据，利用有监督、无监督或半监督的机器学习建立细胞特征模型，并将所述细胞特征模型应用于未知类型或未知状态的细胞的分类或聚类。发明人同时提供了实现上述技术方案的细胞识别的装置以及细胞识别的系统，具有高通量、高分辨率、能实时测量分析活体单细胞的特点。

Description

一种细胞识别的方法、装置和系统 技术领域
本发明涉及细胞分析与鉴定领域，尤其涉及一种细胞识别的方法、装置和系统。
背景技术
目前流行的单细胞层面的定量分析技术为单细胞测序技术(scRNA-seq),其特点在于对单细胞转录组进行定量分析。但缺点在于其相当于对单细胞做了快照，是侵入损毁型的方法，无法对相同的一个细胞进行实时检测。其它方法例如免疫荧光，需要对细胞进行染色，难免对细胞造成影响或损害，过程昂贵且复杂。
发明内容
为此，需要提供一种具有高通量、高分辨率特点，并且能实时测量分析活体单细胞的技术，来解决细胞识别的问题，以进一步帮助科研、药物开发及临床应用。
为实现上述目的，发明人提供了一种细胞识别的方法，包括如下步骤：
获取细胞信息，所述细胞信息包括基于细胞力学传感器获取的细胞中某点的细胞牵引力信息，所述细胞牵引力信息包括该点细胞牵引力的大小；
对细胞信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息；
以结构化细胞信息作为输入数据，利用有监督、无监督或半监督的机器学习建立细胞特征模型，并将所述细胞特征模型应用于未知类型或未知状态的细胞的分类或聚类。
进一步地，所述的细胞识别的方法中，所述细胞牵引力信息还包括该点细胞牵引力的方向。
进一步地，所述的细胞识别的方法中，所述细胞牵引力信息还包括该点细胞牵引力的大小或方向在一定时间间隔内的变化。
进一步地，所述的细胞识别的方法中，所述细胞信息还包括细胞形貌信息。
进一步地，所述的细胞识别的方法中，所述细胞信息是在对细胞进行细胞限定操作下获取的。
发明人同时提供了一种细胞识别的装置，包括信息获取单元、预处理单元、学习单元和识别单元；
所述信息获取单元用于获取细胞信息，所述细胞信息包括基于细胞力学传感器获取的细胞中某点的细胞牵引力信息，所述细胞牵引力信息包括该点细胞牵引力的大小；
所述预处理单元用于对细胞信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息；
所述学习单元用于以结构化细胞信息作为输入数据，利用有监督、无监督或半监督的机器学习建立细胞特征模型；
所述识别单元用于将所述细胞特征模型应用于未知类型或未知状态的细胞的分类或聚类。
进一步地，所述的细胞识别的装置中，所述细胞牵引力信息还包括该点细胞牵引力的方向。
进一步地，所述的细胞识别的装置中，所述细胞牵引力信息还包括该点细胞牵引力的大小或方向在一定时间间隔内的变化。
进一步地，所述的细胞识别的装置中，所述细胞信息还包括细胞形貌信息。
进一步地，所述的细胞识别的装置中，所述细胞信息是在对细胞进行细胞限定操作下获取的。
发明人同时还提供了一种细胞识别的系统，包括细胞力学传感器和如上技术方案所述的细胞识别装置。
进一步地，所述细胞力学传感器包括纳米微柱阵列，或微柱上设有光线反射层的细胞牵引力检测装置。
进一步地，所述微柱上设有光线反射层的细胞牵引力检测装置包括：
基座，以及
设置于基座上的，可受细胞牵引力作用而产生形变的多个微柱构成的微柱阵列，所述微柱的顶部或柱面上部具有光线反射层。
进一步地，所述的细胞识别的系统中，还包括细胞形貌信息获取装置，用于获取细胞形貌信息。
进一步地，所述的细胞识别的系统中，所述细胞形貌信息获取装置包括显微照相机或显微摄像机。
进一步地，所述的细胞识别的系统中，还包括细胞限定装置，用于对细胞进行细胞限定操作。
发明人同时提供了一种细胞状态的检测方法，其包括：通过上述任意一项方案中所述的细胞识别的方法或上述任意一项方案中所述的细胞识别的装置或上述任意一项方案中所述的细胞识别的系统中的任意一项获取细胞牵引力信息，根据所述细胞牵引力信息分析确定细胞状态；
所述细胞状态包括：细胞黏附，细胞活力，细胞分化/活化，细胞增殖和/或细胞迁移。
进一步地，上述任一方案中，细胞可以是单细胞，或者两个以上细胞形成的任何形态的多细胞聚合体。本发明对单细胞或两个以上多细胞形成的各种形态无限定。
区别于现有技术，上述技术方案具有如下优点：本发明利用细胞力学传感器获取细胞力学信息用于细胞识别，这种细胞识别不仅包括细胞的类型，也包括细胞的状态；而且本技术对活细胞无创无影响，具有实时、高通量、高分辨率的显著优势，可基于对每个单细胞的牵引力进行测量，进而识别不同类型细胞，并可进一步用于侦测细胞力受化疗药物的影响，甚至细胞分选，用于生物医药、医疗方面的都具有很大的应用价值；并且，可以对多层细胞，包括肿瘤多聚体等细胞牵引力进行检测，使其可以应用于药物筛选、再生医学、基因编辑、精准医疗、器官发育、疾病建模中的需要对多细胞聚合体表征的场景
附图说明
图1为本发明第四实施例中对某点位位移信息标量化处理的示意图；
图2为本发明第五实施例扩展实施方式中所述将建立的细胞特征模型用于未知细胞或未知细胞表现型的识别的结果图A；
图3为本发明第五实施例扩展实施方式中所述将建立的细胞特征模型用于未知细胞或未知细胞表现型的识别的结果图B；
图4为本发明第十二实施例所述细胞识别的装置的结构示意图；
图5为本发明第十八实施例所述细胞识别的系统的结构示意图；
图6为本发明第二十实施例所述细胞牵引力检测装置的结构示意图；
图7为本发明第二十实施例所述细胞识别的系统的结构示意图；
图8a为本发明第二十实施例所述细胞识别的系统的结构示意图；
图8b为本发明第二十实施例所述细胞识别的系统的信息获取单元获取得到的细胞牵引力检测装置光反射讯号的图像；
图8c为本发明第二十实施例所述细胞识别的系统的预处理单元处理后的力学大小及分布的可视化效果图；
图9a为采用硅薄膜作为细胞限定装置的细胞牵引力检测装置的实物图；
图9b为在光反射下采用硅薄膜作为细胞限定装置的细胞牵引力的检测装置的荧光显微镜图；
图9c为图9b的放大图；
图10a为健康细胞和肺非小细胞癌细胞混合体系的荧光成像图；
图10b为信息获取单元获取得到的细胞牵引力检测装置光反射讯号分布图；
图10c为预处理单元处理后的力学大小及分布的可视化效果图；
图10d为图10c中的健康细胞和肺非小细胞癌细胞的代表性单细胞细胞力分布的放大图；
图10e为健康细胞和肺非小细胞癌细胞在细胞形态上的对比图；
图10f为健康细胞、肺非小细胞癌细胞以及这两种细胞以不同比例混合后的反射信号强弱的对比图；
图10g为将图10c经过结构化处理后，基于结构化细胞信息处理得到的聚类分析图；
图11a为细胞活力检测方法的操作流程示意图；
图11b为A549细胞在不同剂量的5FU处理24h后，MTT法测定得到的细胞活力以及细胞牵引力所反映的细胞活力的对比图；
图11c为A549细胞在不同剂量的5FU处理不同时间后，MTT法测定得到的细胞活力以及细胞牵引力所反映的细胞活力的对比图；
图12a为细胞状态检测方法的操作过程图；
图12b为M0巨噬细胞分化至M1状态的荧光显微镜图；
图12c为M0巨噬细胞分化至M2状态的荧光显微镜图；
图12d为M0巨噬细胞、M1状态以及M2状态的细胞黏附面积对比图；
图12e为M0巨噬细胞、M1状态以及M2状态的细胞圆度对比图；
图12f为M0巨噬细胞、M1状态以及M2状态的牵引力对比图；
图13a为具有第一种形态的肿瘤细胞多聚体在有/无5-Fu作用后的表征图；图中，从左到右分别是细胞膜荧光和反射讯号的混合图像(1)、光反射讯号(2)、细胞核(3)、细胞膜(4)及经过光学图像分析软件(ImageJ)处理后的细胞力可视化图像(5)；
图13b为具有第二种形态的肿瘤细胞多聚体在有/无5-Fu作用后的表征图；图中，从左到右分别是细胞膜荧光和反射讯号的混合图像(1)、光反射讯号(2)、细胞核(3)、细胞膜(4)及经过光学图像分析软件(ImageJ)处理后的细胞力可视化图像(5)。
附图标记说明：
1-信息获取单元；
2-预处理单元；
3-学习单元；
4-识别单元；
10-细胞力学传感器(细胞牵引力检测装置)；
101-基座；102-光信号发生装置；103-微柱；104-分光器；
1031-光线反射层；
20-细胞识别的装置；
201-图像/数据处理装置；
30-细胞形貌信息获取装置；
40-细胞限定装置。
具体实施方式
为详细说明技术方案的技术内容、构造特征、所实现目的及效果，以下结合具体实施例并配合附图详予说明。
第一实施例
一种细胞识别的方法(仅细胞牵引力大小，无标签)，包括如下步骤：
