A multimodal flow microparticle detection method based on PT-symmetric circuits

Abstract

The present invention discloses a multimodal flow particle detection method based on a PT-symmetric circuit. By adjusting the PT-symmetric circuit to the singular point, when the particle passes through the conductive electrode, the output voltage waveform of the circuit can fluctuate extremely sensitively. By processing this voltage waveform, time-domain characteristic information and frequency-domain characteristic information can be obtained. The extreme points of the wave peaks or wave valleys appearing in the characteristic information are used as the characteristic values of a single particle, making the characteristic value information of the particle richer. Thus, a machine learning model trained with the characteristic values of the particle can perform label-free detection and accurate classification of the particle.

Description

Technical Field

[0001] The present invention belongs to the technical field of biomedical detection based on machine learning, and particularly relates to a multi-modal flow particle detection method based on a PT-symmetric circuit. Background Art

[0002] Traditional particle detection technologies mainly include optical flow cytometry and impedance flow cytometry. The patent application with publication number CN104503101A discloses a flow cytometer beam forming system based on a diffractive optical shaping device, including: a laser light source, a diffractive optical shaping device, a reflector, a beam splitter, and a lens; the light emitted by each laser light source forms a light spot with a uniform intensity distribution, including but not limited to a rectangular light spot, after passing through the diffractive optical shaping device, and finally irradiates on the cell liquid flow of the flow cytometer system. Compared with the results of the traditional flow cytometer beam forming system, this patent application applies the diffractive optical shaping device to the beam forming system of the flow cytometer to form an illumination light spot with a uniform intensity distribution, improving the energy utilization rate and measurement and analysis accuracy of the flow cytometer system. However, the optical flow cytometry disclosed in this patent application relies on cell biological markers, and there are problems such as complex operation, difficulty in detecting cells lacking markers, and possible changes in the inherent characteristics of cells.

[0003] The patent application with publication number CN116371494A discloses a microfluidic chip for identifying and separating abnormal cells in CAR-T cells and its preparation method. The microfluidic chip includes: a microchannel, the microchannel includes an inlet for the cells to be detected to enter and at least two outlets for the cells to be detected to leave; the microchannel includes a separation channel and a detection channel; detection electrode pairs, which distinguish whether the cell is an abnormal cell or a normal cell by the impedance detected; positioning electrode pairs, which confirm through which of the outlets the cell leaves according to the impedance detected. However, the impedance flow cytometry disclosed in this patent application relies on external platform parameters, and there are problems such as low detection throughput, insufficient accuracy, poor data consistency, and inability to distinguish cells in different clinical stages.

[0004] Therefore, there is still a need to develop new technologies to meet the requirements of high-throughput, high-sensitivity, low-cost, and label-free cell detection in the biomedical field.

[0005] The singularity effect of the PT (PT-symmetric system) symmetric system has extremely high sensitivity and detection stability, and can produce a large response to the change of a small signal under the condition of gain-loss balance.

[0006] The invention patent application with the publication number CN114812371A discloses a metal thin film thickness detection system based on a PT - symmetric circuit. It includes a first resonant circuit and a second resonant circuit. The first resonant circuit and the second resonant circuit are RLC series resonant circuits with the same natural resonant frequency, and are respectively connected in series at both ends of a coupled inductance coil. By adjusting the capacitance values in the first resonant circuit and the second resonant circuit, the gain - loss coefficient is changed so that the gain - loss coefficients of the two are opposite to each other, thereby obtaining a PT - symmetric resonant system. The self - inductance of the coupled inductance coil and the metal thin film to be measured in the system changes under the influence of the eddy current effect, and the coupling coefficient between the first and second resonant circuits changes, thereby changing the resonant frequency of the PT - symmetric resonant system.

[0007] However, the biological applications of PT - symmetric systems need to be further studied and explored, and rapid detection methods for PT - symmetric systems need to be developed. Summary of the Invention

[0008] The present invention provides a multi - modal flow - type particle detection method based on a PT - symmetric circuit. This method aims to utilize its extremely high sensitivity to small perturbations to achieve label - free detection and precise classification of particles.

