Digestive endoscopy early tumor risk examination and screening system

Abstract

The invention discloses a digestive endoscopy early tumor risk examination and screening system. The system comprises an image acquisition and labeling system, an image digital processing system, an Inception system, an optimized GoogleNet system, a user operation system, a manual screening system and a screening result output system. The image acquisition and labeling system collects digestive endoscope examination images, and a training set and a test set are obtained after labeling; the image digital processing system carries out digital processing on the training set and the test set; the data processing difficulty is reduced by the Inception system; a GoogleNet system is optimized to analyze the data, and the tumor risk is judged; the user operation system has a picture uploading function, an AI identification function and a manual analysis function; the manual screening system realizes a manual analysis function; and the screening result output system outputs a diagnosis reference result, so that early tumor risk examination and screening of the digestive endoscopy are realized. The tumor risk early screening method is easy to operate, high in accuracy, high in practicability and suitable for digestive endoscopy early tumor risk detection and screening.

Description

Technical Field

[0001] This invention relates to the field of medical image recognition technology, and in particular to a digestive endoscopy early tumor risk screening system. Background Technology

[0003] Early gastrointestinal cancers refer to tumors of the digestive tract that invade no more than the submucosa or are confined to the mucosal layer, including early esophageal cancer, early gastric cancer, and early colorectal cancer. Cancer is a new growth formed by the abnormal proliferation and differentiation of cells in local tissues under the influence of various carcinogenic factors. Most gastrointestinal tumors undergo a process of evolution from precancerous lesions to early cancer and then to invasive cancer, which lays the foundation for the endoscopic diagnosis and early treatment of early gastrointestinal cancers. The emergence of new endoscopic diagnostic and treatment technologies has not only significantly improved the detection rate of early gastrointestinal cancers but also provided a basis for our endoscopic treatment of early gastrointestinal cancers.

[0004] Experts recommend that individuals aged 40-79 with any of the following risk factors (precancerous conditions such as chronic atrophic gastritis, gastrointestinal polyps, pernicious anemia; precancerous lesions such as moderate intestinal metaplasia or intraepithelial neoplasia of the gastric mucosa; family history of gastrointestinal cancer (including parents, siblings, etc.); symptoms such as abdominal distension, abdominal pain, nausea, vomiting, difficulty swallowing, melena, hematochezia, heartburn; Helicobacter pylori infection; high-salt diet (average salt intake greater than 20g / day, approximately 3 beer bottle caps); preference for smoked, fried, grilled, or grilled foods (average 3 meals / week); smoking (average >200 cigarettes / year); heavy alcohol consumption (average alcohol intake equivalent to 50g / day, alcohol intake × alcohol concentration × 0.8 = alcohol intake)) are considered high-risk and should undergo gastroscopy or colonoscopy and other related examinations as early as possible to achieve early diagnosis and treatment of gastrointestinal tumors.

[0005] For early-stage gastrointestinal cancers and precancerous lesions, endoscopic minimally invasive treatment can generally achieve a cure. For example, adenomatous polyps in the gastrointestinal tract can be removed using endoscopic mucosal resection (EMR). For early-stage cancers of the digestive tract, such as early-stage esophageal cancer, early-stage gastric cancer, and early-stage colorectal cancer, endoscopic endoscopy (ESD) can be used to remove lesions. Therefore, early endoscopic screening is crucial for improving the survival rate of gastrointestinal malignancies.

[0006] Currently, early screening programs for digestive endoscopy are underway, and the rate of self-administered digestive endoscopy is gradually increasing. Many people are consciously starting to undergo digestive endoscopy examinations regularly. Generally, digestive endoscopy examinations can be completed in tertiary hospitals at the city level. However, due to the inability to fully guarantee the skill level of the doctors performing the examinations, many cases of misdiagnosis and missed diagnosis occur, seriously endangering patients' lives. After the on-site examinations, a large number of medical images are left behind, but these images are often stored in hospital databases, failing to realize their value.

