CN109003659A - Stomach Helicobacter pylori infects pathological diagnosis and supports system and method - Google Patents

Abstract

The invention discloses a pathological diagnosis support system and method for gastric Helicobacter pylori infection based on big data deep learning. The system includes: an image data acquisition unit, which is used to obtain images of normal gastric slices and diagnosed cases of gastric Helicobacter pylori infection The pathological slice image is used as the input image data; the image data labeling unit is used to label the input image data; the image database construction unit is used to classify and organize the labeled image data provided by the image data labeling unit, and construct the pathological image database; a convolutional neural network (CNN) construction unit, used to construct a first convolutional neural network model; and a convolutional neural network model training unit, to obtain an ideal convolutional neural network model. Through the support system and method for the pathological diagnosis of gastric Helicobacter pylori infection of the present invention, accurate and efficient intelligent film reading can be realized to assist the clinical pathological diagnosis of gastric Helicobacter pylori infection, and improve its accuracy, work efficiency and work continuity .

Description

Pathological diagnosis support system and method for gastric Helicobacter pylori infection

technical field

The invention relates to a support system and method for pathological diagnosis of gastric Helicobacter pylori infection based on big data deep learning.

Background technique

Deep learning is currently the most suitable and widely used algorithm for image recognition and speech analysis in the field of artificial intelligence. Its inspiration comes from the working mechanism of the human brain. It is to automatically extract features from externally input data by establishing a convolutional neural network. , so that the machine can understand the learning data, obtain information and output. At present, artificial intelligence based on deep learning has been applied in various industries, including speech recognition, face recognition, car logo recognition, handwritten Chinese character recognition, etc. In recent years, the product research and development of artificial intelligence-assisted medical technology has also made significant progress. For example, the artificial intelligence product developed by Google Brain and Verily for the pathological diagnosis of breast cancer can achieve an accuracy rate of 89% of tumor location; the First Affiliated Hospital of Zhejiang University Using artificial intelligence to quickly analyze the location and benign and malignant of nodules in thyroid B-ultrasound.

In the process of medical diagnosis, pathological tissue biopsy requires extremely high standardization and precision. At present, the vast majority of pathological tissue sections are made manually, and are analyzed and judged by pathologists based on their long-term accumulated clinical diagnosis experience. Modern medicine has proved that gastric Helicobacter pylori (HP) is an important pathogenic factor of chronic gastritis, peptic ulcer and gastric cancer, and has been listed as one of the class I carcinogens by the International Agency for Research on Cancer (IARC). Therefore, early detection of HP infection has important clinical significance for timely eradication of HP and treatment of HP-related diseases. Clinically, gastric mucosal tissue is obtained by gastroscope forceps, paraffin-embedded, sectioned, and stained for histological microscopic examination to detect HP. For experienced pathologists, it is the "gold standard" for diagnosing HP infection. Compared with other HP detection methods, the direct detection of HP by pathological histology has the following advantages: ① When gastroscopy is used, the gross lesions in the stomach, such as ulcers and gastric cancer, can be identified; degree and type; ③Reexamination of patients who received HP eradication therapy can clarify the outcome of gastroduodenal lesions; ④Clinical application can diagnose 98% of HP infections by taking a large enough specimen from the gastric antrum. Direct evidence for the existence of HP. The pathological section can adopt the modified Warthern-Starry silver staining method. Under the 40X microscope, brown or brown curved or short rod-shaped objects can be seen on a light yellow background, with a length of 5 μm. Combined with the characteristics that HP is mostly located in gastric pits or intrinsic glands , easy to identify, high positive rate. Although with the development of modern medical technology, rapid urease test (RUT) and non-invasive detection methods such as serological detection, urea breath test and other detection methods are more widely used, but the above methods can only indirectly reflect the situation of HP infection , and each has limitations. For example, the RUT test results are easily affected by multiple factors such as the patient's medication history (such as recent use of antibiotics, bismuth or proton pump inhibitors can lead to false negative results), the location of the sample under the gastroscope, and the experimental conditions. There must be more than 10 ⁴ bacteria in the test to show positive.