[0009] The present invention provides a multi - modal flow - type particle detection method based on a PT - symmetric circuit, including:

[0010] Connect the conductive electrodes of the microfluidic chip to the PT - symmetric circuit, adjust the parameters of the PT - symmetric circuit to the singular point, segment and process the output voltage waveform of the circuit to obtain time - domain characteristic information and frequency - domain characteristic information. When particles pass between the electrodes, peaks or valleys appear in the characteristic information. Take the extreme points corresponding to the peaks or valleys as the characteristic values of the particles. The number of extreme points corresponds to the number of particles at the same time and can be used for counting;

[0011] Divide the characteristic values of multiple pairs of particles and the corresponding particle labels into a training sample set and a verification sample set, and train a machine - learning model through the training data set to obtain a particle prediction model;

[0012] During application, input the extreme points of the time - domain characteristic information and / or frequency - domain characteristic information of the particles into the particle detection model to obtain the category prediction value, clinical cycle prediction value, variant prediction value or different - state prediction value of the particles. The different states refer to the initial stage, growth stage or apoptosis stage of the biological particles when the particles are biological particles.

[0013] Preferably, the method for respectively counting and calculating the flux of m types of particles includes:

[0014] The multi-modal flow particle detection method based on the PT-symmetric circuit is used to test m kinds of particles separately to obtain the corresponding training sample sets, and machine learning training is carried out to obtain a particle prediction model. Then, the obtained particle prediction model is applied to the detection of samples containing m kinds of particles simultaneously. The characteristic changes caused by each particle when passing through the electrode are input into the particle prediction model for species prediction, and then the number of the i-th kind of particles within the time of is counted. n i ;

[0015] The detection flux corresponding to the i-th kind of particles is calculated as , where n i (i = 1 - m) is the number of peaks or valleys corresponding to the i -th kind of particles within the time;

[0016] The total flux of the particles is K , .

[0017] Preferably, the voltage waveform is segmented and processed to obtain time-domain characteristic information, including:

[0018] The segmented voltage waveforms are analyzed by data processing software such as Matlab or Python, and time-domain characteristic information is constructed based on one or more of the standard deviation, kurtosis, envelope frequency, envelope peak value, envelope valley value, and envelope amplitude information of the voltage waveforms at different times obtained from the analysis.

[0019] Preferably, the voltage waveform is segmented and processed to obtain frequency-domain characteristic information, including:

[0020] The segmented voltage waveform is subjected to Fourier transform, and frequency-domain characteristic information is constructed based on one or more of the frequency shift and amplitude change information compared with the eigenmode when the particle passes through the microfluidic chip.

[0021] Preferably, the time for a single particle to pass between the electrodes is 0.1 - 10 ms.

[0022] Preferably, the PT-symmetric circuit is composed of LC circuits with gain and loss, and the two LC circuits are connected by capacitive coupling, inductive coupling, or an equivalent transmission line, and the parameters of the two LC circuits satisfy the PT-symmetric condition.

[0023] Preferably, the voltage waveform information is obtained by reading the voltage signal of the LC circuit with gain or loss of the PT-symmetric circuit through an oscilloscope, an FPGA circuit, or a single-chip microcomputer circuit.

[0024] Preferably, the microfluidic chip includes:

[0025] Substrate;

[0026] Microchannels, located on the substrate, with grooves on the microchannels, a measurement position is provided between the front and rear ends of the grooves, and the microparticles flow from one end of the groove to the other end and pass through the measurement position;

[0027] And conductive electrodes, the conductive electrodes are located on the substrate and are connected to the PT symmetric circuit, the conductive electrodes are one pair or multiple pairs of electrodes, and the ends of the two electrodes in each pair of electrodes are arranged opposite to each other or on the same side at the measurement position for measuring the impedance when the microparticles pass by.

[0028] Preferably, the method for obtaining the particle size includes:

[0029] When the ratio of the distance between the electrodes, the cross-sectional size of the pipeline to the size of the microparticles is less than 2, the injection pump injects the flow rate Q , and the cross-sectional area near the electrodes S , calculate the flow velocity of the microparticles passing through the electrodes V = Q / S , further combined with the time t when the microparticles pass through the electrodes, the particle diameter size can be calculated X = V × t = Q × t / S , introduce a coefficient a for correction, so the final particle diameter calculation method is X = V × a × t = Q × a × t / S .

[0030] Preferably, a particle detection model is obtained by training a machine learning model with a training sample set, and the machine learning model includes a convolutional neural network, a decision tree, an xgboost or a KNN algorithm.

[0031] Preferably, the multimodal flow particle detection method distinguishes the particles including the number (count), size and type of cells, bacteria, proteins, exosomes or microplastic particles. When the particles specifically refer to cells, their variants, clinical cycles and cell cycles can be further distinguished; when the particles specifically refer to bacteria, their growth states can be further distinguished.

