CN105181649B - A kind of Novel free marking mode identifies cell instrument method - Google Patents

Abstract

The invention discloses a novel label-free pattern recognition cytometer method, comprising: preparing a label-free cell solution to be tested and introducing it into a microfluidic channel or making a cell suspension chip; conducting laser light to excite the cells to be tested to form a scattering pattern distributed in a three-dimensional space Light, scattered light passes through the optical imaging system or does not pass through any optical imaging system, is detected by the two-dimensional CMOS detector and obtains the two-dimensional light scattering pattern corresponding to the cell to be tested; the obtained two-dimensional light scattering pattern is transmitted to the pattern recognition classification system, The system automatically learns the two-dimensional light scattering patterns of known cells belonging to different types, and realizes label-free and automatic identification of unknown cells. The recognition result triggers the corresponding device to realize label-free, automated cell counting or cell classification functions. The present invention does not require complicated fluorescent staining of the cells, can automatically and label-free realize the identification, counting and classification of the cells to be tested, is simple and quick to operate, significantly reduces the analysis cost, and has a wide range of applications.

Description

A Novel Label-Free Pattern Recognition Cytometry Method

technical field

The invention relates to cell classification and identification, in particular to using a label-free cytometer to obtain light scattering pattern information of cells, and then performing pattern recognition on the light scattering pattern to realize label-free, automatic cell counting or classification functions.

Background technique

Traditional flow cytometry can be used for cell analysis and sorting. Generally speaking, traditional flow cytometry requires staining of cells, and fluorescent staining or other biomarkers may have certain effects on cells, especially in the study of the function of living cells. Secondly, the optical path required for fluorescence measurement is complicated, which directly increases the cost of the instrument, and fluorescence measurement requires calibration of the instrument, which is complicated to operate and requires professionals. Finally, in terms of signal processing in the later stage, due to the possible overlap of fluorescence emission spectra, operations such as color complementation are required, and the existing conventional flow cytometers can only classify and identify cells based on predetermined physical and chemical characteristic parameters. The scope identifies the corresponding cell subpopulations, and lacks the function of automatic machine learning and classification recognition, especially in label-free and automated cell classification recognition.

Cervical cancer screening is mainly based on cervical cytology and HPV immunodetection. For cervical cytology examination, firstly, exfoliated cervical cells are collected from cervical tissue, stained and made into slices, and then manually read under a microscope by clinical pathologists. Experienced physicians can more accurately distinguish between normal and cancerous cells in the cervix, and sometimes HPV testing and colposcopy biopsy may be required to confirm the diagnosis. The process of cervical cytology examination is complicated and time-consuming, and manual film reading requires doctors with rich clinical experience, and the results of film reading are highly subjective. HPV detection and colposcopy biopsy are highly accurate but difficult to popularize.

Contents of the invention

The invention discloses a novel label-free pattern recognition cytometer method. On the basis of obtaining a two-dimensional light scattering pattern of label-free cells or cell aggregates, pattern recognition is performed on the scattering pattern to achieve label-free, automatic cell counting and classification purpose of identification. In terms of sample processing, this method overcomes the shortcomings of traditional flow cytometers that require fluorescent staining, and realizes label-free sample processing; it overcomes the shortcomings of traditional flow cytometers in terms of optical path complexity and high cost; In terms of processing, pattern recognition is innovatively used for automatic classification and recognition of scattering patterns. The organic integration of the above technologies has realized the innovation in the method of cell classification. In terms of application effect, this innovative method can be used for rapid identification of the number of cell aggregates, and also achieves label-free identification of normal cervical cells and cancerous cervical HeLa cells.

A novel label-free pattern recognition cytometer method, the specific scheme includes the following steps:

Prepare the label-free cell solution to be tested and introduce it into the microfluidic channel or make a cell suspension chip;

The light emitted by the laser light source is coupled into the optical fiber through the quadruple objective lens, and the optical fiber transmits the laser light and excites single cells or multi-cell aggregates in the microfluidic channel or cell suspension, forming scattered light distributed in three-dimensional space;

The scattered light passes through the optical imaging system and is detected by the two-dimensional CMOS detector to obtain the two-dimensional light scattering pattern corresponding to the cells to be tested;

Alternatively, the scattered light is detected by a two-dimensional CMOS detector without passing through an optical imaging system, and a two-dimensional light scattering pattern corresponding to the cells to be tested is obtained;

The obtained two-dimensional light scattering pattern is transmitted to the pattern recognition system, which realizes label-free and automatic identification of unknown cells by performing machine learning on two-dimensional light scattering patterns of known different cell types;

The recognition results are used to trigger a label-free pattern recognition cytometer automated cell counting or cell sorting system.

