Respiratory tract tumor cell identification method and device based on artificial intelligence algorithm

Abstract

The invention provides a respiratory tract tumor cell identification method and device based on an artificial intelligence algorithm. The method comprises the following steps: acquiring a respiratory tract tumor cell fluorescence staining picture, slitting the respiratory tract tumor cell fluorescence staining picture, performing artificial intelligence identification on sub-pictures and splicing the sub-pictures. According to the respiratory tract tumor cell identification method based on the artificial intelligence algorithm provided by the invention, through the respiratory tract tumor cell identification neural network model, rapid identification of the respiratory tract tumor cells on the respiratory tract tumor cell fluorescent staining picture is realized, the identification efficiency of the respiratory tract tumor cells is greatly improved, and the identification accuracy is improved. The diagnosis efficiency of the respiratory tumor cells is improved; meanwhile, the method realizes the marking of the respiratory tumor cells based on the artificial intelligence network model, thereby facilitating the realization of automatic operation, and effectively reducing the manpower diagnosis cost of the respiratory tumor cells.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for identifying respiratory tumor cells based on an artificial intelligence algorithm. Background Art

[0002] The diagnosis of respiratory tumor cells usually involves quantitative analysis of cell morphology, nuclear morphology, nuclear division, cytoplasmic morphology and other characteristics, combined with technical means such as nuclear staining and immunohistochemical staining to identify respiratory tumor cells. Typical diagnostic methods include cytological examination, histological examination, molecular biological detection and other methods. Through the analysis of cell morphology, tissue structure and biomarkers, doctors can be assisted in the diagnosis and differential diagnosis of respiratory tumors. However, the above typical diagnostic methods have problems such as low detection efficiency, slow diagnosis speed and poor consistency of manual interpretation results. Summary of the invention

[0003] Therefore, the technical problem to be solved by the present invention is to overcome the defects of the prior art such as low efficiency of respiratory tumor cell detection, slow diagnosis speed and poor consistency of manual interpretation results, thereby providing a respiratory tumor cell identification method based on artificial intelligence algorithm, which can effectively improve the detection efficiency of respiratory tumor cells in samples.

[0004] To this end, in a first aspect, the present invention provides a method for identifying respiratory tumor cells based on an artificial intelligence algorithm, comprising the following steps:

[0005] Obtain a fluorescent staining image of respiratory tumor cells, and photograph the immunopurified respiratory tumor cells using a fluorescent dark field detection device to obtain a fluorescent staining image of respiratory tumor cells;

[0006] Cutting the fluorescent staining image of respiratory tract tumor cells into N sub-images, numbering the sub-images, and recording coordinate information of a predetermined position of each sub-image as first coordinate information of the sub-image;

[0007] An artificial intelligence identification sub-image is input into a respiratory tumor cell identification neural network model to perform respiratory tumor cell identification processing on the sub-image to obtain a first sub-image of respiratory tumor cells with a rectangular frame identification; wherein the respiratory tumor cell identification neural network model is trained based on a deep learning target detection algorithm using a fluorescent staining image of manually annotated respiratory tumor cells as input data, wherein the rectangular frame identification includes rectangular frame coordinate information and positive probability data of respiratory tumor cells identified by the rectangular frame, and the fluorescent staining image of the respiratory tumor cells is an image of different types of respiratory tumor cells, such as squamous cell carcinoma cells, adenocarcinoma cells, and small cell lung cancer cells; and images of respiratory tumor cells at different stages of carcinogenesis, such as early, middle, and late stages of carcinogenesis, have different morphological and structural characteristics;

[0008] Splice the sub-images, combine the coordinate information of the rectangular box mark in the first sub-image with the first coordinate information, calculate the coordinate information of the respiratory tumor cells marked by the rectangular box in the first sub-image in the fluorescent staining image, realize the splicing of the first sub-image, and obtain the marked respiratory tumor cell fluorescent staining image.

[0009] Furthermore, before segmenting the respiratory tract tumor cell fluorescence staining image, the respiratory tract tumor cell identification method also includes an image preprocessing step, performing one or more of image enhancement, image size standardization, and denoising operations on the respiratory tract tumor cell fluorescence staining image.

[0010] Furthermore, the deep learning target detection algorithm is a yolov5 algorithm that retains an 8-fold reduction in the image.

[0011] Furthermore, the network loss function of the respiratory tumor cell identification neural network model is: Among them, the coordinate loss function The calculation formula is:

[0012] Among them, N loc is the number of elements in the feature map, is the GIoU loss function, is the manually marked rectangular box identifying the coordinate information, and Represents the coordinate information of the rectangular box output by the neural network model for identifying respiratory tumor cells. The rectangular box indicates that the location is a respiratory tumor cell;

[0013] Classification loss function It is used to determine whether the position marked by the rectangular box is a positive respiratory tumor cell, so as to distinguish the false positive of respiratory tumor cells. The calculation formula is: Among them, ρij Represents the probability value of the respiratory tumor cells marked by the rectangular box being positive, with a value between 0 and 1.0; N a is the feature map height, N o is the feature map width.