[0032] Compared with the prior art, the beneficial effects of the present invention are:

[0033] By adjusting the PT - symmetric circuit to the singular point, when the microparticle passes through the conductive electrode, the output voltage waveform of the circuit can fluctuate extremely sensitively. By processing this voltage waveform, time - domain characteristic information and frequency - domain characteristic information can be obtained. Taking the extreme points of the wave peaks or wave valleys that appear in the characteristic information as the characteristic values of the microparticles makes the characteristic value information of the microparticles richer. Thus, the machine - learning model trained with the characteristic values of the microparticles can perform label - free detection and accurate classification of microparticles. Description of the Drawings

[0034] Figure 1 Schematic diagram of the circuit of the PT - symmetric multi - modal flow - type microparticle detection system provided by a specific embodiment of the present invention;

[0035] Figure 2 Schematic diagram of the negative - resistance circuit based on an operational amplifier provided by a specific embodiment of the present invention;

[0036] Figure 3 Schematic diagram of the microfluidic chip provided by a specific embodiment of the present invention;

[0037] Figure 4 Physical simulation schematic diagram of the PT - symmetric multi - modal flow - type microparticle detection system provided by a specific embodiment of the present invention;

[0038] Figure 5 Flowchart of the multi - modal flow - type microparticle detection method based on the PT - symmetric circuit provided by a specific embodiment of the present invention;

[0039] Figure 6 Graph of the change in the system voltage waveform caused by cells passing through the electrode provided by a specific embodiment of the present invention;

[0040] Figure 7 Graph (1) of the change in the system time - domain characteristics caused by cells passing through the electrode provided by a specific embodiment of the present invention;

[0041] Figure 8 Graph (2) of the change in the system time - domain characteristics caused by cells passing through the electrode provided by a specific embodiment of the present invention;

[0042] Figure 9 Graph of the change in the FFT spectrum corresponding to the change in the system voltage waveform caused by cells passing through the electrode provided by a specific embodiment of the present invention;

[0043] Figure 10 Graph of the change in the system frequency - domain characteristics caused by cells passing through the electrode provided by a specific embodiment of the present invention;

[0044] Figure 11 Graph of the system frequency offset in the high - throughput test of cells provided by a specific embodiment of the present invention;

[0045] Figure 12System frequency offset diagram in the test of different cell survival ratios provided by a specific embodiment of the present invention;

[0046] Figure 13 Prediction result diagram of the time-frequency domain full-feature convolutional neural network provided by a specific embodiment of the present invention;

[0047] Figure 14 Prediction result diagram of the time-domain feature convolutional neural network provided by a specific embodiment of the present invention. Detailed implementation manners

[0048] To deepen the understanding of the present invention, the following will make a detailed description of this embodiment in conjunction with the accompanying drawings.

[0049] In a specific embodiment of the present invention, the voltage waveform when microparticles pass through a microfluidic chip that can be obtained by using a PT-symmetric circuit is processed to obtain time-domain feature information and frequency-domain feature information. When the microparticles pass through the conductive electrodes of the microfluidic chip, peaks or valleys appear in the time-domain feature information and frequency-domain feature information. The extreme points of the appearing peaks or valleys are used as the eigenvalue of the microparticles, so that the eigenvalues of the microparticles are extremely rich. The model trained by using the eigenvalues of the constructed microparticles can quickly and accurately detect the category of the microparticles and the clinical cycle.

[0050] A multi-modal flow cytometry microparticle detection method based on a PT-symmetric circuit provided by a specific embodiment of the present invention includes:

[0051] S1. Connect the conductive electrodes of the microfluidic chip Zcell102 to the PT-symmetric circuit 101, and adjust the parameters of the PT-symmetric circuit to the singular point to obtain the voltage waveform information when the microparticles pass through the microfluidic chip. When the parameters of the PT-symmetric circuit reach the singular point, when the microparticles pass through the conductive electrodes, it can be detected extremely sensitively, and the voltage waveform information changes significantly.

[0052] In a specific embodiment, the PT-symmetric circuit 101 provided in this embodiment is composed of LC circuits with gain and loss. The two LC circuits are connected by capacitive coupling, inductive coupling or equivalent transmission lines. The parameters of the two LC circuits satisfy the PT-symmetric condition, and the two LC circuits are series LC circuits or parallel LC circuits.