[0007] With the rapid development of biotechnology and computer technology, precision medicine has gradually become a hot topic in the medical field. This has led to the rise of medical imaging AI technology. Medical imaging AI technology utilizes artificial intelligence to analyze and interpret medical images to improve the accuracy of medical diagnosis, treatment, and prediction. Its applications are wide-ranging, including various medical imaging modalities such as CT, MRI, and PET. The advantages of medical imaging AI technology include improving the accuracy and efficiency of doctors' diagnoses, shortening patient waiting times, and reducing doctors' workload. Simultaneously, it can mine vast amounts of medical data, promoting the rapid development of medical science. In medical treatment, medical imaging AI technology can assist doctors in developing reasonable treatment plans and monitoring the treatment process and its effects. For example, in cancer treatment, medical imaging AI technology can analyze various medical images to provide doctors with information such as tumor size, location, differentiation degree, and depth of invasion, enabling the development of appropriate surgical resection and radiotherapy plans.

[0008] Chinese invention patent CN118247275A describes a remote processing method for digestive endoscopy based on data analysis. The method involves: acquiring grayscale images of each endoscope; acquiring feature pixels of each grayscale image; calculating the esophageal texture redundancy proximity index of the feature pixels corresponding to each esophageal texture density value in the dense esophageal texture sequence of each grayscale image; calculating the esophageal image redundancy adjacent similarity of each grayscale image; calculating the invalid redundant feature values ​​of each grayscale image; calculating the redundant esophageal image removal index of each grayscale image; and obtaining the remote processing result of digestive endoscopy based on the redundant esophageal image removal index of each grayscale image. This invention is based on image grayscale processing, focuses on remote processing, has low accuracy, and is unusable by ordinary patients.

[0009] Chinese invention patent CN114998260A discloses a vascular enhancement mode for digestive endoscopy images. This mode involves image segmentation, statistical analysis of the color gradient (BGR) and luminal histogram (L-histogram) of each segment, color correction of the segmented BGR and L-histograms, cropping and smoothing of the histograms, and generation of a histogram mapping table for each segment. This vascular enhancement mode analyzes and statistically processes the attenuation characteristics of the image's color channel components, weighting them with the global histogram to adjust the color contrast between blood vessels and surrounding tissues, thereby enhancing image contrast. While this invention provides some reference value for analyzing blood vessels in digestive endoscopy images through image enhancement, it only plays a supplementary role in early screening, and patients, in particular, cannot extract useful information from this analytical method. Summary of the Invention

[0010] This invention provides a digestive endoscopy early tumor risk screening system to address the problems of low accuracy, high complexity, and difficulty in use by existing digestive endoscopy early tumor screening systems.

[0011] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0012] This invention provides a system for screening the early risk of tumors in digestive endoscopy, comprising:

[0013] Image acquisition and annotation system, image digitization processing system, Inception system, optimized GoogleNet system, user operating system, manual screening system, and screening result output system;

[0014] The image acquisition and annotation system collects digestive endoscopy images, annotates the digestive endoscopy images, and obtains training and test sets.

[0015] The image digitization processing system digitizes the training set and the test set to obtain model input data;

[0016] The Inception system is used to reduce the difficulty of processing model input data. P Represented as upper-level data, X F This indicates the data in the lower layer of the channel merging layer;

[0017] The optimized GoogleNet system analyzes the model input data to determine tumor risk;

[0018] The user operating system includes image upload functionality, AI recognition functionality, and manual analysis functionality;

[0019] The manual screening system implements the manual analysis function;

[0020] The screening result output system outputs diagnostic reference results, enabling early tumor risk screening via digestive endoscopy.

[0021] Furthermore, the image acquisition and annotation system, image digitization processing system, Inception system, optimized GoogleNet system, user operating system, manual screening system, and screening result output system include:

[0022] The digestive endoscopy early tumor risk assessment and screening system is built on the basis of medical HIS system, server, PC and mobile phone;

[0023] The image acquisition and annotation system, image digitization processing system, Inception system, and optimized GoogleNet system are deployed on the server;

[0024] The user operating system, manual screening system, and screening result output system are deployed on PCs and mobile phones;

[0025] The server can interact with PCs and mobile phones via the Internet.

