Method for evaluating activity of single CTC cell based on deep learning method

Abstract

The invention discloses a method for evaluating the activity of a single CTC cell based on a deep learning method, and the method is characterized in that the method comprises the steps: 1) carrying out the preprocessing of the initial image data of CTC cells in different states, obtaining the feature image of the single CTC cell, constructing a data set corresponding to a state classification label, randomly dividing the data set of each state into a training set, a verification set and a test set; (2) respectively inputting the training set and the verification set into an OffficientNet-B0 model to carry out vitality interval judgment training so as to obtain an optimal model; 3) inputting the training set and the verification set into a YOLOv5s model to carry out single CTC cell positioning and CTC cell activity value prediction training to obtain an optimal model; and (4) respectively inputting the image data of the cells to be detected into the optimal OffficientNet-B0 model and the optimal YOLOv5s model, and calculating to obtain the activity value of the single CTC cell, and the method has the advantages of high efficiency, rapidness and accuracy.

Description

Technical Field

[0001] The present invention relates to a method for evaluating CTC cell viability, and in particular to a method for evaluating the viability of a single CTC cell based on a deep learning method. Background Art

[0002] Circulating tumor cells (CTCs) are closely related to cancer metastasis and recurrence. They are rare tumor cells that escape immune killing and survive after falling off from the primary or metastatic lesions into the peripheral blood during the development of cancer. As a new tumor marker, CTCs are gradually used for auxiliary staging, prognosis evaluation, postoperative recurrence detection and efficacy evaluation of tumors. At present, the main indicator of CTC detection and monitoring is quantity, but CTC cell viability is worthy of attention as a potential indicator. CTC cell viability can reflect the physiological state and function of CTC, reveal the viability, proliferation rate and metabolic state of CTC, and can obviously provide more abundant reference information for early detection and early diagnosis of tumors, accurate diagnosis of patients with tumor metastasis, monitoring of postoperative tumor recurrence and metastasis of patients, evaluation of the sensitivity of anti-tumor drugs and patient prognosis, and selection of strategies for individualized treatment. Therefore, it is very important to develop an efficient and rapid CTC cell viability assessment method.

[0003] Traditional methods for evaluating cell viability, including the MTT method, MTS method, CCK-8 method, and flow cytometry, are widely recognized and often regarded as evaluation standards. These techniques usually rely on chemical reagents for cell staining, which requires a certain amount of staining time and may also be toxic to cells, affecting long-term observation of their viability. In response to these challenges, a variety of innovative cell viability measurement technologies have emerged in recent years, including cell membrane potential measurement, lens-free shadow imaging technology, and electrochemical measurement. However, both traditional methods and these emerging technologies focus on viability assessment at the cell population level, that is, indirectly assessing cell viability by calculating the ratio of live cells to the total number of cells, and often cannot achieve accurate evaluation at the single-cell level, especially in the field of CTC research. Typically, there are only 1 to 10 CTCs per milliliter of blood, compared with 5×10 9 red blood cells and 7×10 6 The extreme scarcity of CTCs in the blood makes it particularly difficult to accurately assess the vitality of CTCs.

[0004] Deep learning has a powerful feature learning ability. Through a deep nonlinear neural network structure, it effectively combines low-level features from massive data to form dense high-level semantic abstractions, thereby efficiently and accurately extracting deep features from the data. It has been widely used in image processing, natural language understanding, and speech recognition. As an important technical means in the field of computer vision, deep learning has made breakthrough progress in the biomedical field, especially in image analysis, image segmentation, and auxiliary diagnosis and treatment. There have been some reports on deep learning in the assessment of cell viability, such as distinguishing live / dead cells through cell images and evaluating cell metabolic capacity, but it is still difficult to accurately determine the viability of a single cell. At present, there are no reports on the use of deep learning methods to directly assess the viability of a single cell through cell morphological characteristics. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an efficient, fast and accurate method for evaluating the viability of a single CTC cell based on a deep learning method.

[0006] The technical solution adopted by the present invention to solve the above technical problems is: a method for evaluating the viability of a single CTC cell based on a deep learning method, characterized by comprising the following steps:

[0007] (1) Data collection and image preprocessing

[0008] Fluorescence electron microscopy was used to obtain the initial image data of CTC cells in different states. The initial images of CTC cells were processed using the cell segmentation tool Cellpose to outline the morphological contours of individual cells and obtain a mask file. The mask file was imported into the imageJ software to obtain the coordinate information of individual cells. The individual cell coordinates were cut to obtain the characteristic image of individual CTC cells. A data set corresponding to the state classification label was constructed, and the data set of each state was randomly divided into a training set, a validation set, and a test set.

