A deep learning-based method for rapid detection of circulating tumor cells

Abstract

The application provides a kind of deep learning-based circulating tumor cell rapid detection method, first with RetinaNet network structure as foundation, backbone Backbone uses RepVGG, adopts multi-branch structure when training, becomes single-path structure when inference;On the basis of FPN, increase a top-down route, fuse the semantic information of bottom layer and top layer;Simplify detection head Head, optimize classification and regression subnetwork;Reasonably optimize loss function for different stages, that is, new images collected from biological samples are processed by a fast detection model, and CTCs detection results are outputted.Cell recognition, detection and classification are carried out on fluorescent microscopic images.The fast detection model takes into account multiple performance indicators such as cell detection efficiency and accuracy, has high accuracy and strong robustness, and solves the problems of slow circulating tumor cell detection speed and low accuracy in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence and biomedical cross, mainly relates to the technical field of deep learning and image processing, and particularly relates to a circulating tumor cell rapid detection method based on deep learning. BACKGROUND

[0002] Circulating tumor cells (CTCs) are tumor cells that fall off from solid tumor lesions (primary or metastatic lesions) and enter peripheral blood circulation, and survive in a small number after evading immune killing, and are closely related to tumor staging, recurrence and metastasis and drug resistance. As a typical biomarker in liquid biopsy, CTCs have a wide application prospect in early diagnosis of tumors, prognosis, and efficacy monitoring. CTCs are very rare in patient blood, and usually need to be separated and enriched from CTCs in patient blood, then stained and prepared, and finally detected and analyzed on a microscopic image. At present, pathologists generally use manual reading to detect CTCs, and also need to accurately count them. This manual detection method is not only low in efficiency, but also affected by subjective factors such as technical level and experience of pathologists. In recent years, with the continuous expansion of the demand for automatic detection of CTCs, researchers have introduced traditional image processing algorithms into the image detection of CTCs, such as distance transformation, morphological algorithm, maximum stable value region detection, etc., but these methods have poor specificity and low sensitivity, resulting in low detection accuracy. In addition, researchers have applied machine learning to the detection and counting of CTCs. Machine learning is also applied to CTC detection due to its speed and ability to overcome noise. Most of the researches use the size, density, deformability or adhesion of tumor cells to extract features, then select different classifiers for training and classification, and finally obtain the detection result. However, this method needs to manually extract features, and is too dependent on the knowledge and experience of experts, resulting in large deviation in detection.

[0003] Deep learning has been widely applied in biomedical image analysis, and has achieved good results in medical image recognition, segmentation and classification, and has been successfully applied in cancer diagnosis and tissue recognition. Compared with traditional manual feature extraction methods, image processing methods based on convolutional neural network (CNN) can automatically learn complex data features, can be trained end-to-end, and can reduce subjective and consistent bias. At present, only a few studies have explored the use of deep learning for CTC detection. For example, He et al. constructed a new CNN network based on AlexNet, and used fluorescence in situ hybridization (imFISH) to achieve CTC detection and counting on fluorescently labeled images, which demonstrated the advantages of deep learning in image feature extraction; Guo et al. used some classic CNN models, including VGG16, VGG19, ResNet18, AlexNet, and ResNet50, to perform binary classification tasks on multiple fluorescently labeled CTC images, achieving high accuracy. The model structures proposed in the above studies are relatively simple, and only involve single target detection or binary classification problems, and the model performance cannot meet the requirements of clinical application.

[0004] After CTCs are captured, detection and identification become the focus and difficulty. Although the traditional manual film reading CTC detection method can detect CTCs, it has low automation level, long time consumption, low precision, and is affected by subjective errors. The application of deep learning method in CTC detection has the following advantages:

[0005] (1) Using deep learning method to train and identify CTC sample pictures is more efficient and more accurate;

[0006] (2) Compared with traditional CTC detection methods, no prior complex mechanism knowledge is needed, and the CTC detection model based on convolutional neural network has higher accuracy and reliability;

[0007] (3) Using deep learning network for training, prediction and classification can greatly eliminate personal subjective errors.

[0008] At present, the deep learning model applied to CTC detection is relatively simple, and the sensitivity and accuracy are poor, which cannot meet the requirements of clinical application.

