Method and device for rapidly detecting drug-resistant leukemia cells by utilizing optical fluid chip and machine learning

Abstract

The invention discloses a method and a device for rapidly detecting drug-resistant leukemia cells by utilizing an optical fluid chip and machine learning. The method comprises the following steps: S1, obtaining a sample set; s2, injecting each sample into an optical fluid chip; s3, connecting an optical detection module of the optical fluid chip with an external laser and an external optical sensor; s4, processing the optical signals, and extracting peak features; s5, constructing a data set according to the peak characteristic-leukemia cell drug resistance of each sample, and dividing the data set into a training set and a verification set; s6, constructing a machine learning model, training and verifying by using the training set obtained in S5, and obtaining a detection model; and S7, injecting a sample to be detected into the optical fluid chip, obtaining peak value characteristics by referring to the steps S3-S4, and inputting the peak value characteristics into the detection model to obtain a drug resistance detection result of the leukemia cells. According to the method, the required sample size is small, the rapid analysis capacity is achieved, real-time high-throughput analysis can be achieved, and the method has the crucial application in biological diagnosis.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence detection technology, and in particular relates to the rapid detection of drug-resistant leukemia cells, and specifically is a method and device for rapidly detecting drug-resistant leukemia cells using an optofluidic chip and machine learning. Background Art

[0002] Leukemia is a malignancy that arises in the bone marrow and is characterized by the uncontrolled proliferation of abnormal white blood cells. The rapid proliferation of these cells prevents the formation of normal blood cells, leading to complications such as anemia, susceptibility to infection, and bleeding disorders. Effective leukemia treatment strategies often include a combination of chemotherapy, targeted drugs, and even stem cell transplantation. These treatments are tailored to the specific type and stage of the disease.

[0003] Therefore, timely and accurate diagnosis of leukemia is crucial for effective treatment, as it facilitates timely intervention and decision-making about relevant drugs. Leukemia resistance to doxorubicin refers to leukemia cells becoming less responsive or completely resistant to this specific chemotherapy. Rapid identification of leukemia-resistant cells facilitates the use of alternative drugs or combination therapies, thereby providing more effective treatment for leukemia and improving patient outcomes. Summary of the invention

[0004] The present invention aims at the problems existing in the background technology and proposes a method for rapidly detecting drug-resistant leukemia cells.

[0005] Technical solution:

[0006] The present invention first discloses a method for rapidly detecting drug-resistant leukemia cells using an optofluidic chip and machine learning, which comprises the following steps:

[0007] S1. Obtain a sample set and confirm the drug resistance of leukemia cells in each sample;

[0008] S2, each sample is injected into an optofluidic chip, wherein the optofluidic chip integrates a channel module and an optical detection module;

[0009] S3, the optical detection module of the optofluidic chip is connected to an external laser and an optical sensor, and is used to transmit the laser to illuminate the channel module and detect the optical signal of the scattered light of the collection channel module;

[0010] S4, processing the optical signal and extracting peak features;

[0011] S5. Construct a data set based on the peak feature of each sample - leukemia cell resistance, and divide it into a training set and a validation set;

[0012] S6, build a machine learning model, train and verify it using the training set obtained in S5, and obtain a detection model;

[0013] S7, injecting the sample to be tested into the optofluidic chip, obtaining the peak characteristics with reference to steps S3-S4, and inputting the peak characteristics into the detection model to obtain the drug resistance detection result of leukemia cells.

[0014] Preferably, in S1, the method for determining the drug resistance of leukemia cells of each sample is a half-maximal inhibitory concentration determination experiment.

[0015] Preferably, in S2, when each sample is injected into the optofluidic chip, the concentration of each sample is 1×10 5 cells / mL.

[0016] Preferably, in S2, the optofluidic chip includes a channel module and an optical detection module, specifically:

[0017] The channel module comprises: a spiral flow channel for injecting samples, a square sheath flow channel for injecting water and a single straight channel for measuring light scattering signals; wherein: the spiral sample injection channel is located inside the square sheath flow channel, the end of the spiral sample injection channel is connected to a linear single light scattering signal detection straight channel, the single light scattering signal detection straight channel intersects and communicates with the square sheath flow water injection channel; the end of the linear single channel is a wastewater outlet;

[0018] The optical detection module includes: three grooves, one of which is used to insert an optical fiber connected to a laser to transmit laser light, and is arranged on one side of a single straight channel and points vertically to the single straight channel; two grooves are used to insert an optical fiber connected to an optical sensor to detect light signals, and are arranged on the other side of the single straight channel and are arranged in a "Y" shape with the optical fiber of the laser device.