[0009] (2) Optimization of CTC cell viability interval output model

[0010] A. Input the training set and validation set obtained in step (1) into the EffficientNet-B0 model to perform 100-75%, 75-50%, 50-25%, and 25-0% activity interval judgment training to obtain the optimal model weight data; load the optimal model weight data into the EffficientNet-B0 model;

[0011] B. Input the test set obtained in step (1) into the EffficientNet-B0 model obtained in step (2) A for image classification performance testing, and output the activity interval of the CTC image in the test set. If the preset activity value of the CTC image in the test set falls within the output activity interval, the optimal EffficientNet-B0 model is obtained; if the preset activity value of the CTC image in the test set falls outside the output activity interval, repeat step (2) A until the optimal EffficientNet-B0 model is obtained;

[0012] (3) Optimization of CTC cell localization and viability prediction model

[0013] A. Input the training set and validation set obtained in step (1) into the YOLOv5s model to perform localization of a single CTC cell in a cell image and prediction of the vitality value of the CTC cell, and obtain the optimal model weight data; load the optimal model weight data into the YOLOv5s model;

[0014] B. Input the test set obtained in step (1) into the YOLOv5s model for image classification performance detection, and output the location and vitality value of the CTC cells in the test set. If the preset vitality value of the CTC cells in the test set is consistent with the output vitality value, the optimal YOLOv5s model is obtained; if the preset vitality value of the CTC cells in the test set is inconsistent with the output vitality value, repeat step (3) A until the optimal YOLOv5s model is obtained;

[0015] (4) Assessment of single CTC cell viability

[0016] A. Take a fluorescent electron microscope to obtain the image data of the cells to be tested, input the image data of the cells to be tested into the optimal EffficientNet-B0 model, and output the activity range of the CTC cells, with the upper and lower values ​​​​as v1 and v2 respectively; the activity value

[0017] B. Input the cell image data obtained in step (4) A into the optimal YOLOv5s model, output the detection probability P1 based on the data set of the v1 vitality value, output the detection probability P1 based on the data set of the v2 vitality value, and use v1*P1+v2*P2 to calculate the vitality value of a single CTC cell.

[0018] Further, the method for acquiring the initial images of CTC cells in different states in step (1) is as follows: Hela cells are stimulated by exogenous reactive oxygen species to change their viability. First, CTC cells are placed in 2 mL of culture medium and incubated for 18 h in a culture environment of 5% CO2 and 37°C, and 1.5 mL of the upper culture medium is removed; well-grown, adherent CTC cells are added with 1 mL of 1×PBS solution containing H2O2 at concentrations of 0 mmol / L, 1 mmol / L, 8 mmol / L, 15 mmol / L, and 30 mmol / L, respectively, and incubated for 40 min to obtain CTC cells in five classification states i, ii, iii, iv, and v, and the corresponding viability values ​​are 100%, 75%, 50%, 25%, and 0%, respectively.

[0019] Furthermore, the ratio of the number of training sets, validation sets and test sets described in step (1) is 6:3:1.

[0020] Furthermore, in step (2), the parameters of the EffficientNet-B0 model are set as follows: the number of samples in each training batch is specified to be 32, the total number of training rounds is set to 100, and the number of categories in the data set is 5.

[0021] Furthermore, the core structural module of the EffficientNet-B0 model in step (2) is the mobile flip bottleneck convolution MBConv module, which performs 1×1 point-by-point convolution on the input CTC image and changes the output channel dimension according to the expansion ratio, and then performs k×k depth convolution; when compression and excitation operations are introduced, the mobile flip bottleneck convolution will be performed after the depth convolution, and ends with a 1×1 point-by-point convolution to restore the original channel dimension, followed by connection inactivation and input skip connection, so that the model has a random depth.

[0022] Furthermore, in step (3), the YOLOv5s model includes a Conv module and a C3 module. The Conv module performs two upsampling and two downsampling processes on the input. Then, the C3 module is used to extract features. The C3 module defines a neural network module for implementing the CSP Bottleneck structure.