S1、获取细胞信息，所述细胞信息为基于细胞力学传感器获取的细胞中某点的细胞牵引力的大小，具体为：使用细胞力学传感器对多个细胞进行细胞信息采集，其中包括对各个细胞的多点进行细胞牵引力大小信息采集，从而获取多个细胞中的多点细胞牵引力大小数据；
S2、对获取的细胞牵引力大小信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息，此时结构化细胞信息可以视为一个M N×P的二维特征矩阵(Feature matrix)，其中N为细胞数目，P为细胞特征数目，此处P＝1，即细胞特征为细胞牵引力大小；
S3、以上述的结构化细胞信息作为输入数据，利用无监督机器学习建立细胞特征模型，并将所述细胞特征模型应用于未知类型或未知状态的细胞的聚类。
在另外一些与本实施例类似的实施方式里，可以以如下的方式进行优化或改进：对单个细胞而 言，对其获取的多点细胞牵引力大小信息做进一步的信息处理，例如计算出：单位面积细胞牵引力大小的平均值；细胞牵引力大小在细胞内的分布情况；等维度的信息，可以此作为新的细胞特征，添加入步骤S2中所述的二维特征矩阵，即扩充P的内容，然后经由后续的机器学习来获知哪种特征能够更好的将不同类型或状态的细胞区别开来。
第二实施例
一种细胞识别的方法(仅细胞牵引力大小，有标签)，包括如下步骤：
S1、获取若干种已知细胞类型或已知细胞状态的细胞的细胞信息，所述细胞信息为基于细胞力学传感器获取的细胞中某点的细胞牵引力的大小，具体包括：使用细胞力学传感器对上述若干种细胞进行信息采集，其中包括对各个细胞的多点进行细胞牵引力大小信息采集，从而获取多个细胞中的多点细胞牵引力大小数据；
S2、对获取的细胞牵引力大小信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息，此时结构化细胞信息可以视为一个M N×P特征矩阵(Feature matrix)的二维矩阵，其中N为细胞数目，P为细胞特征数目，此处P＝1，即细胞特征为：细胞牵引力大小；
S3、以上述的结构化细胞信息作为输入数据，利用有监督机器学习建立细胞特征模型并利用大量细胞的结构化细胞信息对所述细胞特征模型进行训练，例如，本实施例采用随机森林(Random Forest，RF)算法进行显著特征的提取以及估计模型参数，然后应用于新的细胞，估计新细胞对应的标签，即将所述细胞特征模型应用于未知类型或未知状态的细胞的分类(即识别，本发明所述的“识别”应理解为广义的识别，既包括“分类”，即判定细胞的类型或细胞的状态；也包括“聚类”，即虽然不知具体的细胞状态或类型，但将可能具有相同或相似性质的、可能为相同或相似类型或状态的细胞进行聚类)。而其他实施方式中，还可能采取支持向量机(Support Vector Machine，SVM)或深度学习等机器学习的算法/思路完成相应的模型建立和训练学习工作。
在另外一些与本实施例类似的实施方式里，可以以如下的方式进行优化或改进：对单个细胞而言，对其获取的多点细胞牵引力大小信息做进一步的信息处理，例如计算出：单位面积细胞牵引力大小的平均值；细胞牵引力大小在细胞内的分布情况；等维度的信息，可以此作为新的细胞特征，添加入步骤S2中所述的二维特征矩阵，即扩充P的内容，然后经由后续的机器学习来获知哪种特征能够更好的将不同类型或状态的细胞区别开来。
第三实施例
一种细胞识别的方法(仅细胞牵引力大小，部分数据有标签、部分数据无标签)，包括如下步骤：
S1、获取多个细胞的细胞信息，其中部分细胞为若干种已知细胞类型或已知细胞状态的细胞，其余细胞为未知细胞类型或未知细胞状态的细胞；所述细胞信息为基于细胞力学传感器获取的细胞中某点的细胞牵引力的大小。本步骤具体包括：使用细胞力学传感器对上述若干种细胞进行信息采集，其中包括对各个细胞的多点进行细胞牵引力大小信息采集，从而获取多个细胞中的多点细胞牵引力大小数据；
S2、对获取的细胞牵引力大小信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息，此时结构化细胞信息可以视为一个M N×P特征矩阵(Feature matrix)的二维矩阵，其中N为细胞数目，P为细胞特征数目，此处P＝1，即细胞特征为：细胞牵引力大小；
S3、以上述的结构化细胞信息作为输入数据，利用半监督机器学习建立细胞特征模型并利用大量细胞的结构化细胞信息(同时包括有标签和无标签的细胞)对所述细胞特征模型进行训练，然后将所述细胞特征模型应用于未知类型或未知状态的细胞的分类/聚类。
在另外一些与本实施例类似的实施方式里，可以以如下的方式进行优化或改进：对单个细胞而 言，对其获取的多点细胞牵引力大小信息做进一步的信息处理，例如计算出：单位面积细胞牵引力大小的平均值；细胞牵引力大小在细胞内的分布情况；等维度的信息，可以此作为新的细胞特征，添加入步骤S2中所述的二维特征矩阵，即扩充P的内容，然后经由后续的机器学习来获知哪种特征能够更好的将不同类型或状态的细胞区别开来。
第四实施例
一种细胞识别的方法(细胞牵引力的大小和方向，无标签)，包括如下步骤：
S1、获取细胞信息，所述细胞信息为基于细胞力学传感器获取的细胞中某点的细胞牵引力的大小和方向，具体包括：使用细胞力学传感器(本实施例中为纳米微柱传感器)对多个细胞进行细胞信息采集，其中包括对各个细胞的多点进行细胞牵引力大小和方向的信息采集，从而获取多个细胞中的多点细胞牵引力大小及方向数据；
S2、对获取的细胞牵引力大小和方向的信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息，此时结构化细胞信息可以视为一个M N×P特征矩阵(Feature matrix)的二维矩阵，其中N为细胞数目，P为细胞特征数目，此处P＝2，即细胞特征为：细胞牵引力大小、细胞牵引力方向；
S3、以上述的结构化细胞信息作为输入数据，利用无监督机器学习建立细胞特征模型，并将所述细胞特征模型应用于未知类型或未知状态的细胞的聚类。
具体地，本实施例中的S2步骤对细胞信息做大致如下的处理：假设在一个细胞内共取到n个点的细胞牵引力向量数据(大小和方向)，这些点位为：(i，i∈{1，2，...n)，分别对应两个二维坐标
Figure PCTCN2022121340-appb-000001
以及
Figure PCTCN2022121340-appb-000002
其中t 0和t n分别对应纳米微柱初始以及位移后的点位。这样一来，
Figure PCTCN2022121340-appb-000003
即为每个坐标点上力的方向。此外，每个坐标点位上还应有一个标量信息，即力的大小d。如此即可通过已有数据估计出细胞轴方向以及中心点坐标、并以此为基准将每个细胞整理为一个长度相同的向量。例如，基于每个细胞中的各点位，中心点可以被计算为：
Figure PCTCN2022121340-appb-000004
细胞轴的计算方式为寻找距离最远的两点(x 1，y 1)以及(x 2，y 2)，通过如下公式来得到细胞轴：
Figure PCTCN2022121340-appb-000005
α＝y 2-βx 2，y＝α+βx。
请参见图1，图1为本发明第四实施例中对某点位位移信息标量化处理的示意图，图中每个点代表一个点位，各点位颜色深度由浅至深表示力由小到大。在获得每个细胞的细胞轴之后，可进一步对每个点位量化处理：计算每个点的位移矢量与细胞轴夹角为θ。每个点位可以继而与力的大小(标量)相结合，如将力的大小d视为权重，从而将各点位处理为一个标量s i＝θ i*d i。如此 一来，每个细胞的细胞信息都可以被整理为一个向量z＝(s 1，...s n)。
在另外一些与本实施例类似的实施方式里，可以以如下的方式进行优化或改进：对单个细胞而言，对其获取的多点细胞牵引力大小或细胞牵引力方向信息做进一步的信息处理，例如计算出：单位面积细胞牵引力大小的平均值；细胞牵引力大小在细胞内的分布情况；细胞牵引力向量在细胞内的分布情况；等维度的信息，可以此作为新的细胞特征，添加入步骤S2中所述的二维特征矩阵，即扩充P的内容，然后经由后续的机器学习来获知哪种特征能够更好的将不同类型或状态的细胞区别开来。
第五实施例
一种细胞识别的方法(细胞牵引力的大小和方向，有标签)，包括如下步骤：
S1、获取若干种已知细胞类型或已知细胞状态的细胞的细胞信息，所述细胞信息为基于细胞力学传感器获取的细胞中某点的细胞牵引力的大小和方向，具体包括：使用细胞力学传感器对上述若干种细胞进行信息采集，其中包括对各个细胞的多点进行细胞牵引力大小信息采集，从而获取多个细胞中的多点细胞牵引力大小数据；