Further, the corresponding device for obtaining the two-dimensional light scattering pattern includes: a light source system for generating scattered light from a laser light source to excite the measured cells; a two-dimensional light scattering pattern detection and recording system for recording and collecting the scattered light of the measured cells; pattern recognition System for automatic classification and recognition through data processing and machine learning; cell counting or cell classification system, including digital counters and mechanical classification devices.

Further, when the scattered light passes through the optical imaging system to obtain the two-dimensional light scattering pattern corresponding to the cells to be measured, it is necessary to move and adjust the three-dimensional translation stage to find the laser convergence point, so that the laser light source can enter the optical fiber with the best coupling state after passing through the quadruple objective lens. In this method, the other end of the optical fiber is used as a probe to excite single cells or cell aggregates in microfluidic channels or cell suspensions.

Further, the scattered light generated by the cells in the solution to be tested in the microfluidic channel is excited by the laser, passes through the optical imaging system or does not pass through the optical imaging system;

When passing through the optical imaging system, the cells to be measured in the microfluidic channel or cell suspension are excited by the laser to generate scattered light. The scattered light is observed through the ten times objective lens, and the ten times objective lens is adjusted to observe the original image of the cells. focus, and obtain the scattered light pattern on the CMOS two-dimensional detector.

Without an optical imaging system, that is, without any optical imaging lens, the scattered light passes through a physical aperture to directly form a two-dimensional light scattering pattern on the CMOS plane.

Further, the pattern recognition algorithm adopts the enhanced AdaBoost algorithm to select N-1 patterns of known classification from the detected N two-dimensional light scattering patterns for training to obtain a set of weak classifiers for pattern recognition, and then use this The combination of weak classifiers tests the original Nth pattern, selects the strongest combination of weak classifiers through round-robin testing, and finally obtains a high accuracy rate for the classification and recognition of the cells to be tested.

Furthermore, the above method is used for the classification of yeast cell clusters.

Further, the above method is used for the classification of normal cervical cells and cancerous cervical cells.

Beneficial effects of the present invention:

(1) The new label-free pattern recognition cytometer proposed by the present invention has a simple device, overcomes the shortcomings of traditional flow cytometers such as complex optical path, expensive equipment, and cumbersome operation, and can easily and quickly obtain two-dimensional light scattering patterns.

(2) The label-free technology adopted in the present invention overcomes the problem that traditional flow cytometers need to perform fluorescent staining on cells, which may cause damage to cell samples, and can avoid the interference of fluorescent staining on cells, especially the function of living cells.

(3) The new label-free pattern recognition cytometer proposed by the present invention can realize label-free and automatic classification and recognition of cells or cell groups, that is, the classification and recognition of the cells to be tested can be achieved through an automated pattern recognition algorithm.

(4) The new label-free pattern recognition cytometer proposed by the present invention has strong generalization ability and can be widely used in the classification and recognition of different cells.

(5) The analysis process of the present invention is highly operable, and an appropriate number of weak classifiers can be selected to form a strong classifier to improve the recognition accuracy as much as possible. After the strong classification is obtained, the operator only needs to input the detection image to automatically obtain the classification recognition result.

(6) The label-free and automatic pattern recognition cytometry proposed by the present invention can be used to stimulate corresponding physical counters or sorters to realize the functions of counting and sorting cells.

(7) The present invention provides a novel method for judging the number of cell groups.

(8) The present invention provides a novel judgment method for classifying normal cervical cells and cancerous HeLa cells.

Description of drawings

Fig. 1 is the structure and schematic diagram of device of the present invention,

Fig. 2(a)-Fig. 2(d) Comparison of the simulated diagram of a single yeast cell and the experimental scatter diagram;

Figure 3(a)-Figure 3(d) Comparison of the original and scatter diagrams of different numbers of yeast cell aggregates;

Figure 4(a)-Figure 4(d) Comparison of the original and scatter diagrams of normal cervical cells and HeLa cells;

In Figure 1: 1. Laser light source, 2. Quadruple objective lens, 3. Fiber coupler, 4. Microfluidic channel or cell suspension chip, 5. Ten-fold objective lens or physical aperture, 6. Two-dimensional CMOS detector, 7 , pattern recognition system, 8 classification system;

Table 1 Classification results of different numbers of yeast cell aggregates;

Table 2 The classification results of normal cervical cells and HeLa cells.