[0053] As Figure 1 shown, the PT-symmetric circuit 101 provided by a specific embodiment of the present invention is respectively composed of a detection end and a reading end. The detection end respectively includes a capacitive element ( C s ), a resistive element ( R s ), and an inductive element ( L s ), and has loss ( Rs ) LC parallel resonance circuit; the reading ends respectively include capacitive elements ( C r ), negative resistance elements ( R r ), inductive elements ( L r ), and is an LC parallel resonance circuit with gain ( R r ). The detection end and the reading end are connected by coupling M, and can also be in coupling modes such as capacitive coupling or equivalent transmission lines, forming a PT-symmetric structure, and the parameters satisfy the PT-symmetric condition (i.e., L r = x L s , x C r = C s and R r = -x R s , where x is a scaling factor); the microfluidic chip is connected to the detection end, and there are conductive interdigital electrodes in the chip for detecting the change in electrical signal caused by the passing of particles (here, the impedance of the microfluidic chip is equivalent to a parallel RC circuit, and this has been considered in R s , C s ).

[0054] The part with loss characteristics ( R s ) can be composed of an ordinary resistor or a two-dimensional material with the same impedance value, or one of the metasurfaces; the part with gain characteristics ( R r ) can be composed of an amplifier circuit formed by an operational amplifier, a MOS transistor, a Gunn diode, a tunnel diode, etc., or a two-dimensional material with the same impedance value, or one of the metasurfaces. As shown in Figure 2 is a schematic diagram of a negative resistance circuit based on an operational amplifier. V CC is the positive power supply voltage of the operational amplifier, V EE is the negative power supply voltage of the operational amplifier, R F is the feedback resistor, R G is the input loop resistor, R L is the load resistor or bias resistor, and Z NRC is the negative resistance circuit or bias resistor.

[0055] The microfluidic chip provided in this embodiment includes a substrate 201, a microchannel 202, and a conductive electrode 203, and the specific structure is as shown in Figure 3 .

[0056] The microchannel 202 provided in this embodiment is located on the substrate 201. The microchannel 202 has grooves, and a measurement position is provided between the front and rear ends of the grooves. The microparticles flow from one end of the grooves to the other end and pass through the measurement position.

[0057] The microfluidic chip provided in this embodiment can also be composed of other materials, such as a thermally bonded glass layer fixed on a silicon substrate.

[0058] The conductive electrode 203 provided in this embodiment is located on the substrate 201 and is connected to a PT symmetric circuit. The conductive electrode 203 is one pair or multiple pairs of electrodes. The ends of the two electrodes in each pair of electrodes are arranged opposite to each other left and right at the measurement position. In one embodiment, the distance between the ends of the two oppositely arranged electrodes is 10 nm - 100 μm, which is used to measure the impedance when the microparticles pass by.

[0059] The conductive electrode provided in this embodiment is one pair or multiple pairs of interdigital electrodes and one pair or multiple pairs of positive electrodes.

[0060] The material of the conductive electrode provided in this embodiment is composed of one of conductive nanomaterials, metals, and compounds. The substrate and the microchannel can be composed of one of glass and silicon.

[0061] In a specific embodiment, as Figure 4 shown, the inductance value of the coil ( L s , L r ) is 10 mH, the capacitance ( C s , C r ) is a tunable capacitor (TZB4R500AB10R00, Murata), and the resistance ( R s ) is a tunable resistor (3314J - 1 - 502E, Bourns). Among them, the gain unit (negative resistance) is composed of the positive feedback circuit of an operational amplifier chip (OPA695, Texas instruments) (as Figure 2 shown), R G = R L = 511W. According to the virtual short and virtual open properties of the ideal operational amplifier, it is easy to deduce that the magnitude of the negative resistance is controlled by R F (3314J - 1 - 502E, Bourns). Therefore, when the system parameters satisfy C s = C r , R s =R F When (here, the impedance of the microfluidic chip is equivalent to a parallel RC circuit, and this has been considered in R s , C s ) ， The system operates in a PT-symmetric state.

[0062] The voltage waveform information provided in this embodiment is obtained by reading the voltage signal of the LC circuit that gains or loses in the PT-symmetric circuit through one of an oscilloscope, an FPGA circuit, and a microcontroller circuit.

[0063] The microparticles provided in this embodiment include cells, bacteria, proteins, exosomes, or microplastics.

