Lung cancer cell drug resistance classification method and system based on high content imaging

Abstract

The invention relates to a lung cancer cell drug resistance classification method and system based on high content imaging. The method comprises the following steps: carrying out fluorescent staining and multi-component labeling on cells to be detected; performing high-content imaging on the to-be-detected cells subjected to fluorescent staining to obtain a multi-channel cell image; carrying out drug resistance classification on the multi-channel cell image by utilizing an IRSNet network to obtain a classification result; wherein the IRSNet classification network comprises an Inception-v4 backbone network, an Inception-ResNet module, a Reduction module and a softmax function, and the Inception-v4 backbone network, the Inception-ResNet module, the Reduction module and the softmax function are connected in sequence. The Inception-v4 backbone network is used for extracting the image features of the multi-channel cell image; the Inception-ResNet module is used for improving the depth and the width of the Inception-v4 backbone network and introducing a residual error structure; the Reduction module is used for reducing the network structure of the image features; the softmax function is used for classifying the image features. Compared with the prior art, the method is higher in cell drug resistance classification precision and efficiency.

Description

Technical Field

[0001] The present invention relates to the field of computer-aided medicine, and in particular to a lung cancer cell drug resistance classification method and system based on high-content imaging. Background Art

[0002] Lung cancer has the highest mortality rate among all malignant tumors. Chemotherapy, as an effective treatment, often leads to treatment failure due to drug resistance in patients. In recent years, more and more studies have focused on the field of chemotherapy drug resistance. With the rapid development of medical technology, it has become a trend to guide clinical drug use and formulate personalized chemotherapy through in vitro screening of cells in preclinical settings. In recent years, imaging, as a cell-based detection mode, has provided a new method for measuring biological phenotypes in existing and developing biological models. High-content imaging (HCI) technology can obtain a large amount of relevant information in the cellular field from a single experiment while maintaining the functional and structural integrity of the cells, providing new ideas for related research on cells and drugs.

[0003] Microscope images are a rich source of high-resolution biological data. Microscopic histopathology is the gold standard for diagnosing surgical pathology. In addition, rapid classification of cells based on their phenotype is also a long-standing and important issue. Although diagnosis can be made by relying on genomic analysis, diagnosis based on the cell level has important research value both in terms of accuracy and economic cost. Traditional methods of microscopy and flow cytometry are currently the tools of choice, but the former requires time-consuming manual analysis, while the latter produces very limited morphological information.

[0004] With the advancement of imaging and computing technology, deep learning has been widely used in clinical practice, especially in medical diagnosis. In previous studies, it has been shown that the successful learning of hidden patterns in microscopic images has improved image-based diagnosis in pathology, dermatology, and radiology, for example, detection of organ edges in computed tomography (CT), magnetic resonance imaging (MRI), and malignant skin lesions, classification of skin lesions, and detection of retinopathy. Deep learning is widely used in biological imaging, but there are few studies on high-content data, and a large number of authentic and reliable databases have not yet been established. Therefore, it is of great significance to consider obtaining a large amount of data that can characterize cell phenotypes through biological means, establish a specific database, and further explore the biological functions of cells. For example, Chinese patent CN 112507821 B discloses a method, medium and electronic device for detecting cell drug resistance based on high-content imaging, which includes the following steps: fluorescent staining of cells to be detected; high-content imaging of the stained cells to be detected to obtain corresponding high-content images; preprocessing the high-content images; using the preprocessed images as inputs of a trained convolutional neural network-based drug resistance classification and recognition model to detect the drug resistance category of the cells to be detected. This method uses a convolutional neural network to classify the drug resistance of cells, and there is still room for improvement in classification accuracy. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method and system for classifying lung cancer cell resistance based on high-content imaging, thereby achieving accurate classification of non-small cell lung cancer PC9 cell resistance.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] Conduct fluorescent staining on the cells to be detected;

[0008] Perform high-content imaging on fluorescently stained cells to obtain multi-channel cell images;

[0009] Using the IRSNet classification network to classify the multi-channel cell image according to drug resistance, and obtain a classification result;

[0010] The IRSNet classification network includes an Inception-v4 backbone network, an Inception-ResNet module, a Reduction module, and a softmax function;

[0011] The Inception-v4 backbone network is used to extract image features of the multi-channel cell image;

[0012] The Inception-ResNet module is used to increase the depth and width of the Inception-v4 backbone network and introduce a residual structure;

[0013] The Reduction module is used to reduce the network structure of the image features;

[0014] The softmax function is used to classify the image features.