Detailed ways:

The present invention is described in detail below in conjunction with accompanying drawing:

As shown in Figure 1, a new type of label-free pattern recognition cytometer is mainly composed of a light source system, a two-dimensional light scattering pattern detection and recording system, and a data processing and classification system. The light source system includes a laser light source 1, a quadruple objective lens 2, a fiber coupler 3, a microfluidic channel or a cell suspension chip 4; a two-dimensional light scattering pattern detection and recording system includes a tenfold objective lens or a physical aperture 5, and a two-dimensional CMOS detector 6. Data processing and classification system composition analysis system includes pattern recognition system 7 and classification system 8 .

The two-dimensional light scattering pattern detection system of the present invention includes a light scattering excitation system for single cells and cell aggregates, a microfluidic or cell suspension chip system, and a two-dimensional light scattering pattern acquisition system. The data processing and classification system of the present invention uses a pattern recognition algorithm to perform post-data processing of two-dimensional light scattering patterns, and realizes label-free and automatic classification and recognition of single cells and cell groups. The present invention takes the AdaBoost machine learning algorithm in pattern recognition as an example, but is not limited to using this specific pattern recognition method to identify cell light scattering patterns. The invention does not rely on traditional fluorescent dyeing and manual classification, and realizes label-free and automatic classification and recognition of different numbers of yeast cell groups and different pathological states of cervical cells through pattern recognition of scattering patterns. The application range of the invention can be extended to the physiological and pathological analysis of general biological cells. The novel label-free pattern recognition cytometry method disclosed in the present invention does not require complex fluorescent staining of cells, and can automatically realize the classification and identification of the cells to be tested, as well as the counting and sorting of the cells to be tested without labeling, and the operation is simple and fast. The result is accurate and reliable, the cost of analysis is significantly reduced, and the application range is wide.

Below in conjunction with accompanying drawing 1 the present invention is carried out the detailed description of concrete operation step:

Step 1: prepare the cell solution to be tested, the method of preparing the solution is different according to different cells to be tested.

Step 2: introducing the cell solution to be tested into the microfluidic channel 4 or making a cell suspension chip 4 .

Step 3: Turn on the laser light source 1. The laser light source 1 uses a green laser diode-pumped solid-state laser (DPSS) with a wavelength of 532nm. Diode-pumped solid-state lasers have significant advantages such as long working time, high efficiency, low energy consumption, small thermal effect, and small size. In order to ensure that the laser beam with a diameter of 1.0mm can be coupled into an optical fiber with a diameter of 105 μm and a numerical aperture (NA) of 0.22 to the maximum extent, the present invention selects a quadruple objective lens 2 with a numerical aperture of 0.1 to improve the coupling efficiency.

Step 4: Constantly move and adjust the calibration of the laser coupler so that the laser light source can enter the fiber end of the fiber coupler 3 in the best coupling state after passing through the quadruple objective lens 2 . The other end of the optical fiber of the optical fiber coupler 3 is used as a probe to excite cells or cell groups in the microfluidic channel 4 or the cell suspension chip 4 .

Step 5: The laser excites the cells to be tested in the microfluidic channel 4 or the cell suspension chip 4 to generate scattered light. First looking through the objective lens, move the fiber to position the cell until the cell to be measured is in the middle of the laser beam and fully excites the side scattered light.

Step 6: Obtain a two-dimensional light scattering pattern of the cells to be tested. Adjust the tenfold objective lens 5 so that the cell image in the field of view is the clearest, and what is obtained at this time is the original image of the cell. The image to be acquired by the present invention is not the image formed by the cells themselves, but the two-dimensional scattered light pattern formed by them. On the basis of focusing, adjust the objective lens in the same direction and the same distance, that is, "defocus", at this time, a two-dimensional light scattering pattern will be formed on the plane of the COMS two-dimensional detector 6. The size of the CMOS two-dimensional detector of the present invention is 22.3×14.9 mm, and the recording pixels are about 17.9 million. The use of CMOS two-dimensional detectors has the advantages of high integration, low power consumption, low cost, and easy integration with other chips. Alternatively, the light scattered by the cells does not pass through any optical imaging system, that is, does not pass through any lens, and can also use the physical aperture 5 to form a two-dimensional light scattering pattern on the plane of the CMOS detector 6 .