[0064] S2. Segment and process the voltage waveform to extract the peak value, and then remove the baseline to obtain the time-domain waveform information and the frequency-domain waveform information. When the microparticle passes between the electrodes, peaks or valleys appear in the time-domain characteristic information and the frequency-domain characteristic information. Take the extreme points corresponding to the peaks or valleys as the characteristic values of the microparticle, such as Figure 5 Characteristics 1-10 of cell 1 in

[0065] In one embodiment, this embodiment provides the specific steps for segmenting and processing the voltage waveform to obtain the time-domain waveform characteristics, including:

[0066] Use data processing software such as matlab or python to analyze and remove the baseline from the segmented voltage waveform respectively. Take one or more of the standard deviation, kurtosis, envelope frequency, envelope peak value, envelope valley value, and envelope amplitude of the voltage waveform at different times obtained after removing the baseline as the time-domain characteristic information. Since the time-domain characteristic information is more diverse, it is beneficial to increase the accuracy of the training model. At the same time, the processing speed of the time-domain information is faster, which is beneficial to further improving the detection speed in the later stage.

[0067] In the embodiment of the present invention, the counting of microparticles is realized by counting the number of peaks or valleys in the standard deviation, kurtosis, envelope frequency, envelope peak value, envelope valley value, or envelope amplitude characteristic information of the voltage waveform.

[0068] In one embodiment, this embodiment provides the steps for segmenting and processing the voltage waveform to obtain the frequency-domain information, including:

[0069] The Fourier transform of the segmented voltage waveform information gives a system operating frequency range between 1-100 MHz. Take one or more of the characteristic information of the frequency shift and amplitude change compared with the eigenmode when the microparticle passes through the microfluidic chip as the frequency-domain characteristic information.

[0070] In the embodiments of the present invention, the counting of single particles is achieved by statistically counting the number of peaks or valleys in the characteristic information of frequency offset or amplitude change.

[0071] S3. Divide the eigenvalue pairs (eigenfeature 1 - eigenfeature 10) of multiple pairs of particles and their corresponding particle labels into a training set and a validation set, and train a machine learning model through the training set to obtain a particle detection model. Since the eigenvalue pairs of the particles provided in the specific embodiments of the present invention are extremely rich, a model capable of accurately identifying particle categories and clinical cycles can be trained. The machine learning model includes a convolutional neural network, a decision tree, an xgboost, or a KNN algorithm.

[0072] During application, input the extreme points of the time domain features and / or frequency domain features of the particles into the particle detection model to obtain the predicted particle category value, clinical cycle prediction value, variant prediction value, or different state prediction values. The different states refer to the initial stage, growth stage, apoptosis stage, or clinical cycle of the biological particle when the particle is a biological particle. It can be understood that as long as the present invention can provide different categories of particles, as well as the clinical cycle, variant, or different states of biological particles, and train in the system provided in the specific embodiments of the present invention, a particle detection model capable of accurately predicting the particle category prediction value, clinical cycle prediction value, variant prediction value, or different state prediction values can be obtained.

[0073] In a specific embodiment, the convolutional neural network provided in this embodiment includes 10 input neurons, 100, 50, and 30 neurons in the three hidden layers respectively, and 5 output neurons, which is used for cell recognition to obtain the predicted cell category or clinical cycle, that is, the predicted label, and construct a confusion matrix through the true label and the predicted label.

[0074] The specific embodiments of the present invention also provide a method for obtaining the particle size, including:

[0075] When the ratio of the distance between the electrodes, the cross-sectional size of the pipeline to the size of the particle is less than 2, the injection pump injects flow Q , and the cross-sectional area near the electrode S , we can calculate the flow velocity of the particle passing through the electrode V = Q / S . Further, in combination with the time t for the particle to pass through the electrode, the particle diameter size can be calculated X = V × t = Q × t / S . Considering the influence of the actual width of the electrode on the calculation of time t, a coefficient a is introduced for correction. Therefore, the final particle diameter calculation method is X = V× a × t = Q × a × t / S 。

[0076] Using the above multi-modal flow particulate detection method based on a PT-symmetric circuit, the time for a single particulate to pass between the electrodes is 0.1 - 10 ms.

