Rare cell detection method, device and system

Abstract

The invention provides a rare cell detection method, device and system, and relates to the technical field of rare cell detection. The method comprises the following steps: acquiring a multi-frequency impedance signal of a to-be-detected cell sample; extracting a plurality of electrical phenotypic features based on the multi-frequency impedance signal; inputting the various electrical phenotypic characteristics of the cell sample to be detected into a pre-trained cell detection model, and outputting a rare cell detection result; wherein the multi-frequency impedance signal is obtained by the following steps of: injecting a prepared cell sample into the microfluidic device, realizing single cell arrangement of the cell sample to be detected through hydrodynamic focusing, and synchronously measuring the impedance signal of the single cell at different frequencies to obtain the multi-frequency impedance signal. The electrical phenotypic features are extracted from the multi-frequency impedance signals, the extracted electrical phenotypic features can reflect subtle differences of the cells under various electrical frequencies, the discrimination degree of a classification model is enhanced, accurate detection of the rare cells is achieved, the rare cells and other cells are accurately distinguished, and the classification accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of rare cell detection technology, and in particular to a method, apparatus and system for rare cell detection. Background Technology

[0002] Single-cell analysis has become a powerful tool in biological and medical research, enabling in-depth insights into cellular heterogeneity and precise characterization of individual cells. This approach is particularly important for identifying rare cell subpopulations, which, despite their scarcity, play a crucial role in the development and progression of diseases. Detecting these rare cells is essential for advancing medical diagnosis and treatment, including cancer prognosis, prenatal screening, and viral infection surveillance. For example, in clinical studies, counting and characterizing rare cells such as circulating tumor cells in peripheral blood can provide valuable information about disease progression in patients. However, the detection and analysis of rare cells present significant challenges due to the sheer number of other cell types, especially white blood cells (WBCs). Even after WBC removal, their numbers remain tens to hundreds of times greater than rare cells, making accurate detection difficult. Traditional immunofluorescence-based techniques rely on prior knowledge of specific antibodies and receptors and require complex staining procedures, which can cause irreversible cell damage and limit further analysis. Summary of the Invention

[0003] This invention provides a method, apparatus, and system for detecting rare cells, addressing the shortcomings of existing technologies that require fluorescent labeling, which can cause irreversible damage to cells. This invention avoids impacting cell viability while enabling precise detection of rare cells, accurately distinguishing them from other cells, and improving classification accuracy. The technical solution proposed by this invention is as follows: In a first aspect, the present invention provides a method for detecting rare cells, comprising: Acquire multi-frequency impedance signals from the cell sample to be tested; Multiple electrical phenotypic features are extracted based on the multi-frequency impedance signal; The various electrical phenotypic features of the cell sample to be tested are input into a pre-trained cell detection model, which outputs rare cell detection results. The multi-frequency impedance signal is obtained in the following way: The prepared cell sample is injected into a microfluidic device, and the cell sample is focused by fluid dynamics to achieve single-cell arrangement. The impedance signal of the single cell is measured synchronously at different frequencies to obtain the multi-frequency impedance signal.

[0004] Optionally, the multi-frequency impedance signal includes a first frequency band impedance signal, a second frequency band impedance signal, and a third frequency band impedance signal, wherein the first frequency band, the second frequency band, and the third frequency band belong to low frequency, medium frequency, and high frequency, respectively; The various electrical phenotypic features include electrical diameter, opacity X and opacity Y in the second frequency band, and opacity X and opacity Y in the third frequency band; The extraction of various electrical phenotypic features based on the multi-frequency impedance signal includes: Calculate the electrical diameter based on the impedance signal of the first frequency band; Calculate the opacity X and opacity Y in the second frequency band based on the impedance signal of the first frequency band and the impedance signal of the second frequency band; The opacity X and opacity Y in the third frequency band are calculated based on the impedance signal of the first frequency band and the impedance signal of the third frequency band.

[0005] Optionally, the cell detection model is trained in the following manner: Multiple electrical phenotypic features of each cell sample were extracted from the raw microfluidic impedance flow cytometry data of the training dataset and the test sample dataset. Multiple electrical phenotypic features of the training dataset are input into a machine learning model for training, and then applied to the test sample dataset to generate initial prediction results. The initial prediction results include: the initial predicted number of rare cells and the initial predicted number of other cells. The training dataset is randomly divided into a training subset and a test subset according to a preset ratio. The hyperparameters of the machine learning model are optimized based on the training subset, and the false positive rate and false negative rate of each prediction are calculated based on the test subset. The iteration is repeated until the kernel density estimation distribution of the false positive rate and false negative rate tends to be stable. The stabilized false positive rate and false negative rate are used as the target false positive rate and target false negative rate, respectively. The initial prediction results are corrected based on the target false positive rate and target false negative rate, and the corrected number of rare cells and the corrected number of other cells are output.

[0006] Optionally, the initial prediction results are corrected based on the target false positive rate and the target false negative rate according to the following formula, and the corrected number of rare cells and the corrected number of other cells are output: ; ; in, This represents the initially predicted number of rare cells. For the initially predicted number of other cells, This represents the corrected number of rare cells. For the corrected number of other cells, To target the false positive rate, The target is the false negative rate.

[0007] Optionally, the kernel density estimates of the false positive rate and false negative rate are considered stable if the peak error fluctuation range is less than a preset threshold.

[0008] Secondly, the present invention also provides a rare cell detection device, comprising the following modules: The signal acquisition module is used to acquire the multi-frequency impedance signal of the cell sample to be tested; wherein, the multi-frequency impedance signal is acquired by the following method: injecting the prepared cell sample into a microfluidic device, focusing the cell sample to be tested into single-cell arrangement through fluid dynamics, and simultaneously measuring the impedance signal of the single cell at different frequencies to obtain the multi-frequency impedance signal; The feature extraction module is used to extract various electrical phenotypic features based on the multi-frequency impedance signal; The cell classification module is used to input the various electrical phenotypic features of the cell sample to be tested into a pre-trained cell detection model and output the rare cell detection results.