[0009] The foregoing background knowledge is intended to help those of ordinary skill in the art to understand the prior art close to the present application, and to facilitate the understanding of the inventive concept and technical scheme of the present application. It should be clear that in the absence of explicit evidence that the above-mentioned content was disclosed before the filing date of the present patent application, the above-mentioned background technology should not be used to evaluate the novelty of the technical scheme of the present application. SUMMARY

[0010] Technical problem

[0011] In order to solve the above problems, the purpose of the present application is to provide a deep learning-based circulating tumor cell rapid detection method, a new rapid detection model is proposed for the rapid detection of CTCs, involving a tumor cell rapid detection method based on deep learning and image processing, cell recognition, detection and classification are performed on fluorescence microscopic images, the rapid detection model takes into account cell detection efficiency, accuracy and other performance indicators, has high accuracy and strong robustness, greatly eliminates personal subjective errors, and solves the problems of slow circulating tumor cell detection speed and low accuracy in the prior art.

[0012] Technical scheme

[0013] That is, the present application comprises the following technical solutions.

[0014] Scheme one, the construction method of the circulating tumor cell rapid detection model, comprising:

[0015] (1) Based on the RetinaNet network structure, the backbone network Backbone adopts RepVGG, and a multi-branch structure is used to enhance feature extraction during training, and the structure is changed into a single-path structure by using the structure reparameterization idea during inference; reduce memory occupation and speed up the inference speed;

[0016] (2) On the basis of the traditional feature pyramid network FPN, a top-down route is added to fuse the semantic information of the bottom and the top; thereby compensating and enhancing the positioning information;

[0017] (3) Simplify the detection head Head, optimize the classification and regression subnetwork, reduce the number of serial convolution and the number of channels on the basis of the traditional RetinaNet detection head, and use large separable convolution; while reducing the model parameters and the amount of calculation, increase the receptive field of the feature map, and then enhance the model detection performance;

[0018] (4) Reasonably optimize the loss function: Smooth L1 loss, GIoU loss and varifocalloss for different stages; obtain a rapid detection model.

[0019] Further, the multi-branch structure of step (1) comprises at least one of Conv-BN structure and BN structure.

[0020] Further, the step (1) sets a weight adjustment mechanism to dynamically adjust the weights of different branch structures; the weight adjustment mechanism enables the model to automatically adjust the weights of each branch structure according to different characteristics of the input data, dynamically select which branch is active at a specific moment, and this selection can be based on the characteristics of the data or the needs of the current training stage, which helps to enhance the generalization ability of the model, thereby improving the adaptability and efficiency of the model. The multi-branch structure needs to process more parameters and calculations during training, which will make the training process more complex and time-consuming. In addition, the multi-branch structure needs to save the intermediate results of each branch until the final fusion operation, which increases the memory access cost during inference, further slowing down the processing speed. Further verification shows that for the same batch of training set data, setting the weight adjustment mechanism helps to reduce the training time by at least 10% while maintaining the model accuracy. The cycle tumor cell rapid detection model obtained in this way has the advantages of high accuracy and fast inference speed, and can realize efficient cell multi-classification recognition and CTC detection.

[0021] Further, in the step (4), IoU-Aware Clas Quartification Score (IACS) is used in the training process to represent the confidence of object existence and the positioning accuracy, so as to produce more accurate detection levels in the dense object detector.

[0022] Further, in the step (4), the reasonable optimization of the loss function is specifically: for boundary box regression, the loss function in the early training stage is dominated by Smooth L1 loss; the loss function after the boundary box prediction is stable is dominated by GIoU loss; for target classification, varifocal loss is used.

[0023] Further, the method for constructing the cycle tumor cell rapid detection model further comprises:

[0024] The VOC format labeled data set obtained by the following steps is used to train the rapid detection model: CTCs are separated and enriched from the blood of tumor patients, and after immunofluorescence staining, multi-channel fluorescence images are collected using a fluorescence microscope. The fluorescence images are preprocessed to make various cells more visually distinguishable, and the images are manually labeled to mark the position and category of each cell, thereby obtaining a VOC format labeled data set.

[0025] Further, the pre-processing of the fluorescence image comprises:

[0026] A median filter is applied to the fluorescence image to reduce salt and pepper noise;

[0027] Bilinear downsampling is used to reduce shot noise;

[0028] The pixel intensity of each image is normalized to have the same mean and standard deviation.

[0029] Scheme II, a rapid detection model of circulating tumor cells obtained by the construction method described above.

[0030] Scheme III, a rapid detection method of circulating tumor cells based on deep learning for non-disease diagnosis and treatment, comprising applying the rapid detection model of circulating tumor cells described above to process new images collected from biological samples, and outputting the CTC detection results.