S2、对获取的细胞牵引力大小信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息，此时结构化细胞信息可以视为一个M N×P特征矩阵(Feature matrix)的二维矩阵，其中N为细胞数目，P为细胞特征数目，此处P＝2，即细胞特征为：细胞牵引力大小、细胞牵引力方向；
S3、以上述的结构化细胞信息作为输入数据，利用有监督机器学习建立细胞特征模型并利用大量细胞的结构化细胞信息对所述细胞特征模型进行训练，然后将所述细胞特征模型应用于未知类型或未知状态的细胞的分类。例如，本实施例采用随机森林(Random Forest，RF)算法进行显著特征的提取以及估计模型参数，然后应用于新的细胞，估计新细胞对应的标签，即将所述细胞特征模型应用于未知类型或未知状态的细胞的分类。而其他实施方式中，还可能采取支持向量机(Support Vector Machine，SVM)或深度学习等机器学习的算法/思路完成相应的模型建立和训练学习工作。
请参见图2和图3，图2和图3分别为本发明第五实施例扩展实施方式中所述将建立的细胞特征模型用于未知细胞或未知细胞表现型的识别的结果图A以及结果图B，图A中不同行代表不同细胞类型，不同列代表不同样本，图中黑点为采用随机森林所算法提取的前50个显著特征，或者可以称为显著点位，此处的点位是指一个细胞中的某个位置，从不同的位置获取的细胞牵引力信息不同。而图B则展示了利用前50显著特征(显著点位)对三种不同的细胞类型的显著区分效果。基于有标签数据学习到的显著特征可对数据降维可视化。后续亦可经由聚类算法，基于降维数据，对细胞进行分类和识别。
在另外一些与本实施例类似的实施方式里，可以以如下的方式进行优化或改进：对单个细胞而言，对其获取的多点细胞牵引力大小或细胞牵引力方向信息做进一步的信息处理，例如计算出：单位面积细胞牵引力大小的平均值；细胞牵引力大小在细胞内的分布情况；细胞牵引力向量在细胞内的分布情况等维度的信息，可以此作为新的细胞特征，添加入步骤S2中所述的二维特征矩阵，即扩充P的内容，然后经由后续的机器学习来获知哪种特征能够更好的将不同类型或状态的细胞区别开来。
第六实施例
一种细胞识别的方法(细胞牵引力大小和方向，部分数据有标签、部分数据无标签)，包括如下步骤：
S1、获取多个细胞的细胞信息，其中部分细胞为若干种已知细胞类型或已知细胞状态的细胞，其余细胞为未知细胞类型或未知细胞状态的细胞；所述细胞信息为基于细胞力学传感器获取的细胞 中某点的细胞牵引力的大小。具体包括：使用细胞力学传感器对上述若干种细胞进行信息采集，其中包括对各个细胞的多点进行细胞牵引力大小信息采集，从而获取多个细胞中的多点细胞牵引力大小数据；
S2、对获取的细胞牵引力大小信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息，此时结构化细胞信息可以视为一个M N×P特征矩阵(Feature matrix)的二维矩阵，其中N为细胞数目，P为细胞特征数目，此处P＝2，即细胞特征为：细胞牵引力大小、细胞牵引力方向；
S3、以上述的结构化细胞信息作为输入数据，利用半监督机器学习建立细胞特征模型并利用大量细胞的结构化细胞信息对所述细胞特征模型进行训练，然后将所述细胞特征模型应用于未知类型或未知状态的细胞的分类。
在另外一些与本实施例类似的实施方式里，可以以如下的方式进行优化或改进：对单个细胞而言，对其获取的多点细胞牵引力大小或细胞牵引力方向信息做进一步的信息处理，例如计算出：单位面积细胞牵引力大小的平均值；细胞牵引力大小在细胞内的分布情况；细胞牵引力向量在细胞内的分布情况；等维度的信息，可以此作为新的细胞特征，添加入步骤S2中所述的二维特征矩阵，即扩充P的内容，然后经由后续的机器学习来获知哪种特征能够更好的将不同类型或状态的细胞区别开来。
第七实施例
一种细胞识别的方法(细胞牵引力向量的瞬间值、细胞牵引力向量在一定时间间隔内的变化情况，无标签)，包括如下步骤：
S1、获取细胞信息，所述细胞信息为基于细胞力学传感器获取的细胞中某点的细胞牵引力的向量的瞬时值和该点细胞牵引力向量在一定时间间隔内的变化情况；具体包括：使用细胞力学传感器对多个细胞进行细胞信息采集，其中包括对各个细胞的多点进行细胞牵引力大小和方向的信息采集，从而获取多个细胞中的多点细胞牵引力大小及方向数据；
S2、对获取的细胞牵引力大小和方向的信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息，此时结构化细胞信息可以视为一个M N×P特征矩阵(Feature matrix)的二维矩阵，其中N为细胞数目，P为细胞特征数目。
S3、以上述的结构化细胞信息作为输入数据，利用无监督机器学习建立细胞特征模型，并将所述细胞特征模型应用于未知类型或未知状态的细胞的聚类。
本实施方式中，由于获取的不仅是细胞内某个点位的细胞牵引力向量的瞬时值，还包括了其在一定时间间隔内的间隔情况，因此所获取的单个细胞的力学数据实际上可以类比为图像(瞬时值)或者视频(时间维度内的变化情况)，如此一来，如果将当前细胞覆盖下的点阵类比为图像中的像素点，而将各点位所记录的信息(力的大小、方向等多个细胞特征)可以类比为像素点所对应的色彩，就可以在后续机器学习中借鉴图像和视频数据处理领域的机器学习算法，例如的采取广泛应用在图像识别当中的机器学习，更确切点说，深度学习(deep learning)中的卷积神经网络(Convolutional Neural Network，CNN)来对数据建模分析。
第八实施例
一种细胞识别的方法(细胞牵引力向量的瞬间值、细胞牵引力向量在一定时间间隔内的变化情况，有标签)，包括如下步骤：
S1、获取若干种已知细胞类型或已知细胞状态的细胞的细胞信息，所述细胞信息为基于细胞力学传感器获取的细胞中某点的细胞牵引力的向量的瞬时值和该点细胞牵引力向量在一定时间间隔内的变化情况；具体包括：使用细胞力学传感器对多个细胞进行细胞信息采集，其中包括对各个细胞的多点进行细胞牵引力大小和方向的信息采集，从而获取多个细胞中的多点细胞牵引力大小及方向数据；
S2、对获取的细胞牵引力大小和方向的信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息，此时结构化细胞信息可以视为一个M N×P特征矩阵(Feature matrix)的二维矩阵，其中N为细胞数目，P为细胞特征数目。
S3、以上述的结构化细胞信息作为输入数据，利用有监督机器学习建立细胞特征模型并利用大量细胞的结构化细胞信息对所述细胞特征模型进行训练，然后将所述细胞特征模型应用于未知类型或未知状态的细胞的分类(即识别，本发明所述的“识别”应理解为广义的识别，既包括“分类”，即判定细胞的类型或细胞的状态；也包括“聚类”，即虽然不知具体的细胞状态或类型，但将可能具有相同或相似性质的、可能为相同或相似类型或状态的细胞进行聚类)。例如，本实施例采用随机森林(Random Forest，RF)算法进行显著特征的提取以及估计模型参数，然后应用于新的细胞，估计新细胞对应的标签，即将所述细胞特征模型应用于未知类型或未知状态的细胞的分类。而其他实施方式中，还可能采取支持向量机(Support Vector Machine，SVM)或深度学习等机器学习的算法/思路完成相应的模型建立和训练学习工作。
本实施方式中，由于获取的不仅是细胞内某个点位的细胞牵引力向量的瞬时值，还包括了其在一定时间间隔内的间隔情况，因此所获取的单个细胞的力学数据实际上可以类比为图像(瞬时值)或者视频(一定时间维度内的多个瞬时值)，如此一来，如果将当前细胞覆盖下的点阵类比为图像中的像素点，而将各点位所记录的信息(力的大小、方向等多种细胞特征)可以类比为像素点所对应的色彩，就可以在后续机器学习中借鉴图像和视频数据处理领域的机器学习算法，例如的采取广泛应用在图像识别当中的机器学习，更确切点说，深度学习(deep learning)中的卷积神经网络(Convolutional Neural Network，CNN)来对数据建模分析。
第九实施例
一种细胞识别的方法(细胞牵引力向量的瞬间值、细胞牵引力向量在一定时间间隔内的变化情况，部分数据有标签，部分数据无标签)，包括如下步骤：
S1、获取若干种已知细胞类型或已知细胞状态的细胞的细胞信息，所述细胞信息为基于细胞力学传感器获取的细胞中某点的细胞牵引力的向量的瞬时值和该点细胞牵引力向量在一定时间间隔内的变化情况；具体包括：使用细胞力学传感器对多个细胞进行细胞信息采集，其中包括对各个细胞的多点进行细胞牵引力大小和方向、并在一定时间范围内持续进行该信息采集，从而获取多个细胞中的多点细胞牵引力大小及方向数据，以及所述多点的细胞牵引力向量在一定时间间隔内的变化信息；