[0009] Thirdly, the present invention also provides a rare cell detection system, comprising: a microfluidic device, an impedance analyzer, and a rare cell detection device as described in the second aspect; A microfluidic device includes a microfluidic channel and a sensing electrode; the microfluidic channel is used to focus a sample of cells to be tested into single-cell arrangement through hydrodynamics. An impedance analyzer, connected to the sensing electrode, is used to simultaneously measure the impedance signal of a single cell at different frequencies to obtain a multi-frequency impedance signal, which is then used by a rare cell detection device to perform cell detection and obtain rare cell detection results.

[0010] Fourthly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the rare cell detection method as described in the first aspect above.

[0011] Fifthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the rare cell detection method as described in the first aspect above.

[0012] In a sixth aspect, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the rare cell detection method as described in the first aspect above.

[0013] Based on the above technical solution, the beneficial effects of the present invention compared with the prior art are as follows: The rare cell detection method, apparatus, and system provided by this invention achieve single-cell arrangement through a microfluidic device, ensuring independent measurement of the impedance signal of each cell and avoiding interference from cell aggregation. Single-cell resolution makes the detection results more accurate, aiding in the discovery of rare cells (such as circulating tumor cells, CTCs). Multi-frequency impedance signal extraction through multi-frequency measurement comprehensively captures the electrical properties of cells, improving the ability to distinguish rare cells from other cells. Extracting electrical phenotypic features from the multi-frequency impedance signals reflects subtle differences in cells at various electrical frequencies, enhancing the discriminative power of the classification model, achieving precise detection of rare cells, accurately distinguishing rare cells from other cells, and improving classification accuracy. The detection method based on electrical properties eliminates the need for cell staining or labeling, avoiding any impact on cell viability.

[0014] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is one of the flowcharts illustrating the rare cell detection method provided by the present invention.

[0018] Figure 2a This is the second flowchart illustrating the rare cell detection method provided by the present invention.

[0019] Figure 2b This is a schematic diagram of the original impedance signal at different frequencies provided by the present invention.

[0020] Figure 3 This is a schematic diagram of the average profile coefficients of each cancer cell type and Jurkat cells at a series of frequencies provided by the present invention; wherein each frequency is performed in three independent experiments.

[0021] Figure 4a , Figure 4b , Figure 4c , Figure 4d The characteristic distribution maps and confusion matrices of MDA-MB-231 cells and Jurkat cells under different characteristics are shown.

[0022] Figure 5a , Figure 5b , Figure 5c , Figure 5d The characteristic distribution maps and confusion matrices of A549 cells and Jurkat cells under different characteristics are shown.

[0023] Figure 6a , Figure 6b , Figure 6c , Figure 6d The distribution maps and confusion matrices of HELA cells and Jurkat cells under different characteristics are shown.

[0024] Figure 7a , Figure 7b , Figure 7c , Figure 7d The characteristic distribution maps and confusion matrices of MDA-MB-231 cells and PBMCs under different characteristics are shown.

[0025] Figure 8a , Figure 8b , Figure 8c , Figure 8d The characteristic distribution maps and confusion matrices of A849 cells and PBMCs under different characteristics are shown.

[0026] Figure 9a , Figure 9b , Figure 9c , Figure 9d The characteristic distribution maps and confusion matrices of HELA cells and PBMCs under different characteristics are shown.

[0027] Figure 10a , Figure 10b , Figure 10c , Figure 10d This image shows the differentiation results between HELA cells and A1049 cells under different quantities of electrical phenotypic characteristics at medium, low, and high frequencies.

[0028] Figure 11a , Figure 11b , Figure 11c , Figure 11d This image shows the differentiation results between MDA-MB-231 cells and HELA cells under different electrical phenotypic characteristics at medium, low, and high frequencies.

[0029] Figure 12a , Figure 12b , Figure 12c , Figure 12d The image shows the differentiation results between MDA-MB-231 cells and A1249 cells under different electrical phenotypic characteristics at medium, low and high frequencies.

[0030] Figure 13a This is a schematic diagram of the training process.

[0031] Figure 13b The graph shows the results of rare cell detection with and without prediction post-correction.

[0032] Figure 13c Comparison of flow cytometry results and the results obtained using this method when A549 cells and Jurkat cells are mixed.

[0033] Figure 13d A comparison of flow cytometry results and the results obtained using this method when MDA-MB-231 cells are mixed with PBMCs.

[0034] Figure 14 This is a schematic diagram of the rare cell detection device provided by the present invention.

[0035] Figure 15 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0037] The following is combined Figures 1-15 This invention describes a method, apparatus, and system for detecting rare cells. In this invention, rare cells refer to cells that constitute less than 10% of a cell sample.

[0038] Reference Figure 1 As shown, the method includes the following: S110. Obtain the multi-frequency impedance signal of the cell sample to be tested.

[0039] In this invention, the multi-frequency impedance signals of the cell samples to be tested and the cell samples used for training and testing the cell detection model are obtained by the following method: the prepared cell samples are injected into a microfluidic device, and the cell samples to be tested are aligned into single cells by focusing through fluid dynamics. The impedance signals of the single cells are measured synchronously at different frequencies to obtain the multi-frequency impedance signals.

[0040] The microfluidic device includes an impedance-based microfluidic cytometry chip, which serves as the detection platform. The chip incorporates a hydrodynamic focusing structure, precisely controlling the fluid flow rate to achieve single-cell alignment of cells in the sample as they flow through the detection area. The chip consists of two main parts: polydimethylsiloxane (PDMS) microfluidic channels and sensing electrodes (such as three coplanar gold electrodes). The microfluidic channels have a width ranging from 10-80 μm and a height ranging from 10-50 μm (e.g., 30 μm wide, 20 μm high). Each electrode has a width and spacing ranging from 10-50 μm (e.g., 20 μm wide, 20 μm spacing). It should be noted that this is merely an example; those skilled in the art can configure it according to actual needs. The channels are fabricated on a silicon wafer using standard photolithography techniques and a negative photoresist. The PDMS prepolymer mixture (monomer to curing agent weight ratio of 10:1) is thoroughly mixed, degassed under vacuum, and cured at 60°C for 2 hours. The cured PDMS layer is removed from the mold and perforated to form inlets and outlets. After oxygen plasma treatment of the PDMS channel surface and electrode substrate, the components are precisely aligned and bonded. The assembled device is baked at 85°C for 20 minutes to ensure a stable seal, completing the preparation process. The above monomer to curing agent weight ratio, curing temperature, baking temperature, and baking time are only examples; those skilled in the art can set them according to actual needs. The monomer to curing agent weight ratio, curing temperature, baking temperature, and baking time can be set to a range of 5:1-15:1, 20-120°C, 40-120°C, and 10-1000 minutes.