[0014] Preferably, the value of N is not less than 4, and more preferably, the value of N is equal to 4.

[0015] Preferably, the predetermined position is the upper left corner of the sub-image.

[0016] In a second aspect, the present invention provides a respiratory tumor cell identification device based on an artificial intelligence algorithm, comprising:

[0017] An acquisition module is used to acquire a fluorescence staining image of respiratory tumor cells, and to photograph the immunopurified respiratory tumor cells using a fluorescence dark field detection device to obtain a fluorescence staining image of the respiratory tumor cells;

[0018] A cutting module is used to cut the fluorescent staining image of respiratory tumor cells into N*N sub-images, and to number the sub-images, and to record the coordinate information of the predetermined position of each sub-image as the first coordinate information of the sub-image;

[0019] An identification module, an artificial intelligence identification sub-image, inputs the sub-image into a respiratory tumor cell identification neural network model to perform respiratory tumor cell identification processing on the sub-image to obtain a first sub-image of respiratory tumor cells with a rectangular frame identification; wherein the respiratory tumor cell identification neural network model is trained based on a deep learning target detection algorithm using a fluorescent staining image of manually annotated respiratory tumor cells as input data, wherein the rectangular frame identification includes rectangular frame coordinate information and positive probability data of respiratory tumor cells identified by the rectangular frame, and the fluorescent staining image of the respiratory tumor cells is an image of different types of respiratory tumor cells, such as squamous cell carcinoma cells, adenocarcinoma cells, and small cell lung cancer cells; and images of respiratory tumor cells at different stages of carcinogenesis, such as early, middle, and late stages of carcinogenesis, whose morphological and structural characteristics are different;

[0020] The stitching module combines the coordinate information of the rectangular box mark in the first sub-image with the first coordinate information, calculates the coordinate information of the respiratory tumor cells marked by the rectangular box in the first sub-image in the fluorescent staining image, realizes the stitching of the first sub-image, and obtains the marked respiratory tumor cell fluorescent staining image.

[0021] Furthermore, the respiratory tract tumor cell identification device also includes an image preprocessing module, which performs one or more of image enhancement, image size standardization, and denoising operations on the fluorescent staining image of the respiratory tract tumor cells.

[0022] Furthermore, the respiratory tumor cell identification device also includes an artificial intelligence interpretation module, which extracts features from the fluorescent staining images of respiratory tumor samples: it involves identifying key properties in the image such as cell type, morphology, size and color intensity, etc. These features are crucial for subsequent interpretation and identification; cell recognition: the algorithm will distinguish and identify different cell and tissue types based on the extracted features, thereby accurately marking cancer cells and their development stages; image analysis: the AI ​​module uses a deep learning network to conduct a deeper analysis and understanding of these features, which can evaluate the heterogeneity of cells, the degree of lesions, and the expression of other important biomarkers; data processing: all analysis results are integrated and processed to generate reports that are easy for doctors to understand, providing clinical decision support on pathological status, lesion type and possible treatment directions. This series of processes ultimately realizes the interpretation of fluorescent staining images of respiratory tumor samples.

[0023] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for identifying respiratory tumor cells based on an artificial intelligence algorithm is implemented.

[0024] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for identifying respiratory tumor cells based on an artificial intelligence algorithm.

[0025] The technical solution of the present invention has the following advantages:

[0026] 1. The respiratory tract tumor cell identification method based on artificial intelligence algorithm provided by the present invention realizes rapid identification of respiratory tract tumor cells on the fluorescent staining images of respiratory tract tumor cells through the respiratory tract tumor cell identification neural network model, which greatly improves the identification efficiency of respiratory tract tumor cells and improves the diagnosis efficiency of respiratory tract tumor cells; at the same time, since the method realizes respiratory tract tumor cell labeling based on the artificial intelligence network model, it is easy to realize automated operation and effectively reduce the human diagnosis cost of respiratory tract tumor cells; the fluorescent staining images of respiratory tract tumor cells include images of tumor cells of different types and different stages, which can greatly improve the recognition ability of the respiratory tract tumor cell identification neural network model for different types of respiratory tract tumor cells and tumor cells of different stages.

[0027] 2. The respiratory tract tumor cell identification method based on artificial intelligence algorithm provided by the present invention can effectively reduce hardware requirements and save processor costs by segmenting the respiratory tract tumor cell fluorescence image and then inputting it into the respiratory tract tumor cell identification neural network model for computer processing;

[0028] 3. The respiratory tumor cell identification method based on artificial intelligence algorithm provided by the present invention can effectively improve the quality and accuracy of the fluorescence image data of respiratory tumor cells by performing image preprocessing on the fluorescence imaging images of respiratory tumor samples, including image enhancement, size standardization, noise removal, etc.

[0029] 4. The method for identifying respiratory tumor cells based on artificial intelligence algorithm provided by the present invention can directly mark the respiratory tumor cells with rectangular frames and display the positive probability of respiratory tumor cells in the pictures after being processed by the neural network model of respiratory tumor cells, so as to facilitate users such as patients and testing doctors to interpret the pictures.