S2、对获取的细胞牵引力大小和方向的信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息，此时结构化细胞信息可以视为一个M N×P特征矩阵(Feature matrix)的二维矩阵，其中N为细胞数目，P为细胞特征数目。
S3、以上述的结构化细胞信息作为输入数据，利用半监督机器学习建立细胞特征模型并利用大量细胞的结构化细胞信息对所述细胞特征模型进行训练，然后将所述细胞特征模型应用于未知类型或未知状态的细胞的分类/聚类。
本实施方式中，由于获取的不仅是细胞内某个点位的细胞牵引力向量的瞬时值，还包括了其在一定时间间隔内的间隔情况，因此所获取的单个细胞的力学数据实际上可以类比为图像(瞬时值)或者视频(一定时间维度内的多个瞬时值)，如此一来，如果将当前细胞覆盖下的点阵类比为图像中的像素点，而将各点位所记录的信息(力的大小、方向等多种细胞特征)可以类比为像素点所对应的色彩，就可以在后续机器学习中借鉴图像和视频数据处理领域的机器学习算法，例如的采取广泛应用在图像识别当中的机器学习，更确切点说，深度学习(deep learning)中的卷积神经网络(Convolutional Neural Network，CNN)来对数据建模分析。
第十实施例
一种细胞识别的方法，其与第一至第九实施例的区别在于，所述细胞信息还包括细胞形貌信息， 所述细胞形貌信息由显微照相机或显微摄像机获取(当需要对一定时间间隔内的信息进行连续采集时)。此时，数据预处理步骤中的细胞特征还包括细胞形貌信息或经由对细胞形貌信息做进一步处理或分析所得出的进一步信息，包括：细胞尺寸、细胞形状、细胞核尺寸、细胞核形状、细胞颜色中的一种或若干种。
第十一实施例
一种细胞识别的方法，其与第一至第十实施例的区别在于，所述细胞信息是在对细胞进行限制的前提下采集的。对细胞进行形态限制的方法包括物理方法，如在若干个细胞力学传感器周围建立起可包绕一定区域(面积)的限制墙，这种限制墙的高度高于其内的细胞力学传感器的高度，故可以视为一种细胞限定装置，由于其设立了一定的空间区域，所以只有相应大小的细胞可以落入其中与所述细胞力学传感器接触，也可以认为只有相应个数(大部分情况下，该限制墙可容纳1个细胞落入其中)。在特殊情况下，该限制墙还可以通过挤压作用让进入其中的细胞在一定程度上适应为限制墙所包围的横截面的形状，从而达到一定的细胞形态限定效果。
与不对细胞进行限制之下进行本发明所述的细胞识别的方案相比，对细胞进行限制(个数限制或形态限制)后，首先能够减少细胞间的接触，使绝大多数细胞都处于单细胞状态；其次能够减少细胞大小、形状这一维度，即对细胞特征数目进行了降维工作，能够一定程度上降低分析过程的难度；再者，利用特定的细胞形状也能控制细胞的分化状态，例如免疫细胞可被形态限定在M1状态，在特定的场景下运用能够得到更有价值的结果。
当然，若不对细胞进行形态限制，则将使细胞处在自然状态，能够减少外来干预对细胞的影响；并且，也必须在不加形态限制的时候才能对细胞进行边缘检测从而得到细胞形态，向性等特征，从而协助对细胞的辨识与预测。简言之，本实施方式所述的细胞限制方案在特定技术场景下运用有其独到优势。
第十二实施例
请参阅图4，本实施例提供了一种细胞识别的装置，包括信息获取单元1、预处理单元2、学习单元3和识别单元4；
所述信息获取单元1用于获取细胞信息，所述细胞信息包括基于细胞力学传感器获取的细胞中某点的细胞牵引力信息，所述细胞牵引力信息包括该点细胞牵引力的大小；
所述预处理单元2用于对细胞信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息；
所述学习单元3用于以结构化细胞信息作为输入数据，利用有监督、无监督或半监督的机器学习建立细胞特征模型；
所述识别单元4用于将所述细胞特征模型应用于未知类型或未知状态的细胞的分类或聚类。
本实施方式所述的细胞识别装置可用于实现如第一至第三实施例所述的细胞识别方法技术方案。
第十三实施例
一种细胞识别的装置，其与第十二实施例不同之处在于，所述信息获取单元1所获取的细胞牵引力信息还包括该点细胞牵引力的方向。本实施方式所述的细胞识别装置可用于实现如第四至第六实施例所述的细胞识别方法技术方案。
第十四实施例
一种细胞识别的装置，其与第十二、第十三实施例不同之处在于，所述信息获取单元1所获取的细胞牵引力信息还包括该点细胞牵引力的大小或方向在一定时间间隔内的变化。本实施方式所述的细胞识别装置可用于实现如第七至第九实施例所述的细胞识别方法技术方案。
第十五实施例
一种细胞识别的装置，其与第十二至第十四实施例不同之处在于，所述信息获取单元1所获取的细胞牵引力信息还包括细胞形貌信息。本实施方式所述的细胞识别装置可用于实现如第十实施例所述的细胞识别方法技术方案。
第十六实施例
一种细胞识别的装置，其与第十二至第十五实施例不同之处在于，所述细胞信息是在对细胞进行细胞限定操作下获取的。对细胞进行形态限制的方法包括物理方法，如在若干个细胞力学传感器周围建立起可包绕一定区域(面积)的限制墙，这种限制墙的高度高于其内的细胞力学传感器的高度，故可以视为一种细胞限定装置，由于其设立了一定的空间区域，所以只有相应大小的细胞可以落入其中与所述细胞力学传感器接触，也可以认为只有相应个数(大部分情况下，该限制墙可容纳1个细胞落入其中)。在特殊情况下，该限制墙还可以通过挤压作用让进入其中的细胞在一定程度上适应为限制墙所包围的横截面的形状，从而达到一定的细胞形态限定效果。
与不对细胞进行限制之下进行本发明所述的细胞识别的方案相比，对细胞进行限制(个数限制或形态限制)后，首先能够减少细胞间的接触，使绝大多数细胞都处于单细胞状态；其次能够减少细胞大小、形状这一维度，即对细胞特征数目进行了降维工作，能够一定程度上降低分析过程的难度；再者，利用特定的细胞形状也能控制细胞的分化状态，例如免疫细胞可被形态限定在M1状态，在特定的场景下运用能够得到更有价值的结果。
当然，若不对细胞进行形态限制，则将使细胞处在自然状态，能够减少外来干预对细胞的影响；并且，也必须在不加形态限制的时候才能对细胞进行边缘检测从而得到细胞形态，向性等特征，从而协助对细胞的辨识与预测。简言之，本实施方式所述的细胞限制方案在特定技术场景下运用有其独到优势。
与不对细胞进行限制之下进行本发明所述的细胞识别的方案相比，对细胞进行限制(个数限制或形态限制)后，首先能够减少细胞间的接触，使绝大多数细胞都处于单细胞状态；其次能够减少细胞大小、形状这一维度，即对细胞特征数目进行了降维工作，能够一定程度上降低分析过程的难度；再者，利用特定的细胞形状也能控制细胞的分化状态，例如免疫细胞可被形态限定在M1状态，在特定的场景下运用能够得到更有价值的结果。
当然，若不对细胞进行限制，则将使细胞处在自然状态，能够减少外来干预对细胞的影响；并且，也必须在不加形态限制的时候才能对细胞进行边缘检测从而得到细胞形态，向性等特征，从而协助对细胞的辨识与预测。简言之，本实施方式所述的细胞限制方案在特定技术场景下运用有其独到优势。
本实施方式所述的细胞识别装置可用于实现如第十一实施例所述的细胞识别方法技术方案。
第十七实施例
一种细胞识别的系统，包括细胞力学传感器10和细胞识别的装置20。所述细胞识别的装置为上述第十二至第十六实施例里描述的细胞识别装置，用于实现上述第一至第九实施例所描述的细胞识别方法所述的技术方案。
本实施方式中的细胞力学传感器10为纳米微柱阵列，其他实施方式中还可以采取任意能够获取细胞牵引力信息的设备作为细胞力学传感器。
第十八实施例
请参阅图5，本实施例提供了一种细胞识别的系统，其与第十七实施例不同之处在于，还包括细胞形貌信息获取装置30，用于获取细胞形貌信息。本实施例所述的细胞形貌信息获取装置30为显微照相机，其他实施方式中，所述细胞形貌信息获取装置30还可以是显微摄像机或其他能够获取细胞形貌信息的设备。本实施例所述的细胞识别系统可用于实现上述第十实施例所描述的细胞识别方法所述的技术方案。
第十九实施例
一种细胞识别的系统，其与第十七、第十八实施例不同之处在于，还包括细胞限定装置40，用于对细胞进行细胞限定操作。本实施方式中细胞限定装置40具体为如下结构：在若干个细胞力学传感器周围建立起可包绕一定区域(面积)的限制墙，这种限制墙的高度高于其内的细胞力学传感器的高度，故可以视为一种细胞限定装置，由于其设立了一定的空间区域，所以只有相应大小的细胞可以落入其中与所述细胞力学传感器接触，也可以认为只有相应个数(大部分情况下，该限制 墙可容纳1个细胞落入其中)。在特殊情况下，该限制墙还可以通过挤压作用让进入其中的细胞在一定程度上适应为限制墙所包围的横截面的形状，从而达到一定的细胞形态限定效果。本实施例所述的细胞识别系统可用于实现上述第十一实施例所描述的细胞识别方法所述的技术方案。