[0031] Further, the biological sample is selected from blood or lymph.

[0032] Scheme IV, application of the rapid detection method of circulating tumor cells based on deep learning for non-disease diagnosis and treatment described above in biological sample detection.

[0033] Further, the biological sample is selected from blood or lymph.

[0034] Scheme V, application of the rapid detection model of circulating tumor cells obtained by the construction method described above in biological sample detection.

[0035] Further, the biological sample is selected from blood or lymph.

[0036] Scheme VI, a computer device comprising a memory, a processor, a communication interface, and a communication bus; wherein the memory, the processor, and the communication interface communicate with each other through the communication bus; the memory is used to store computer programs; the processor is used to execute the computer programs stored in the memory, and the processor executes the computer programs to implement at least one step of the construction method of the rapid detection model of circulating tumor cells or the rapid detection method of circulating tumor cells based on deep learning.

[0037] Scheme VII, a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement at least one step of the construction method of the rapid detection model of circulating tumor cells or the rapid detection method of circulating tumor cells based on deep learning.

[0038] On the basis of common sense in the art, the above-mentioned preferred conditions can be combined with each other to obtain the specific embodiments.

[0039] Beneficial effects

[0040] According to the application, by taking the RetinaNet network structure as the basis, the backbone network Backbone adopts RepVGG, the semantic information is fused by increasing the top-down routing on the basis of the traditional feature pyramid network FPN, the detection head Head is simplified, the classification and regression sub-networks are optimized, the loss function is reasonably optimized for different stages, and the VOC format annotation data set is used to train the fast detection model, so as to obtain the circulating tumor cell fast detection model, which has the advantages of high accuracy and fast reasoning speed when detecting circulating tumor cells, can realize efficient cell multi-classification recognition and CTC detection, and greatly eliminates personal subjective errors, and solves the problems of slow circulating tumor cell detection speed and low accuracy in the prior art.

[0041] The application adopts the technical scheme, makes up for the defects of the prior art, is reasonable in design, and convenient to operate. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to make the above and / or other purposes, features, advantages and examples of the application more obvious and easy to understand, the drawings needed in the specific embodiments of the application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the application, and other drawings can be obtained from these drawings without creative labor for those skilled in the art.

[0043] Figure 1 A construction route schematic diagram of the circulating tumor cell fast detection model is shown.

[0044] Figure 2 A backbone network Backbone schematic diagram of the circulating tumor cell fast detection model is shown.

[0045] Figure 3 A feature pyramid FPN structure schematic diagram of the circulating tumor cell fast detection model is shown.

[0046] Figure 4 A detection head Head schematic diagram of the circulating tumor cell fast detection model is shown.

[0047] Figure 5 A data set cell allocation schematic diagram is shown.

[0048] Figure 6 An example of a fluorescence image and a detection result are shown (wherein A represents an image example, B represents a test result, R (red) in the middle represents a CTC, G (green) represents a normal cell, and U (blue) represents a cell nucleus).

[0049] Figure 7 An Epoch loss curve is shown.

[0050] Figure 8 Validation Set Map Curve 1 is shown.

[0051] Figure 9 Validation Set Map Curve 2 is shown. DETAILED DESCRIPTION

[0052] Those skilled in the art can readily make appropriate substitutions and / or modifications to the process parameters described herein, with the benefit of the present disclosure, without departing from the scope of the application. However, it is expressly noted that all such substitutions and / or modifications are deemed to be within the scope of those skilled in the art, in light of the present disclosure, and are thus expressly intended to be within the scope of the application. The products and methods of making described herein have been described by way of preferred examples, and it will be apparent to those skilled in the art that modifications or appropriate variations and combinations of the products and methods described herein can be made without departing from the spirit and scope of the present application.

[0053] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as those commonly understood by one of ordinary skill in the art to which this application belongs. Methods and materials are described herein for use in the present application; however, other methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present application. The materials, methods, and examples described herein are illustrative only and not intended to be limiting. All publications, patent applications, patents, database entries, and other references mentioned herein are incorporated by reference in their entirety. In case of conflict, the present specification, including definitions, will control.

[0054] The materials, methods, and examples described herein are illustrative only and not intended to be limiting. Except in the Examples, or where otherwise explicitly indicated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. Although various methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present application, suitable methods and materials are described herein.

[0055] The technical solutions in the embodiments of the present application will be clearly and completely described below. Obviously, the described embodiments are only a part of embodiments of the present application, rather than all embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application. Meanwhile, the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0056] It should be understood that any technical solution claimed by the present application does not involve the diagnosis and treatment of diseases.