[0015] Furthermore, the Inception-v4 backbone network includes multiple convolutional layers, maximum pooling layers and concatenation operations.

[0016] Furthermore, the Inception-v4 backbone network outputs image features of the multi-channel cell image through a ReLU activation function.

[0017] Furthermore, the IRSNet classification network also includes a SeNet module.

[0018] Furthermore, the fluorescent dyes for fluorescent staining include DAPI, MitoTracker Deep Red, Concanavalin A AlexaFluor488 conjugate and YF594-Phalloidin conjugate.

[0019] Further, the cell nucleus, mitochondria, endoplasmic reticulum and actin are labeled using the fluorescent dye;

[0020] Wherein, the DAPI is used to mark the cell nucleus, the MitoTracker Deep Red is used to mark the mitochondria, the Concanavalin A AlexaFluor488 conjugate is used to mark the endoplasmic reticulum, and the YF594-Phalloidinconjugate is used to mark actin.

[0021] Furthermore, the specific process of the fluorescent staining is as follows:

[0022] Discard the culture medium and wash with PBS, take Mito Tracker Deep Red FM and dilute it with PBS, add it to each well and place it in the incubator for culture;

[0023] The staining solution was discarded, the cells were washed with PBS, the washing solution was aspirated, the cells were fixed with paraformaldehyde, the washing solution was discarded with PBS, and the cells were permeabilized with Triton X-100 at room temperature;

[0024] Discard the upper layer of liquid, wash with PBS, take YF594-Phalloidin and dilute it with PBS, add it to each well, incubate at room temperature, discard the staining solution and wash with PBS;

[0025] Concanavalin A488 was taken and diluted with PBS, added to each well, incubated at room temperature, and the staining solution was discarded and washed with PBS;

[0026] DAPI dye was added to each well, the dye was aspirated and washed with PBS, PBS was added to each well, the cells were kept moist and placed in a high-content imaging system for imaging.

[0027] Furthermore, the imaging steps of the high-content imaging are as follows:

[0028] Place the labeled cell samples into the item bin, select the imaging objective, imaging mode, and corresponding plate type;

[0029] Select the field of view within the well and the desired fluorescence channel;

[0030] Adjusting the autofocus setting so that a preliminary image can be formed in the shooting field of view;

[0031] Set the focus offset value so that the sample can be imaged clearly;

[0032] Adjust the exposure time of each fluorescent channel to make each fluorescent channel display evenly, and save the above settings;

[0033] Set the experiment name, experimental group and experimental time, check the parameter information of each channel, and start image acquisition.

[0034] Furthermore, the fluorescent channels include four channels: DAPI, Cy5, FITC and Texas Red.

[0035] According to another aspect of the present invention, a lung cancer cell drug resistance classification system based on high-content imaging is provided, comprising:

[0036] Cell staining module, for fluorescent staining of cells to be detected;

[0037] High-content imaging module, which performs high-content imaging on fluorescently stained cells to obtain multi-channel cell images;

[0038] A drug resistance classification module, which uses an IRSNet classification network to classify the drug resistance of the multi-channel cell image to obtain a classification result;

[0039] The IRSNet classification network includes an Inception-v4 backbone network, an Inception-ResNet module, a Reduction module, and a softmax function;

[0040] The Inception-v4 backbone network is used to extract image features of the multi-channel cell image;

[0041] The Inception-ResNet module is used to increase the depth and width of the Inception-v4 backbone network and introduce a residual structure;

[0042] The Reduction module is used to reduce the network structure of the image features;

[0043] The softmax function is used to classify the image features.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. The present invention proposes an optimal IRSNet classification network based on the collected cell image data, which can accurately classify the drug resistance of PC9 and PC9 / GR cells and improve the classification accuracy.