第二十实施例
一种细胞识别的系统，其与第十七、第十八、第十九实施例不同之处在于，本实施方式中的细胞力学传感器10为微柱上设有光线反射层的细胞牵引力检测装置，其他实施方式中还可以采取任意能够获取细胞牵引力信息的设备作为细胞力学传感器。
具体的，请参阅图6，为细胞牵引力检测装置的结构示意图。如图6所示，本实施例中的细胞牵引力检测装置包括透光的基座101以及设置在该基座101上的、可受细胞牵引力作用而产生形变的微柱103，微柱103的顶部涂覆有光线反射层1031，光线反射层1031的厚度为5nm(在其他一些实施方式中，光线反射层1031的厚度可以在5nm-20nm之间——涂层的厚度和涂层材料有关，在应用同种涂层材料的前提下，涂层厚度的选择应以保证透光效果、微柱柱体稳定性、保证与微柱柱体的连接不脱落为限)。微柱103的柱体可以透射光线，图中方向相反的箭头簇表示入射光线和反射光线。(注意：本实施例中用了“涂层”一词，仅表示本实施例中的光线反射层1031可以是涂覆工艺制备的，并不限定光线反射层1031一定是涂覆工艺制备的)
当本实施方式所述的细胞牵引力的检测装置1在投入使用时，微柱103的数量将不止一个。请参阅图7，为本发明第二十一实施例相关的细胞识别系统的结构示意图。图7展示的系统除了涉及本实施例阐述的细胞牵引力检测装置1、细胞识别的装置(包括信息获取单元、预处理单元、学习单元、识别单元)之外，还涉及：在基座101的下方设置的具有光源的光信号发生装置102，所述光源发出的光线通过入射光路从细胞牵引力检测装置10的透光的基座101照射到微柱103的光线反射层；所述信息获取单元1用于检测从微柱103顶部的光线反射层1031反射后的光线，所述光线反射层1031反射的光线经过反射光路并经过分光器104作用之后进入细胞识别的装置20。细胞识别的装置20的信息获取单元1(光信号检测装置)在获得反射光信号后，由预处理单元2对信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息，此时结构化细胞信息可以视为一个的二维特征矩阵(Feature matrix)，其中N为细胞数目，P为细胞特征数目，此处P＝1或2，即细胞特征为细胞牵引力大小和/或细胞牵引力在细胞内的分布；接着，以上述的结构化细胞信息作为输入数据，利用有监督、无监督或半监督的学习单元3学习建立细胞特征模型，并利用大量细胞的结构化细胞信息对所述细胞特征模型进行训练；最后，通过识别单元4将所述细胞特征模型应用于未知类型或未知状态的细胞的分类或聚类。当微柱12未受力的情况下微柱理应保持直立状态，从而能最大限度地反射探测光；而当微柱103与细胞相接触时，在细胞牵引力的作用下，微柱103发生弯曲，导致光反射水平降低。所以，当细胞牵引力越大时，所得到的光反射信号就应越小，这样藉由对光反射信号强度的观测即可轻易反推该点细胞牵引力的大小。
在本发明的一些实施例中，分光器104可以为半透半反或其它等效的光学元件，主要目的是简化光路的设计。
在本发明的一些实施例中，光信号发生装置102可以是LED，卤素灯，激光(例如，红外激光)，或其他光源，或其他具有这些光源的装置，本发明不作具体限制。
在本发明的一些实施例中，细胞识别的装置中的信息获取单元1可以为显微镜，电荷耦合元件CCD，互补金属氧化物半导体CMOS，光电倍增管PMT及光电转换器PT，胶片，或者其它具有相同功能的光信号检测元件，本发明不作具体限制。
在本发明的一些实施例中，细胞识别的装置中的预处理单元、学习单元、识别单元为集成了这些单元的图像/数据处理装置201，例如其可以是光学图像分析软件ImageJ，Matlab，Fluoview，Python，或者其它具有相同功能的光学图像/数据分析元件，或者这些分析软件的联合使用，本发明不作具体限制。
下面对本实施例相关的一种细胞识别的系统的牵引力检测、细胞识别分析过程作具体介绍。
请参阅图8a-图8c，图8a为细胞识别的系统的结构示意图；图8b为信息获取单元获取得到的细胞牵引力检测装置光反射讯号的图像；图8c为力学大小及分布的可视化效果图。
如图8a所示，在细胞牵引力检测装置上，每根微柱的顶部均设有金属反射层，侧面则设有抗反射层；在无细胞时，光线由下方照射微柱，会完全反射而被细胞识别的装置中的信息获取单元(例如CCD相机)完全接收；但当细胞贴附在微柱上时，细胞移动产生的细胞力会使得微柱发生倾斜，从而降低了反射讯号，经过光反射讯号分析后，即可计算出细胞力强度。
进一步地，通过信息获取单元(例如CCD相机)采集细胞牵引力检测装置光反射信号的图像以及局部细胞黏附区域放大图像(如图8b所示)。接着，通过预处理单元对图8b的图像进行进一步的处理，使其转化为更为直观的力学大小及分布的可视化效果图8c。
具体处理过程如下：首先，基于图8b，获得明视野反射信号图(I，对焦于细胞上)；接着，通过对图像进行傅立叶变换后过滤高频信号，并进行逆傅立叶变换操作，从而计算获得到未偏移情况下的微柱反射讯号图像(I 0)；接着，将I和I 0图像进一步处理，以将反射讯号图转为更为直观的细胞力学图(I 0讯号值减去I讯号值)后标准化获得更为直观的细胞牵引力强度图j。
第二十一实施例
本实施例与第二十实施例不同之处在于，本实施例中的细胞限定装置40为硅薄膜。
具体的，请参照图9a-图9c，图9a为采用硅薄膜作为细胞限定装置的细胞牵引力检测装置的实物图，图9a中，硅薄膜在使用激光打孔后，黏附在基座上，每个孔中均设有微柱，通过硅薄膜限制细胞的形态和迁移，同时控制细胞与细胞之间存在接触或黏连；图9b为在光反射下采用硅薄膜作为细胞限制机构的细胞牵引力检测装置的荧光显微镜图，图9c为图9b的放大图。在一些实施例中，硅薄膜的每个孔的大小可以设置为与单细胞大小相适配，适合单一细胞贴附，从而限制细胞接触、细胞形态及其迁移范围。
第二十二实施例
本实施例具体提供第二十实施例中所述的一种细胞识别的系统获取得到的细胞牵引力信息，应用于识别细胞的方法。
请参照图10a至图10g，图10a为健康细胞和肺非小细胞癌细胞混合体系的荧光成像图；图10b为信息获取单元获取得到的细胞牵引力检测装置光反射讯号分布图；图10c为力学大小及分布的可视化效果图；图10d为图10c中的健康细胞和肺非小细胞癌细胞的代表性单细胞细胞力分布的放大图；图10e为健康细胞和肺非小细胞癌细胞在细胞形态上的对比图；图10f为健康细胞、肺非小细胞癌细胞以及这两种细胞以不同比例混合后的反射信号强弱的对比图；图10g为将图10c经过结构化处理后，基于结构化细胞信息处理得到的聚类分析图。
具体的，本实施例以健康细胞(Normal)和肺非小细胞癌细胞系(Cancer)作为检测对象，使用两种不同荧光染料(Dil&DIO)对健康细胞和肺癌细胞的细胞膜预染色，并以一定比例混合后添加到同一个细胞牵引力检测装置(在另一些实施方式中，可以是添加到不同的独立的细胞牵引力检测装置)上。
进一步地，通过信息获取单元(本实施例使用显微镜)采集了细胞牵引力检测装置光反射讯号的图像(如图10b所示)，而两种细胞内高分辨率的力场分布被信息获取单元直接渲染即可转化成可读取的光强度衰减信号(反映细胞力强度)并显示在图片内(如图10c所示)。根据图10c图片显示的两种细胞的光衰减程度的不同，可以通过肉眼观察对两种细胞进行直观的区分(定性分析)。
进一步地，通过光信号分析装置对图10c中的光反射讯号进行进一步处理。具体地，本实施例使用ImageJ及Python分析软件(在其它实施例中还可以使用其他图像/数据分析软件)对所获得的细胞力场的图10c进行信息采集，其中包括对各个细胞的多点进行细胞牵引力大小信息采集，从而获取多个细胞中的多点细胞牵引力大小数据；对获取的细胞牵引力大小信息做预处理，形成结构化细胞信息；并根据结构化细胞信息分析得到健康细胞和肺非小细胞癌细胞在细胞形态上的对比结果图(如图10e所示)。
所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息(例如本实施例中 的细胞黏附面积及细胞圆度)，此时结构化细胞信息可以视为一个特征矩阵(Feature Matrix)的二维矩阵，其中N为细胞数目，P为细胞特征数目，此处P＝2，即细胞特征为：细胞牵引力大小、细胞牵引力在细胞内的分布。