[0026] Furthermore, the image acquisition and annotation system collects digestive endoscopy images, annotates the digestive endoscopy images, and obtains a training set and a test set, including:

[0027] Based on existing digestive endoscopy results, collect the images and actual test results from the examination results. The existing digestive endoscopy results should include normal digestive tract examination images of different age groups and different locations, as well as examination images of tumors that have been diagnosed.

[0028] Based on the images in the examination results and the actual test results, professional doctors will annotate them to obtain an annotated dataset.

[0029] The resolution of the labeled dataset images was adjusted to 224×224×3 using linear interpolation, and the labeled dataset images were divided into training and testing sets.

[0030] Furthermore, the image digitization processing system digitizes the training set and test set to obtain model input data, including:

[0031] The input 224×224×3 image is convolved with a kernel size of 7×7, a stride of 2, padding of 3, and 64 output channels. The output size is (224-7+3×2) / 2+1=112.5(rounded down)=112, and the output is 112×112×64. ReLU operation is performed after convolution.

[0032] Max pooling is performed with a window size of 3×3 and a stride of 2. The output size is ((112 -3) / 2)+1=55.5(rounded up)=56, and the output is 56×56×64.

[0033] Dimensionality reduction is performed by using 64 1×1 convolutional kernels (dimensionality reduction before 3×3 convolutional kernels) to transform the input feature map (56×56×64) into 56×56×64, and then performing ReLU operation.

[0034] Perform convolution with a kernel size of 3×3, a stride of 1, and 192 output channels. The output size is (56-3+1×2) / 1+1=56, and the output is 56×56×192. Then perform ReLU operation.

[0035] Max pooling is performed with a window size of 3×3 and a stride of 2. The number of output channels is 192, and the output is ((56-3) / 2)+1=27.5(rounded up)=28. The output feature map dimension is 28×28×192, which yields the model input data.

[0036] Furthermore, the Inception system is used to reduce the difficulty of processing model input data, X P Represented as upper-level data, X F This indicates the data below the channel merging layer, including:

[0037] For upper-level data X P Dimensionality reduction is performed, with a convolution kernel size of 1×1, to obtain X. P1 ;

[0038] For upper-level data X P1 Perform convolution with a 3×3 kernel to obtain X. P2 ;

[0039] For upper-level data X P1 Perform convolution with a kernel size of 5×5 to obtain X. P3 ;

[0040] For upper-level data X P Max pooling is performed with a 3×3 window and a stride of 2, followed by dimensionality reduction with a 1×1 kernel to obtain X. P4 ;

[0041] For X P1 X P2 X P3 X P4 Link them by depth to get the lower layer data X of the channel merging layer. F .

[0042] Furthermore, the optimized GoogleNet system analyzes the model input data to determine tumor risk, including:

[0043] The input data of the model is max-pooled with a window size of 3×3 and a step size of 2.

[0044] Perform the Inception system process five times;

[0045] Perform max pooling with a window size of 3×3 and a step size of 2;

[0046] Perform two Inception system processes;

[0047] Perform average pooling with a 7x7 window and a stride of 1 to obtain data X. B ;

[0048] Regarding the obtained data X B Discard randomly with a fixed value of P;

[0049] Softmax regression was performed to assess tumor risk.

[0050] Furthermore, the optimized GoogleNet system analyzes the model input data to determine tumor risk, including:

[0051] The model described above is trained using the training set and tested using the test set to select the optimal network parameters.

[0052] The X B Given a linked list of data with a window size of 1×1 and a step size of 1024, let X in the test set... B satisfy:

[0053]

[0054] The risk level for gastrointestinal tumor detection is categorized as low risk, medium risk, or high risk, and the output result is counted as O. i (i = 1, 2, 3);

[0055] Adjust the hyperparameter P, 0≦P≦1, for the X B By performing fully connected layers through hidden layers with hyperparameter P, the test set is optimized to form the optimal network parameters.