[0030] 5. The respiratory tumor cell identification device based on artificial intelligence algorithm provided by the present invention realizes rapid identification of respiratory tumor cells in respiratory tumor cell fluorescence staining images through an identification module, which greatly improves the identification efficiency of respiratory tumor cells in respiratory tumor cell fluorescence staining images; improves the diagnostic efficiency of respiratory tumor cells; at the same time, because this method realizes respiratory tumor cell labeling based on an artificial intelligence network model, it is easy to realize automated operation and effectively reduce the human diagnosis cost of respiratory tumor cells; after the image is processed by the respiratory tumor cell neural network model, the respiratory tumor cells can be directly marked with rectangular frames and the positive probability of respiratory tumor cells can be displayed, which is convenient for users such as patients and testing doctors to interpret the images.

[0031] 6. The respiratory tumor cell identification device based on artificial intelligence algorithm provided by the present invention performs feature extraction on the fluorescent staining images of respiratory tumor samples by adding an artificial intelligence interpretation module: it involves identifying key properties in the image such as cell type, morphology, size and color intensity, etc. These features are crucial for subsequent interpretation and identification; cell identification: the algorithm will distinguish and identify different cell and tissue types based on the extracted features, thereby accurately marking cancer cells and their development stages; image analysis: the AI ​​module uses a deep learning network to perform a deeper level of analysis and understanding of these features, which can evaluate the heterogeneity of cells, the degree of lesions, and the expression of other important biomarkers; data processing: all analysis results are integrated and processed to generate reports that are easy for doctors to understand, providing clinical decision support on pathological status, lesion type and possible treatment direction. This series of processes ultimately realizes the interpretation of fluorescent staining images of respiratory tumor samples.

[0032] 7. The electronic device provided by the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the electronic device executes a respiratory tumor cell identification method based on an artificial intelligence algorithm, it realizes rapid identification of respiratory tumor cells in fluorescent staining images of respiratory tumor cells, thereby effectively improving the identification efficiency of respiratory tumor cells.

[0033] 8. The non-transitory computer-readable storage medium provided by the present invention stores a computer program thereon. When the computer program is executed in a respiratory tumor cell identification method based on an artificial intelligence algorithm, the computer program can realize rapid identification of respiratory tumor cells in fluorescent staining images of respiratory tumor cells, thereby effectively improving the identification efficiency of respiratory tumor cells. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 This is a flow chart of the respiratory tract tumor cell identification method based on artificial intelligence algorithm in Example 1 of the present invention;

[0036] Figure 2 This is a fluorescence staining picture of respiratory tumor cells obtained in an embodiment of the present invention;

[0037] Figure 3 This is a manually labeled fluorescent staining image of respiratory tumor cells;

[0038] Figure 4 for Figure 2 Image processed by the respiratory tumor cell identification neural network model

[0039] Figure 5 This is a flow chart of a respiratory tumor cell identification method based on an artificial intelligence algorithm in Example 2 of the present invention;

[0040] Figure 6 for Figure 2 The cut pictures;

[0041] Figure 7 This is a principle block diagram of a respiratory tumor cell identification device based on an artificial intelligence algorithm in Example 3 of the present invention;

[0042] Figure 8 This is a principle block diagram of a respiratory tumor cell identification device based on an artificial intelligence algorithm in Example 4 of the present invention;

[0043] Fig. 9 This is a principle block diagram of an electronic device in Embodiment 5 of the present invention. DETAILED DESCRIPTION

[0044] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0046] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, it can also be the internal connection of two components, it can be a wireless connection, or it can be a wired connection. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0047] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0048] Example 1

[0049] This embodiment provides a method for identifying respiratory tumor cells based on an artificial intelligence algorithm, which is applied to a respiratory tumor fluorescence staining image detection device, such as Figure 1 As shown, the following steps are included:

[0050] Step 101: Obtain a fluorescence staining image of respiratory tumor cells, and photograph the immunopurified respiratory tumor cell smear using a fluorescence dark field detection device to obtain a fluorescence staining image of respiratory tumor cells, such as Figure 2 shown.

[0051] Step 102: Input the respiratory tract tumor cell fluorescence staining picture into the respiratory tract tumor cell identification neural network model to perform respiratory tract tumor cell identification processing on the respiratory tract tumor cell fluorescence staining picture to obtain a respiratory tract tumor cell picture with a rectangular frame identification; wherein the respiratory tract tumor cell identification neural network model is a fluorescent staining picture of respiratory tract tumor cells that is manually annotated (such as Figure 3 As shown) is used as input data, and is trained based on a deep learning target detection algorithm. In the present invention, training is online learning. After the labeling personnel complete a certain amount of labeling, the training will be automatically triggered. Then, after the training is completed, the model will be automatically updated as the latest pre-labeling to assist the labeling personnel in labeling and reviewing more efficiently. The whole process is a closed loop. The achieved effect is that the labeling personnel label faster and faster, the quality is getting higher and higher, and the pre-labeling and algorithm detection effects are getting better and better. The whole process is a closed loop, and the algorithm is automatically iterated and updated with the labeling. Among them, the rectangular box mark includes the rectangular box coordinate information and the positive probability data of the respiratory tumor cells marked by the rectangular box. The present invention also adds a labeling function for suspected tumor cells to solve the problem that some manual work cannot distinguish between miscellaneous bacteria and acid-fast. As Figure 4 shown.