Therefore, the histopathological detection of HP, as a sensitive and specific detection method that can visually detect pathogens, still has very important clinical value for the diagnosis, treatment and follow-up of HP infection and related diseases. Therefore, the shortcomings of manual detection of HP infection in pathological histology are mainly: the results of pathological slides are judged by pathologists' naked eyes. error. Although the morphology of Helicobacter pylori is clear under improved W-S staining, inexperienced pathologists are likely to miss and misdiagnose complicated cases such as small number of bacteria, uneven section quality, complex and diverse gastric mucosal lesions, and even interference with pathogen observation. At the same time, pathologists are responsible for checking all visible biological tissues on the slices, and each patient has many slices, and each slice has more than 10 billion pixels when magnified by 40 times, so the workload of manual reading is extremely high. Large, easily affected by factors such as the reader's subjective emotions and fatigue reading. Furthermore, different pathologists may come to quite different diagnostic conclusions for the same patient. Therefore, this pathological and histological diagnosis method, which is highly dependent on human factors, has subjective differences, coupled with its shortcomings such as high work intensity, high time cost, and inconsistency in diagnosis, which will largely affect the early diagnosis and treatment of HP infection and thus affect patients. prognosis. In addition, cultivating qualified professional pathologists requires long-term professional training and practice. The training cycle is long and is easily affected by social factors such as the current socio-economic and cultural factors. This means that the number of pathologists in my country and the world is in short supply. The severe situation with a large gap needs to be resolved urgently.

Contents of the invention

Aiming at the shortcomings of manual reading in pathological histology, this invention intends to use a computer to conduct deep learning on a large number of pathological images of gastric Helicobacter pylori infection to establish an intelligent mathematical model for the pathological diagnosis of gastric Helicobacter pylori infection, and to build a model based on big data and deep learning. The artificial intelligence platform for gastric Helicobacter pylori infection-assisted pathological diagnosis based on algorithms can realize high-accuracy and high-efficiency intelligent film reading to assist clinical pathological diagnosis of gastric Helicobacter pylori infection and improve its accuracy, work efficiency and work efficiency. persistent state.

Based on this, the object of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a support system for pathological diagnosis of gastric Helicobacter pylori infection that can improve the efficiency of clinical diagnosis of gastric Helicobacter pylori infection and reduce medical costs.

In order to achieve the above object, the technical solution adopted by the present invention is: a support system for pathological diagnosis of gastric Helicobacter pylori infection based on big data deep learning, the support system includes: an image data acquisition unit for obtaining normal slice images of the stomach and The pathological slice image of the confirmed gastric Helicobacter pylori infection case is used as the input image data; the image data labeling unit is used to label the input image data and ensure that the label of the image is consistent with the real pathological diagnosis result of the image; The image database construction unit is used to classify and organize the marked image data provided by the image data labeling unit to construct a pathological image database; the convolutional neural network construction unit is used to construct the first convolutional neural network model; and convolution The neural network model training unit adjusts the parameters of the first convolutional neural network model by using the image data of the pathological image database, and trains the first convolutional neural network model to obtain pathological image data that can be used to detect patients The second convolutional neural network model.

Thus, the doctor can quickly diagnose whether the patient suffers from gastric Helicobacter pylori infection by combining the classification results and corresponding probabilities given by the support device for the input patient pathological images, as well as the doctor's professional knowledge and experience. Improve the efficiency of clinical diagnosis, thereby reducing medical costs; Among them, in order to ensure the accuracy of the collected image data, you can use the image annotation tool ASAP to annotate each pathological slice image to ensure that the label of the image is consistent with the real value; To speed up the training of the network model, you can use the GPU with high-speed parallel computing instead of the CPU for training; in order to speed up the detection speed of the convolutional neural network model, based on the convolutional neural network training unit, you can remodel the trained network model It is a CNN classification model structure with a variable step size, which is used in the detection method in the actual operation; the model will block the huge full-slice image, and divide the pre-selected living tissue area into ROI points of the same size. block, because the detection between blocks can be highly parallelized, the speed of detection is significantly improved under the parallelism of multiple GPUs, and then the prediction probability distribution image is generated through the detection of the CNN classification model with variable step size; the image database will pathological Image data is divided into training set, test set and verification set, etc.; the parameters of the first convolutional neural network model include network parameters such as learning rate, number of training times, and how many layers of the network. Training refers to automatically adjusting parameters when seeking the optimal solution the process of.

Preferably, the support system further includes a convolutional neural network model checking unit, configured to obtain an ideal convolutional neural network model. It should be noted that "ideal" here refers to the high accuracy and "robustness" of the convolutional neural network model.