[0056] Furthermore, the user operating system includes image upload functionality, AI recognition functionality, and manual analysis functionality, including:

[0057] The image upload function allows users to upload images to the server via a webpage or mini-program.

[0058] The AI ​​recognition function is an AI screening result obtained after the user analyzes the uploaded image using the optimized GoogleNet system;

[0059] The manual analysis results are obtained by users submitting a manual analysis request on a PC or mobile phone, and then a professional medical institution conducts a manual analysis of the uploaded images, along with the AI ​​screening results.

[0060] Furthermore, the manual screening system implements the manual analysis function, including:

[0061] Doctors can access images uploaded to the server online via PC or mobile phone and use their professional medical knowledge to assess the risk of artificial tumors.

[0062] The results of the artificial tumor risk assessment are uploaded to the server.

[0063] Furthermore, the screening result output system outputs diagnostic reference results to achieve early tumor risk screening in digestive endoscopy, including:

[0064] After a user uploads an image through their operating system, if the user only uses the AI ​​recognition function, the screening result output system will output the AI ​​screening result.

[0065] After users upload images through their operating system, when users use the manual analysis function, the screening result output system outputs AI screening results and artificial tumor risk results, realizing early tumor risk detection and screening in digestive endoscopy.

[0066] The beneficial effects of the technical solution provided by this invention include at least the following:

[0067] This invention comprises an image acquisition and annotation system, an image digitization processing system, an Inception system, an optimized GoogleNet system, a user operating system, a manual screening system, and a screening result output system. The image acquisition and annotation system collects images from digestive endoscopy examinations, annotating them to obtain training and test sets. The image digitization processing system digitizes the training and test sets. The Inception system reduces data processing complexity. The optimized GoogleNet system analyzes the data to determine tumor risk. The user operating system includes image upload, AI recognition, and manual analysis functions. The manual screening system implements manual analysis. The screening result output system outputs diagnostic reference results, enabling early tumor risk screening via digestive endoscopy. This invention offers simple operation, high accuracy, and strong practicality for early tumor risk screening via digestive endoscopy. It considers the most readily available examination image information from digestive endoscopy results, ensures standardized data processing, optimizes the GoogLeNet model to effectively prevent overfitting and underfitting, and optimizes the model by changing hyperparameters during training. The resulting model can be lightweightly deployed on computing devices with sufficient GPU performance and can make relatively accurate judgments based on newly added images. Therefore, the present invention can accurately determine the risk of early gastrointestinal tumors. Thus, the method of the present invention can be used in Internet data services such as big data and medical HIS system databases. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1A flowchart of the method for early tumor risk screening system in digestive endoscopy provided in this embodiment of the invention;

[0070] Figure 2 A schematic diagram of an image digitization processing system provided in an embodiment of the present invention;

[0071] Figure 3 A diagram illustrating the ReLU operation on the data;

[0072] Figure 4 A flowchart of the Inception system provided in this embodiment of the invention;

[0073] Figure 5 A flowchart of the method of the GoogleNet system provided in the embodiments of the present invention.

[0074] Figure 6 This is a schematic diagram of the training process of the early tumor risk screening system for digestive endoscopy provided in an embodiment of the present invention. Detailed Implementation

[0075] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0076] Example

[0077] This embodiment provides a system for screening the early tumor risk in digestive endoscopy, including:

[0078] Image acquisition and annotation system, image digitization processing system, Inception system, optimized GoogleNet system, user operating system, manual screening system, and screening result output system.

[0079] Please refer to Figure 1 The flowchart shown is a method for screening early tumor risk in digestive endoscopy provided in an embodiment of the present invention.

[0080] The image acquisition and annotation system collects digestive endoscopy images, annotates the digestive endoscopy images, and obtains training and test sets.