[0052] The respiratory tumor cell identification neural network model used in the present invention is trained by a large amount of respiratory tumor cell fluorescence staining image data, including respiratory tumor cell fluorescence staining images of different types and stages, which can greatly improve the recognition ability of the respiratory tumor cell identification neural network model for different types of respiratory tumor cells and tumor cells at different stages; before identifying the respiratory sample fluorescence staining picture, the respiratory sample fluorescence staining picture is preprocessed, including image enhancement, size standardization and noise removal, which can greatly improve the quality of the respiratory sample fluorescence staining picture and improve the accuracy of respiratory tumor identification; in addition, the training of the respiratory tumor cell identification neural network model used in this embodiment is online learning. After the labeling personnel complete a certain amount of labeling, the training will be automatically triggered, and then the model will be automatically updated as the latest pre-labeling after the training is completed, assisting the labeling personnel to label and review more efficiently; the whole process is closed-loop. The effect achieved is that the labeling personnel label faster and faster, the quality is getting higher and higher, and the pre-labeling and algorithm detection effects are getting better and better. The whole process is closed-loop, and the algorithm is automatically iterated and updated with the labeling. The training method of the neural network model for respiratory tumor cell identification can greatly increase the data volume and accuracy of model training, while significantly reducing the workload of manually annotating training images.

[0053] Select a deep learning model suitable for image recognition tasks, such as a convolutional neural network (CNN); design and build multiple layers of convolutional layers, pooling layers, and fully connected layers to extract features from images and classify them; divide the dataset into training and test sets to ensure that the model is tested on unseen data, use the training set to train the deep learning model, and use the back-propagation algorithm and optimizer to update the model parameters and minimize the loss function; use cross-validation and other techniques to tune the model to prevent overfitting; use the test set to evaluate the trained model, including accuracy, recall, and precision indicators to evaluate the performance of the model.

[0054] Among them, accuracy is the most intuitive performance indicator, which measures the proportion of correct predictions (whether positive or negative) made by the model. The formula is:

[0055]

[0056] In respiratory tumor cell identification, high accuracy means that the model can effectively distinguish tumor cells from non-tumor cells, which is crucial for subsequent treatment selection and disease management.

[0057] The recall rate, also known as the true positive rate, measures the proportion of positive samples (tumor cells) correctly identified by the model to all actual positive samples. This is especially important for medical diagnosis, as missed diagnoses (false negatives) may lead to delayed treatment. The formula is:

[0058]

[0059] In the test set, high recall indicates that the vast majority of tumor cells are correctly identified by the model, which helps ensure that patients receive necessary treatment.

[0060] The precision rate is the proportion of samples predicted as positive by the evaluation model that are actually positive. In tumor identification, a high precision rate means fewer false positive results, thereby reducing the risk of unnecessary treatment due to misdiagnosis. The formula for calculating the precision rate is:

[0061]

[0062] The accuracy of the neural network model for respiratory tumor cell identification (test set) is 90%, and the actual accuracy using manual comparison is about 93% to 96%; the recall rate is 100%, and the precision rate is 85%. 20,000 images were used as the training set, and the test set was 2,000 images; the optimizer was SGD, the batch size was 8 (two 1080ti graphics cards), the input image resolution was: 2736*1824 (width*height), the learning rate decayed from 0.01 to 0.0001, and epoch=400.

[0063] The above-mentioned respiratory tumor cell identification method based on artificial intelligence algorithm realizes the rapid identification of respiratory tumor cells on the fluorescent staining images of respiratory tumor cells through the respiratory tumor cell identification neural network model, which greatly improves the identification efficiency of respiratory tumor cells and improves the diagnostic efficiency of respiratory tumor cells; at the same time, because this method realizes respiratory tumor cell labeling based on the artificial intelligence network model, it is easy to realize automated operation and effectively reduce the human diagnosis cost of respiratory tumor cells.

[0064] Since the coordinates of the rectangular box in the fluorescent staining image of respiratory tumor cells are small, the coordinate regression loss of the rectangular box is negligible after processing by the deep learning target detection algorithm, which easily causes network redundancy. Therefore, this embodiment improves the deep learning target detection algorithm and only retains the deep learning target detection algorithm that reduces the input image by 8 times, which can effectively improve the image processing efficiency and effectively avoid network redundancy.

[0065] To calculate the network loss function for each input image, and update the network parameters through back propagation, after multiple training iterations, a neural network model for respiratory tumor cell identification was obtained.