[0041] The impedance analyzer operates in differential mode and is connected to the sensing electrodes of the microfluidic device, recording impedance data with a 3V input voltage. A pressure pump controls fluid flow, applying a positive pressure of 180 mbar at the sample inlet to drive cells through the microfluidic channel. The input voltage and positive pressure at the sample inlet range from 0-10V and 20-2000 mbar, respectively.

[0042] Specifically, refer to Figure 2a As shown, the rare cell analysis process is as follows: A sample of cells is fed into a microfluidic device for single-cell impedance characterization. Hydrodynamic focusing within the channel allows the cell to pass through the center. Sensing electrodes embedded in a glass substrate ensure high sensitivity and stability of the impedance measurement. The middle electrode is connected to an AC voltage input, while the two side electrodes are connected to a differential transimpedance amplifier (TA), which transmits the signal to an impedance analyzer for further processing. When a cell passes between the first two electrodes, it displaces the surrounding conductive medium, generating a peak differential signal, while an opposite peak is generated when it passes through the remaining electrodes. These differential signals are simultaneously demodulated at multiple frequencies to capture the cell's biophysical properties.

[0043] In the single-cell array region, the impedance signal of each single cell is simultaneously measured at multiple preset frequencies (e.g., 100 kHz to 50 MHz) using an electrode array integrated on the chip. The impedance response at different frequencies reflects the response characteristics of different structures such as the cell membrane, cytoplasm, and nucleus to an electric field. The acquired raw impedance signals are amplified, filtered, and preprocessed before being converted into digital signals for storage, forming a multi-frequency impedance signal. The impedance signal preprocessing employs a recursive baseline correction method to dynamically update the baseline value and reduce noise. Those skilled in the art can use existing peak detection algorithms to identify positive and negative peaks while ensuring their significance and independence. Peaks that do not reach the preset transition time and amplitude threshold are filtered to improve data accuracy. These preprocessing steps enable the extraction of multi-frequency impedance peaks, facilitating the calculation of opacity and cell diameter as classification features. This comprehensive workflow provides a solid foundation for subsequent machine learning predictions, ensuring high reliability and accuracy.

[0044] S120. Extract various electrical phenotypic features based on the multi-frequency impedance signal.

[0045] Based on multi-frequency impedance signals, various electrical phenotypic features are extracted, including but not limited to: electrical diameter, opacity at different frequencies, impedance amplitude and phase, dispersion characteristics, and statistical features. Impedance amplitude and phase represent the changes in impedance amplitude and phase at different frequencies, reflecting the dielectric properties of the cell. Dispersion characteristics describe the dispersion behavior of impedance as it changes with frequency, extracting characteristic frequency points. Statistical features, such as mean, variance, and skewness, describe the distribution characteristics of the impedance signal.

[0046] S130. Input the various electrical phenotypic features of the cell sample to be tested into the pre-trained cell detection model and output the rare cell detection results.

[0047] The aforementioned cell detection model can be trained using a machine learning model based on Support Vector Machine (SVM). Alternatively, it can be trained using models such as Quadratic Discriminant Analysis (QDA), Random Forest, Logistic Regression, K-Nearest Neighbors (KNN), and Naive Bayes. SVM performs exceptionally well in handling high-dimensional data (such as electrophysiological features), especially suitable for small datasets. It is insensitive to noise and outliers, making it suitable for the noise that may exist in biomedical data. It handles non-linearly separable data through kernel functions (such as the RBF kernel), adapting to the complex differences in features between rare cells and other cells. The goal of training an SVM is to find an optimal hyperplane that separates data points of different classes (such as tumor cells and white blood cells) while maximizing the margin between the two classes of data points and the hyperplane.

[0048] The cell detection model can be obtained by scaling the electrical phenotypic features extracted from the multi-frequency impedance signal to the same range (e.g., [0,1]) to avoid the impact of feature dimension differences on model performance. The hyperparameters of the machine learning model (e.g., regularization parameter C, kernel function parameter γ) are optimized through cross-validation (e.g., grid search) to finally obtain the pre-trained cell detection model.

[0049] The extracted electrophenotypic features of the cell samples to be tested are used as input to a pre-trained cell detection model. Based on the input features, the model outputs rare cell detection results, presented in the form of probability or category labels.

[0050] This invention provides a rare cell detection method that utilizes a microfluidic device to achieve single-cell arrangement, ensuring independent measurement of the impedance signal for each cell and avoiding interference from cell aggregation. Single-cell resolution leads to more accurate detection results, aiding in the discovery of rare cells (such as circulating tumor cells, CTCs). Multi-frequency impedance signal extraction through multi-frequency measurements comprehensively captures the electrical properties of cells, improving the ability to distinguish rare cells from other cells. Extracting electrical phenotypic features from the multi-frequency impedance signals reflects subtle differences in cells at various electrical frequencies, enhancing the discriminative power of the classification model, achieving precise detection of rare cells, accurately distinguishing them from other cells, and improving classification accuracy. The entire process, from sample injection to result output, is highly automated, reducing manual intervention and improving detection efficiency. The microfluidic device design allows for the simultaneous processing of multiple cells, suitable for rapid screening of large-scale samples. The detection method based on electrical properties eliminates the need for cell staining or labeling, avoiding any impact on cell viability. It can detect rare cells in the blood, potentially enabling early diagnosis of tumors.

[0051] In an optional embodiment, the present invention extracts key features such as electrical diameter and electrical properties of different intracellular structures by combining low-frequency, mid-frequency, and high-frequency impedance signals. These features can comprehensively reflect the electrical properties of cells, improve the accuracy of classifying rare cells from other cells, enhance the robustness of the model, and reduce computational complexity.