Preferably, the convolutional neural network model verification unit includes a convolutional neural network model verification unit and a convolutional neural network model testing unit, and the convolutional neural network model verification unit is used to detect the second convolutional neural network The accuracy rate of the network model; the convolutional neural network model testing unit is used to detect whether the second convolutional neural network model is overfitting, so as to screen out a robust third convolutional neural network model; it should be explained Yes, if the accuracy rate of the model on the test set is significantly different from the accuracy rate in the verification set training, it means that the model is overfitting. At this time, you can return to the convolutional neural network training unit and adjust the network structure or parameters. Retrain to get a better network model; if the accuracy on the test set is very close to the accuracy in the validation set training, the model is more robust.

Preferably, the support system further includes a convolutional neural network model pre-training unit, configured to perform pre-training on the first convolutional neural network model when the input image data collected by the image data acquisition unit is insufficient .

Preferably, the supporting system further includes a pathological image data preprocessing unit, which is used for screening and displaying the region to be detected in the patient's pathological image.

Preferably, in order to ensure the validity of the detection, the preprocessing unit uses an adaptive threshold method to filter out the region to be detected.

Preferably, the convolutional neural network training unit uses a fine-tuning method to train the first convolutional neural network model.

As another aspect of the present invention, the present invention also provides a method for supporting the pathological diagnosis of gastric Helicobacter pylori infection, the supporting method comprising the following steps:

Image data acquisition: Obtain images of normal gastric slices and pathological slice images of confirmed cases of gastric Helicobacter pylori infection as input image data;

Image data labeling: labeling the input image data, and ensuring that the label of the image is consistent with the real pathological diagnosis result of the image;

Image database construction: classifying and sorting the labeled image data provided by the image data labeling unit, and constructing a pathological image database;

Convolutional neural network construction: constructing a first convolutional neural network model; and

Convolutional neural network model training: use the image data of the pathological image database to adjust the parameters of the first convolutional neural network model, and train the first convolutional neural network model to obtain pathological images that can be used to detect patients The second convolutional neural network model for the data.

It should be noted that image data annotation and image database construction can be regarded as the stage of pathological image database construction. Preferably, the supporting method also includes a convolutional neural network model verification step: obtaining an ideal convolutional neural network model; the convolutional neural network model verification step includes convolutional neural network model verification and convolutional neural network model testing , the convolutional neural network model verification is used to detect the accuracy of the second convolutional neural network model; the convolutional neural network model test is used to detect whether the second convolutional neural network model is overfitting combined to screen out a robust third convolutional neural network model. It should be noted that the convolutional neural network construction, convolutional neural network model training, and convolutional neural network model verification can be regarded as the training phase of the convolutional neural network, which is used to obtain an ideal convolutional neural network model.

As the third aspect of the present invention, the present invention also relates to the clinical application of the above support system in the pathological diagnosis of gastric Helicobacter pylori infection.

In summary, the beneficial effects of the present invention are:

Compared with the existing manual reading by pathologists, the supporting equipment for the pathological diagnosis of gastric Helicobacter pylori infection based on big data and deep learning algorithms in the present invention has the advantages of high accuracy, short time-consuming and long working duration, and this The promotion and cloud service of the invention in major hospitals, including the top three and grassroots hospitals, will help to solve the uneven distribution of medical resources, realize remote high-quality medical care, etc., and provide more convenient and accurate pathological diagnosis services for the majority of patients; the advantages of the above This is achieved because the device and method of the present invention utilize the advantages of deep learning algorithms in image recognition, allowing the computer to perform deep learning of pathological sections of gastric Helicobacter pylori infection at the large data level, so as to train a pathological doctor who can simulate the reading of slices and is comparable to it. The intelligent neural network model, after continuous learning and verification, the neural network model can realize the intelligent reading, rapid identification and scientific conclusion of the pathological sections of gastric Helicobacter pylori infection.

Description of drawings

Fig. 1 is the block diagram of structure of gastric Helicobacter pylori infection pathological diagnosis support system of the present invention;

2 is a flow chart of a method for supporting pathological diagnosis of gastric Helicobacter pylori infection according to Example 2 of the present invention;

Fig. 3 is the slice diagram of gastric Helicobacter pylori infection;

Fig. 4 is the schematic diagram of the rapid detection model of the second embodiment of the present invention;

Fig. 5 is the flowchart of the application of the gastric Helicobacter pylori infection pathological diagnosis support system of the present invention;

Among them, 1. Gastric Helicobacter pylori infection pathological diagnosis support system, 2. Image data acquisition unit, 3. Image data labeling unit, 4. Convolutional neural network construction unit, 5. Convolutional neural network model training unit, 6. Volume 7. Image database construction unit, 8. Pathological image data preprocessing unit, 9. Input terminal, 10. Output terminal.