[0057] In order to facilitate the understanding of the embodiments of the present application, first, the abbreviations and key terms that may be involved in the embodiments of the present application are explained and defined. For the abbreviations or key terms not defined, they are commonly understood by those skilled in the art.

[0058] In addition, the test methods used in the embodiments are conventional methods unless otherwise specified; the materials, reagents, etc. used are commercially available unless otherwise specified. The reagents or instruments not specified by the manufacturer are conventional products that can be obtained by purchase. All publications and other references mentioned herein are incorporated by reference in their entirety.

[0059] The following detailed description of the application

[0060] Example 1

[0061] This embodiment provides a rapid detection model of circulating tumor cells (CTCs) which does not involve the diagnosis and treatment method of diseases, which is based on the detection of CTCs by deep learning. Specifically, a high-efficiency target detection model is constructed based on the classical RetinaNet network structure. The construction route of the rapid detection model of CTCs based on deep learning is shown in Figure 1 , which first acquires the original data set of CTCs, including blood sample preparation, CTC separation and enrichment, immunofluorescence staining, image acquisition and preprocessing, which are used as the original data for training and testing of the deep learning model, and the data set is divided; then, the deep learning model is established, and the model is trained, tested and evaluated.

[0062] The specific construction steps of the rapid detection model of CTCs are as follows:

[0063] (1) The backbone network (Backbone) adopts RepVGG, as shown in Figure 2 , a multi-branch structure is used to enhance feature extraction during training, and during inference, the structure is reparameterized to a single-path structure, reducing memory usage while speeding up inference;

[0064] (2) Based on the traditional feature pyramid network (FPN) structure, a top-down routing is added to fuse the semantic information of the bottom and top layers, thereby compensating and enhancing the positioning information, as shown in Figure 3 ;

[0065] (3) Simplify the detection head (Head), optimize the classification and regression subnetwork, as shown in Figure 4 , based on the traditional RetinaNet detection head, reduce the number of serial convolutions and the number of channels, and use large separable convolutions to reduce model parameters and computational complexity while increasing the receptive field of the feature map, thereby enhancing the detection performance of the model. At the same time, the corresponding loss function is optimized;

[0066] (4) Finally, the loss function is optimized. For bounding box regression, Smooth L1 loss is combined with GIoU loss. In the early training stage, the loss function is dominated by Smooth L1 loss, which accelerates the convergence of the model. After the bounding box prediction is stable, GIoU loss can make the positioning more accurate. For target classification, varifocal loss is used. This loss function is based on the Focal loss solution to the imbalance between positive and negative samples. In the training process, the IoU-aware clas quartification score (IACS) that can represent the object existence confidence and positioning accuracy at the same time is used to produce more accurate detection levels in dense object detectors.

[0067] The labeled dataset is obtained by the following steps to train the fast detection model:

[0068] Step one: CTCs are separated and enriched from the blood of tumor patients, and then immunofluorescence staining is performed. Finally, multi-channel fluorescence images are collected using a fluorescence microscope.

[0069] Step two: The fluorescence images are preprocessed to make various cells more visually distinguishable. First, a median filter is applied to the fluorescence images to reduce salt and pepper noise. Then, bilinear downsampling is used to reduce shot noise. Finally, the pixel intensity of each image is normalized to have the same mean and standard deviation.

[0070] Step three: After image processing, LabelImg is used to manually annotate the images to mark the location and category of each cell, and obtain a VOC format labeled dataset. Then, the obtained CTCs dataset is divided to obtain the training set, validation set and test set.

[0071] Step four: The dataset obtained in step three is used to train, test and optimize the target detection model. Finally, the optimized model is deployed to the target platform to process new images collected from the blood of tumor patients and output the CTCs detection results.

[0072] This embodiment uses a public dataset to train and test the model. The dataset is divided into two parts: training set and test set. The training set is used for model training, and the test set is used for model evaluation.

[0073] The production process of this dataset: This dataset uses a cholangiocarcinoma cell line, and the specific allocation is as follows: Figure 5shown; then the CTCs were mixed with normal cells at a ratio of 1:1, the CTCs were stained with deep red fluorescence (Cytopainter ab 176736), the normal cells were stained with green fluorescence (Cytopainter ab 176735), and the cell nuclei were stained with blue fluorescence (Hoechst), as shown in Figure 6 Finally, they were dropped on a 96-well plate and imaged by the Opera Phenix instrument (Perkin Elmer) for fluorescence. Ultimately, 1200 fluorescence images of CTCs and normal cells were obtained. Each image consisted of 1080x1080 pixels and contained an average of 20-30 individual cells.