[0066] The network loss function is Among them, the coordinate loss function The calculation formula is: Among them, N loc is the number of elements in the feature map, is the GIoUloss function, is the manually marked rectangular box identifying the coordinate information, and Represents the coordinate information of the rectangular box output by the neural network model for identifying respiratory tumor cells. The rectangular box marks the position of the respiratory tumor cell. The network loss function provided by the present invention fully considers the inhibitory effect of negative samples on the false positive rate, supports the addition of negative sample training during training, and controls and adjusts the size of the inhibitory effect of negative samples on the false positive rate in model training through hyper-parameters, and can adaptively adjust the false positive rate inhibitory effect according to the negative sample image.

[0067] Classification loss function It is used to determine whether the position marked by the rectangular box is a positive respiratory tumor cell, so as to distinguish the false positive of respiratory tumor cells. The calculation formula is: Among them, ρ ij Represents the probability value of the respiratory tumor cells marked by the rectangular box being positive, with a value between 0 and 1.0; N a is the feature map width, N o is the feature map height.

[0068] Example 2

[0069] Compared with Example 1, the difference is that the respiratory tumor cell identification method based on artificial intelligence algorithm provided in this embodiment adds the operation of segmenting and splicing the fluorescent staining image of respiratory tumor cells. By segmenting the fluorescent image of respiratory tumor cells and then inputting it into the neural network model for respiratory tumor cell identification for computer processing, the hardware requirements can be effectively reduced and the processor cost can be saved. It mainly includes the following steps, such as Figure 5 As shown:

[0070] Step 101: Obtain a fluorescence staining image of respiratory tumor cells, and photograph the immunopurified respiratory tumor cell smear using a fluorescence dark field detection device to obtain a fluorescence staining image of respiratory tumor cells, such as Figure 2 As shown, then right.

[0071] Step 102: Slice the respiratory tract tumor cell fluorescence staining image with a resolution of 2736×1824, slice the respiratory tract tumor cell fluorescence staining image into 4 sub-images, number the sub-images, and record the coordinate information of a predetermined position (the specific position is not limited, and in this embodiment, it is the upper left corner) of each sub-image as the first coordinate information of the sub-image; Figure 6 As shown, the coordinates of sub-image 1 are (0, 0), the coordinates of sub-image 2 are (1368, 0), the coordinates of sub-image 3 are (0, 912), and the coordinates of sub-image 4 are (1368, 912).

[0072] Step 103: Artificial intelligence identifies a sub-image, and inputs the sub-image into a respiratory tumor cell identification neural network model to perform respiratory tumor cell identification processing on the sub-image to obtain a first sub-image of respiratory tumor cells with a rectangular frame identification; wherein the respiratory tumor cell identification neural network model is trained based on a deep learning target detection algorithm using a fluorescent staining image of manually annotated respiratory tumor cells as input data. wherein the rectangular frame identification includes rectangular frame coordinate information and positive probability data of respiratory tumor cells identified by the rectangular frame.

[0073] Step 104

[0074] The sub-images are stitched together, the coordinate information of the rectangular frame mark in the sub-image is combined with the first coordinate information, and the coordinate information of the respiratory tumor cells marked by the rectangular frame in the sub-image in the fluorescent staining image is calculated to stitch together the sub-images. The stitched image is as follows Figure 3 As shown, the rectangular box indicates the location of the respiratory tract tumor cells, and the data above the rectangular box indicates the probability value of the respiratory tract tumor cells indicated by the rectangular box being positive.

[0075] The above-mentioned respiratory tumor cell identification method based on artificial intelligence algorithm realizes the rapid identification of respiratory tumor cells on the fluorescent staining images of respiratory tumor cells through the respiratory tumor cell identification neural network model, which greatly improves the identification efficiency of respiratory tumor cells and improves the diagnostic efficiency of respiratory tumor cells; at the same time, because this method realizes respiratory tumor cell labeling based on the artificial intelligence network model, it is easy to realize automated operation and effectively reduce the human diagnosis cost of respiratory tumor cells.

[0076] In addition, the fluorescence staining images of respiratory tumor cells are generally large. By segmenting the fluorescence staining images of respiratory tumor cells and then inputting them into the respiratory tumor cell identification neural network model for processing, the hardware requirements can be effectively reduced and the processor cost can be saved.

[0077] Since the coordinates of the rectangular box in the fluorescent staining image of respiratory tumor cells are small and belong to a small target, the coordinate regression loss of the rectangular box is negligible after processing by the deep learning target detection algorithm, which easily causes network redundancy. Therefore, this embodiment improves the deep learning target detection algorithm and only retains the deep learning target detection algorithm that reduces the input image by 8 times, which can effectively improve the image processing efficiency and effectively avoid network redundancy.

[0078] To calculate the network loss function for each input image, and update the network parameters through back propagation, after multiple training iterations, a neural network model for respiratory tumor cell identification was obtained.

[0079] The network loss function is Among them, the coordinate loss function The calculation formula is: Among them, N loc is the number of elements in the feature map, is the GIoU loss function, is the manually marked rectangular box identifying the coordinate information, and Represents the coordinate information of the rectangular box output by the neural network model for identifying respiratory tumor cells. The rectangular box indicates that the position is a respiratory tumor cell. The network loss function provided by the present invention fully considers the inhibitory effect of negative samples on the false positive rate, supports adding negative sample training to the training, and controls and adjusts the magnitude of the inhibitory effect of negative samples on the false positive rate in model training through hyperparameters, and can adaptively adjust the false positive rate inhibitory effect according to the negative sample images.