进一步地，以上述的结构化细胞信息作为输入数据，利用有监督的图像/数据处理装置的学习单元(与不同的细胞膜染料(Dil&DIO)对两种细胞系进行预染色进行对比)学习建立细胞特征模型，并利用大量细胞的结构化细胞信息对所述细胞特征模型进行训练，得到如图10g所示的聚类分析图，然后将所获得的细胞特征模型应用于未知类型或未知状态的细胞的分类识别。由此可知，根据结构化的细胞特征数据(细胞牵引力大小、细胞牵引力在细胞内的分布)作为输入数据，可以利用图像/数据处理装置的识别单元对正常的健康细胞及癌症细胞进行聚类分型，从而实现对未知细胞种类的识别。
图10e示出了不同细胞在形态(包括细胞黏附面积及细胞圆度)上没有统计学意义上的明显区别；图10f示出了正常细胞及肿瘤细胞反射讯号强度(反映细胞力)的明显差异，以及把正常细胞及肿瘤细胞按一定比例混合在一起后反射讯号强度和混合比例呈一定线性关系。由此可知，相比于细胞的其它特征(例如图10e中的细胞黏附面积、细胞圆度等形态信息)，基于本发明的细胞牵引力检测装置测得的细胞力学特征可对细胞的状态及类型进行更直观精确的识别(定量和定性分析)。
此外，从图10d和图10f的数据得出，肿瘤细胞比正常细胞显示出更高的牵引力大小，并且分布更不均匀。可见，细胞牵引力以图像方式可视化后，即可通过肉眼直观地看出不同细胞的力场特征具有明显区别；并且进一步地，通过图像分析软件对不同细胞各点的力场大小进行结构化处理后，综合分析得到图10e的细胞形态信息，图10f的反射讯号强度(反映细胞力)以及图10g的聚类分析图。本发明通过细胞各点的力场结构化信息的综合分析可以对不同细胞(例如本实施例中健康细胞以及非小细胞肺癌细胞)进行聚类分型及定量分析，从而实现对细胞种类的精确识别。
综上可知，基于包括本实施例的细胞牵引力检测装置的细胞识别系统，不仅可以通过肉眼直观区分以进行定性分析，还可以基于测得的细胞力学特征对细胞的状态及类型进行更直观精确的识别(定量和定性分析)，并且证实了以细胞力场作为标志物能够更好地对细胞类型进行区分。
第二十三实施例
本实施例具体提供第二十实施例中所述的一种细胞识别的系统获取得到的细胞牵引力信息，应用于监测细胞的活力。
请参阅图11a-图11c，图11a为细胞活力检测方法的操作流程示意图；图11b为A549细胞在不同剂量的5FU处理24h后，MTT法测定得到的细胞活力以及基于本实施例的细胞识别系统获取得到的细胞牵引力的对比图；图11c为A549细胞在不同剂量的5FU处理不同时间后，MTT法测定得到的细胞活力以及本实施例的细胞识别系统获取得到的细胞牵引力的对比图。
具体地，本实施例将非小细胞肺癌细胞A549培养在多个细胞牵引力检测装置上，分别加入不同剂量的抑制细胞增殖的药物5-氟尿嘧啶(5-FU)进行处理，并通过第二十实施例中所述的一种细胞识别的系统监测不同时间点的细胞牵引力，通过CCK-8试剂盒监测不同时间点的细胞增殖和细胞毒性，同时以MTT测定法测定的细胞活力作为对照组，得到图11b和图11c的数据。
从图11b和图11c显示，通过传统MTT测定法和本实施例细胞识别系统进行测定后，MTT测定法测定的细胞活力以及细胞牵引力所反映的细胞活力都是以剂量依赖性呈逐渐降低的趋势，即细胞牵引力与细胞活力呈正相关。
此外，如图11b所示，在用不同剂量的5FU处理24h后，与对照组DMSO相比，通过牵引力可以以更大的降低幅度反映细胞活力的降低，从而更加直观地对细胞活力进行评估。如图11c所示，在用5FU处理12h内，MTT法测定的细胞活力变化不明显；而通过测定牵引力，可以在MTT法检测到细胞代谢活性降低之前的更早时间点即观测到细胞牵引力的降低，具体的在0.5μM处理剂量下即可在6h时出现明显的降低趋势，在1μM处理剂量下即可在3h时出现明显的降低趋势，从而可以更加灵敏地表征细胞活力的降低。
综上可知，本实施例通过包括本实施例的细胞牵引力检测装置的细胞识别系统直接检测细胞牵 引力是评估细胞对药物反应活力的一种高度敏感且有效的方法。
第二十四实施例
本实施例具体提供第二十实施例中所述的一种细胞识别的系统获取得到的细胞牵引力信息，并根据所述细胞牵引力信息分析确定细胞状态。
请参阅图12a至图12f，图12a为细胞状态检测方法的操作过程图；图12b为M0巨噬细胞分化至M1状态的荧光显微镜图；图12c为M0巨噬细胞分化至M2状态的荧光显微镜图；图12d为M0巨噬细胞、M1状态以及M2状态的细胞黏附面积对比图；图12e为M0巨噬细胞、M1状态以及M2状态的细胞圆度对比图；图12f为M0巨噬细胞、M1状态以及M2状态的牵引力对比图。
具体的，本实施例以巨噬细胞作为检测对象，将其分别添加到不同的独立的细胞牵引力检测装置的微柱上，并分别采用内毒素LPS和白介素IL4引导巨噬细胞从M0分化为M1和M2状态，以M0状态作为对照组。在细胞分化后，通过信息获取单元(本实施例采用显微镜)采集图12b和图12c，并通过图像/数据处理装置(ImageJ及Python)对图12b和图12c作进一步的图像处理和数据分析，将其转化为结构化信息，并分析制得图12d-图12f的数据。由图12a至图f的数据可知，不管是从直观的观测(图12b和图12c)还是经过结构的数据量化处理(图12d-图12f)，M0巨噬细胞以及分化后的M1和M2状态之间都具有明显的区别。
第二十五实施例
本实施例具体提供第二十实施例中所述的一种细胞识别的系统获取得到的细胞牵引力信息，根据所述细胞牵引力信息分析确定细胞状态。
具体地，本实施例提供的多细胞聚合体以多种方式结合在细胞牵引力检测装置上，本实施例提供其中的两种具体结合方式：
第一种结合方式：在细胞牵引力检测装置的微柱上设置培养基，将细胞移植到微柱上的培养基中培养得到多细胞聚合体；在另一些实施方式中，这种结合方式，在细胞牵引力信息以可视化形式输出下，可以实时监控细胞的培养过程，以应用于如培养基、药物等化学、生物和物理外界刺激对细胞生长的影响；
第二种结合方式：直接将培养好的多细胞聚合体粘附于细胞牵引力检测装置的微柱上，进行检测。
更具体地，本实施例提供一种肿瘤细胞多聚体在细胞牵引力检测装置上培养，应用于药物敏感试验，包括以下步骤：
S1、肿瘤细胞多聚体的生成：在细胞牵引力检测装置的微柱的顶部涂FN 50μg/mL，并紫外线杀菌30分钟；取乳腺癌细胞MCF-7(大约1×10 5～9×10 5个)种植于细胞牵引力的检测装置的微柱(无限定微柱的数量)的顶部表面，再将细胞牵引力监测装置浸置在3dGRO TM Spheroid Medium(S3077)培养液中培养3天以上，引导肿瘤细胞多聚体生成；
S2、将上述生成的肿瘤细胞多聚体应用于5-Fu药物敏感实验中：在上述肿瘤细胞多聚体加入200μM的5-Fu后培养1天；试验中，在加5-Fu前和加5-Fu后培养一天的细胞牵引力检测装置(生成有肿瘤细胞多聚体，可以连同培养液一起)，通过细胞识别的装置中的信息获取单元(CCD感光元件)获取光反射讯号，并使用光学图像分析软件(Image J)获得细胞力分布图像，如图13所示。
请参阅图13a-图13b，图13a为具有第一种形态的肿瘤细胞多聚体在有/无5-Fu作用后的表征图；图13b为具有第二种形态的肿瘤细胞多聚体在有/无5-Fu作用后的表征图；从左到右分别是细胞膜荧光和反射讯号的混合图像(1)、光反射讯号(2)、细胞核(3)、细胞膜(4)及经过光学图像分析软件(Image J)处理后的细胞力可视化图像(5)。由于细胞具有异质性，肿瘤细胞各不相同，多聚体也有各种各样的形态，因此细胞多聚体会以不同的形态黏附在一起。本实施例选取具有代表性的两种形态进行细胞形态检测，所述第一种形态系指较大的两个细胞黏在一起的细胞形态；所述第二种形态系指一群小细胞黏在一起的细胞形态。
由图13a-图13b可以看出，经抗肿瘤药物处理细胞使之活力降低后，反射讯号明显减弱；并且，从5-Fu作用前后的细胞机械力的改变可以看出两种不同形态的肿瘤细胞多聚体对药物的敏感 性也有明显差异。由此可知本发明的细胞牵引力检测装置能对多细胞聚合体(如肿瘤多聚体)进行细胞牵引力测定，并可通过细胞牵引力用于监测细胞多聚体的活力状态，并可区分不同的细胞形态。
在本发明的定义中，细胞多聚体指的是：
细胞是生物体基本的结构和功能单位，细胞通常会繁殖或分化形成两个以上细胞团聚一起形成细胞群体，即：多细胞聚合体；多细胞聚合体包括肿瘤多聚体等体外或体内培养所得到的细胞团体。
需要说明的是，尽管在本文中已经对上述各实施例进行了描述，但并非因此限制本发明的专利保护范围。因此，基于本发明的创新理念，对本文所述实施例进行的变更和修改，或利用本发明说明书及附图内容所作的等效结构或等效流程变换，直接或间接地将以上技术方案运用在其他相关的技术领域，均包括在本发明的专利保护范围之内。