[0052] The aforementioned multi-frequency impedance signals include a first-band impedance signal, a second-band impedance signal, and a third-band impedance signal. The first, second, and third bands belong to low frequency, mid frequency, and high frequency, respectively. The first band (low frequency) mainly reflects cell size. The second (mid frequency) and third (high frequency) bands reflect the electrical properties of different structures within the cell.

[0053] The various electrical phenotypic characteristics include electrical diameter, opacity X and opacity Y in the second frequency band, and opacity X and opacity Y in the third frequency band. Electrical diameter reflects the equivalent electrical size of the cell at low frequencies and is related to the actual physical size of the cell. Opacity X and opacity Y in the second and third frequency bands characterize the electrical properties of different structures within the cell and are key features for distinguishing rare cells from other cells.

[0054] The extraction of various electrical phenotypic features based on the multi-frequency impedance signal described in S120 above includes: Calculate the electrical diameter based on the impedance signal of the first frequency band. ; Calculate the opacity in the second frequency band based on the impedance signal of the first frequency band and the impedance signal of the second frequency band. With Opacity ; Calculate the opacity in the third frequency band based on the impedance signal of the first frequency band and the impedance signal of the third frequency band. With Opacity The calculation formula is as follows:

[0055] in and These are the current amplitudes of the impedance signal in the first frequency band. The absolute values ​​of the positive and negative peaks; G is the calibration coefficient used to convert the current amplitude of the impedance signal into the electrical diameter of the cell. These are the real and imaginary parts of the impedance signal in the first frequency band, respectively. These are the real and imaginary parts of the impedance signal in the second frequency band, respectively. These are the real and imaginary parts of the impedance signal in the third frequency band, respectively.

[0056] This invention extracts the overall electrical properties (electrical diameter) and internal structural characteristics of cells by combining low-frequency, mid-frequency, and high-frequency impedance signals. Rare cells exhibit significant differences in electrical phenotypic features compared to other cells. Multi-band feature extraction can capture these differences more comprehensively, thereby improving classification accuracy. Low-frequency signals are more robust to noise, while high-frequency signals provide more refined features. Through multi-band fusion, the model can maintain stable performance under different noise levels. The electrical properties of rare cells and other cells exhibit heterogeneity across different frequency bands; multi-band feature extraction can better adapt to this heterogeneity, improving the model's generalization ability. By using multi-band feature extraction, the original multi-frequency impedance signals are transformed into more discriminative electrical phenotypic features, reducing the computational complexity of subsequent machine learning models.

[0057] The optimal frequencies used to distinguish rare cells from other cells, namely the first frequency band, the second frequency band, and the third frequency band, are determined in the following manner: Within a predefined frequency range, multiple excitation frequencies are scanned at fixed intervals to excite cell samples; the predefined frequency range covers low, medium, and high frequencies. Based on the impedance response data of cell samples collected at different frequencies, cell discrimination indexes (such as average profile coefficient) are calculated for each frequency. Based on the cell discrimination indexes, multiple key frequencies capable of distinguishing rare cells from other cells are selected from the predefined frequency range to obtain the first frequency band, the second frequency band, and the third frequency band.

[0058] Specifically, using microfluidic impedance flow cytometry, impedance response data of the two cell samples to be distinguished were measured at each frequency. The corresponding profile coefficients were calculated based on the impedance response data. For each frequency, three measurements were performed on the two cell types to be distinguished, and the average profile coefficients obtained from the three measurements were averaged to obtain the mean profile coefficient for that frequency. Key frequencies with an average profile coefficient greater than a preset coefficient (e.g., 0.7) were selected from a predefined frequency range; these frequencies belonged to low, medium, and high frequencies, respectively, thus obtaining the optimal frequencies used to distinguish rare cells from other cells.

[0059] This invention uses tumor cells as rare cells and white blood cells as other cells for illustration. To determine the optimal frequency for distinguishing tumor cells from white blood cells (WBCs), excitation frequencies were scanned at 2 MHz intervals within the range of 6 MHz to 30 MHz. Human immortalized T lymphocytes (Jurkat cells) were used as the white blood cell model. To quantitatively assess the distinguishing ability across the entire spectral range, the average silhouette coefficient (reference) of three independent experiments was calculated. Figure 3 (As shown). Figure 3From left to right, the figures show statistics on MDA-MB-231 and Jurkat cells, statistics on A549 and Jurkat cells, and statistics on Hela and Jurkat cells. The results confirm that specific excitation frequencies can effectively distinguish cancer cells from white blood cells (WBCs). Among the tested frequencies, 10 MHz (mid-frequency) and 28 MHz (high-frequency) consistently effectively distinguished all three cancer cell lines, highlighting their universal potential in this application. Therefore, to achieve comprehensive impedance-based detection of rare cancer cells in the WBC population, three key frequencies were selected: 500 kHz (low-frequency), 10 MHz (mid-frequency), and 28 MHz (high-frequency). These frequencies can extract five different electrical phenotypic features: the electrical diameter at low frequency reflects cell size, and the opacities X and Y at mid-frequency and high frequency assess the electrical properties of different intracellular structures, respectively.

[0060] Reference Figure 2b As shown, the original impedance signals at the three key frequencies mentioned above are illustrated. The diagram highlights the real and imaginary parts of the original impedance signal. At low frequencies (kilohertz, e.g., 500 kHz), the cell membrane can be considered an electrical insulator, allowing cell size to be inferred from the amplitude of the differential current. As the frequency increases to mid-frequency (several megahertz, e.g., 10 MHz), the cell membrane capacitance is short-circuited, making cytoplasmic conductivity the dominant factor. At high frequencies (tens of megahertz, e.g., 28 MHz), the nuclear membrane capacitance is further short-circuited, enabling characterization of nuclear properties. To reduce the influence of cell size on impedance information, the opacity at mid- and high frequencies is calculated by normalizing the differential current to the low-frequency signal.

[0061] The electrical diameter (Diameter) at low frequencies is calculated as: G multiplied by (current amplitude at 500kHz) The cube root of the equation is given, where G is a calibration coefficient and the current amplitude at 500 kHz. It is determined by the absolute value of the positive peak. and absolute value of negative peak The sum is obtained by dividing by 2.