[0081] The image digitization processing system digitizes the training set and the test set to obtain model input data;

[0082] The Inception system is used to reduce the difficulty of processing model input data. P Represented as upper-level data, X F This indicates the data in the lower layer of the channel merging layer;

[0083] The optimized GoogleNet system analyzes the model input data to determine tumor risk;

[0084] The user operating system includes image upload functionality, AI recognition functionality, and manual analysis functionality;

[0085] The manual screening system implements the manual analysis function;

[0086] The screening result output system outputs diagnostic reference results, enabling early tumor risk screening via digestive endoscopy.

[0087] The system includes an image acquisition and annotation system, an image digitization processing system, an Inception system, an optimized GoogleNet system, a user operating system, a manual screening system, and a screening result output system.

[0088] The aforementioned early tumor risk screening system for digestive endoscopy is built upon a medical HIS system, server, PC, and mobile phone. The Hospital Information System (HIS) is the primary function used in this embodiment. This includes: inputting patient examination information, issuing electronic prescriptions, writing patient medical records, editing outpatient surgical treatment application forms, and various medical technology examination application forms; and storing and retrieving images and written descriptions of patients' digestive endoscopy results.

[0089] The image acquisition and annotation system, image digitization processing system, Inception system, and optimized GoogleNet system are deployed on a server. Because AI analysis of a large number of images requires extensive parallel computing, a server with a high-performance GPU is needed. In this embodiment, cloud server resources rented by the hospital from a network operator are used.

[0090] The user operating system, manual screening system, and screening result output system are deployed on PCs and mobile phones. In this embodiment, users can operate them through an app and a WeChat mini-program.

[0091] The server can interact with PCs and mobile phones via the Internet.

[0092] The image acquisition and annotation system collects images from digestive endoscopy examinations, annotates the images, and obtains training and testing sets.

[0093] Based on existing digestive endoscopy results, collect the images and actual test results from the examination results. The existing digestive endoscopy results should include normal digestive tract examination images of different age groups and different locations, as well as examination images of tumors that have been diagnosed.

[0094] Based on the images and actual test results from the examination, a professional physician annotates the data to obtain an annotated dataset. In this embodiment, since the digestive endoscopy results and diagnostic cases from the hospital's HIS system are used, the actual annotation work has already been completed through text; only a professional physician needs to complete the annotation based on the existing images, diagnostic results, and subsequent biopsy results.

[0095] The existing gastroscopy images already have high resolution. In order to reduce the image processing difficulty of subsequent models, it is necessary to standardize images of different specifications.

[0096] Linear interpolation refers to an interpolation method where the interpolation function is a first-order polynomial, and the interpolation error at the interpolation nodes is zero. Geometrically, linear interpolation means that the original function is approximated using a straight line passing through points A and B in the overview diagram. Linear interpolation can be used to approximate the original function and can also be used to calculate values ​​not found in a lookup table.

[0097] For example:

[0098]

[0099] Where X1 and Y1 are the x and y values ​​of known data points, X2 and Y2 are the x and y values ​​of another known data point, X is the x value to be interpolated, and Y is the calculated difference result.

[0100] In this embodiment, 224×224 is the resolution, and a 3D vector is also needed to represent the color.

[0101] The resolution of the labeled dataset images was adjusted to 224×224×3 using linear interpolation, and the labeled dataset images were divided into training and test sets with a ratio of 7:3.

[0102] The image digitization processing system digitizes the training set and the test set to obtain model input data.

[0103] The input 224×224×3 image is convolved with a kernel size of 7×7, a stride of 2, padding of 3, and 64 output channels. The output size is (224-7+3×2) / 2+1=112.5(rounded down)=112, and the output is 112×112×64. ReLU operation is performed after convolution.

[0104] Max pooling is performed with a window size of 3×3 and a stride of 2. The output size is ((112 -3) / 2)+1=55.5(rounded up)=56, and the output is 56×56×64.

[0105] Dimensionality reduction is performed by using 64 1×1 convolutional kernels (dimensionality reduction before 3×3 convolutional kernels) to transform the input feature map (56×56×64) into 56×56×64, and then performing ReLU operation.