[0074] The rapid detection model was trained using the above dataset. The model training strategy adopted transfer learning and frozen training. Using the pre-trained model on ImageNet, the backbone network was first trained on the dataset for 50 epochs, and then unfreezed for 250 epochs, i.e. a total of 300 epochs. The batch size was set to 4, the Adam optimizer was used, and the learning rate was set to 3e-4. The results are shown in Figure 7 and Figure 8 The training and validation losses quickly converged, and the Map value on the validation set was high. This indicates that compared to classic models such as Faster RCNN, SSD, RetinaNet, etc., the rapid detection model of the present application converges significantly faster and the Map value during training is significantly improved.

[0075] As a further optimization of the training mode, an interference subset was additionally added to the training set used for model training, which included: (1) fluorescence images that were not preprocessed with a median filter and bilinear downsampling, and the various cells were relatively difficult to distinguish visually; (2) the fluorescence images were first preprocessed independently in the channel using Morphology Filter and Denoising to preserve spherical structures and remove background noise, and then pixel intensity normalization was performed; (3) first perform desaturation processing, find the darkest center pixel point by Kmean or brightness, lower the pixel value close to this center pixel point, and make the very dark area that should be background into black, and then perform pixel intensity normalization; the proportion of each interference subset in the training set was 2-10%. When the proportion of each interference subset in the training set was set to 10%, the same training method was used to obtain the validation set results, as shown in Map curve 2 in Figure 9 Figure 8 ​After adding the interference subset to the training set, the model converges faster, and the Map value in the training process increases faster, which indicates that expanding the training set with the new images of the interference subset helps to improve the generalization ability and robustness of the model, and significantly improves the performance of the model. Moreover, the setting of the weight adjustment mechanism helps the model to adjust different branch structures when facing input images that have never been seen before, thereby improving the generalization ability of the model.

[0076] In order to verify the detection ability of the rapid detection model for CTCs, the performance of the rapid detection model is evaluated by using the test set, and is compared with classic models such as Faster R-CNN, SSD, YOLOv3, RetianNet and Ours, and the evaluation results are shown in Table 1. The Map value of the rapid detection model of the present application reaches 90.43%, and the F1 score reaches 0.882, which indicates that the rapid detection model of the present application has better detection accuracy and faster model inference speed, and the comprehensive performance of the model reaches the optimum.

[0077] Table 1: Comparison results of detection performance of the rapid detection model and other classic models

[0078]

[0079] Therefore, the present embodiment also provides a rapid detection model of circulating tumor cells obtained by the aforementioned construction method.

[0080] Also provided is a rapid detection method of circulating tumor cells for non-disease diagnosis and treatment based on deep learning, which comprises applying the aforementioned rapid detection model of circulating tumor cells to process new images collected from biological samples, and outputting CTC detection results. The biological samples are selected from blood or lymph.

[0081] Also provided is an application of the aforementioned rapid detection method of circulating tumor cells for non-disease diagnosis and treatment based on deep learning in biological sample detection, and the application does not involve the diagnosis and treatment of diseases. The biological samples are selected from blood or lymph.

[0082] Also provided is an application of the rapid detection model of circulating tumor cells obtained by the aforementioned construction method in biological sample detection, and the application does not involve the diagnosis and treatment of diseases. The biological samples are selected from blood or lymph.

[0083] Example 2:

[0084] Also provided is a computer readable storage medium, which stores a computer program executable by a processor, the computer program, when executed by the processor, performing at least one step in the specific construction method of the circulating tumor cell rapid detection model and achieving the same technical effects. To avoid repetition, the present embodiment will not be described again.

[0085] Embodiment 3:

[0086] A computer device includes a memory, a processor, a communication interface and a communication bus; wherein the memory, the processor and the communication interface communicate with each other through the communication bus; the memory is used to store a computer program; the processor is used to execute the computer program stored in the memory, and the processor executes the computer program to implement at least one step in the specific construction method of the circulating tumor cell rapid detection model and achieve the same technical effects. To avoid repetition, the present embodiment will not be described again.

[0087] Computer readable media includes permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. The information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage device, or any other non-transmission medium that can be used to store information accessible by a computing device. According to the definition herein, computer readable media does not include transitory computer readable media, such as modulated data signals and carriers.