[0023] Compared with the prior art, the advantages of the present invention are: the present invention is a method for evaluating the viability of a single CTC cell based on a deep learning method, collects images of CTCs with different viabilities, trains a deep learning model, extracts its morphological features, and then applies it to the identification of single CTC cells and the evaluation of single cell viability. The method uses a deep learning model jointly constructed by EfficientNet-B0 and YOLOv5s, has good prediction performance, has the advantages of being label-free, non-toxic, and accurately evaluating the viability of a single CTC, and has important potential application value in the medical field. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 (A) The relationship between H2O2 concentration and cell viability determined by CCK-8 method, (B) cell image in state i, (C) cell morphology in state ii, (D) cell morphology in state iii, (E) cell morphology in state iv, (F) cell morphology in state v, scale bar, 10 μm;

[0025] Figure 2 The activity evaluation of CTCs treated with different concentrations of H2O2 was performed, where the AF concentrations were 20mmol / L, 12mmol / L, 10mmol / L, 6mmol / L, 5mmol / L, and 0.6mmol / L.

[0026] Figure 3 The cell viability under different H2O2 concentrations was measured simultaneously by the deep learning model and CCK-8 method;

[0027] Figure 4 Bland-Altman plot for deep learning model and CCK-8 method;

[0028] Figure 5 Performance evaluation of the deep learning model jointly constructed by EfficientNet-B0 and YOLOv5s of the present invention, including (A) precision, (B) recall, (C) F1 score, (D) AP, and (E) mAP;

[0029] Figure 6 Real sample CTC viability measurement for the deep learning model constructed for the present invention, scale bar, 10 μm. DETAILED DESCRIPTION

[0030] The present invention is further described in detail below with reference to the accompanying drawings. 1. Specific embodiments

[0032] 1. Data collection and image preprocessing

[0033] Human cervical cancer cells Hela were selected as the research object. Hela cells were stimulated by exogenous reactive oxygen species (ROS) to change their activity. First, Hela cells were placed in 2mL of culture medium and incubated for 18h in a culture environment of 5% CO2 and 37℃, and 1.5mL of the upper culture medium was aspirated. Well-growing and adherent Hela cells were added with 1mL of 1×PBS solution containing H2O2 at concentrations of 0mmol / L, 1mmol / L, 8mmol / L, 15mmol / L, and 30mmol / L, respectively, and incubated for 40min to obtain CTC cells of five classification states i, ii, iii, iv, and v, with corresponding activity values ​​of 100%, 75%, 50%, 25%, and 0%, respectively;

[0034] Fluorescence electron microscopy was used to obtain the initial image data of CTC cells in the corresponding states. The initial images of CTC cells were processed using the cell segmentation tool Cellpose to outline the morphological contours of individual cells and obtain a mask file. The mask file was imported into the imageJ software to obtain the coordinate information of individual cells, and the individual cell coordinates were cut to obtain the characteristic image of individual CTC cells. A data set corresponding to the state classification label was constructed, where there were 33358, 39952, 17594, 28821 and 23078 individual cell images of states i, ii, iii, iv and v, respectively. The data set for each state was randomly divided into training set, validation set and test set in a ratio of 6:3:1.

[0035] 2. Optimization of CTC cell viability interval output model

[0036] A. Input the training set and validation set obtained in step 1 into the EffficientNet-B0 model for 100-75%, 75-50%, 50-25%, and 25-0% activity interval judgment training to obtain the optimal model weight data; load the optimal model weight data into the EffficientNet-B0 model;

[0037] B. Input the test set obtained in step 1 into the EffficientNet-B0 model obtained in step 2A for image classification performance detection, and output the activity interval of the CTC image in the test set. If the preset activity value of the CTC image in the test set falls within the output activity interval, the optimal EffficientNet-B0 model is obtained; if the preset activity value of the CTC image in the test set falls outside the output activity interval, repeat step 2A until the optimal EfficientNet-B0 model is obtained;

[0038] The above process first sets parameters: for example, the number of samples in each training batch is specified to be 32, that is, the number of samples used each time the model is updated; the total number of training rounds is set to 100, that is, the number of times the entire training data set is traversed; the number of categories in the data set is 5, which is used to specify the number of output categories of the model.

[0039] The core structural module of the above EffficientNet-B0 model is the mobile flip bottleneck convolution (MBConv) module, which performs 1×1 point-by-point convolution on the input CTC image and changes the output channel dimension according to the expansion ratio, and then performs k×k depth convolution; when compression and excitation operations are introduced, the mobile flip bottleneck convolution will be performed after the depth convolution, and end with 1×1 point-by-point convolution to restore the original channel dimension, followed by connection inactivation and input skip connection, so that the model has random depth.