[0106] Perform convolution with a kernel size of 3×3, a stride of 1, and 192 output channels. The output size is (56-3+1×2) / 1+1=56, and the output is 56×56×192. Then perform ReLU operation.

[0107] Max pooling is performed with a window size of 3×3 and a stride of 2. The number of output channels is 192, and the output is ((56-3) / 2)+1=27.5(rounded up)=28. The output feature map dimension is 28×28×192, which yields the model input data.

[0108] This step digitizes the image data into a digital structure suitable for model processing.

[0109] The Inception system is used to reduce the difficulty of processing model input data. P Represented as upper-level data, X F This indicates the data in the lower layer of the channel merging layer.

[0110] The Inception architecture is a highly efficient neural network architecture designed to address overfitting, computational complexity, and vanishing gradient problems in deep learning. This architecture reduces the number of parameters and improves computational efficiency by integrating multi-scale convolutions and parallel computation, making it particularly suitable for resource-constrained devices. The design philosophy of the Inception architecture is to combine multiple convolutional or pooling operations into a network module, assembling the entire network structure on a module-by-module basis to form a sparse network structure. This increases the expressiveness of the neural network while ensuring efficient use of computational resources. The Inception architecture in this embodiment uses the Inception V1 model, stacking 1×1, 3×3, and 5×5 convolutions and 3×3 pooling together. This increases both the network width and its adaptability to scale.

[0111] For upper-level data X P Dimensionality reduction is performed, with a convolution kernel size of 1×1, to obtain X. P1 ;

[0112] For upper-level data X P1 Perform convolution with a 3×3 kernel to obtain X. P2 ;

[0113] For upper-level data X P1 Perform convolution with a kernel size of 5×5 to obtain X. P3;

[0114] For upper-level data X P Max pooling is performed with a 3×3 window and a stride of 2, followed by dimensionality reduction with a 1×1 kernel to obtain X. P4 ;

[0115] For X P1 X P2 X P3 X P4 Link them by depth to get the lower layer data X of the channel merging layer. F .

[0116] The optimized GoogleNet system analyzes the input data of the model to determine the risk of tumors.

[0117] The optimized GoogleNet system primarily uses the Inception architecture, which significantly reduces the number of deep learning layers compared to the traditional VGG architecture. This reduces gradient vanishing and gradient exploding, allowing the optimized GoogleNet system to utilize computational resources more efficiently and extract more features with the same computational cost, thereby improving training results.

[0118] Compared to the traditional GoogleNet system, this embodiment optimizes the GoogleNet system by pruning unnecessary auxiliary classifiers. The original two auxiliary classifiers were deleted, and only the core part was retained, which greatly reduces the amount of computation and improves the analysis efficiency.

[0119] The input data of the model is max-pooled with a window size of 3×3 and a step size of 2.

[0120] Perform the Inception system process five times;

[0121] Perform max pooling with a window size of 3×3 and a step size of 2;

[0122] Perform two Inception system processes;

[0123] Perform average pooling with a 7x7 window and a stride of 1 to obtain data X. B ;

[0124] Regarding the obtained data X B Discard randomly with a fixed value of P;

[0125] Softmax regression was performed to assess tumor risk.

[0126] The optimized GoogleNet system analyzes the input data of the model to determine the risk of tumors.

[0127] The model described above is trained using the training set and tested using the test set to select the optimal network parameters.

[0128] The X B Given a linked list of data with a window size of 1×1 and a step size of 1024, let X in the test set... B satisfy:

[0129]

[0130] The risk level for gastrointestinal tumor detection is categorized as low risk, medium risk, or high risk, and the output result is counted as O. i (i = 1, 2, 3);

[0131] Adjust the hyperparameter P, 0≦P≦1, for the X B A fully connected layer with hyperparameter P is used to optimize the test set and form the optimal network parameters. In this embodiment, the hyperparameter P is set to 0.57.

[0132] The user operating system includes image upload functionality, AI recognition functionality, and manual analysis functionality.