[0077] The specific embodiments of the present invention also provide a method for separately counting different particulates, including:

[0078] By separately testing m types of particulates individually, obtaining the corresponding training sample sets, and performing machine learning training to obtain a particulate prediction model, then applying the obtained particulate prediction model to the detection of a sample containing m types of particulates simultaneously. Input the characteristic changes caused by each particulate when passing through the electrodes into the particulate prediction model for type prediction, and then count the number of the i-th type of particulate within the i-th type of particulate within n i 。Calculate the detection flux corresponding to the i-th type of particulate as ,where n i (i = 1 - m) is the number of peaks or valleys corresponding to the i i-th type of particulate within the time. The total flux of the particulates is K , 。

[0079] In one embodiment, the detection sample is breast cancer cells (MCF-7). In this embodiment, an injection pump is used to inject the cell solution, an oscilloscope is used to record and save the voltage waveform at the reading end, and then offline data processing is performed. According to the values of the LC components of the system, it can be estimated that the operating frequency of the system is less than 10 MHz. Therefore, the sampling rate of the oscilloscope is set to 40 M / s, and the number of sampling points is set to 100 Mpts. After saving the data, signal analysis is performed using matlab, or real-time data processing is directly performed using the FPGA + ADC method, such as Figure 6 First, the waveform signal changes during the process of cells passing through the electrodes are shown (to more clearly show the signal changes caused by the cells flowing through the electrodes, only the voltage waveform information with a duration of 8 ms (320 pts) is shown here). In this embodiment, the graph between two adjacent valleys of the outer envelope is regarded as a small envelope, and its duration is T. Then the envelope frequency of the outer envelope is 1 / T, Figure 6 the waveforms in (a) of Figure 6 and Figure 6 the waveform in (c) of Va It decreases, and the number of small envelopes within the same time increases (i.e., the envelope frequency increases), indicating that the cell passing through the electrode will cause obvious waveform changes.

[0080] To perform signal analysis quantitatively, in this embodiment, the waveform signal with a length of 100 Mpts is segmented into segments with a length of 20 kpts (corresponding to 0.5 ms), and finally divided into 5000 segments. For each segment of the signal with a length of 20 kpts, its mean square error, kurtosis, and outer envelope frequency are calculated, so as to obtain the variation of these parameters with time. As shown in Figure 7 the (a) in Figure 7 shows the standard deviation analysis result of 11 data files of 100 Mpts collected continuously. Each downward peak represents the change caused by a cell passing through the electrode. According to the size difference between different downward peaks, the response differences between different cells are reflected. Figure 6 The (b) in Figure 7 shows the standard deviation change when a single cell passes through the electrode after magnification. Combining with V a the waveform changes at three moments when the cell passes through the electrode, it can be seen that when the cell passes through the electrode, the number of envelopes per unit time increases, the waveform density increases, the standard deviation decreases, and the minimum reaches 0.352 V. When the cell is far from the electrode, the standard deviation value gradually returns to the baseline, that is, 0.372 V. From V b the (b) in V b - V a / 2), etc. time domain characteristics, as shown in Figure 7 the (c) in Figure 7 - Figure 8 the (f) in Figure 8 and the (a) in

[0081] These characteristics all reflect the influence of the impedance change of the cell passing through the electrode on the system. Figure 9 For the frequency domain changes during the process of the cell passing through the electrode, corresponding to Figure 7 the three moments in the (b) in f 1, A 1), and the original frequency and amplitude of the right peak are denoted as (f 2, A 2), and record the frequency and amplitude of the left peak when the cell passes through the electrode as ( f 1 ’ , A 1 ’ ), and record the original frequency and amplitude of the right peak as ( f 2 ’ , A 2 ’ ). Then the signal change of the left peak is D f 1 = f 1 - f 1 ’ , D A 1 = A 1 - A 1 ’ . The signal change of the right peak is D f 2 = f 2 – f 2 ’ , D A 2 = A 2 – A 2 ’ . Figure 10 in (a)- Figure 10 in (d) respectively show the corresponding frequency-domain signal changes, and at the same time show the high signal-to-noise ratio of this system. According to each peak, cell counting and analysis of cells can be carried out.

[0082] Furthermore, in this embodiment, the solution injection speed of the injection pump is increased for high-throughput cell analysis. As Figure 11 in (a) and Figure 11 in (b) show, the change of the system frequency-domain characteristics with time after the flow rate is increased is shown. It can be seen from the enlarged view on the right that the time for a single cell to pass through the electrode is 0.1 ms, showing a test throughput of 10,000 cells per second.