[0019] Preferred:

[0020] In the channel module, all channels are 100 μm wide and 50 μm deep;

[0021] In the optical detection module, all grooves are 120 μm wide and 120 μm deep.

[0022] Preferably, in S2, the sample and water are injected into the optofluidic chip through a syringe pump, and the injection speed is 200 μL / h.

[0023] Preferably, in S3, the laser emits laser light having a wavelength of 532 nm.

[0024] Preferably, in S2, the optical signal is processed, including noise filtering and redundant data removal.

[0025] Preferably, in S6, an SVM classifier is selected to build a machine learning model, the training set is imported into the model for training, the trained model is imported into the validation set and validated using a five-fold cross validation method, and a grid search is used to find the best hyperparameters to obtain the best model.

[0026] The present invention also discloses a device for rapidly detecting drug-resistant leukemia cells using an optofluidic chip and machine learning, wherein the device comprises at least an optofluidic chip, a processor, and a memory connected to the processor, wherein:

[0027] The memory is used to store computer programs or instructions;

[0028] The processor is used to obtain and execute the computer program or instruction in the memory so that the device can perform the method of the present invention;

[0029] The optofluidic chip includes a channel module and an optical detection module, specifically:

[0030] The channel module comprises: a spiral sample injection channel, a square sheath flow water injection channel, and a single light scattering signal detection straight channel, wherein: the spiral sample injection channel is located inside the square sheath flow channel, the end of the spiral sample injection channel is connected to the linear single light scattering signal detection straight channel, the single light scattering signal detection straight channel intersects and communicates with the square sheath flow water injection channel; the end of the linear single channel is a wastewater outlet;

[0031] The optical detection module includes: three grooves, one of which is used to insert an optical fiber connected to a laser to transmit laser light, and is arranged on one side of a single straight channel and points vertically to the single straight channel; the other two grooves are used to insert an optical fiber connected to an optical sensor to detect light signals, and are arranged on the other side of the single straight channel and are arranged in a "Y" shape with the optical fiber of the laser device.

[0032] Beneficial effects of the present invention

[0033] The present invention was used to classify leukemia cells and drug-resistant leukemia cells by injecting the test sample and obtaining the scattered light signal. A spiral microchannel was designed based on the Dean effect to manipulate the streamlined flow of cells. An integrated optofluidic chip was developed by embedding optical fibers coupled with a photosensitive sensor and a laser source. Peaks were detected when the cells passed through the light source. These peaks were separated with a window size of 241 data points and used as features for training and testing machine learning classifiers. The SVM classifier achieved the highest classification accuracy of 91.1% in analyzing leukemia cells and drug-resistant leukemia cells compared to LR, RF, and KNN classifiers. The proposed method can complete the detection within 10 minutes, and the total experimental duration is 20 minutes. The presented results show that it is feasible to detect and classify biological entities with subtle differences based on scattered light signals using optofluidic technology and machine learning methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Schematic diagram of the structure and use of the optofluidic chip of the present invention.

[0035] Figure 2 Schematic diagram of an experimental device in an embodiment of the present invention.

[0036] Figure 3 This is a verification diagram of the design effect of the channel module in the embodiment of the present invention.

[0037] Figure 4 Schematic diagram of cell drug resistance detection of leukemia cell samples in an embodiment of the present invention.

[0038] Figure 5 Schematic diagram of processed signals obtained after eliminating irrelevant data from a leukemia cell sample in an embodiment of the present invention.

[0039] Figure 6 This is a result diagram of the confusion matrix evaluation of leukemia (K562) and drug-resistant leukemia cells (K562 / ADR) using four machine learning models in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The present invention will be further described below in conjunction with embodiments, but the protection scope of the present invention is not limited thereto:

[0041] Combination Figure 1 The present invention first designs an optofluidic chip, including a channel module and an optical detection module, specifically:

[0042] The channel module comprises: a spiral flow channel for injecting samples, a square sheath flow channel for injecting water and a single straight channel for measuring light scattering signals; wherein: the spiral sample injection channel is located inside the square sheath flow channel, the end of the spiral sample injection channel is connected to a linear single light scattering signal detection straight channel, the single light scattering signal detection straight channel intersects and communicates with the square sheath flow water injection channel; the end of the linear single channel is a wastewater outlet;

[0043] The optical detection module includes: three grooves, one of which is used to insert an optical fiber connected to a laser to transmit laser light, and is arranged on one side of a single straight channel and points vertically to the single straight channel; two grooves are used to insert an optical fiber connected to an optical sensor to detect light signals, and are arranged on the other side of the single straight channel and are arranged in a "Y" shape with the optical fiber of the laser device.