Step 7: Transfer the obtained two-dimensional light scattering pattern collected by the COMS two-dimensional detector 6 to the pattern recognition system 7 .

Step 8: Standardize the obtained scattering patterns and unify them into 220×220 pixels. The normalized scatter patterns are classified and identified using a pattern recognition algorithm.

Step 9: In the pattern recognition system 7, the corresponding pattern recognition algorithm can be used to perform classification and recognition according to specific conditions. The present invention takes the AdaBoost algorithm as an example, uses the leave-one-out method for the detected N scattering patterns, selects N-1 known classification patterns for training, and then uses the remaining Nth The pattern debugs this combination of weak classifiers to obtain a set of weak classifier combinations. Next, replace one of the original N-1 training samples with the original Nth pattern, so as to train another group of weak classifiers. According to this method, multi-layer weak classifiers are obtained, a corresponding number of multi-layer weak classifiers is selected, and a strong classifier is constructed by effectively combining the classification results of the weak classifiers. The number of multi-layer weak classifiers contained in the strong classifier can make the classification result optimal. When there are enough training samples, the final classifier can be obtained through a single training for packaging, so as to meet the classification recognition requirements.

Step 10: Automatically learn the two-dimensional light scattering pattern of the cells to be tested through the pattern recognition system, and realize label-free and automatic identification of unknown cells. The recognition result triggers the corresponding sorting system 8 device to realize label-free, automatic cell counting or cell sorting functions.

Example 1

Prepare a yeast solution with a suitable concentration, use the device of the present invention to obtain a two-dimensional light scattering pattern of the yeast solution, and compare and verify the experimental results and theoretical simulation results. Figure 2 shows the comparison of the theoretical simulation results of the two-dimensional scattering pattern formed by a single yeast cell in Figure 2(a) and Figure 2(b) with the experimental results in Figure 2(c) and Figure 2(d). Yeast is a single-celled microorganism with a cell diameter of about 3-6 μm. The scatter diagram of the experiment presents two different morphologies as shown in Figure 2(c) and Figure 2(d). In the simulation, yeast cells are assumed to be spherical particles with different diameters, the refractive index is 1.42, the incident wavelength is 532nm, and the refractive index of the surrounding medium is 1.334. The diameter of the cells in Figure 2 (a) is 3.8 μm, and the diameter in (b) is 5.0 μm. It can be observed from the figure that the experimental results and the theoretical simulation results are consistent with each other in terms of the number of fringes and the positions of the fringes, which can verify the accuracy of the device of the present invention.

Example 2

Using the device of the present invention, 60 corresponding two-dimensional light scattering patterns of aggregates are obtained from 60 groups of yeast cell aggregates. Among them, 30 groups are patterns of 3 yeast cell aggregations, and the other 30 groups are patterns of 4 yeast cell aggregations. The size of each scattering pattern is 220×220 pixels, and has been normalized, as shown in Figure 3(a)-Figure 3(d).

The present invention uses the AdaBoost method to perform a leave-one-out experiment. The specific implementation steps are: train the light scattering patterns of 59 known classifications (3 yeast cell aggregations or 4 yeast cell aggregations), obtain a group of weak classifiers for pattern recognition, and then use the 60th pattern to weakly classify this The classifier is tested. Record the correct number (CN) of the test data, and calculate the correct rate (AR) through the formula AR=CN/TN. TN represents the number of all 2D light scattering patterns. As shown in Table 1, the study found that when the number of layers of the weak classifier increases, the AR changes accordingly. The present invention uses three layers of weak classifiers, and can obtain the maximum correct rate of 86.7%. Among them, the AR of 3 yeast cell aggregates was 93.3%, and the AR of 4 yeast cell aggregates was 80%.

Table 1 Classification results of yeast cell aggregates

Example 3

The device of the invention is used for classifying and identifying normal cervical cells and cancerous cervical cells (HeLa cells). For the classification and identification of cervical cells, a total of 92 two-dimensional light scattering patterns were obtained using the label-free pattern recognition cytometer method of the present invention, 54 of which were normal cervical cell light scattering patterns, and 38 were HeLa cell light scattering patterns. As shown in Figure 4(a)-Figure 4(d). When using the AdaBoost method for pattern recognition, 91 scattering patterns are used to train the weak classifier, and the 92nd one is used for testing. As shown in Table 2, the study found that when the number of weak classifier layers is 7, the classification accuracy AR reaches a maximum value of 90.2%, among which the correct rate AR of normal cervical cells is 90.7%, and the correct rate AR of HeLa cells is 89.5%. Normal cervical cells and cancerous cervical cells are observed under a 400-fold microscope. Their appearance is similar, but their internal structure has changed. The pattern of two-dimensional light scattering contains information about the changes inside the cell, and the label-free pattern recognition cytometer can automatically classify and identify the two types of cells. The 90.2% correct rate of cervical cell classification and recognition in the present invention indicates that the label-free pattern recognition cytometer has good clinical application prospects.