[0062]

[0063] OpacityX at mid-frequency 10MHz The numerator is the real part of the intermediate frequency current signal. ) multiplied by the real part of the low-frequency current signal ( ), plus the imaginary part of the intermediate frequency current signal ( Multiply by the imaginary part of the low-frequency current signal ( The denominator is the real part of the low-frequency current signal. The square of ) plus the imaginary part of the low-frequency current signal ( The square of ).

[0064]

[0065] Opacity Y at mid-frequency 10MHz The numerator is the imaginary part of the intermediate frequency current signal. ) multiplied by the real part of the low-frequency current signal ( ), minus the real part of the intermediate frequency current signal ( Multiply by the imaginary part of the low-frequency current signal ( The denominator is the real part of the low-frequency current signal. The square of ) plus the imaginary part of the low-frequency current signal ( The square of ).

[0066]

[0067] OpacityX at high frequencies 28MHz The molecule is the real part of the high-frequency current signal. ) multiplied by the real part of the low-frequency current signal ( ), plus the imaginary part of the high-frequency current signal ( Multiply by the imaginary part of the low-frequency current signal ( The denominator is the real part of the low-frequency current signal. The square of ) plus the imaginary part of the low-frequency current signal ( The square of ).

[0068]

[0069] Opacity Y at high frequencies 28MHz The molecule is the imaginary part of the high-frequency current signal. ) multiplied by the real part of the low-frequency current signal ( ), minus the real part of the high-frequency current signal ( Multiply by the imaginary part of the low-frequency current signal ( The denominator is the real part of the low-frequency current signal. The square of ) plus the imaginary part of the low-frequency current signal ( The square of ).

[0070]

[0071] In one specific embodiment, the present invention utilizes five electrical phenotypic features extracted by microfluidic impedance-based flow cytometry (μIFC) to demonstrate an effective distinction between three cancer cell types and white blood cells (WBCs). Figure 4a , Figure 4b , Figure 4c , Figure 4d The distribution maps and confusion matrices of MDA-MB-231 cells and Jurkat cells under different characteristics are shown. Figure 5a , Figure 5b , Figure 5c , Figure 5d The characteristic distribution maps and confusion matrices of A549 cells and Jurkat cells under different characteristics are shown; Figure 6a , Figure 6b , Figure 6c , Figure 6d The characteristic distribution maps and confusion matrices of HELA cells and Jurkat cells under different characteristics are shown. Figure 7a , Figure 7b , Figure 7c , Figure 7d The distribution maps and confusion matrices of MDA-MB-231 cells and PBMCs under different characteristics are shown. Figure 8a , Figure 8b , Figure 8c , Figure 8d The diagram shows the characteristic distribution maps and confusion matrices of A849 cells and PBMCs under different characteristics. Figure 9a , Figure 9b , Figure 9c , Figure 9d The diagrams show the characteristic distribution maps and confusion matrices of HELA cells and PBMCs under different characteristics. The first row of these diagrams represents the characteristic distribution maps, and the second row represents the confusion matrices. The characteristic distribution maps highlight the low-frequency, medium-frequency, and high-frequency data, as well as the comprehensive 3D scatter plot in the first row.

[0072] At low frequencies, Jurkat cells exhibit a smaller size than cancer cells, which is reflected in their lower impedance amplitude. At mid-frequency frequencies, cancer cells have lower opacity (X) but higher opacity (Y) compared to white blood cells (WBCs). At high frequencies, both X and Y opacities of cancer cells are higher than those of WBCs. These unique opacity characteristics allow for clear differentiation between cancer cells and WBCs. To further enhance visualization, impedance measurements at each frequency were converted to a one-dimensional space and combined into a three-dimensional representation. This transformation effectively clustered the data, demonstrating that the extracted five electrophenotypic features provide a solid foundation for distinguishing cancer cells from WBCs.

[0073] Furthermore, features extracted at different frequencies are used as input to the machine learning model to evaluate classification accuracy. Support vector machines are used as an example. Figures 4a-9d The confusion matrix shown in the second row directly illustrates the classification performance, where each column represents the predicted class and each row represents the true class. At low frequencies, the classification accuracy based on the electrical diameter parameter exceeds 80%, but this accuracy is limited by the partial overlap in size between cancer cells and WBCs. At mid- or high frequencies, using two opacity parameters, the accuracy is approximately 97%–99%. (See reference...) Figure 4d , Figure 5d , Figure 6d As shown, when all five electrical phenotypic features are used in combination, the classification accuracy for distinguishing all three cancer cell types (MDA-MB-231 cells, A549 cells, and HELA cells) from Jurkat cells is improved to over 99%, demonstrating the enhanced performance of the integrated multi-frequency impedance signal of this invention.

[0074] Under clinical conditions, white blood cells (WBCs) may exist in multiple subsets. To evaluate the capability of the cell detection model of this invention, its ability to distinguish between three cancer cell types (MDA-MB-231 cells, A549 cells, and HELA cells) and peripheral blood mononuclear cells (PBMCs), which are composed of lymphocytes and monocytes, was tested. (Refer to...) Figures 7a to 9dAs shown, the five electrophenotypic features across three frequency ranges effectively distinguished lymphocyte and monocyte populations in PBMCs from various cancer cell types. Due to the smaller size of PBMCs compared to Jurkat cells, classification accuracy was improved at lower frequencies. When all five electrophenotypic features were used in combination, the classification accuracy for all three cancer cell lines exceeded 99%, demonstrating the robustness and future potential of the method. Compared to models relying on a single parameter or two parameters at a single frequency, integrating all five electrophenotypic features significantly improved classification accuracy. This result highlights the effectiveness of the method in accurately distinguishing different cancer cell types.