[0044] exist Figure 1 The enlarged schematic part a shows the sample being injected into the chip; Figure 1 The enlarged schematic diagram of part b shows that the spiral microchannel forms a "dean" flow, which affects the flow of cells to a specific trajectory; Figure 1 In the enlarged schematic diagram of part c, deionized water is injected from the side channel to form a streamline, guiding the particles to flow through the center of the channel and maintain the center trajectory; Figure 1 The enlarged schematic (d) shows cells flowing through the optical detection module, where laser light passes through the cells and the scattered light is collected by the photosensor.

[0045] The optofluidic chip is designed to control the flow of particles and enter the optical detection module. The combination of "dean" flow and sheath flow can precisely manipulate the particles to flow regularly on the central axis of the microfluidic channel. "Dean" flow is used to precisely manipulate microparticles by designing circular and curved microchannels. "Dean" flow is beneficial for focusing the particles in the channel and reducing dispersion. After inducing the movement of the particles to the central area of ​​the channel, sheath flow technology is applied to surround and confine the sample in the microchannel. The sheath fluid enters the main channel through two surrounding channels (Y-shaped). This configuration produces a laminar flow profile and confines the particles to the center of the channel surrounded by the sheath fluid. The sheath flow design controls the velocity of the microparticles, reduces particle dispersion, and prevents the sample from contacting the side walls of the channel.

[0046] In this embodiment, the assembly design of the optofluidic chip was completed using AutoCAD software (Autodesk, USA). The channel module of the chip consists of three parts, including Dean vortex, sheath flow channel and a single channel with a cross-sectional area of ​​100 μm wide and 50 μm high. In addition, three slots with a width of about 120 μm and a depth of about 120 μm were made for inserting optical fibers in the optofluidic chip as optical detection modules. Two slots are used to insert optical fibers connected to the optical sensor, while the remaining slot is used to insert the optical fiber connected to the laser.

[0047] To fabricate the optofluidic chip, a 4-inch silicon wafer was spin-coated with a 50 μm thick layer of negative photoresist SU-8 2050, followed by mask alignment photolithography. The second layer was spin-coated with negative photoresist SU-8 2050 again to obtain a thickness of about 100 μm. To make the PDMS prepolymer layer, PDMS and curing agent were mixed in a ratio of 10:1 and applied to the mold, followed by degassing in a vacuum oven and curing at 65°C overnight. The microfluidic chip made using PDMS was removed from the mold and cut. Holes with a diameter of 1.4 mm were used as the inlet and outlet of the sample. Finally, the prepared PDMS chip was bonded to a glass slide using plasma treatment. Kolliphore P188 (1%) was used as a surfactant and applied to the prepared chip channels. The surfactant treatment reduced the surface tension within the channel and prevented particles from adhering to the microchannel surface.

[0048] The optical detection module in the optofluidic chip consists of three main parts: laser source, photoelectric sensor, and signal acquisition system. The particle sample with optimized concentration was injected into the optofluidic chip through a split syringe pump (WH-SSP-08, Suzhou Wenhao Microfluidic Technology, China). The laser source and the photoelectric sensor were coupled through an optical fiber embedded in the chip. A laser light source with a wavelength of 532 nm was used, and the output power was about 1 mW. This laser wavelength was chosen because it should be smaller than the size of the particles to be measured. Lower light wavelengths have several advantages, including maximum scattering from microparticles, high sensitivity, reduced environmental interference, and improved microfluidic performance. Both the laser source and the photoelectric sensor were customized using SMA905 to connect to the optical fiber. The wavelength range of the photoelectric sensor is 400 to 1100 nm, and the response time is 6 ns. The interface of the photoelectric sensor was modified through an FC connector to provide a portable fiber-optic coupling interface. The output value of the photoelectric sensor was manually calibrated to a zero voltage signal under dark conditions. When the incident light directly irradiated the photoelectric sensor in an empty microfluidic channel, the output value was 3 V. The photoelectric sensor was positioned at an angle of 45° relative to the channel in the microfluidic chip. The scattered light signals were acquired with a sampling frequency of 10 kHz using the LabView platform and processed by a data acquisition system (NI-USB 6001, National Instruments). Detecting scattered light signals from microparticles passing light through a channel is challenging. The application of microfluidic chips has the advantages of small sample requirements, controllable particle flow, and narrow sample flow in microfluidic channels. An important aspect of optofluidic chips is the integration of an optical system to collect scattered light patterns in real time.