[0046] 2. By optimizing the experimental process, the present invention constructs a staining scheme that can simultaneously label the cell nucleus, mitochondria, endoplasmic reticulum, and actin and perform uniform imaging, and preliminarily establishes a lung cancer cell data set, laying the foundation for subsequent research on drug resistance based on cell phenotype. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the process of the present invention;

[0048] Figure 2 This is the overall network structure diagram of IRSNet;

[0049] Figure 3 The results of labeling of various organelles of PC9, where (a) is the labeling result of the nucleus of PC9 cells imaged by the DAPI channel; (b) is the labeling result of the mitochondria of PC9 cells imaged by the CY3 channel; (c) is the labeling result of the endoplasmic reticulum of PC9 cells imaged by the FITC channel; (d) is the labeling result of the actin of PC9 cells imaged by the TexasRed channel;

[0050] Figure 4 This is the mixed labeling result of each channel of PC9 cells;

[0051] Figure 5The results of labeling of various organelles of PC9 / GR, where (a) is the labeling result of the nucleus of PC9 / GR cells imaged by the DAPI channel, (b) is the labeling result of the mitochondria of PC9 / GR cells imaged by the CY3 channel, (c) is the labeling result of the endoplasmic reticulum of PC9 / GR cells imaged by the FITC channel, and (d) is the labeling result of the actin of PC9 / GR cells imaged by the TexasRed channel;

[0052] Figure 6 This is the mixed labeling result of each channel of PC9 / GR cells;

[0053] Figure 7 Loss function curves for CNN, VGG, and IRSNet;

[0054] Figure 8 Classification accuracy curves of CNN, VGG, and IRSNet. DETAILED DESCRIPTION

[0055] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0056] Example 1

[0057] This embodiment provides a method for classifying drug resistance of lung cancer cells based on high content imaging, comprising the following steps:

[0058] S1. Fluorescence staining of cells to be detected;

[0059] PC9 cells and PC9 / GR cells (PC9 gefitinib-resistant cells) were evenly plated in a 24-well plate at a density of 50,000 per well and cultured for 24 hours to allow them to adhere;

[0060] In this embodiment, the cell culture medium required in the experimental process is first prepared to prepare for the subsequent cell culture experiment; then the cells are revived and the cell state is restored to ensure the experimental results; finally, the cells are passaged to expand the cell number for multiple experiments, or the cells are frozen to preserve the cell seeds for subsequent verification experiments.

[0061] Four fluorescent dyes, DAPI, MitoTracker Deep Red, Concanavalin A AlexaFluor488 conjugate, and YF594-Phalloidin conjugate, were used to label the nucleus, mitochondria, endoplasmic reticulum, and actin of PC9 cells and PC9 / GR cells, respectively. Through multiple experiments, the cell number gradient, the time gradient, and the concentration gradient of the dye were set to make the cell plating density appropriate, and the staining time and concentration of each dye were optimized to ensure uniform imaging in each imaging channel with low background noise.

[0062] The specific process of fluorescent staining is as follows:

[0063] When performing multi-component cell labeling and simultaneous imaging, in order to avoid or reduce fluorescence bleed-through, fluorescent dyes with strong fluorescence should be selected as much as possible when selecting fluorescent agents, and the emission spectrum overlap of the selected multiple fluorescent dyes should be minimized. Based on multiple optimizations, the following staining agents were selected in this embodiment: DAPI, MitoTracker Deep Red FM, Concanavalin AAlex 488, and YF594-Phalloidin Conjugate for multi-component cell labeling.

[0064] The culture medium was discarded, and the cells were washed twice with 2 mL of PBS. The Mito Tracker Deep Red FM stock solution (prepared with DMSO to 1 mM) was diluted to 250 nM in PBS and added to each well. The cells were placed in the incubator and cultured for 20 minutes.

[0065] The staining solution was discarded, and the cells were washed three times with PBS for 10 minutes each time. The washing solution was aspirated, and the cells were fixed with 4% paraformaldehyde for 20 minutes, then the solution was discarded and washed three times with PBS, and the cells were permeabilized with 0.5% Triton X-100 at room temperature for 10 minutes;

[0066] The upper layer of liquid was discarded, and the wells were washed three times with PBS. 1 μL of YF594-Phalloidin stock solution was diluted with 200 μL PBS per well, and after incubation at room temperature for 20 minutes for staining, the staining solution was discarded and the wells were washed three times with PBS;

[0067] Take the Concanavalin A 488 working solution (prepared with 0.1M NaHCO3 solution to 5mg / mL) and add PBS to dilute to 50μg / mL. Add 200μL of staining solution to each well and incubate at room temperature for 15 minutes. After discarding the staining solution, wash three times with PBS.