[0075] Without loss of generality, by employing five electrical phenotypic characteristics, the method of the present invention can distinguish not only various cancer cells from Jurkat cells and PBMCs, but also various other cancer cells. To evaluate the capability of the cell detection model of the present invention, its ability to distinguish various cancer cells was also tested. 10a. Figure 10b , Figure 10c , Figure 10d A schematic diagram showing the differentiation results of HELA cells and A549 cells under different quantities of electrical phenotypic characteristics at medium, low and high frequencies; Figure 11a , Figure 11b , Figure 11c , Figure 11d This diagram illustrates the differentiation results between MDA-MB-231 cells and HELA cells under different quantities of electrical phenotypic characteristics at medium, low, and high frequencies. Figure 12a , Figure 12b , Figure 12c , Figure 12d This diagram illustrates the differentiation results between MDA-MB-231 cells and A549 cells under different numbers of electrophenotypic features at medium, low, and high frequencies. When all five electrophenotypic features are used in combination, the classification accuracy of the three cancer cell combinations exceeds 90%.

[0076] In an alternative embodiment, despite the high accuracy achieved by the machine learning algorithm, misclassification remains a problem, especially in rare cell detection. Even a small false positive rate (FPR) or false negative rate (FNR) can significantly impact the identification of rare cells. To address this issue, the present invention implements a post-prediction correction strategy to recalibrate misclassified cells. (Refer to...) Figure 13a As shown, firstly, five electrical phenotypic features were extracted from the raw μIFC data of the training dataset and the test sample dataset. Then, the electrical phenotypic features from the training dataset were input into a machine learning model (using a support vector machine model as an example) and applied to the test sample dataset to generate initial prediction results for two cell populations. and Then, the original training dataset was randomly divided into a training subset and a test subset. The training subset was used for model optimization, and the test subset was used for performance evaluation. The fully labeled test subset was used to compute the FPR and FNR for each prediction. After several iterations, the kernel density estimates of FPR and FNR showed obvious peaks, indicating stable values ​​in the error distribution. Subsequently, these stable FPR and FNR (i.e., the target false positive rate) were... and target false negative rate The results were used to adjust the initial predictions, resulting in corrected counts for both cell types: and This improves classification accuracy. The ratio of the training subset to the test subset can be set according to actual needs, such as 4:1 or 5:1. The number of iterations can also be set according to actual needs, such as 200 times.

[0077] Figure 13a Taking tumor cells and white blood cells as examples, the above five electrophenotype characteristics are: , , , , The generated initial prediction result is and The corrected result is and . This represents the initial predicted number of tumor cells. This is the initial predicted white blood cell count. This represents the corrected number of tumor cells. This is the corrected white blood cell count.

[0078] Specifically, the training process of the cell detection model is as follows: S210. Extract multiple electrical phenotypic features for each cell sample from the raw microfluidic impedance flow cytometry data of the training dataset and the test sample dataset.

[0079] Information is extracted from the raw microfluidic impedance flow cytometry data (i.e., the multi-frequency impedance signals mentioned above) of the training dataset and the test sample dataset. Cell-related information is obtained by measuring the impedance changes generated when cells flow in the microfluidic channel. For each cell sample, multiple electrical phenotypic features are extracted. These reflect the physical and electrical properties of the cells and form the basis for subsequent model training and prediction. The extraction process is described above for S120 and will not be repeated here.

[0080] S220. Input the various electrical phenotypic features of the training dataset into the machine learning model for training, and apply it to the test sample dataset to generate an initial prediction result. The initial prediction result includes: the initial predicted number of rare cells and the initial predicted number of other cells.

[0081] A machine learning model was chosen as the base model. Multiple electrical phenotypic features from the training dataset were input into the machine learning model for training. During training, the model learned the feature patterns of different cell types (rare cells and other cells) and established classification boundaries.

[0082] The trained model is applied to the test sample dataset to generate initial predictions. These initial predictions include the initial predicted number of rare cells and the initial predicted number of other cells. This is a preliminary classification and counting of unknown samples based on the feature patterns learned by the model on the training dataset.

[0083] S230. The training dataset is randomly divided into a training subset and a test subset according to a preset ratio. The hyperparameters of the machine learning model are optimized based on the training subset. The false positive rate and false negative rate of each prediction are calculated based on the test subset. The iteration is repeated until the kernel density estimation distribution of the false positive rate and false negative rate tends to stabilize.

[0084] The training dataset is randomly divided into a training subset and a test subset according to a preset ratio. The training subset is used to optimize the hyperparameters of the machine learning model, such as regularization parameters and kernel function parameters; the test subset is used to evaluate the model's performance. Based on the training subset, the hyperparameters of the machine learning model are optimized using methods such as grid search and random search to improve the model's classification performance. Based on the test subset, the false positive rate (the probability of misclassifying other cells as rare cells) and false negative rate (the probability of misclassifying rare cells as other cells) are calculated for each prediction. The false positive rate and false negative rate are important indicators for evaluating model performance, reflecting the errors made by the model in the classification process. To optimize the performance of the classifier in the machine learning model, this invention performs a grid search with 5-fold cross-validation to adjust key hyperparameters, including regularization parameters and kernel function parameters.

[0085] Repeat the hyperparameter cross-validation process described above until the kernel density estimates of the false positive and false negative rates exhibit a clear peak characteristic in their error distributions. This indicates that these error indicators are relatively concentrated around specific values, demonstrating stable distribution. Kernel density estimation is a nonparametric statistical method used to estimate the probability density function of a random variable. When the kernel density estimates of the false positive and false negative rates tend to stabilize, it indicates that the model's performance has reached a relatively stable state. The kernel density estimates of the false positive and false negative rates are considered stable if the peak error fluctuation range is less than a preset threshold. Those skilled in the art can set the preset threshold according to actual needs, such as 2%.

[0086] S240. The stabilized false positive rate and false negative rate are used as the target false positive rate and target false negative rate, respectively. The initial prediction results are corrected based on the target false positive rate and target false negative rate, and the corrected number of rare cells and the corrected number of other cells are output.

[0087] The initial prediction results can be corrected using the following formulas, outputting the corrected number of rare cells and the corrected number of other cells:

[0088] in, This represents the initially predicted number of rare cells. For the initially predicted number of other cells, This represents the corrected number of rare cells. For the corrected number of other cells, To target the false positive rate, The target is the false negative rate.

[0089] Taking the distinction between tumor cells and white blood cells as an example, the above formula can be converted to:

[0090] in, This represents the initial predicted number of tumor cells. This is the initial predicted white blood cell count. This represents the corrected number of tumor cells. This is the corrected white blood cell count. To target the false positive rate, The target is the false negative rate.