[0083] Then, in this embodiment, formalin solution is used to inactivate MCF-7 cells, and it is mixed with live cells to obtain cell solutions with different survival ratios, which are used for testing. We extract Figure 10 the extreme points in (b) and [[ID= in (d) (i.e., the moment when the cell is at the electrode) as the final response of the system, so as to obtain the response data sets of different cells. The frequency-domain test results are as ​ in (a) and ​As shown in (b), the box plot shows the 95% confidence ellipse distribution intervals of the three cell solutions. The pentagrams within the distribution intervals represent the centroid points of the ellipses. The box plots on the opposite sides of the xy-axis correspond to statistical parameters such as the minimum value, lower quartile, median, upper quartile, and maximum value of the frequency domain response data sets for each cell solution.

[0084] As can be seen from ​ (a), as the cell survival ratio decreases, D at the centroid point f 1 deflects from negative to positive, showing an obvious trend. As can be seen from ​ (b), as the cell survival ratio decreases, D at the centroid point A 1 and D A 2 both gradually increase, also showing an obvious trend. This is mainly because, as cells die, the permeability of the cell membrane changes, and its resistance characteristics gradually weaken at a test frequency of 7 MHz. The above results show the potential of the PT system in cell activity analysis.

[0085] The multi-modal flow particle detection method provided in this embodiment distinguishes the number (count), size, and type of the particles including cells, bacteria, proteins, exosomes, or microplastic particles. When the particles specifically refer to cells, their variants, clinical cycles, and cell cycles can be further distinguished; when the particles specifically refer to bacteria, their growth states can be further distinguished.

[0086] In a specific embodiment, different types of cancer cells - cervical cancer cells (Hela), breast cancer cells (MCF-7), and small cell lung cancer cells at different clinical cycles - early stage (PC-9), middle stage (NCI-1373), and late stage (NCI-1792) are used as test samples. The voltage waveforms during the test process are obtained, and the extreme value points of the time-frequency domain feature changes when the cells pass through the electrodes are extracted as feature values, including 6 time-domain features such as standard deviation, kurtosis, envelope frequency, envelope peak value, envelope valley value, and envelope amplitude, and 4 frequency features such as the frequency offset and amplitude change of two intrinsic modes. Each cell data is 1700. We use matlab to add labels to them and randomly sample them according to the ratio of training set: test set = 7:3 to obtain the training set and the test set for convolutional neural network training. Among them, the number of input neurons of the convolutional neural network is 10, the number of neurons in the 3 hidden layers is 100, 50, and 30 respectively, and the number of output neurons is 5. The trained convolutional neural network is used for cell recognition. The results are shown in ​ , and the classification accuracy rate can reach over 97.13%.

[0087] To show the importance of the newly introduced frequency domain features of the present invention, the frequency domain features are further removed, and only the time domain features are used for training. The results are as​ As shown, the accuracy rates of various types of cells have all significantly decreased, and the accuracy rate of 1792 cells has directly dropped to 81.12%, resulting in a large recognition error.

[0088] The application potential of the method provided by the specific embodiments of the present invention in label-free, high-throughput, and highly sensitive single-cell detection and classification is applicable to fields such as cancer diagnosis, drug screening, and immunoassay.

[0089] It should be noted that the above embodiments are not used to limit the protection scope of the present invention. Equivalent transformations or substitutions made on the basis of the above technical solutions all fall within the protection scope of the claims of the present invention.

Claims

1. A multi-modal flow particle detection method based on a PT-symmetric circuit, characterized in that, Including: Connect the conductive electrodes of the microfluidic chip to the PT-symmetric circuit, adjust the parameters of the PT-symmetric circuit to the singularity point, segment and process the output voltage waveform of the circuit to obtain time-domain characteristic information and frequency-domain characteristic information. When the microparticles pass between the electrodes, peaks or valleys appear in the characteristic information. Take the extreme points corresponding to the peaks or valleys as the characteristic values of the microparticles. The number of extreme points corresponds to the number of microparticles at the same time and can be used for counting; Divide the characteristic values of multiple pairs of microparticles and the corresponding microparticle labels into a training sample set and a validation sample set, and train a machine learning model through the training data set to obtain a microparticle prediction model; During application, input the extreme points of the time-domain characteristic information and / or frequency-domain characteristic information of the microparticles into the microparticle detection model to obtain the class prediction value, clinical cycle prediction value, variant prediction value or different state prediction values of the microparticles. The different states are the initial stage, growth stage or apoptosis stage of the biological microparticles when the microparticles are biological microparticles.