Detailed ways

In order to better illustrate the purpose, technical solutions and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

Example 1

Referring to Fig. 1, a kind of embodiment of gastric Helicobacter pylori infection pathological diagnosis support system 1 of the present invention, it comprises:

The image data obtaining unit 2 is used to obtain the normal slice image of the stomach and the pathological slice image of the confirmed gastric Helicobacter pylori infection case as the input image data;

An image data labeling unit 3, configured to label the input image data, and ensure that the label of the image is consistent with the real pathological diagnosis result of the image;

The image database construction unit 7 is used to classify and organize the labeled image data provided by the image data labeling unit, and construct a pathological image database;

A convolutional neural network construction unit 4, configured to construct a first convolutional neural network model;

The convolutional neural network model training unit 5 uses the image data of the pathological image database to adjust the parameters of the first convolutional neural network model, and trains the first convolutional neural network model to obtain the second convolutional neural network model that can be used to detect the patient's pathological image data. Convolutional neural network model;

The convolutional neural network model verification unit 6 is used to obtain an ideal convolutional neural network model, including a convolutional neural network model verification unit (not shown in the figure) and a convolutional neural network model test unit (not shown in the figure), The convolutional neural network model check unit is used to detect the accuracy of the second convolutional neural network model; the convolutional neural network model test unit is used to detect whether the second convolutional neural network model is overfitting, to screen out robust The third convolutional neural network model.

Convolutional neural network model pre-training unit (not shown in the figure), used for pre-training the first convolutional neural network model when the input image data collected by the image data acquisition unit is insufficient; and

The pathological image data preprocessing unit 8 is used for screening and showing the region to be detected in the patient's pathological image.

Wherein, the pathological image data preprocessing unit 8 adopts the self-adaptive threshold method to filter out the area to be detected; the convolutional neural network model training unit 5 adopts the fine-tuning (fine-tune) method to train the first convolutional neural network model; the database includes the following four Class datasets: training set, validation set, test set, and regular published pathology image datasets.

In addition, the input terminal 9 is used to input the existing normal gastric slice images and pathological slice images of confirmed gastric Helicobacter pylori infection cases into the image data acquisition unit 2, and these input data will be finally classified by the image database construction unit 7 Collected to support subsequent clinical diagnosis;

And input the pathological slice image of the patient to be detected into the pathological image data preprocessing unit 8;

The output terminal 10 is used for the result of the pathological slice image classification detection (pathological type) of the patient to be detected by the robust convolutional neural network model obtained by the convolutional neural network model training unit 5 input to the pathological image data preprocessing unit 8 and the corresponding probability) are presented to doctors for reference in clinical diagnosis.

Example 2

Referring to Fig. 2, an embodiment of the diagnosis support method of gastric Helicobacter pylori infection of the present invention, it comprises the following steps:

(1) Collect image data

Using the medical biobank data of the Sixth Affiliated Hospital of Sun Yat-sen University as the data source, 14,000 pathological slice images were collected, including 7,000 gastric normal tissue slice images and 7,000 gastric Helicobacter pylori infected tissue slices, and respectively according to the training set: verification set : test set = 3:1:1 ratio of the number of random groups. As shown in Table 1 below:

Table 1 Specific data of pathological slice images.

The collected images were digitally scanned and stored, serial numbered and archived to create a pathological image database of gastric Helicobacter pylori infection.

(2) Annotate image information

Use the existing ASAP image labeling software to perform data labeling on the pathological slice images of the training set, verification set and test set collected in step (1). In order to ensure the accuracy of information labeling, it is necessary to optimize the image before labeling. The image labeling work mainly includes: outline various pathological structure areas in the image with different colors/thickness/virtual and solid lines, and determine the degree of inflammation and Type; under the microscope, brown or brown curved or short rod-shaped objects can be seen on a light yellow background; HP is mostly located in gastric pits or intrinsic glands. Tab naming. The correctly labeled pathological images are digitally stored for the next step of network model training and verification. Figure 3 is an annotated map of the atypical hyperplasia area in gastric Helicobacter pylori infection.