[0091] This invention extracts multiple electrophenotypic features to more comprehensively describe cell characteristics, thereby improving the model's ability to distinguish different cell types. Iterative optimization based on the false positive and false negative rates of the test subset allows the model to better adapt to the actual data distribution, reducing classification errors and improving prediction accuracy. Correcting the initial prediction results based on the target false positive and false negative rates further reduces classification errors and improves the reliability of the prediction results. By dividing the training dataset into training and test subsets and performing multiple iterative optimizations, the model learns more generalized feature patterns, enhancing its adaptability to different datasets and improving its generalization ability. When the kernel density estimation distributions of the false positive and false negative rates tend to stabilize, the model's performance reaches a relatively stable state, providing reliable prediction results in practical clinical applications. By outputting the corrected number of rare cells and other cell counts, doctors can make more accurate diagnostic and treatment decisions based on these results, improving the feasibility and effectiveness of clinical applications.

[0092] To verify the reliability of the correction strategy of this invention, a mixed sample was generated from the labeled dataset, with HeLa cells as the rare cells and Jurkat cell samples as the WBCs, and the proportion of rare cells was set to 1%, 5%, and 10%. A total of 50 randomized experiments were conducted to compare the prediction results with and without post-correction. Figure 13b As shown, without correction (i.e., without post-prediction correction), rare cell counts are overestimated, especially in the 1% pooled sample. However, post-prediction correction effectively brings the predicted rare cell values ​​closer to the true values. These results confirm that the correction strategy proposed in this invention can improve prediction accuracy under different rare cell proportions.

[0093] To evaluate the detection performance of the cell detection model of this invention, mixed samples containing less than 10% rare cells were prepared. Fluorescence-based flow cytometry was used as the validation benchmark. The results showed strong consistency between the two methods even at a rare cell proportion of 1%, highlighting the accuracy of the cell detection model of this invention in rare cell detection. Figure 13c An example is shown where A549 cells were mixed with Jurkat cells, and flow cytometry detected 1.15%, while the cell detection model of this invention predicted 1.24%. Similarly, Figure 13dThe study showed that a mixture of MDA-MB-231 cells and PBMCs yielded 5.63% of the rare cell count, as detected by flow cytometry, compared to the predicted 5.51%. Notably, the method described in this invention is label-free and requires no immunostaining. This capability highlights the broader potential of the method to achieve accurate, high-throughput detection of rare cells at single-cell resolution.

[0094] Accurate detection of rare cancer cells in heterogeneous populations is crucial, as misclassification can lead to misdiagnosis and inappropriate treatment, posing significant risks to patients. Despite achieving high classification accuracy, rare cell detection remains challenging, with even a 1% misclassification rate potentially resulting in significant errors. To address this issue, the predictive post-correction strategy of this invention improves the reliability of detection.

[0095] The cell samples in this invention are prepared in the following manner: Breast cancer (MDA-MB-231) cells, lung cancer (A549) cells, cervical cancer (HeLa) cells, and human T lymphocytes (Jurkat) were cultured at 37°C and 5%... Under controlled conditions, Jurkat cells were cultured in suspension medium on RPMI-1640 containing 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin (PS). Adherent cancer cell lines (MDA-MB-231, A549, and HeLa) were cultured in high-glucose DMEM medium containing 10% FBS and 1% PS. Adherent cells were digested with 0.25% trypsin every two days and washed with Duchenne phosphate-buffered saline (DPBS). Primary human peripheral blood mononuclear cells (PBMCs) were purchased from STEMCELL Technologies (CAT #70025.1) and thawed according to the manufacturer's handling protocol.

[0096] To prepare mixed cell samples, cancer cells and white blood cells (WBCs) were collected separately from culture flasks, washed once with DPBS, and resuspended in 1.5 wt% polyethylene oxide (PEO, molecular weight 600 kDa, Sigma-Aldrich) solution. Cell density was measured using a ThermoFisher Countess II and mixed proportionally to ensure that cancer cells comprised less than 10% of the total cell count.

[0097] To verify the proportion of cancer cells using fluorescence-based flow cytometry, both cancer cells and white blood cells (WBCs) were stained nuclearly with 1 μM Hoechst 33342 (Beyotime). Additionally, cancer cells were selectively labeled with 5 μM red fluorescent membrane dye DID (Beyotime) for identification. Immunostaining was performed at 37°C for 10 min, followed by three washes with DPBS and centrifugation at 1200 rpm for 3 min. The final mixed cell suspension was aliquoted into two fractions in 1.5 wt% PEO for analysis using this method and conventional flow cytometry (MACSQuant Analyzer), respectively, to assess the concordance of the assays.

[0098] Compared to conventional flow cytometry, the cell detection model of this invention exhibits high consistency in detecting rare cancer cells in WBC populations ranging from 1% to 10%. Although circulating tumor cells typically comprise less than 1% of blood samples, most detection methods rely on WBC removal and cell concentration, which still suffer from interference from residual WBCs. The method of this invention provides a potential solution for analyzing rare cancer cells in concentrated samples. Furthermore, this method paves the way for future research on the characterization of rare cells under various pathological conditions, broadening its potential and scope of biomedical applications.

[0099] To evaluate the cell detection model of this invention, the ability of this method to distinguish between three cancer cell types (MDA-MB-231 cells, HELA cells, and A549 cells) and peripheral blood mononuclear cells (PBMCs) was tested. Under different proportions of rare cells, this method showed strong consistency with flow cytometry, highlighting the accuracy of this invention in rare cell detection. For example, when MDA-MB-231 cells were mixed with PBMCs, flow cytometry detected 5.63%, while the method of this invention predicted 5.51%.

[0100] The ability of this method to distinguish between three cancer cell types (A549 cells, MDA-MB-231 cells, and HELA cells) and Jurkat cells was also tested. Even at a rare cell proportion of 1%, this method showed strong consistency with flow cytometry, highlighting the accuracy of the method in rare cell detection. For example, when A549 cells and Jurkat cells were mixed, flow cytometry detected 1.15%, while the method of this invention predicted 1.24%.