[0080] Classification loss function It is used to determine whether the position marked by the rectangular box is a positive respiratory tumor cell, so as to distinguish the false positive of respiratory tumor cells. The calculation formula is: Among them, ρ ij Represents the probability value of the respiratory tumor cells marked by the rectangular box being positive, with a value between 0 and 1.0; N a is the feature map height, N o is the feature map width.

[0081] Example 3

[0082] This embodiment provides a respiratory tumor cell identification device based on an artificial intelligence algorithm, such as Figure 7 As shown, it includes: an acquisition module 201, a cutting module 202, a marking module 203 and a splicing module 204, wherein:

[0083] Acquisition module 201, acquiring a fluorescence staining picture of respiratory tumor cells, photographing the immunopurified respiratory tumor cells through a fluorescence dark field detection device, and obtaining a fluorescence staining picture of the respiratory tumor cells;

[0084] An image preprocessing module (not shown in the figure) performs one or more of image enhancement, image size standardization, and denoising operations on the fluorescent staining image of respiratory tumor cells.

[0085] A segmentation module 202 is configured to segment the respiratory tract tumor cell fluorescence staining image into four sub-images, number the sub-images, and record the coordinate information of the upper left corner of each sub-image as the first coordinate information of the sub-image;

[0086] Identification module 203, artificial intelligence identification sub-image, input the sub-image into the respiratory tumor cell identification neural network model to perform respiratory tumor cell identification processing on the sub-image to obtain a first sub-image of respiratory tumor cells with a rectangular frame identification; wherein the respiratory tumor cell identification neural network model is trained based on a deep learning target detection algorithm using a fluorescent staining image of manually annotated respiratory tumor cells as input data. wherein the rectangular frame identification includes rectangular frame coordinate information and positive probability data of respiratory tumor cells identified by the rectangular frame;

[0087] The stitching module 204 combines the coordinate information of the rectangular frame mark in the first sub-image with the first coordinate information, calculates the coordinate information of the respiratory tumor cells marked by the rectangular frame in the first sub-image in the fluorescent staining image, and stitches the first sub-image.

[0088] The above-mentioned device realizes rapid identification of respiratory tumor cells in fluorescent staining images of respiratory tumor cells through the identification module, greatly improving the identification efficiency of respiratory tumor cells in fluorescent staining images of respiratory tumor cells; improving the diagnostic efficiency of respiratory tumor cells; at the same time, because the method realizes respiratory tumor cell labeling based on an artificial intelligence network model, it is easy to realize automated operation and effectively reduce the human diagnosis cost of respiratory tumor cells; by adding a cutting module and a splicing module, the hardware requirements of the respiratory tumor cell identification device for the computer processor are effectively reduced, saving processor costs.

[0089] The present invention integrates equipment and algorithms and does not require networking. The algorithm in this invention detects and analyzes images collected from an entire well, and the returned result is a comprehensive test result of all images of the entire well (dozens to hundreds of images), directly returning the pathological analysis result of the sample, which is closer to the doctor's judgment result and can better and more intelligently assist the doctor in completing the pathological analysis.

[0090] Example 4

[0091] This embodiment provides a respiratory tumor cell identification device based on an artificial intelligence algorithm, which is different from Embodiment 3 in that Figure 8 As shown, an artificial intelligence interpretation module is added after the splicing module, including: an acquisition module 201, a cutting module 202, an identification module 203, a splicing module 204, and an artificial intelligence interpretation module, wherein:

[0092] Acquisition module 201, acquiring a fluorescence staining picture of respiratory tumor cells, photographing the immunopurified respiratory tumor cells through a fluorescence dark field detection device, and obtaining a fluorescence staining picture of the respiratory tumor cells;

[0093] An image preprocessing module (not shown in the figure) performs one or more of image enhancement, image size standardization, and denoising operations on the fluorescent staining image of respiratory tumor cells.

[0094] A segmentation module 202 is configured to segment the respiratory tract tumor cell fluorescence staining image into four sub-images, number the sub-images, and record the coordinate information of the upper left corner of each sub-image as the first coordinate information of the sub-image;

[0095] Identification module 203, artificial intelligence identification sub-image, input the sub-image into the respiratory tumor cell identification neural network model to perform respiratory tumor cell identification processing on the sub-image to obtain a first sub-image of respiratory tumor cells with a rectangular frame identification; wherein the respiratory tumor cell identification neural network model is trained based on a deep learning target detection algorithm using a fluorescent staining image of manually annotated respiratory tumor cells as input data. wherein the rectangular frame identification includes rectangular frame coordinate information and positive probability data of respiratory tumor cells identified by the rectangular frame;

[0096] The stitching module 204 combines the coordinate information of the rectangular frame mark in the first sub-image with the first coordinate information, calculates the coordinate information of the respiratory tumor cells marked by the rectangular frame in the first sub-image in the fluorescent staining image, and stitches the first sub-image.