(3) Training convolutional neural network

①Design model

(a) Construct a convolutional neural network according to the cascade of convolutional layers, maximum sampling layers, nonlinear functions, and fully connected layers;

(b) Using a multi-layer network to enhance the fitting ability of the network;

(c) The number of channels of the output of the last fully connected layer of the network is set to 2, respectively representing that the image is a normal gastric slice image and a gastric Helicobacter pylori infected tissue slice image.

② training network

(a) According to the image data collected in steps (1), (2), the parameters of the convolutional neural network model are adjusted, and the accuracy of classification is observed on the verification set;

(b) In order to speed up the training network, use a GPU with high-speed parallel computing instead of a CPU for training;

(c) The method of updating the weight parameters of the convolutional neural network uses SGD to solve, if the convergence speed is slow, use Adadelta, Adam and other optimization methods to solve;

(d) If the number of training data (i.e., image data) collected in step (1) is too small, the convolutional neural network model is first pre-trained on a conventional public pathological image data set, and then the collected image data is fine- tune (fine-tuning) method to train the convolutional neural network model;

(e) If training on the existing convolutional neural network model, the classification accuracy cannot be increased, the fitting ability of the convolutional neural network model can be increased by increasing the depth of the convolutional neural network model.

③ Design a rapid detection model (as shown in Figure 4)

1. In order to improve the detection efficiency, an adaptive threshold method is used in the preprocessing stage to preselect the living tissue area from the whole slice image as the detection object of the convolutional neural network (as shown by the arrow 101 in Figure 4, representing the preprocessing process) .

2. In order to improve the accuracy and flexibility of detection, based on step (3), the trained convolutional neural network can be remodeled as a variable step-size CNN classification model for the detection method in actual operation; This model will block the huge full-slice image, and divide the pre-selected living tissue area into ROI blocks of the same size; since the detection between blocks can be highly parallelized, the detection speed is faster than that of multi-GPU. In this case, it is effectively improved (as shown by arrow 102 in FIG. 4 , which represents a fast model detection process). Through the detection of the convolutional neural network model, an image of the predicted probability distribution is generated.

3. Based on the predicted probability distribution image of step 2, in the post-processing, after screening out the scattered points, analyze the predicted probability distribution map to obtain the predicted result of the final full-slice image (as shown in Fig. 4 arrow 103, representing post-processing process).

(4) Test set verification

(a) Based on the variable step size classification model structure of step ③, use the convolutional neural network model trained in step (3) to test the test set, and observe the accuracy of the model on the test set.

(b) If the accuracy rate of the convolutional neural network model trained in step (3) on the test set is significantly different from the accuracy rate of the verification set in training, it means that the model is overfitting; at this time, return to In step (3), adjust the network structure or parameters of the convolutional neural network model to obtain a better network model.

(c) If the accuracy rate of the trained convolutional neural network model on the test set in step (3) is very close to the accuracy rate of the verification set in training, it means that the trained convolutional neural network model is relatively robust. Rod, can be used as a suitable network model for detecting pathological images of patients.

Example 3

An application example of the gastric Helicobacter pylori infection pathological diagnosis support system of the present invention, the pathological image to be detected is input into the pathological image data preprocessing unit 8 in the gastric Helicobacter pylori infection diagnosis support device of the present invention through the input terminal 9, See Figure 5 for the subsequent operation process, in which,

(a) In order to ensure the validity of the detection, an adaptive threshold method is used in the initial stage to pre-select the living tissue area from the whole slice image, and then select the area to be detected (ie the patient's pathological tissue area) based on the threshold result area center frame;

(b) After that, preprocessing is performed on the image of the patient's pathological tissue area, and the preprocessing includes steps such as denoising, histogram equalization, and normalization;

(c) Use the previously trained convolutional neural network model (i.e. the second convolutional neural network model in Example 1) to classify and detect the area to be detected in the preprocessed picture, thereby obtaining the infection rate of gastric Helicobacter pylori The prediction results include the category of gastric Helicobacter pylori infection to which the pathological section belongs and the corresponding probability.

Example 4 Comparison between the method for supporting the pathological diagnosis of gastric Helicobacter pylori infection of the present invention and existing methods

At present, the clinical pathological diagnosis is performed by the pathologists who have undergone standardized training to manually read the pathological tissue slides, and combine their long-term accumulated clinical diagnosis experience to make analysis and diagnosis. Since this method of manual naked eye image reading is closely related to the pathologist's own experience, working status, subjective emotions and other factors, the accuracy rate is not high, but it takes a long time and the working duration is limited, which is prone to missed diagnosis, misdiagnosis and inconsistent diagnosis. The present invention uses a computer to perform deep learning on a large number of standardized pathological images of gastric Helicobacter pylori infection, and performs parameter adjustment and fitting training on the convolutional neural network, thereby obtaining a relatively robust network model. This kind of neural network based on big data and deep learning can simulate manual film reading, and obtain the corresponding output value, that is, the pathological diagnosis conclusion, according to the new pathological image input. Furthermore, through model reconstruction, the detection speed is greatly improved without affecting the detection accuracy.