[0049] Combination Figure 2 , then a complete experimental setup is given. Figure 2Figure 1 includes (a) the experimental setup, including the optofluidic chip connected to a syringe pump, a laser source, a data acquisition system, and a computer; (b) a close-up image of the optofluidic chip, which integrates optical fibers coupled to lasers and light sensors (PS1 and PS2). The inlet is used to inject the sample, and the outlet is used to discharge the waste sample from the chip; (c) a microscopic image of the optical detection module in the chip, showing the embedded optical fibers coupled to the lasers and light sensors (PS1 and PS2), surrounding the microfluidic channel; (d) a fluorescence microscopic image showing the manipulation of the sample in the sheath flow (red line), and a magnified image showing that the sheath flow controls the flow of the sample, with a width of 10.82 μm. (d) The magnified image shows that a sheath flow (red line) of 10.82 μm is generated by adjusting the flow rates of water and sample. The purpose of inducing the sheath flow is to provide a controlled environment to align the particles correctly so that the laser can be focused on a specific location of the particles. In particular, the width of the sheath flow can be controlled by adjusting the flow rates of the sample and water. However, reducing the sheath flow width requires higher water flow and lower sample flow, which also increases the flow rate of the particles. Higher particle flow rates may reduce data accuracy due to sensor limitations. To obtain the best results, it is necessary to strike a balance between the flow rates of the two channels, properly sizing and balancing the two channels to obtain the largest significant peak with the smallest sheath flow width.

[0050] In the embodiment, three different assembly forms of optofluidic chip microchannels were manufactured and tested to evaluate the feasibility of the design for detecting particles. Three microfluidic chip designs were studied, including a single-channel chip, a Y-shaped channel chip, and a chip containing a combination of a Dean flow effect circular channel and a Y-shaped channel. Each assembly form manufactured was tested separately to determine the best design. A test sample containing 1 μm diameter polystyrene particles was injected into each manufactured chip to obtain light scattering data, such as Figure 3 (ac) A peak detection algorithm was applied to calculate the total number of peaks in the data obtained from each chip. Figure 3 (df) shows the histograms of the peak intensity detection results obtained from three chips. The blue bars indicate the number of different normalized peak intensities. The red curve is a Gaussian (normal) distribution fitted to the histogram data, providing a smooth and idealized representation of the data distribution. The parameters evaluated from each histogram include the mean (μ), standard deviation (σ), and determination coefficient (R 2 ). The results showed that the microfluidic chip assembly containing Dean effect circular channels and Y-shaped channels showed the best effect, such as Figure 3 f ((d) single-channel chip, the coefficient of variation is about 39.18%, (e) Y-shaped channel, the coefficient of variation is about 36.82%, (f) combined component including Dean effect circular channel and Y-shaped channel, the coefficient of variation is about 26.64%).

[0051] In order to verify the effect of the present invention, the example prepared the leukemia detection sample:

[0052] Human chronic myeloid leukemia / adriamycin-resistant strain (K562 / ADR) and human chronic myeloid leukemia (K562) cell lines were purchased from Cellverse. K562 cell lines were cultured in RPMI-1640 medium containing 10% fetal bovine serum (Gibco, USA) and 1% penicillin-streptomycin solution (Gibco, USA). K562 / ADR cell lines were cultured in RPMI-1640 medium containing 1 μg / mL adriamycin (ADR), and both cells were grown in an incubator at 37°C, constant humidity, and 5% CO2.