[0040] 3. Optimization of CTC cell localization and vitality value prediction model

[0041] A. Input the training set and validation set obtained in step 1 into the YOLOv5s model to perform localization of a single CTC cell in the cell image and prediction of the vitality value of the CTC cell to obtain the optimal model weight data; load the optimal model weight data into the YOLOv5s model;

[0042] B. Input the test set obtained in step 1 into the YOLOv5s model for image classification performance detection, and output the location and vitality value of the CTC cells in the test set. If the preset vitality value of the CTC cells in the test set is consistent with the output vitality value, the optimal YOLOv5s model is obtained; if the preset vitality value of the CTC cells in the test set is inconsistent with the output vitality value, repeat step 3A until the optimal YOLOv5s model is obtained.

[0043] The above YOLOv5s model includes a Conv module and a C3 module. The Conv structure performs two upsampling and two downsampling processes on the input: the upsampling process doubles the size of the feature map; the first downsampling process uses a convolutional layer to reduce the size of the feature map by half; the second downsampling uses a convolutional layer to reduce the size of the feature map by half again.

[0044] The C3 module is then used to extract features. The C3 module defines a neural network module for implementing the CSP Bottleneck structure. CSPdarknet first extracts three different levels of feature maps feat1, feat2, and feat3 through the backbone model, and then performs a series of convolution operations on the extracted features according to different conditions in order to carry out the next step of feature processing and fusion.

[0045] 4. Single CTC cell viability assessment

[0046] A. Use a fluorescent electron microscope to photograph the sample containing CTC cells to obtain the image data of the cells to be tested, input the image data of the cells to be tested into the optimal EffficientNet-B0 model, and output the activity range of CTC cells, with the upper and lower values ​​correspondingly as v1 and v2;

[0047] B. Input the cell image data obtained in step 4A into the optimal YOLOv5s model, output the detection probability P1 based on the data set of the v1 vitality value, output the detection probability P1 based on the data set of the v2 vitality value, and calculate the single CTC cell vitality value according to the formula v1*P1+v2*P2.

[0048] The above detection probability is the inherent output value of the YOLOv5s model. It compares the target CTC image with the images in the classification data set v1 or v2 (for example, 25%) to observe how similar the target image is to the images in the data set, that is, to evaluate and output a probability value, usually between 0 and 1.

[0049] 2. Experimental Results Analysis

[0050] 1. Viability assessment based on cell morphology

[0051] Hela cells were cultured in a culture flask containing DMEM medium (containing 10% fetal bovine serum, 100U / mL streptomycin, and 100U / mL penicillin) in a culture environment of 5% CO2 and 37°C. Afterwards, the old culture medium was discarded and the cells were rinsed with 1mL 1×PBS. Then, 1mL 0.25% trypsin digestion solution was added to the culture flask for 1.5min, the trypsin digestion solution was discarded and placed in a 5% CO2, 37°C culture environment for 1min, and then 2mL of new culture medium was added and repeatedly blown with a pipette to suspend the cells. Finally, the cells were transferred to a 35mm×12mm culture dish for use. Hela cells were stimulated by exogenous reactive oxygen species (ROS) to change their viability. First, Hela cells were placed in 2mL of culture medium and incubated for 18h in a 5% CO2, 37°C culture environment, and 1.5mL of the upper culture medium was aspirated. Then, 1 mL of 1× PBS solution containing a certain concentration of H2O2 was added and incubated for 40 min to investigate the relationship between the H2O2 concentration and the cell viability determined by the CCK-8 method.

[0052] The steps for measuring cell viability using the CCK-8 method are as follows:

[0053] (1) Treat well-adherent cells with trypsin digestion solution to obtain a cell suspension and count them. In a 96-well culture plate, inoculate 100 μL of the cell suspension into each well to ensure that the number of cells is approximately 1,000;

[0054] (2) Place the culture plate in a 5% CO2, 37°C culture environment for 2 h to allow cells to adhere;

[0055] (3) 75 μL of the upper culture medium was removed, and 50 μL of 1× PBS solution containing different concentrations of H2O2 was added to each well and incubated for 40 min;

[0056] (4) Transfer the suspension mixture in the well to a 2 mL filter centrifuge tube and centrifuge at 800 × g for 5 min. Redisperse the cells on the filter membrane in 100 μL of cell culture medium and transfer them to the original well.