[0068] Add 200 μL DAPI dye to each well, remove the dye after 5 minutes, and wash three times with PBS. Add 200 μL PBS to each well, keep the cells moist, and place them in a high-content imaging system for imaging.

[0069] S2, perform high-content imaging on the fluorescently stained cells to be detected to obtain multi-channel cell images;

[0070] Obtain fluorescent images of the nucleus, mitochondria, endoplasmic reticulum, and actin of PC9 and PC9 / GR cells; perform high-content imaging of stained cells, select a 20x objective in wide-field mode, select a 6×6 field of view for each well, and use four channels of DAPI, Cy5, FITC, and Texas Red for imaging according to the excitation and emission wavelength ranges of the dyes used. The specific imaging steps are as follows:

[0071] Place the fluorescently labeled samples into the object compartment, select the imaging lens, imaging mode, and corresponding plate type;

[0072] Select the field of view to be photographed within the well, and the desired fluorescence channel;

[0073] Adjust the autofocus setting so that a preliminary image can be seen in the field of view;

[0074] Set the focus offset value so that the sample can be imaged clearly;

[0075] Adjust the exposure time of each channel so that each channel can be displayed evenly, and save the above settings;

[0076] Set the experiment name, experimental group and experimental time, check the parameter information of each channel, and start image acquisition.

[0077] In this example, a total of 3456 groups of cell images were obtained, including 1728 groups of PC9 and PC9 / GR cells, each group including four channels. The cell nucleus labeling results of PC9 cells imaged by DAPI channel are shown in Figure 2. Figure 3 As shown in (a), the size is uniform, generally round or oval; the mitochondrial labeling results of PC9 cells imaged by CY3 channel are shown in Figure 3 As shown in (b), one section is mostly rod-shaped or filamentous and distributed throughout the cell. The stronger the brightness, the more mitochondria there are in this area. The endoplasmic reticulum labeling results of PC9 cells imaged by FITC channel are shown in Figure 3 As shown in (c), they are generally dot-shaped and distributed throughout the cell; the actin labeling results of PC9 cells imaged by the TexasRed channel are shown in Figure 3 As shown in (d), the shape of the entire cell is outlined, with obvious boundaries; the mixed labeling results of each channel of PC9 cells are shown in Figure 4 The results of PC9 / GR cell organelle labeling are shown in Figure 5As shown, the nuclear labeling results of PC9 / GR cells imaged by DAPI channel are as follows Figure 5 As shown in (a), the mitochondrial labeling results of PC9 / GR cells imaged by CY3 channel are as follows Figure 5 As shown in (b), the endoplasmic reticulum labeling results of PC9 / GR cells imaged by FITC channel are as follows Figure 5 As shown in (c), the actin labeling results of PC9 / GR cells imaged by the TexasRed channel are as follows Figure 5 (d) shows the mixed labeling results of PC9 / GR cells in each channel. Figure 6 shown.

[0078] S3. Using the Inception-ResNet-SeNet network (IRSNet for short), by marking various organelles, the obtained multi-channel images are used to classify drug-resistant cells and normal cells to obtain classification results.

[0079] The IRSNet network is mainly designed by 1 Inception-v4 backbone network, 22 SeNets, 20 Inception-ResNets, and different types of Inception and Reduction modules combined according to different hyperparameters, such as Figure 2 As shown in the figure. This model combines the features of Inception-v4, ResNet and SeNet. The Inception network increases the width of the network while increasing the depth, obtaining more local information. The introduced residual structure can increase the training speed. At the same time, the introduction of the attention network makes the network focus more on more effective information, thereby improving the effectiveness of the network. The embedded ResNet network module greatly improves the training speed. The introduction of the SeNet channel attention network makes the network focus more on more discriminative image features, reducing parameters and calculations, thereby improving the classification performance of the network. The Inception-v4 backbone network is used for feature extraction, which is mainly composed of 10 convolutional layers, 2 maximum pooling layers and 2 Concatenation operations, and finally output by the ReLU function. The Inception-ResNet module introduces the residual module while widening the network, and the Reduction module is used to reduce the network structure of the feature map. At the end of the network, softmax is used to achieve binary classification.