[0097] Artificial intelligence interpretation module 205, the artificial intelligence interpretation module performs feature extraction, cell recognition, image analysis and data processing on the fluorescent staining images of respiratory tumor samples to realize the judgment of the fluorescent staining images of respiratory tumor samples. Among them, feature extraction: involves identifying key properties in the image such as cell type, morphology, size and color intensity, etc. These features are crucial for subsequent interpretation and recognition; cell recognition: the algorithm will distinguish and identify different cell and tissue types based on the extracted features, so as to accurately mark cancer cells and their development stages; image analysis: the AI ​​module uses a deep learning network to perform a deeper analysis and understanding of these features, which can evaluate the heterogeneity of cells, the degree of lesions and other important biomarker expressions; data processing: all analysis results are integrated and processed to generate reports that are easy for doctors to understand, providing clinical decision support on pathological status, lesion type and possible treatment direction. This series of processes ultimately realizes the interpretation of fluorescent staining images of respiratory tumor samples.

[0098] The above-mentioned device realizes rapid identification of respiratory tumor cells in fluorescent staining images of respiratory tumor cells through the identification module, greatly improving the identification efficiency of respiratory tumor cells in fluorescent staining images of respiratory tumor cells; improving the diagnostic efficiency of respiratory tumor cells; at the same time, because the method realizes respiratory tumor cell labeling based on an artificial intelligence network model, it is easy to realize automated operation and effectively reduce the human diagnosis cost of respiratory tumor cells; by adding a cutting module and a splicing module, the hardware requirements of the respiratory tumor cell identification device for the computer processor are effectively reduced, saving processor costs.

[0099] Example 5

[0100] Based on the same inventive concept, this embodiment provides an electronic device, see Fig. 9, including a memory 302, a processor 301, a communication interface 303, a communication bus 304, and a computer program stored in the memory and executable on the processor, wherein the processor 301, the communication interface 303, and the memory 302 communicate with each other via the communication bus 304; the communication interface 303 is used to realize information transmission between various modeling software and intelligent equipment module libraries and other related devices; the processor 301 is used to call the computer program in the memory 302, and the processor implements the methods provided by the above-mentioned various method embodiments when executing the computer program. For example, when the processor executes the computer program, it performs the following operations:

[0101] Obtain a fluorescent staining image of respiratory tumor cells, and photograph the immunopurified respiratory tumor cells using a fluorescent dark field detection device to obtain a fluorescent staining image of respiratory tumor cells;

[0102] Cutting the fluorescent staining image of respiratory tract tumor cells into 4 sub-images, numbering the sub-images, and recording coordinate information of a predetermined position of each sub-image as first coordinate information of the sub-image;

[0103] Artificial intelligence identifies a sub-image, and inputs the sub-image into a respiratory tumor cell identification neural network model to perform respiratory tumor cell identification processing on the sub-image to obtain a first sub-image of respiratory tumor cells with a rectangular frame identification; wherein the respiratory tumor cell identification neural network model is trained based on a deep learning target detection algorithm using a fluorescent staining image of manually annotated respiratory tumor cells as input data. wherein the rectangular frame identification includes rectangular frame coordinate information and positive probability data of respiratory tumor cells identified by the rectangular frame;

[0104] The sub-images are stitched together, the coordinate information of the rectangular frame mark in the first sub-image is combined with the first coordinate information, and the coordinate information of the respiratory tumor cells marked by the rectangular frame in the first sub-image in the fluorescent staining image is calculated to stitch together the first sub-image.

[0105] Example 6

[0106] This embodiment provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following operations are performed:

[0107] Obtain a fluorescent staining image of respiratory tumor cells, and photograph the immunopurified respiratory tumor cells using a fluorescent dark field detection device to obtain a fluorescent staining image of respiratory tumor cells;

[0108] Cutting the respiratory tract tumor cell fluorescence staining picture into 4 sub-pictures, numbering the sub-pictures, and recording the coordinate information of the upper left corner of each sub-picture as the first coordinate information of the sub-picture;

[0109] Artificial intelligence identifies a sub-image, and inputs the sub-image into a respiratory tumor cell identification neural network model to perform respiratory tumor cell identification processing on the sub-image to obtain a first sub-image of respiratory tumor cells with a rectangular frame identification; wherein the respiratory tumor cell identification neural network model is trained based on a deep learning target detection algorithm using a fluorescent staining image of manually annotated respiratory tumor cells as input data. wherein the rectangular frame identification includes rectangular frame coordinate information and positive probability data of respiratory tumor cells identified by the rectangular frame;

[0110] The sub-images are stitched together, the coordinate information of the rectangular frame mark in the first sub-image is combined with the first coordinate information, and the coordinate information of the respiratory tumor cells marked by the rectangular frame in the first sub-image in the fluorescent staining image is calculated to stitch together the first sub-image.