Table 2 Results of cervical cell classification

The above description of the specific embodiments of the present invention in conjunction with the accompanying drawings is not a limitation of the protection scope of the present invention. On the basis of the technical solution of the present invention, various modifications or modifications can be made by those skilled in the art without creative labor. Deformation is still within the protection scope of the present invention.

Claims (8)

1. a kind of label-free pattern-recognition cell instrument method, it is characterized in that, comprise the following steps：

Make label-free cell solution to be measured and import in microchannel or be made cell suspension chip；

The light that LASER Light Source is sent is coupled into optical fiber by four times of object lens, and fiber optic conduction laser simultaneously excites microchannel or cell Unicellular or many cells gathering groups in suspension, distribute in the scattering light of three dimensions；

Scattering light passes through optical imaging system, is detected by two-dimentional cmos detector and obtains two-dimentional light scattering corresponding to cell to be measured Pattern；

Or scattering light needs not move through optical imaging system, is detected and is obtained corresponding to cell to be measured by two-dimentional cmos detector Two-dimentional light scattering pattern；

The two-dimentional light scattering pattern of acquisition is transmitted to PRS, and the system passes through the two dimension to known different cell categories Light scattering pattern carries out machine learning, realizes unmarked, the automatic identification to unknown cell；

Recognition result is used to trigger label-free pattern-recognition cell instrument automatic cytological counting or cellular classification system.

2. a kind of label-free pattern-recognition cell instrument method as claimed in claim 1, it is characterized in that, two-dimentional light scattering pattern obtains Device corresponding to taking includes：Light-source system, the scattering light of tested cell is excited for producing LASER Light Source；Two-dimentional light scattering pattern Record system is detected, record collects the scattering light of tested cell；PRS, carried out by data processing and machine learning Automatic Classification and Identification；Cell count or cellular classification system, including digit counter and mechanical sorter.

3. a kind of label-free pattern-recognition cell instrument method as claimed in claim 1, it is characterized in that, scattering light by optics into As system, it is necessary to which mobile adjustment three-D displacement platform finds laser convergence when obtaining two-dimentional light scattering pattern corresponding to cell to be measured Point, LASER Light Source is set to enter after four times of object lens with Best Coupling state in optical fiber, the other end of optical fiber is used as probe In exciting the unicellular or cell aggregation group in microchannel or cell suspension.

4. a kind of label-free pattern-recognition cell instrument method as claimed in claim 1, it is characterized in that, microchannel or cell hang Cell to be measured produces scattering light by laser excitation in liquid, and the scattering light is observed by ten times of object lens, adjusts ten times of object lens, observation To the original image of cell, the object lens are defocused, scattering optical pattern is obtained on COMS two-dimensional detectors.

5. a kind of label-free pattern-recognition cell instrument method as claimed in claim 1, it is characterized in that, it is to be measured molten in microchannel Cell in liquid scatters light caused by laser excitation, without optical imaging system, that is, does not need any optical imaging lens, Scattering light passes through a physical pore size, and two-dimentional light scattering pattern is directly formed in CMOS planes.

6. a kind of label-free pattern-recognition cell instrument method as claimed in claim 1, it is characterized in that, algorithm for pattern recognition uses Strengthen AdaBoost algorithms, to N number of unicellular or cell aggregation group the two-dimentional light scattering pattern detected, select N-1 The pattern of known classification is trained, and the Weak Classifier of one group of pattern-recognition is obtained, then with the n-th pattern pair originally left This Weak Classifier is tested, and a strong classifier for including a number of Weak Classifier is obtained by testing.

7. a kind of label-free pattern-recognition cell instrument method as claimed in claim 1, it is characterized in that, the above method is used for yeast The classification of cell aggregation group.

8. a kind of label-free pattern-recognition cell instrument method as claimed in claim 1, it is characterized in that, the above method is used for normal The classification of cervical cell and canceration cervical cell.