[0101] The rare cell detection device provided by the present invention will be described below. The rare cell detection device described below can be referred to in correspondence with the rare cell detection method described above.

[0102] The rare cell detection device provided by this invention refers to... Figure 14 As shown, it includes: The signal acquisition module 310 is used to acquire the multi-frequency impedance signal of the cell sample to be tested; wherein, the multi-frequency impedance signal is acquired by the following method: injecting the prepared cell sample into a microfluidic device, focusing the cell sample to be tested into single-cell arrangement through fluid dynamics, and simultaneously measuring the impedance signal of the single cell at different frequencies to obtain the multi-frequency impedance signal. Feature extraction module 320 is used to extract various electrical phenotypic features based on the multi-frequency impedance signal; The cell classification module 330 is used to input the various electrical phenotypic features of the cell sample to be tested into a pre-trained cell detection model and output the rare cell detection results.

[0103] The present invention also provides a rare cell detection system, comprising: a microfluidic device, an impedance analyzer, and the rare cell detection device; A microfluidic device includes a microfluidic channel and a sensing electrode; the microfluidic channel is used to focus a sample of cells to be tested into single-cell arrangement through hydrodynamics. An impedance analyzer, connected to the sensing electrode, is used to simultaneously measure the impedance signal of a single cell at different frequencies to obtain a multi-frequency impedance signal, which is then used by a rare cell detection device to perform cell detection and obtain rare cell detection results.

[0104] Figure 15 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 15 As shown, the electronic device may include a processor 410, a communications interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a rare cell detection method.

[0105] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0106] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute the rare cell detection methods provided by the above methods.

[0107] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the rare cell detection methods provided by the methods described above.

[0108] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0109] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting rare cells, characterized in that, include: Acquire multi-frequency impedance signals from the cell sample to be tested; Multiple electrical phenotypic features are extracted based on the multi-frequency impedance signal; The various electrical phenotypic features of the cell sample to be tested are input into a pre-trained cell detection model, which outputs rare cell detection results. The multi-frequency impedance signal is obtained in the following way: The prepared cell sample is injected into a microfluidic device, and the cell sample is focused by fluid dynamics to achieve single-cell arrangement. The impedance signal of the single cell is measured synchronously at different frequencies to obtain the multi-frequency impedance signal.

2. The method for detecting rare cells according to claim 1, characterized in that, The multi-frequency impedance signal includes a first frequency band impedance signal, a second frequency band impedance signal, and a third frequency band impedance signal, wherein the first frequency band, the second frequency band, and the third frequency band belong to low frequency, medium frequency, and high frequency, respectively; The various electrical phenotypic features include electrical diameter, opacity X and opacity Y in the second frequency band, and opacity X and opacity Y in the third frequency band; The extraction of various electrical phenotypic features based on the multi-frequency impedance signal includes: Calculate the electrical diameter based on the impedance signal of the first frequency band; Calculate the opacity X and opacity Y in the second frequency band based on the impedance signal of the first frequency band and the impedance signal of the second frequency band; The opacity X and opacity Y in the third frequency band are calculated based on the impedance signal of the first frequency band and the impedance signal of the third frequency band.

3. The method for detecting rare cells according to claim 1, characterized in that, The cell detection model was trained in the following manner: Multiple electrical phenotypic features of each cell sample were extracted from the raw microfluidic impedance flow cytometry data of the training dataset and the test sample dataset. Multiple electrical phenotypic features of the training dataset are input into a machine learning model for training, and then applied to the test sample dataset to generate initial prediction results. The initial prediction results include: the initial predicted number of rare cells and the initial predicted number of other cells. The training dataset is randomly divided into a training subset and a test subset according to a preset ratio. The hyperparameters of the machine learning model are optimized based on the training subset, and the false positive rate and false negative rate of each prediction are calculated based on the test subset. The iteration is repeated until the kernel density estimation distribution of the false positive rate and false negative rate tends to be stable. The stabilized false positive rate and false negative rate are used as the target false positive rate and target false negative rate, respectively. The initial prediction results are corrected based on the target false positive rate and target false negative rate, and the corrected number of rare cells and the corrected number of other cells are output.

4. The method for detecting rare cells according to claim 3, characterized in that, The initial prediction results are corrected based on the target false positive rate and target false negative rate using the following formulas, and the corrected number of rare cells and the corrected number of other cells are output: ； ； in, This represents the initially predicted number of rare cells. For the initially predicted number of other cells, This represents the corrected number of rare cells. For the corrected number of other cells, To target the false positive rate, The target is the false negative rate.

5. The method for detecting rare cells according to claim 3, characterized in that, The kernel density estimates of the false positive rate and false negative rate are considered stable if the peak error fluctuation range is less than a preset threshold.

6. A rare cell detection device, characterized in that, include: The signal acquisition module is used to acquire the multi-frequency impedance signal of the cell sample to be tested; wherein, the multi-frequency impedance signal is acquired by the following method: injecting the prepared cell sample into a microfluidic device, focusing the cell sample to be tested into single-cell arrangement through fluid dynamics, and simultaneously measuring the impedance signal of the single cell at different frequencies to obtain the multi-frequency impedance signal; The feature extraction module is used to extract various electrical phenotypic features based on the multi-frequency impedance signal; The cell classification module is used to input the various electrical phenotypic features of the cell sample to be tested into a pre-trained cell detection model and output the rare cell detection results.

7. A rare cell detection system, characterized in that, include: Microfluidic devices, impedance analyzers, and rare cell detection devices as described in claim 6; Microfluidic devices, including microfluidic channels and sensing electrodes; Microfluidic channels are used to focus the sample of cells to be tested into single-cell arrangements through fluid dynamics. An impedance analyzer, connected to the sensing electrode, is used to simultaneously measure the impedance signal of a single cell at different frequencies to obtain a multi-frequency impedance signal, which is then used by a rare cell detection device to perform cell detection and obtain rare cell detection results.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the rare cell detection method as described in any one of claims 1 to 5.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the rare cell detection method as described in any one of claims 1 to 5.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the rare cell detection method as described in any one of claims 1 to 5.

Images