Select 30 doctors with more than 3 years of experience in the diagnosis and treatment of gastric Helicobacter pylori infection, each provide 40 pathological images of suspected gastric Helicobacter pylori infection, determine its type, then calculate the accuracy rate and average time, and count the doctor's diagnosis State, compared with the diagnostic support method of the present invention, the results are shown in Table 2 below.

Table 2 Comparison of diagnostic results of gastric Helicobacter pylori infection

It can be known from the above table 2 that the reading of pathological tissue slices by the method of the present invention has a higher accuracy rate than that of professional pathologists, and the time-consuming is shorter and the working duration is longer.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than limit the protection scope of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that, The technical solution of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solution of the present invention.

Claims (10)

1. gastric Helicobacter pylori infection pathological diagnosis support system, it is characterized in that, described support system comprises: The image data obtaining unit is used to obtain the normal slice image of the stomach and the pathological slice image of the confirmed gastric Helicobacter pylori infection case as the input image data; An image data labeling unit, configured to label the input image data, and ensure that the label of the image is consistent with the real pathological diagnosis result of the image; An image database construction unit, configured to classify and organize the labeled image data provided by the image data labeling unit, and construct a pathological image database; A convolutional neural network construction unit, configured to construct a first convolutional neural network model; and The convolutional neural network model training unit uses the image data of the pathological image database to adjust the parameters of the first convolutional neural network model, and trains the first convolutional neural network model to obtain pathological parameters that can be used to detect patients. A second convolutional neural network model for image data. 2. The support system according to claim 1, characterized in that, the support system also includes a convolutional neural network model checking unit for obtaining an ideal convolutional neural network model. 3. The support system according to claim 2, wherein the convolutional neural network model verification unit comprises a convolutional neural network model verification unit and a convolutional neural network model testing unit, and the convolutional neural network model The verification unit is used to detect the accuracy rate of the second convolutional neural network model; the convolutional neural network model testing unit is used to detect whether the second convolutional neural network model is overfitting, to screen out the Great third convolutional neural network model. 4. The support system according to claim 1, characterized in that, the support system also includes a convolutional neural network model pre-training unit for when the input image data collected by the image data obtaining unit is insufficient, Pre-training the first convolutional neural network model. 5. The support system according to claim 1, characterized in that, the support system further comprises a pathological image data preprocessing unit for screening and displaying the region to be detected in the patient's pathological image. 6. The support system according to claim 5, wherein the pre-processing unit uses an adaptive threshold method to filter out the region to be detected. 7. The support system according to claim 1, wherein the convolutional neural network training unit trains the first convolutional neural network model using a fine-tuning method. 8. A support method for the pathological diagnosis of gastric Helicobacter pylori infection, characterized in that the support method comprises the following steps: Image data acquisition: Obtain images of normal gastric slices and pathological slice images of confirmed cases of gastric Helicobacter pylori infection as input image data; Image data labeling: labeling the input image data, and ensuring that the label of the image is consistent with the real pathological diagnosis result of the image; Image database construction: classifying and sorting the labeled image data provided by the image data labeling unit, and constructing a pathological image database; Convolutional neural network construction: constructing a first convolutional neural network model; and Convolutional neural network model training: use the image data of the pathological image database to adjust the parameters of the first convolutional neural network model, and train the first convolutional neural network model to obtain pathological images that can be used to detect patients The second convolutional neural network model for the data. 9. The support method according to claim 8, characterized in that, the support method also includes a convolutional neural network model checking step: obtain an ideal convolutional neural network model; the convolutional neural network model checking step includes convolution Convolutional neural network model verification and convolutional neural network model test, the convolutional neural network model verification is used to detect the accuracy of the second convolutional neural network model; the convolutional neural network model test is used for Detecting whether the second convolutional neural network model is overfitting, so as to screen out a robust third convolutional neural network model. 10. Use of the supporting device according to any one of claims 1-6 in diagnosing gastric Helicobacter pylori infection.