[0053] K562 and K562 / ADR cells were treated with different concentration gradients of chemotherapeutic drugs (ADR) at 37°C for 48 h. Cells were incubated with enhanced cell counting kit-8 (CCK-8, beyotime, China) for 1 h. Optical density (OD) was measured at 450 nm using a spectrophotometer (Molecular Devices, USA). IC was calculated using the logarithmic method 50 The value is the ADR concentration that inhibits cell growth by 50%. Each experiment was repeated three times.

[0054] Cultured K562 and K562 / ADR cell samples were used to prepare suspensions for analysis. First, centrifuge at 1000rpm for 5 minutes and discard the supernatant. After centrifugation, add 1mL of 4% paraformaldehyde (Biosharp, China) to resuspend and fix the cells for 10-15 minutes, and then centrifuge at 3000rpm for 5 minutes. Discard the supernatant, add 1mL of sterile ddH2O to wash the cells, and then centrifuge at 3000rpm for 5 minutes. After discarding the supernatant, resuspend the precipitate in 1mL ddH2O and mix thoroughly. Adjust the cell concentration to 1×10 by adding an appropriate amount of ddH2O. 5 The samples were sealed and stored at 4°C, or immediately used for subsequent testing.

[0055] Combination Figure 4 The cell viability was detected by CCK-8 method, and the IC 50 Values ​​were used to evaluate the drug resistance status of drug-resistant leukemia cells. Figure 4 (a) shows the drug concentration-dependent response to ADR after 48 h of culture, with the OD value decreasing as the drug concentration increased. The proliferation rate of K562 / ADR cells was higher than that of K562 cells. Figure 4(b) shows the comparison of the average IC50 values ​​of K562 cell line and K562 / ADR cell line to 0.22 μg / mL and 7.99 μg / mL of ADR, respectively. The results show that the drug resistance index RI = IC 50 (K562 / ADR) / IC 50 (K562) = 36.32 times (P < 0.0001). This indicates that K562 / ADR cells have good drug resistance.

[0056] The machine learning model of the present invention uses support vector machine (SVM). SVM is an efficient learning classifier suitable for linear and nonlinear classification problems because it can find the best hyperplane to maximize the classification of data into different categories. SVM is an effective classifier that can handle complex decision boundaries. Logistic regression (LR) is a supervised learning model for binary classification problems. The LR model determines the probability value of the category to which the features of the input training sample belong based on a logical function. This technology classifies data by evaluating the decision boundaries of different categories of data. Random forest (RF) is an integrated learning algorithm that improves classification results by combining multiple decision trees. Each tree is trained on a random subset of data and features, and the prediction result is finally determined by majority voting. The randomization method reduces overfitting and improves the versatility of the model, especially when processing high-dimensional data sets. K nearest neighbor (KNN) is an effective learning technique based on the similarity principle, the principle of which is that the category of a data point is determined by the majority category of its k nearest neighbors in the feature space. KNN is very effective when the data distribution is irregular or the decision boundary is nonlinear.

[0057] The combined features are divided into training and test sets by cross validation. In cross validation, the dataset is divided into S subsets and a total of S models are trained in each iteration. Each model is trained using the data from S-1 subsets and the remaining subset is used to test the trained algorithm. This process is repeated S times and different samples are retained for validation in each iteration. The average classification accuracy is obtained by calculating the average of the S classification accuracies. The performance of the trained classifier is evaluated by testing and analyzing the classification parameters.

[0058] The performance of the machine learning model is evaluated by analyzing different parameters in the obtained confusion matrix. A confusion matrix is ​​a table that visualizes the results of a classification model for known data set values. The number of rows and columns of a confusion matrix is ​​equal to the number of classes. The rows represent instances of known classes, and the columns represent instances of the classes predicted by the classifier. In a binary classification task, the confusion matrix consists of two rows and two columns. The four basic components of a confusion matrix are: True Positives (TP), True Negatives (TN), False Positives (FP), and False Negatives (FN). True Positives refer to the results that the model correctly identifies as positive, while True Negatives refer to the correctly predicted negative results. False Positives refer to the results that are incorrectly labeled as positive, while False Negatives are the results that are incorrectly classified as negative.