[0133] The image upload function allows users to upload images to the server via a webpage or mini-program.

[0134] The AI ​​recognition function is an AI screening result obtained after the user analyzes the uploaded image using the optimized GoogleNet system;

[0135] The manual analysis results are obtained by users submitting a manual analysis request on a PC or mobile phone, and then a professional medical institution conducts a manual analysis of the uploaded images, along with the AI ​​screening results.

[0136] The manual screening system implements the manual analysis function.

[0137] Doctors can access uploaded images from the server online via PC or mobile phone and use their professional medical knowledge to assess the risk of artificial tumors. This is a paid feature, primarily used when AI screening determines a risk of malignancy, or when a user has doubts about the AI ​​screening results.

[0138] The results of the artificial tumor risk assessment are uploaded to the server.

[0139] The screening result output system outputs diagnostic reference results to achieve early tumor risk screening in digestive endoscopy.

[0140] After a user uploads an image through their operating system, if the user only uses the AI ​​recognition function, the screening result output system will output the AI ​​screening result.

[0141] After users upload images through their operating system, when users use the manual analysis function, the screening result output system outputs AI screening results and artificial tumor risk results, realizing early tumor risk detection and screening in digestive endoscopy.

[0142] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0143] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0144] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0145] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0146] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A system for screening the early risk of tumors in digestive endoscopy, characterized in that, include: Image acquisition and annotation system, image digitization processing system, Inception system, optimized GoogleNet system, user operating system, manual screening system, screening result output system; The image acquisition and annotation system collects digestive endoscopy images, annotates the digestive endoscopy images, and obtains training and test sets. The image digitization processing system digitizes the training set and the test set to obtain model input data; The Inception system is used to reduce the difficulty of processing model input data. P Represented as upper-level data, X F This indicates the data in the lower layer of the channel merging layer; The optimized GoogleNet system analyzes the model input data to determine tumor risk; The user operating system includes image upload functionality, AI recognition functionality, and manual analysis functionality; The manual screening system implements the manual analysis function; The screening result output system outputs diagnostic reference results, enabling early tumor risk screening via digestive endoscopy.

2. The early tumor risk screening system for digestive endoscopy as described in claim 1, characterized in that, The image acquisition and annotation system, image digitization processing system, Inception system, optimized GoogleNet system, user operating system, manual screening system, and screening result output system include: The digestive endoscopy early tumor risk assessment and screening system is built on the basis of medical HIS system, server, PC and mobile phone; The image acquisition and annotation system, image digitization processing system, Inception system, and optimized GoogleNet system are deployed on the server; The user operating system, manual screening system, and screening result output system are deployed on PCs and mobile phones; The server can interact with PCs and mobile phones via the Internet.

3. The early tumor risk screening system for digestive endoscopy as described in claim 1, characterized in that, The image acquisition and annotation system collects images from digestive endoscopy examinations, annotates these images, and obtains training and testing sets, including: Based on existing digestive endoscopy results, collect the images and actual test results from the examination results. The existing digestive endoscopy results should include normal digestive tract examination images of different age groups and different locations, as well as examination images of tumors that have been diagnosed. Based on the images in the examination results and the actual test results, professional doctors will annotate them to obtain an annotated dataset. The resolution of the labeled dataset images was adjusted to 224×224×3 using linear interpolation, and the labeled dataset images were divided into training and testing sets.

4. The early tumor risk screening system for digestive endoscopy as described in claim 1, characterized in that, The image digitization processing system digitizes the training set and test set to obtain model input data, including: The input 224×224×3 image is convolved with a kernel size of 7×7, a stride of 2, padding of 3, and 64 output channels. The output size is (224-7+3×2) / 2+1=112.5(rounded down)=112, and the output is 112×112×64. ReLU operation is performed after convolution. Max pooling is performed with a window size of 3×3 and a stride of 2. The output size is ((112-3) / 2)+1=55.5(rounded up)=56, and the output is 56×56×64. Dimensionality reduction is performed by using 64 1×1 convolutional kernels (dimensionality reduction before 3×3 convolutional kernels) to transform the input feature map (56×56×64) into 56×56×64, and then performing ReLU operation. Perform convolution with a kernel size of 3×3, a stride of 1, and 192 output channels. The output size is (56-3+1×2) / 1+1=56, and the output is 56×56×192. Then perform ReLU operation. Max pooling is performed with a window size of 3×3 and a stride of 2. The number of output channels is 192, and the output is ((56-3) / 2)+1=27.5(rounded up)=28. The output feature map dimension is 28×28×192, which yields the model input data.