2. The multimodal flow particulate detection method based on a PT-symmetric circuit according to claim 1, wherein A method for separately counting and calculating the flux of m types of microparticles, including: Test m types of particles separately through the above-mentioned multi-modal flow particle detection method based on PT symmetric circuit, obtain the corresponding training sample sets, and perform machine learning training to obtain a particle prediction model. Then apply the obtained particle prediction model to the detection of samples containing m types of particles simultaneously. Input the characteristic changes caused by each particle when passing through the electrode into the particle prediction model for species prediction, and then count the number of the i-th type of particle within the time of n i ; Calculate the detection flux corresponding to the i-th particle as , where n i is the number of peaks or valleys corresponding to the i i-th particle within the time, where i = 1 - m; The total flux of the particles is K , .

3. The multimodal flow particulate detection method based on a PT-symmetric circuit according to claim 1, characterized in that Segment and process the voltage waveform to obtain time-domain characteristic information, including: Analyze the segmented voltage waveforms respectively through data processing software such as matlab or python, and construct time-domain characteristic information based on one or more of the standard deviation, kurtosis, envelope frequency, envelope peak value, envelope valley value, and envelope amplitude information of the voltage waveforms at different times obtained from the analysis.

4. The multimodal flow particle detection method based on a PT-symmetric circuit according to claim 1, wherein Segment and process the voltage waveform to obtain frequency-domain characteristic information, including: Perform Fourier transform on the segmented voltage waveform, and construct frequency-domain characteristic information based on one or more of the frequency offset and amplitude change information compared with the eigenmode when the microparticles pass through the microfluidic chip.

5. The multimodal flow particle detection method based on a PT-symmetric circuit according to claim 1, wherein The time for a single microparticle to pass between the electrodes is 0.1 - 10 ms.

6. The multimodal flow particulate detection method based on a PT-symmetric circuit according to claim 1, wherein, The PT-symmetric circuit is composed of LC circuits with gain and loss. Two LC circuits are connected through capacitive coupling, inductive coupling or equivalent transmission lines, and the parameters of the two LC circuits satisfy the PT-symmetric condition.

7. The multimodal flow particle detection method based on a PT-symmetric circuit according to claim 1 or 6, characterized in that The voltage waveform information is obtained by reading the voltage signal of the LC circuit with gain or loss of the PT-symmetric circuit through an oscilloscope, an FPGA circuit, or a single-chip microcomputer circuit.

8. The multimodal flow particle detection method based on a PT-symmetric circuit according to claim 1, characterized in that The microfluidic chip includes: A substrate; Microchannels, located on the substrate. The microchannels have grooves, and a measurement position is provided between the front and rear ends of the grooves. The microparticles flow from one end of the groove to the other end and pass through the measurement position; And conductive electrodes, the conductive electrodes are located on the substrate and are connected to the PT-symmetric circuit. The conductive electrodes are one pair or multiple pairs of electrodes. The ends of the two electrodes in each pair of electrodes are arranged opposite to each other or on the same side at the measurement position for measuring the impedance when the microparticles pass by.

9. The multimodal flow particulate detection method based on a PT-symmetric circuit according to claim 2, wherein A method for obtaining the size of microparticles, including: When the ratio of the distance between the electrodes, the pipe cross-sectional size to the particle size is less than 2, the injection pump injection flow rate Q , and the cross-sectional area near the electrodes S , calculate the flow velocity of the particles passing through the electrodes V = Q / S . Further combined with the time t for the particles to pass through the electrodes, the particle diameter size can be calculated X = V × t = Q × t / S . Introduce a coefficient a for correction. Therefore, the final particle diameter calculation method is X = V × a × t = Q × a × t / S .

10. The multimodal flow particulate detection method based on a PT-symmetric circuit according to claim 1, wherein Train a machine learning model through the training sample set to obtain a microparticle detection model. The machine learning model includes a convolutional neural network, a decision tree, an xgboost or a KNN algorithm.

11. The multimodal flow particle detection method based on a PT-symmetric circuit according to claim 1, characterized in that, The multi-modal flow particle detection method differentiates the particles including the quantity, size and type of cells, bacteria, proteins, exosomes or microplastic particles. When the particles specifically refer to cells, their variants, clinical cycles and cell cycles can be further differentiated; when the particles specifically refer to bacteria, their growth states can be further differentiated.

Images