[0059] The data obtained from different samples are classified using machine learning classifiers. Data classification and evaluation are based on various data preprocessing steps, including noise filtering, redundant data removal (such as Figure 5 As shown, Figure 5 (a) shows the processed signal obtained after eliminating irrelevant data from a sample containing leukemia cells. Figure 5 (b) shows the processed signal obtained after eliminating irrelevant data from a sample containing drug-resistant leukemia cells), peak feature extraction, training and testing of the machine learning algorithm. The classification results obtained are analyzed using the performance evaluation parameters of the trained classifier. Specifically:

[0060] 1) Feature extraction

[0061] When a cell or particle intersects the incident light path, light scattering occurs, causing the incident light to deviate in different directions. The collected scattered light data exhibits a non-steady-state waveform characteristic with multiple peaks or local maxima. These peaks are extracted from the waveform with a window size of 241 points. Each peak corresponds to the moment when a particle or cell flows through the laser light path in the channel. The window size of 241 points is determined experimentally because the peak is always within a given time domain range. 241 point segments are collected from each of the two sensors and then merged with 482 data points from both sensors. The collected 482 point segments are used as features for training and testing various machine learning classifiers. These features provide information about the peak changes in the signal. The features are collected from two different types of samples, so the features of each class of samples are labeled. The features obtained from the data of the sample containing K562 cells are labeled as "0", while the features obtained from the data containing K562 / ADR cells are labeled as "1". The data points extracted from the two classes of samples are merged and randomly permuted to avoid bias during model training. The features are normalized and then the machine learning classifier is modeled and trained.

[0062] 2) Machine Learning Classification

[0063] The performance analysis of the trained model is evaluated through classification parameters including accuracy, precision, recall, specificity, and F-score. Classification accuracy refers to the number of instances correctly identified among all instances. Precision is the ratio of true positive results to all positive results, indicating the ability of a trained model to avoid false positive results. Recall, also known as sensitivity, quantifies the number of true positive results among all actual positive results. Specificity indicates how effectively the model avoids false negatives by calculating the ratio of accurate negative values ​​to all actual negative values. The F1 score is calculated by the harmonic mean of precision and recall. Classification parameters provide a comprehensive understanding of the performance of a trained machine learning model. The following is a mathematical representation of evaluating classification parameters using confusion matrix values:

[0064]

[0065] 3) Results display

[0066] This method was applied in biosensing applications to analyze leukemia (K562) and drug-resistant leukemia cells (K562 / ADR). The technique was used to classify two classes, so the confusion matrix consists of two rows and two columns, such as Figure 6 As shown. The rows of the confusion matrix represent the actual categories, and the columns represent the categories predicted by the classifier. The first row of the confusion matrix contains actual instances of the first category (K562). Similarly, the second row contains actual instances of the second category (K562 / ADR). The first column of the confusion matrix corresponds to instances predicted to be the first category (K562). Similarly, the second column corresponds to instances predicted to be the second category (K562 / ADR). Machine learning models, including SVM, LR, RF, and KNN, showed classification accuracy rates of 91.1%, 83%, 86.7%, and 79.5%, respectively. The classification accuracy predicted by the SVM classifier is higher than that of LR, RF, and KNN. Therefore, this application conducts a detailed analysis of the confusion matrix based on the SVM model. As shown Figure 6 As shown in (a), the values ​​in the first row predict that 1410 K562 cells are correctly identified, while the remaining 163 cells are incorrectly identified as K562 / ADR. Similarly, the values ​​in the second row predict that 115 K562 / ADR cells are incorrectly identified as K562, while 1432 instances in the second row are correctly identified. Based on the confusion matrix ( Figure 6 (a)), the performance indicators of the SVM classifier include TP = 1410, FN = 163, FP = 115 and TN = 1432. Table 1 shows the performance indicators of all evaluated classifiers based on the values ​​extracted from the confusion matrix ( Figure 6). Higher values ​​of the classifier performance metrics indicate that the trained classifier performs better in terms of correctly identifying the results. The results show that the developed optofluidic platform is effective in identifying leukemia and drug-resistant leukemia cells using the SVM classifier with a higher classification accuracy of 91.1%. Based on the SVM classifier, the precision of K562 was 92.5%, indicating that the model rarely misclassified K562 / ADR as K562. The recall rate of K562 was 89.6%, indicating that most K562 cells were correctly identified. The F1-score of K562 was 91%, reflecting a good balance between precision and recall. The total duration of conducting the test experiment was approximately 20 minutes, including 10 minutes of sample preparation time, 5 minutes of data acquisition time, and 5 minutes of data classification time to obtain the final results. The optofluidic chip-based method requires a small sample size and has rapid analysis capabilities, which enables real-time high-throughput analysis and has vital applications in biological diagnosis.