[0057] (5) Add 10 μL CCK-8 solution and continue culturing for 2.5 h;

[0058] (6) The absorbance (A) at 450 nm was measured using an ELISA instrument, and the cell viability value of the CCK-8 method was calculated according to the following formula:

[0059] Cell viability = [(A (实验组) -A (空白组) ) / (A (对照组) -A (空白组) )]×100%

[0060] A (实验组) : Absorbance of experimental wells (CCK-8 solution, H2O2-treated cells);

[0061] A (对照组) : Absorbance of control well (CCK-8 solution, cells);

[0062] A (空白组) : Blank well absorbance (CCK-8 solution).

[0063] The relationship between H2O2 concentration and cell viability determined by CCK-8 method is as follows Figure 1 As shown in A, when the H2O2 concentration is 0, 1, 8, 15, and 30 mmol / L, the cell viability values ​​corresponding to states i, ii, iii, iv, and v are 100%, 75%, 50%, 25%, and 0%, respectively. Figure 1 As shown in BF, there are five types of CTC cells after treatment: State i, the cells show the best growth state with good adhesion ( Figure 1 B); State ii, the cell membrane is vacuolated, accompanied by budding of some cells ( Figure 1 C); State iii, the cell volume shrinks, and apoptotic bodies initially appear around each cell ( Figure 1 D); State iv, the volume of apoptotic bodies around cells increases significantly ( Figure 1 E); State v, cell membrane ruptures, cells cannot adhere to the wall and are in a suspended state ( Figure 1 F). The results showed that it is feasible to evaluate cell viability based on cell morphology.

[0064] 2. Model Performance Evaluation

[0065] (1) Model accuracy

[0066] The deep learning model constructed by the specific embodiment was used to analyze the cell images obtained by treating CTC with H2O2 at different concentrations (20, 12, 10, 6, 5, 600 mmol / L) to analyze the CTC viability. The results are as follows Figure 2 As shown in (AF), each CTC in the image has a specific vitality value output. The model results are then compared with the cell vitality results measured by CCK-8, as shown in Figure 3 As shown in Figure 2, the two methods have a high degree of consistency in CTC viability analysis. Figure 4 The Bland-Altman analysis shown shows that the ordinates of all points are within this range, and the average difference between this measurement and CCK-8 is -0.255. In addition, the 95% confidence interval of the difference between the two measures is between -3.042 and 2.532, further indicating that there is good consistency between the constructed deep learning model and the CCK-8 method. The above results show that the deep learning model can accurately distinguish cells with different viability states and accurately determine the viability of each CTC through cell morphological characteristics. These analysis results verify the effectiveness of the deep learning model based on cell morphology for cell viability assessment.

[0067] (2) Comparative analysis

[0068] As shown in Table 1, compared with baseline models such as VGG-16, ResNet18, ResNeXt50_32X4D, and DenseNet-121, the EfficientNet-B0 model used in the experiment can extract cell state features and accurately classify cell images, and all performance evaluation indicators perform best.

[0069] Table 1 Comparison of the accuracy and loss rate of different CNN models in classifying cell viability status.

[0070]

[0071] Then, a series of standard indicators commonly used to evaluate machine learning models were applied to evaluate the overall performance of the deep learning model constructed by EfficientNet-B0 and YOLOv5s. The results of the performance indicators are shown in Figure 5 As shown, including an accuracy rate of 90.12% ( Figure 5 A) Recall rate 42.99% ( Figure 5 B) F1 score is close to 1 ( Figure 5 C) AP 82.31% ( Figure 5 D) and mAP82.31% ( Figure 5 E). The above analysis results all show that the model has good prediction performance.

[0072] 3. CTC Viability Determination in Real Samples

[0073] Cell images of whole blood samples were collected and imported into the constructed deep learning model to perform vitality analysis on CTCs in whole blood. The results are as follows: Figure 6 As shown in (AF), the model can quickly and accurately identify CTCs from whole blood samples and output viability values ​​based on cell morphology. The results show that the cell viability assessment model based on deep learning can be applied to the analysis of real whole blood samples.

[0074] The above description is not intended to limit the present invention, and the present invention is not limited to the above examples. Any changes, modifications, additions or substitutions made by a person skilled in the art within the spirit and scope of the present invention shall also fall within the protection scope of the present invention.