Claims (15)

  1. 一种细胞识别的方法，其特征在于，包括如下步骤：
    获取细胞信息，所述细胞信息包括基于细胞力学传感器获取的细胞中某点的细胞牵引力信息，所述细胞牵引力信息包括该点细胞牵引力的大小；
    对细胞信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息；
    以结构化细胞信息作为输入数据，利用有监督、无监督或半监督的机器学习建立细胞特征模型，并将所述细胞特征模型应用于未知类型或未知状态的细胞的分类或聚类。
  2. 如权利要求1所述的细胞识别的方法，其特征在于，所述细胞牵引力信息还包括该点细胞牵引力的方向。
  3. 如权利要求1所述的细胞识别的方法，其特征在于，所述细胞牵引力信息还包括该点细胞牵引力的大小或方向在一定时间间隔内的变化。
  4. 如权利要求1所述的细胞识别的方法，其特征在于，所述细胞信息还包括细胞形貌信息。
  5. 如权利要求1所述的细胞识别的方法，其特征在于，所述细胞信息是在对细胞进行细胞限定操作下获取的。
  6. 一种细胞识别的装置，其特征在于，包括信息获取单元、预处理单元、学习单元和识别单元；
    所述信息获取单元用于获取细胞信息，所述细胞信息包括基于细胞力学传感器获取的细胞中某点的细胞牵引力信息，所述细胞牵引力信息包括该点细胞牵引力的大小；
    所述预处理单元用于对细胞信息做预处理，形成结构化细胞信息；所述结构化细胞信息包括细胞数目、细胞特征数目和各细胞特征的特征信息；
    所述学习单元用于以结构化细胞信息作为输入数据，利用有监督、无监督或半监督的机器学习建立细胞特征模型；
    所述识别单元用于将所述细胞特征模型应用于未知类型或未知状态的细胞的分类或聚类。
  7. 如权利要求6所述的细胞识别的装置，其特征在于，所述细胞牵引力信息还包括该点细胞牵引力的方向。
  8. 如权利要求6所述的细胞识别的装置，其特征在于，所述细胞牵引力信息还包括该点细胞牵引力的大小或方向在一定时间间隔内的变化。
  9. 如权利要求6所述的细胞识别的装置，其特征在于，所述细胞信息还包括细胞形貌信息。
  10. 如权利要求6所述的细胞识别的装置，其特征在于，所述细胞信息是在对细胞进行细胞限定操作下获取的。
  11. 一种细胞识别的系统，其特征在于，包括细胞力学传感器和如权利要求6-10中任一项所述的细胞识别装置。
  12. 如权利要求11所述的细胞识别的系统，其特征在于，还包括细胞形貌信息获取装置，用于获取细胞形貌信息。
  13. 如权利要求12所述的细胞识别的系统，其特征在于，所述细胞形貌信息获取装置包括显微照相机或显微摄像机。
  14. 如权利要求11所述的细胞识别的系统，其特征在于，还包括细胞限定装置，用于对细胞进行细胞限定操作。
  15. 一种细胞状态的检测方法，其特征在于，包括：通过如权利要求1-5中任意一项所述的细胞识别的方法或权利要求6-10中任意一项所述的细胞识别的装置或权利要求11-14中任意一项所述的细胞识别的系统获取细胞牵引力信息，根据所述细胞牵引力信息分析确定细胞状态；
    所述细胞状态包括：细胞黏附，细胞活力，细胞分化/活化，细胞增殖和/或细胞迁移。
PCT/CN2022/121340 2021-09-26 2022-09-26 一种细胞识别的方法、装置和系统 WO2023046167A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
CN202280064594.5A CN118076981A (zh) 2021-09-26 2022-09-26 一种细胞识别的方法、装置和系统
AU2022350581A AU2022350581A1 (en) 2021-09-26 2022-09-26 Method, device, and system for cell identification
EP22872210.4A EP4407571A1 (en) 2021-09-26 2022-09-26 Cell recognition method, apparatus, and system