[0067] Table 1: Performance indicators of each classifier

[0068]

[0069] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A method for rapidly detecting drug-resistant leukemia cells using an optofluidic chip and machine learning, characterized in that It includes the following steps: S1. Obtain a sample set and confirm the drug resistance of leukemia cells in each sample; S2, each sample is injected into an optofluidic chip, wherein the optofluidic chip integrates a channel module and an optical detection module; S3, the optical detection module of the optofluidic chip is connected to an external laser and an optical sensor, and is used to transmit the laser to illuminate the channel module and detect the optical signal of the scattered light of the collection channel module; S4, processing the optical signal and extracting peak features; S5. Construct a data set based on the peak feature of each sample - leukemia cell resistance, and divide it into a training set and a validation set; S6. Build a machine learning model, train and verify it using the training set obtained in S5, and obtain a detection model; S7, inject the sample to be tested into the optofluidic chip, and obtain the peak characteristics by referring to steps S3-S4, The peak features are input into the detection model to obtain the drug resistance detection results of leukemia cells.

2. The method according to claim 1, characterized in that In S1, the method for determining the drug resistance of leukemia cells of each sample was a half-maximal inhibitory concentration determination experiment.

3. The method according to claim 1, characterized in that In S2, when each sample was injected into the optofluidic chip, the concentration of each sample was 1×10 5 cells / mL.

4. The method according to claim 1, characterized in that In S2, the optofluidic chip includes a channel module and an optical detection module, specifically: The channel module comprises: a spiral flow channel for injecting samples, a square sheath flow channel for injecting water and a single straight channel for measuring light scattering signals; wherein: the spiral sample injection channel is located inside the square sheath flow channel, the end of the spiral sample injection channel is connected to a linear single light scattering signal detection straight channel, the single light scattering signal detection straight channel intersects and communicates with the square sheath flow water injection channel; the end of the linear single channel is a wastewater outlet; The optical detection module includes: three grooves, one of which is used to insert an optical fiber connected to a laser to transmit laser light, and is arranged on one side of a single straight channel and points vertically to the single straight channel; two grooves are used to insert an optical fiber connected to an optical sensor to detect optical signals, and are arranged on the other side of the single straight channel and are arranged in a "Y" shape with the optical fiber of the laser device.

5. The method according to claim 4, characterized in that: In the channel module, all channels are 100 μm wide and 50 μm deep; In the optical detection module, all grooves are 120 μm wide and 120 μm deep.

6. The method according to claim 4, characterized in that In S2, the sample and water were injected into the optofluidic chip through a syringe pump, and the injection speed was 200 μL / h.

7. The method according to claim 1, characterized in that In S3, the laser emits laser light, and the wavelength of the laser light is 532nm.

8. The method according to claim 4, characterized in that In S2, the optical signal is processed, including noise filtering and redundant data removal.

9. The method according to claim 1, characterized in that In S6, the SVM classifier is used to build a machine learning model, the training set is imported into the model for training, the trained model is imported into the validation set for validation using the five-fold cross-validation method, and the grid search is used to find the best hyperparameters to obtain the best model.

10. A device for rapid detection of drug-resistant leukemia cells using an optofluidic chip and machine learning, characterized in that: The device comprises at least one optofluidic chip, a processor, and a memory connected to the processor, wherein: The memory is used to store computer programs or instructions; The processor is used to obtain and execute a computer program or instruction in the memory so that the device performs the method according to claims 1-8; The optofluidic chip includes a channel module and an optical detection module, specifically: The channel module comprises: a spiral sample injection channel, a square sheath flow water injection channel, and a single light scattering signal detection straight channel, wherein: the spiral sample injection channel is located inside the square sheath flow channel, the end of the spiral sample injection channel is connected to the linear single light scattering signal detection straight channel, the single light scattering signal detection straight channel intersects and communicates with the square sheath flow water injection channel; the end of the linear single channel is a wastewater outlet; The optical detection module includes: three grooves, one of which is used to insert an optical fiber connected to a laser to transmit laser light, and is arranged on one side of a single straight channel and points vertically to the single straight channel; the other two grooves are used to insert an optical fiber connected to an optical sensor to detect optical signals, and are arranged on the other side of the single straight channel and are arranged in a "Y" shape with the optical fiber of the laser device.

Images