[0111] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0113] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0114] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0115] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A method for identifying respiratory tumor cells based on an artificial intelligence algorithm, characterized in that: The steps include: Obtain a fluorescent staining image of respiratory tumor cells, and photograph the immunopurified respiratory tumor cells using a fluorescent dark field detection device to obtain a fluorescent staining image of respiratory tumor cells; Cut the respiratory tract tumor cell fluorescence staining picture into N sub-pictures, and record the coordinate information of the predetermined position of each sub-picture as the first coordinate information of the sub-picture; An artificial intelligence identification sub-image is input into a respiratory tumor cell identification neural network model to perform respiratory tumor cell identification processing on the sub-image to obtain a first sub-image of respiratory tumor cells with a rectangular frame identification; wherein the respiratory tumor cell identification neural network model is trained based on a deep learning target detection algorithm using a fluorescent staining image of manually annotated respiratory tumor cells as input data, wherein the rectangular frame identification includes rectangular frame coordinate information and positive probability data of respiratory tumor cells identified by the rectangular frame, and the fluorescent staining images of respiratory tumor cells are images of different types of respiratory tumor cells and images of respiratory tumor cells at different stages of canceration; Splice the sub-images, combine the coordinate information of the rectangular box mark in the first sub-image with the first coordinate information, calculate the coordinate information of the respiratory tumor cells marked by the rectangular box in the first sub-image in the fluorescent staining image, realize the splicing of the first sub-image, and obtain the marked respiratory tumor cell fluorescent staining image.

2. The method for labeling respiratory tumor cells according to claim 1, characterized in that: Before cutting the respiratory tract tumor cell fluorescence staining picture, it also includes a respiratory tract tumor cell fluorescence staining image preprocessing step, which performs one or more of image enhancement, image size standardization, and denoising operations on the respiratory tract tumor cell fluorescence staining picture.

3. The method for labeling respiratory tract tumor cells according to claim 2, characterized in that: The deep learning target detection algorithm is the yolov5 algorithm that retains the image 8 times smaller.

4. The method for labeling respiratory tract tumor cells according to claim 3, characterized in that: The network loss function of the respiratory tumor cell identification neural network model is: Among them, the coordinate loss function The calculation formula is: Among them, N loc is the number of elements in the feature map, is the GIoU loss function, is the manually marked rectangular box identifying the coordinate information, and Represents the coordinate information of the rectangular box output by the neural network model for identifying respiratory tumor cells. The rectangular box indicates that the location is a respiratory tumor cell; Classification loss function It is used to determine whether the position marked by the rectangular box is a positive respiratory tumor cell, so as to distinguish the false positive of respiratory tumor cells. The calculation formula is: Among them, ρ ij Represents the probability value of the respiratory tumor cells marked by the rectangular box being positive, with a value between 0 and 1.0; N a is the feature map height, N o is the feature map width.

5. The method for labeling respiratory tract tumor cells according to claim 4, characterized in that: The N value is not less than 4, and preferably, the N value is equal to 4.

6. The method for labeling respiratory tract tumor cells according to claim 4, characterized in that: The predetermined position is the upper left corner of the sub-image.

7. A respiratory tumor cell identification device based on artificial intelligence algorithm, characterized in that: include: An acquisition module is used to acquire a fluorescence staining image of respiratory tumor cells, and to photograph the immunopurified respiratory tumor cells using a fluorescence dark field detection device to obtain a fluorescence staining image of the respiratory tumor cells; A cutting module, cutting the respiratory tract tumor cell fluorescence staining picture, cutting the respiratory tract tumor cell fluorescence staining picture into N sub-pictures, and recording the coordinate information of the predetermined position of each sub-picture as the first coordinate information of the sub-picture; An identification module, an artificial intelligence identification sub-image, inputting the sub-image into a respiratory tumor cell identification neural network model to perform respiratory tumor cell identification processing on the sub-image to obtain a first sub-image of respiratory tumor cells with a rectangular frame identification; wherein the respiratory tumor cell identification neural network model is trained based on a deep learning target detection algorithm using a fluorescent staining image of manually annotated respiratory tumor cells as input data, wherein the rectangular frame identification includes rectangular frame coordinate information and positive probability data of respiratory tumor cells identified by the rectangular frame, and the fluorescent staining images of respiratory tumor cells are images of different types of respiratory tumor cells; and images of respiratory tumor cells at different stages of carcinogenesis; The stitching module combines the coordinate information of the rectangular box mark in the first sub-image with the first coordinate information, calculates the coordinate information of the respiratory tumor cells marked by the rectangular box in the first sub-image in the fluorescent staining image, realizes the stitching of the first sub-image, and obtains the marked respiratory tumor cell fluorescent staining image.

8. The respiratory tract tumor cell identification device according to claim 7, characterized in that: It also includes an image preprocessing module, which performs one or more of image enhancement, image size standardization, and denoising operations on the fluorescent staining image of respiratory tumor cells.

9. The respiratory tumor cell identification device according to claim 8, characterized in that: It also includes an artificial intelligence interpretation module, which extracts features from the fluorescent staining images of respiratory tumor samples to interpret the fluorescent staining images of respiratory tumor samples.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method for identifying respiratory tumor cells according to any one of claims 1 to 6 is implemented.

Images