[0088] The conventional techniques in the above embodiments are known to those skilled in the art, and therefore will not be described in detail here.

[0089] The specific embodiments described herein are merely illustrative of the spirit of the present application. Those skilled in the art of the present application can make various modifications or supplements to the described specific embodiments or replace them with similar ways, without departing from the spirit of the present application or exceeding the scope defined by the appended claims.

[0090] While the application has been described and illustrated with reference to specific embodiments, it is understood that various modifications can be made by those skilled in the art without departing from the spirit and scope of the application.

[0091] While the forgoing detailed description has shown, described, and pointed out the novel features of the various embodiments, it will be understood that various omissions, substitutions, and changes in the form and details of the device or process illustrated can be made by those skilled in the art without departing from the spirit of the disclosure. The various features and methods described above can be used independently of one another, or can be combined in various ways. All possible combinations and subcombinations are intended to fall within the scope of the present disclosure. Many embodiments have been described in great detail herein. While the embodiments described above have specific components, features and / or arrangements, embodiments having different components, features and / or arrangements can also be constructed according to the principles of the present disclosure. It is therefore understood that the application can be practiced otherwise than as specifically explained and illustrated herein. This includes structural and / or logical changes in the various components and / or methods described herein and combinations of components and / or methods described herein with other components and / or methods not specifically described herein. The scope of the application should therefore be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.

[0092] The details of this application are subject to known art.

Claims

1. A method for constructing a rapid detection model of circulating tumor cells, characterized in that The method comprises the following steps: (1) based on the RetinaNet network structure, the backbone network Backbone adopts RepVGG, and a multi-branch structure is adopted during training to enhance feature extraction, and the structure is changed into a single-path structure by using the structure reparameterization idea during inference; (2) based on the traditional feature pyramid network FPN, a top-down route is added to fuse the semantic information of the bottom layer and the top layer; (3) simplify the detection head Head, optimize the classification and regression subnetwork, reduce the number of serial convolution and the number of channels based on the traditional RetinaNet detection head, and adopt a large separable convolution; (4) reasonably optimize the loss function: Smooth L1 loss, GIoU loss and varifocal loss for different stages to obtain a fast detection model; The multi-branch structure of step (1) comprises at least one of a Conv-BN structure and a BN structure; In step (1), a weight adjustment mechanism is arranged to dynamically adjust the weights of different branch structures; In step (4), the reasonable optimization of the loss function is as follows: for the boundary box regression, the loss function in the early training stage is dominated by Smooth L1 loss; the loss function after the boundary box prediction is stable and is dominated by GIoU loss; for target classification, varifocal loss is adopted; Further comprising: using the VOC format labeled data set obtained by the following steps to train the fast detection model: separating and enriching CTCs from tumor patient blood, collecting multi-channel fluorescence images using a fluorescence microscope after immunofluorescence staining, preprocessing the fluorescence images to make various cells more visually distinguishable, manually labeling the images to mark the position and category of each cell, and obtaining a VOC format labeled data set.

2. The method for constructing a circulating tumor cell fast detection model according to claim 1, wherein: The preprocessing of the fluorescence image comprises: Applying a median filter to the fluorescence image to reduce salt and pepper noise; Using bilinear downsampling to reduce shot noise; Normalizing the pixel intensity of each image to have the same mean value and standard deviation.

3. A system of a circulating tumor cell fast detection model obtained by the construction method of claim 1 or 2.

4. Application of the circulating tumor cell fast detection model obtained by the construction method of claim 1 or 2 in biological sample detection. 5.A deep learning-based rapid detection method for non-disease diagnosis and treatment of circulating tumor cells, characterized by: The application comprises processing a new image collected from a biological sample using the system of the circulating tumor cell fast detection model of claim 3, and outputting a CTC detection result.

6. Application of the deep learning-based non-disease diagnosis and treatment circulating tumor cell fast detection method of claim 5 in biological sample detection.

7. A computer device comprising a memory, a processor, a communication interface, and a communication bus; wherein, The memory, the processor and the communication interface communicate with each other through the communication bus; the memory is used for storing a computer program; the processor is used for executing the computer program stored on the memory, and the computer program is characterized in that: when the processor executes the computer program, the method for constructing the circulating tumor cell rapid detection model of claim 1 or 2 or the method for rapidly detecting circulating tumor cells based on deep learning for non-disease diagnosis and treatment of claim 5 is implemented.

Images