Claims

1. A method for evaluating the viability of a single CTC cell based on a deep learning method, characterized in that The following steps are involved: (1) Data collection and image preprocessing Fluorescence electron microscopy was used to obtain the initial image data of CTC cells in different states. The initial images of CTC cells were processed using the cell segmentation tool Cellpose to outline the morphological contours of individual cells and obtain a mask file. The mask file was imported into the imageJ software to obtain the coordinate information of individual cells. The individual cell coordinates were cut to obtain the characteristic image of individual CTC cells. A data set corresponding to the state classification label was constructed, and the data set of each state was randomly divided into a training set, a validation set, and a test set. (2) Optimization of CTC cell viability interval output model A. Input the training set and validation set obtained in step (1) into the EffficientNet-B0 model to perform 100-75%, 75-50%, 50-25%, and 25-0% activity interval judgment training to obtain the optimal model weight data; load the optimal model weight data into the EffficientNet-B0 model; B. Input the test set obtained in step (1) into the EffficientNet-B0 model obtained in step (2) A for image classification performance testing, and output the activity interval of the CTC image in the test set. If the preset activity value of the CTC image in the test set falls within the output activity interval, the optimal EffficientNet-B0 model is obtained; if the preset activity value of the CTC image in the test set falls outside the output activity interval, repeat step (2) A until the optimal EffficientNet-B0 model is obtained; (3) Optimization of CTC cell localization and viability prediction model A. Input the training set and validation set obtained in step (1) into the YOLOv5s model to perform localization of a single CTC cell in a cell image and prediction of the vitality value of the CTC cell, and obtain the optimal model weight data; load the optimal model weight data into the YOLOv5s model; B. Input the test set obtained in step (1) into the YOLOv5s model for image classification performance detection, and output the location and vitality value of the CTC cells in the test set. If the preset vitality value of the CTC cells in the test set is consistent with the output vitality value, the optimal YOLOv5s model is obtained; if the preset vitality value of the CTC cells in the test set is inconsistent with the output vitality value, repeat step (3) A until the optimal YOLOv5s model is obtained; (4) Assessment of single CTC cell viability A. Use a fluorescent electron microscope to photograph the sample containing CTC cells to obtain the image data of the cells to be tested, input the image data of the cells to be tested into the optimal EffficientNet-B0 model, and output the activity range of CTC cells, with the upper and lower values ​​correspondingly as v1 and v2; B. Input the cell image data obtained in step (4) A into the optimal YOLOv5s model, output the detection probability P1 based on the data set of the v1 vitality value, output the detection probability P1 based on the data set of the v2 vitality value, and use v1*P1+v2*P2 to calculate the vitality value of a single CTC cell.

2. A method for evaluating the viability of a single CTC cell based on a deep learning method according to claim 1, characterized in that The method for acquiring the initial images of CTC cells in different states in step (1) is as follows: Hela cells are stimulated by exogenous reactive oxygen species to change their viability. First, CTC cells are placed in 2 mL of culture medium and incubated for 18 h in a culture environment of 5% CO2 and 37°C, and 1.5 mL of the upper culture medium is removed; well-grown, adherent CTC cells are added with 1 mL of 1×PBS solution containing H2O2 at concentrations of 0 mmol / L, 1 mmol / L, 8 mmol / L, 15 mmol / L, and 30 mmol / L, respectively, and incubated for 40 min to obtain CTC cells in five classification states i, ii, iii, iv, and v, and the corresponding viability values ​​are 100%, 75%, 50%, 25%, and 0%, respectively.

3. The method for evaluating the viability of a single CTC cell based on a deep learning method according to claim 1, characterized in that: The ratio of the number of training sets, validation sets and test sets described in step (1) is 6:3:

1.

4. A method for evaluating the viability of a single CTC cell based on a deep learning method according to claim 1, characterized in that In step (2), the parameters of the EffficientNet-B0 model are set as follows: the number of samples in each training batch is specified to be 32, the total number of training rounds is set to 100, and the number of categories in the data set is 5.

5. The method for evaluating the viability of a single CTC cell based on a deep learning method according to claim 1, characterized in that: The core structural module of the EffficientNet-B0 model in step (2) is the mobile flip bottleneck convolution MBConv module. The mobile flip bottleneck convolution module performs 1×1 point-by-point convolution on the input CTC image and changes the output channel dimension according to the expansion ratio, and then performs k×k depth convolution. When compression and excitation operations are introduced, the mobile flip bottleneck convolution will be performed after the depth convolution, and the original channel dimension will be restored at the end of 1×1 point-by-point convolution, followed by connection inactivation and input skip connection, so that the model has a random depth.

6. The method for evaluating the viability of a single CTC cell based on a deep learning method according to claim 1, characterized in that: In step (3), the YOLOv5s model includes a Conv module and a C3 module. The Conv module performs two upsampling and two downsampling processes on the input; then the C3 module is used to extract features. The C3 module defines a neural network module for implementing the CSP Botleneck structure.

Images