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111127053.6A CN115880689A (zh) 2021-09-26 2021-09-26 细胞识别的方法、装置和系统
CN202111127053.6 2021-09-26

Publications (1)

Publication Number Publication Date
WO2023046167A1 true WO2023046167A1 (zh) 2023-03-30

Family

ID=85720149

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/121340 WO2023046167A1 (zh) 2021-09-26 2022-09-26 一种细胞识别的方法、装置和系统

Country Status (4)

Country Link
EP (1) EP4407571A1 (zh)
CN (2) CN115880689A (zh)
AU (1) AU2022350581A1 (zh)
WO (1) WO2023046167A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104359876A (zh) * 2014-10-14 2015-02-18 厦门大学 细胞牵引力显微镜及其在抗癌药物药效及药理检测中的应用
US20190317050A1 (en) * 2016-08-29 2019-10-17 Hunan Agricultural University Real-time and quantitative measurement method for cell traction force
CN110647874A (zh) * 2019-11-28 2020-01-03 北京小蝇科技有限责任公司 一种端到端的血细胞识别模型构造方法及应用
CN111666895A (zh) * 2020-06-08 2020-09-15 上海市同济医院 基于深度学习的神经干细胞分化方向预测系统及方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104359876A (zh) * 2014-10-14 2015-02-18 厦门大学 细胞牵引力显微镜及其在抗癌药物药效及药理检测中的应用
US20190317050A1 (en) * 2016-08-29 2019-10-17 Hunan Agricultural University Real-time and quantitative measurement method for cell traction force
CN110647874A (zh) * 2019-11-28 2020-01-03 北京小蝇科技有限责任公司 一种端到端的血细胞识别模型构造方法及应用
CN111666895A (zh) * 2020-06-08 2020-09-15 上海市同济医院 基于深度学习的神经干细胞分化方向预测系统及方法

Also Published As

Publication number Publication date
AU2022350581A1 (en) 2024-05-09
EP4407571A1 (en) 2024-07-31
CN115880689A (zh) 2023-03-31
CN118076981A (zh) 2024-05-24

Similar Documents

Publication Publication Date Title
Meijering et al. Tracking in molecular bioimaging
Roy et al. A review of recent progress in lens-free imaging and sensing
US20140139625A1 (en) Method and system for detecting and/or classifying cancerous cells in a cell sample
Bordescu et al. Fractal analysis of Neuroimagistic. Lacunarity degree, a precious indicator in the detection of Alzheimer’s disease
Legesse et al. Texture analysis and classification in coherent anti-Stokes Raman scattering (CARS) microscopy images for automated detection of skin cancer
Klyuchko On the mathematical methods in biology and medicine
Rejintal et al. Image processing based leukemia cancer cell detection
CN111797786B (zh) 用于离体生物样本的检测方法及四分类、计算机设备和计算机可读存储介质
Li et al. An approach for cell viability online detection based on the characteristics of lensfree cell diffraction fingerprint
Chen et al. Computer-aided detection (CADe) system with optical coherent tomography for melanin morphology quantification in melasma patients
Herold et al. Automated detection and quantification of fluorescently labeled synapses in murine brain tissue sections for high throughput applications
Swiderska-Chadaj et al. Convolutional neural networks for lymphocyte detection in immunohistochemically stained whole-slide images
WO2023046167A1 (zh) 一种细胞识别的方法、装置和系统
Li et al. Detection and tracking of overlapping cell nuclei for large scale mitosis analyses
Liu et al. Fast 3D cell tracking with wide-field fluorescence microscopy through deep learning
Burgemeister et al. CellViCAM—cell viability classification for animal cell cultures using dark field micrographs
CN113178228B (zh) 基于细胞核dna分析的细胞分析方法、计算机设备、存储介质
Saribudak et al. Spatial heterogeneity analysis in evaluation of cell viability and apoptosis for colorectal cancer cells
CN114460005A (zh) 动态显微成像的红细胞多角度形态学测量方法及系统
JP2020509347A (ja) 細胞分析方法及び装置
US20140273075A1 (en) Methods, systems and devices for determining white blood cell counts for radiation exposure
Gao et al. Evaluation of GAN architectures for visualisation of HPV viruses from microscopic images
Bischin et al. Computerized morphometric assessment of the human red blood cells treated with cisplatin
WO2018180012A1 (ja) 情報処理装置、情報処理システム、及び情報処理方法
Vicar et al. Detection and characterization of apoptotic and necrotic cell death by time-lapse quantitative phase image analysis

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22872210

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2024542229

Country of ref document: JP

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 202280064594.5

Country of ref document: CN

WWE Wipo information: entry into national phase

Ref document number: AU2022350581

Country of ref document: AU

WWE Wipo information: entry into national phase

Ref document number: 2022872210

Country of ref document: EP

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2022872210

Country of ref document: EP

Effective date: 20240426

ENP Entry into the national phase

Ref document number: 2022350581

Country of ref document: AU

Date of ref document: 20220926

Kind code of ref document: A