5. The early tumor risk screening system for digestive endoscopy as described in claim 1, characterized in that, The Inception system is used to reduce the difficulty of processing model input data. P Represented as upper-level data, X F This indicates the data below the channel merging layer, including: For upper-level data X P Dimensionality reduction is performed, with a convolution kernel size of 1×1, to obtain X. P1 ; For upper-level data X P1 Perform convolution with a 3×3 kernel to obtain X. P2 ; For upper-level data X P1 Perform convolution with a kernel size of 5×5 to obtain X. P3 ; For upper-level data X P Max pooling is performed with a 3×3 window and a stride of 2, followed by dimensionality reduction with a 1×1 kernel to obtain X. P4 ; For X P1 X P2 X P3 X P4 Link them by depth to get the lower layer data X of the channel merging layer. F .

6. The early tumor risk screening system for digestive endoscopy as described in claim 1, characterized in that, The optimized GoogleNet system analyzes the model input data to determine tumor risk, including: The input data of the model is max-pooled with a window size of 3×3 and a step size of 2. Perform the Inception system process five times; Perform max pooling with a window size of 3×3 and a step size of 2; Perform two Inception system processes; Perform average pooling with a 7x7 window and a stride of 1 to obtain data X. B ; Regarding the obtained data X B Discard randomly with a fixed value of P; Softmax regression was performed to assess tumor risk.

7. The early tumor risk screening system for digestive endoscopy as described in claim 1, characterized in that, The optimized GoogleNet system analyzes the model input data to determine tumor risk, including: The model described above is trained using the training set and tested using the test set to select the optimal network parameters. The X B Given a linked list of data with a window size of 1×1 and a step size of 1024, let X in the test set... B satisfy: The risk level for gastrointestinal tumor detection is categorized as low risk, medium risk, or high risk, and the output result is counted as O. i (i = 1, 2, 3); Adjust the hyperparameter P, 0≦P≦1, for the X B By performing fully connected layers through hidden layers with hyperparameter P, the test set is optimized to form the optimal network parameters.

8. The early tumor risk screening system for digestive endoscopy as described in claim 1, characterized in that, The user operating system includes image upload functionality, AI recognition functionality, and manual analysis functionality, including: The image upload function allows users to upload images to the server via a webpage or mini-program. The AI ​​recognition function is an AI screening result obtained by the user after analyzing the uploaded image using the optimized GoogleNet system; The manual analysis results are obtained by users submitting a manual analysis request on a PC or mobile phone, and then a professional medical institution conducts a manual analysis of the uploaded images, along with the AI ​​screening results.

9. The early tumor risk screening system for digestive endoscopy as described in claim 1, characterized in that, The manual screening system implements the manual analysis function, including: Doctors can access images uploaded to the server online via PC or mobile phone and use their professional medical knowledge to assess the risk of artificial tumors. The results of the artificial tumor risk assessment are uploaded to the server.

10. The early tumor risk screening system for digestive endoscopy as described in claim 1, characterized in that, The screening result output system outputs diagnostic reference results to achieve early tumor risk screening in digestive endoscopy, including: After a user uploads an image through their operating system, if the user only uses the AI ​​recognition function, the screening result output system will output the AI ​​screening result. After users upload images through their operating system, when users use the manual analysis function, the screening result output system outputs AI screening results and artificial tumor risk results, realizing early tumor risk detection and screening in digestive endoscopy.

Images