[0080] When evaluating the model results, the four main indicators used are accuracy, precision, recall, and F1 value. Accuracy refers to the ratio of correctly classified samples to the total samples, where TP (True Positive) is true positive, FP (False Positive) is false positive, TN (True Negative) is true negative, and FN (False Negative) is false negative. The specific formula is as follows:

[0081]

[0082] Precision is generally defined as the ratio of true positive samples to all predicted positive samples. The specific formula is as follows:

[0083]

[0084] The recall rate is mainly defined for positive samples, which refers to the ratio of true positive samples to actual samples in the positive category. The formula is as follows:

[0085]

[0086] The F1 value is expressed as the weighted harmonic mean of precision and recall. P refers to precision and R refers to recall. The formula is as follows:

[0087]

[0088] The drug resistance of PC9 and PC9 / GR cells was classified using the IRSNet network, CNN, and VGG networks. The data was divided into a training set and a test set at a ratio of 4:1. After 300 iterations, the test set was input into the network. The loss function of the test set obtained by the classification network varied with the number of iterations. Figure 7 shown.

[0089] After 70 iterations, each network began to converge, among which VGG and IRSNet built in this embodiment converged faster. The curve of the classification accuracy of the test set changing with the number of iterations is shown in Figure 8 As shown, the classification accuracy of the classification network constructed in this embodiment is significantly higher than that of traditional CNN and VGG networks.

[0090] The classification accuracy of the CNN network is 79.17%, the classification accuracy of the VGG network is 81.67%, and the classification accuracy of the IRSNet network is 85.83%. In addition, the binary classification is performed based on the obtained database, and the precision, recall rate and F1 value are listed in Table 1. Table 1 is shown below.

[0091] Table 1 Classification results of three types of networks

[0092]

[0093] From the values ​​in Table 1, we can see that the IRSNet network is also superior to the traditional CNN and VGG networks in terms of accuracy and sensitivity.

[0094] By using three different deep learning networks, CNN, VGG and IRSNet, to classify cell drug resistance, and comparing the accuracy, precision and sensitivity of each network, it is shown that IRS can converge quickly and the classification accuracy is significantly better than the other two types of networks. In order to further focus on the role of organelles in drug resistance classification, the optimal classification network IRSNet is used to train the four channels of w1 nucleus, w2 mitochondria, w3 endoplasmic reticulum and w4 actin. The final classification accuracy based on w1 is 63.33%, the classification accuracy based on w2 is 80.83%, the classification accuracy based on w3 is 58.33%, and the classification accuracy based on w4 is 72.50%. The results of the accuracy, recall rate and F1 value based on the classification of each channel are listed in Table 2, as shown below.

[0095] Table 2 Classification results based on each channel

[0096]

[0097] In summary, this embodiment proposes an optimal classification network based on the collected cell image data, which can accurately classify PC9 and PC9 / GR cells, and verifies the experimental feasibility of drug resistance identification based on cell phenotype. In addition, by conducting drug resistance classification experiments on each component separately, it is found that the results based on mitochondrial classification alone are closest to the results based on multi-component labeling. This seems to indicate that mitochondria contain more information that is conducive to cell resistance discrimination, which is worthy of further research and is a comprehensive application of emerging biological technologies.

[0098] Example 2

[0099] This embodiment provides a lung cancer cell drug resistance classification system based on high-content imaging, including:

[0100] Cell staining module, for fluorescent staining of cells to be detected;

[0101] High-content imaging module, which performs high-content imaging on fluorescently stained cells to obtain multi-channel cell images;

[0102] A drug resistance classification module, which uses an IRSNet classification network to classify the drug resistance of the multi-channel cell image to obtain a classification result;

[0103] The IRSNet classification network includes an Inception-v4 backbone network, an Inception-ResNet module, a Reduction module, and a softmax function;

[0104] The Inception-v4 backbone network is used to extract image features of the multi-channel cell image;

[0105] The Inception-ResNet module is used to increase the depth and width of the Inception-v4 backbone network and introduce a residual structure;

[0106] The Reduction module is used to reduce the network structure of the image features;

[0107] The softmax function is used to classify the image features.

[0108] The rest is the same as in Example 1.

[0109] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for classifying drug resistance of lung cancer cells based on high-content imaging, characterized in that: The following steps are involved: Conduct fluorescent staining on the cells to be detected; Perform high-content imaging on fluorescently stained cells to obtain multi-channel cell images; Using the IRSNet classification network to classify the multi-channel cell image according to drug resistance, and obtain a classification result; The IRSNet classification network includes an Inception-v4 backbone network, an Inception-ResNet module, a Reduction module, and a softmax function; The Inception-v4 backbone network is used to extract image features of the multi-channel cell image; The Inception-ResNet module is used to increase the depth and width of the Inception-v4 backbone network and introduce a residual structure; The Reduction module is used to reduce the network structure of the image features; The softmax function is used to classify the image features.

2. The method for classifying drug resistance of lung cancer cells based on high content imaging according to claim 1, characterized in that: The Inception-v4 backbone network includes multiple convolutional layers, maximum pooling layers, and concatenation operations.

3. The method for classifying drug resistance of lung cancer cells based on high content imaging according to claim 1, characterized in that: The Inception-v4 backbone network outputs the image features of the multi-channel cell image through a ReLU activation function.

4. The method for classifying drug resistance of lung cancer cells based on high content imaging according to claim 1, characterized in that: The IRSNet classification network also includes a SeNet module.

5. The method for classifying drug resistance of lung cancer cells based on high content imaging according to claim 1, characterized in that: The fluorescent dyes for fluorescent staining include DAPI, MitoTracker Deep Red, Concanavalin AAlexaFluor488 conjugate and YF594-Phalloidin conjugate.

6. The method for classifying drug resistance of lung cancer cells based on high content imaging according to claim 5, characterized in that: Using the fluorescent dye to label the cell nucleus, mitochondria, endoplasmic reticulum and actin; Wherein, the DAPI is used to mark the cell nucleus, the MitoTracker Deep Red is used to mark the mitochondria, the Concanavalin A AlexaFluor488 conjugate is used to mark the endoplasmic reticulum, and the YF594-Phalloidinconjugate is used to mark actin.

7. The method for classifying drug resistance of lung cancer cells based on high content imaging according to claim 1, characterized in that: The specific process of the fluorescent staining is as follows: Discard the culture medium and wash with PBS, take Mito Tracker Deep Red FM and dilute it with PBS, add it to each well and place it in the incubator for culture; The staining solution was discarded, the cells were washed with PBS, the washing solution was aspirated, the cells were fixed with paraformaldehyde, the washing solution was discarded with PBS, and the cells were permeabilized with Triton X-100 at room temperature; Discard the upper layer of liquid, wash with PBS, take YF594-Phalloidin and dilute it with PBS, add it to each well, incubate at room temperature, discard the staining solution and wash with PBS; Concanavalin A488 was taken and diluted with PBS, added to each well, incubated at room temperature, and the staining solution was discarded and washed with PBS; DAPI dye was added to each well, the dye was aspirated and washed with PBS, PBS was added to each well, the cells were kept moist and placed in a high-content imaging system for imaging.

8. The method for classifying drug resistance of lung cancer cells based on high content imaging according to claim 1, characterized in that: The imaging steps of the high-content imaging are as follows: Place the labeled cell samples into the item bin, select the imaging objective, imaging mode, and corresponding plate type; Select the field of view within the well and the desired fluorescence channel; Adjusting the autofocus setting so that a preliminary image can be formed in the shooting field of view; Set the focus offset value so that the sample can be imaged clearly; Adjust the exposure time of each fluorescent channel to make each fluorescent channel display evenly, and save the above settings; Set the experiment name, experimental group and experimental time, check the parameter information of each channel, and start image acquisition.

9. The method for classifying drug resistance of lung cancer cells based on high content imaging according to claim 8, characterized in that: The fluorescent channels include four channels: DAPI, Cy5, FITC and Texas Red.

10. A lung cancer cell drug resistance classification system based on high content imaging, characterized in that: include: Cell staining module, for fluorescent staining of cells to be detected; High-content imaging module, which performs high-content imaging on fluorescently stained cells to obtain multi-channel cell images; A drug resistance classification module, which uses an IRSNet classification network to classify the drug resistance of the multi-channel cell image to obtain a classification result; The IRSNet classification network includes an Inception-v4 backbone network, an Inception-ResNet module, a Reduction module, and a softmax function; The Inception-v4 backbone network is used to extract image features of the multi-channel cell image; The Inception-ResNet module is used to increase the depth and width of the Inception-v4 backbone network and introduce a residual structure; The Reduction module is used to reduce the network structure of the image features; The softmax function is used to classify the image features.

Images