Endoscopic Image Recognition System and Device Based on Deep Learning

Abstract

The present invention discloses an endoscopic image recognition system and device based on deep learning. The system includes a preliminary screening unit for substituting an original gastroscope image into a preset convolutional neural network model to obtain a qualified gastric image; a secondary screening unit for substituting the qualified gastric image into a preset deep learning classification model to obtain an abnormal image; and an identification unit for substituting the abnormal image into an infection feature recognition model to obtain the infection features of Helicobacter pylori, and determining the current infection status according to the infection features. It can improve the accuracy and effectiveness of Helicobacter pylori infection judgment, provide a strong diagnostic basis for endoscopic physicians to diagnose HP infection, and at the same time assist endoscopic physicians to more effectively and accurately analyze the disease risk.

Description

Technical Field

[0001] The present invention relates to the technical field of medical auxiliary detection, and particularly to an endoscopic image recognition system and device based on deep learning. Background Art

[0002] Helicobacter pylori (H.pylori, HP) infection is one of the most common infectious diseases globally. It is estimated that the number of people infected with HP accounts for 40% to 50% of the global population; research shows that any type of gastritis associated with HP infection has the risk of developing into gastric cancer, and eradicating HP is an effective strategy for preventing gastric cancer; with the progress of endoscopic technology, endoscopic examination has been used to diagnose gastritis to determine the presence of HP infection and evaluate the risk of gastric cancer.

[0003] The characteristics of HP infection under endoscopy are usually non-specific and are distributed in multiple lesions, making it difficult to identify; according to the "Kyoto Gastritis Protocol" published in Japan in 2014, the characteristics of HP infection under endoscopy were analyzed and summarized, making HP infection easier to identify under endoscopy and facilitating the evaluation of the risk of gastric cancer, and Zhao et al. showed in "Accuracy of Endoscopic Diagnosis of Helicobacter pylori Based on the Kyoto Classification of Gastritis: A Multicenter Study" that the Kyoto gastritis classification can be used to assist in the diagnosis of HP infection in the Chinese population. It can be seen that the existing method is to analyze and summarize the characteristics of HP infection under endoscopy, making HP infection easier to identify under endoscopy and facilitating the evaluation of the risk of gastric cancer; however, the effects of different characteristics on the sensitivity, specificity, and accuracy of indicating HP infection are different, and there is a lack of a unified standard to judge whether there is HP infection. Summary of the Invention

[0004] The main object of the present invention is to provide an endoscopic image recognition system and device based on deep learning, aiming to solve the technical problem in the prior art that due to the different effects of the sensitivity, specificity, and accuracy of different characteristics of HP infection under endoscopy, the accuracy and effectiveness of HP infection judgment are relatively low.

[0005] In a first aspect, the present invention provides an endoscopic image recognition system based on deep learning. The endoscopic image recognition system based on deep learning includes the following steps:

[0006] A preliminary screening unit for substituting the original gastroscope image into a preset convolutional neural network model to obtain a qualified gastric image;

[0007] A secondary screening unit for substituting the qualified gastric image into a preset deep learning classification model to obtain an abnormal image;

[0008] An identification unit, configured to substitute the abnormal image into an infection feature recognition model to obtain the infection features of Helicobacter pylori, and determine the current infection status according to the infection features.

[0009] Optionally, the endoscopic image recognition system based on deep learning further includes: a model construction unit,

[0010] The model construction unit is configured to obtain an input item data set of Helicobacter pylori infection features, and construct a preset convolutional neural network model, the preset deep learning classification model, and the infection feature recognition model based on the residual network according to the input item data set.

[0011] Optionally, the preliminary screening unit is further configured to construct a preset convolutional neural network model based on the residual network, and substitute the original gastroscope image into the preset convolutional neural network model to obtain a qualified gastric image.

[0012] Optionally, the secondary screening unit is further configured to substitute the qualified gastric image into a preset deep learning classification model to filter out normal gastric images and obtain abnormal images of the stomach.

[0013] Optionally, the identification unit is further configured to substitute the abnormal image into the infection feature recognition model, obtain the features that match the preset Helicobacter pylori symptom set as the infection features of Helicobacter pylori, and determine the current infection status according to the infection features.

[0014] Optionally, the identification unit is further configured to determine the current infection status according to a preset intelligent diagnosis formula and the infection features.

[0015] Optionally, the identification unit is further configured to determine the number of infection features according to the infection features, and obtain an infection index through the following preset intelligent diagnosis formula;

[0016]

[0017] where f is the infection index, exp is the exponential function, b is the regression coefficient, x i represents the number of the i-th related feature of Helicobacter pylori infection, and lambda is the adjustment parameter;

[0018] Compare the infection index with a preset infection threshold. When the infection index is greater than the preset infection threshold, determine that the current infection status is the infected state;

[0019] When the infection index is not greater than the preset infection threshold, determine that the current infection status is the non-infected state.

[0020] Optionally, the recognition unit is further configured to obtain the correlation between each single feature in the infection features and Helicobacter pylori infection;

[0021] The recognition unit is further configured to perform regression processing on the correlations of the single features to obtain the regression coefficients of the single features;

[0022] The recognition unit is further configured to perform fitting and cross-validation on each single feature through logistic regression, select the adjustment parameter with the smallest cross-validation error, and add the adjustment parameter to the regression coefficient to obtain the log odds;

[0023] And obtain a preset infection threshold according to the log odds and the following formula:

[0024]

[0025] where Probablity is the preset infection threshold, exp is the exponential function, and Log Odds is the log odds.

[0026] In a second aspect, to achieve the above object, the present invention further provides an endoscopic image recognition device based on deep learning, characterized in that the endoscopic image recognition device based on deep learning includes: a memory, a processor, and an endoscopic image recognition program based on deep learning stored on the memory and executable on the processor, and the endoscopic image recognition program based on deep learning is configured to implement the functions of the endoscopic image recognition system based on deep learning as described above.

[0027] Optionally, the endoscopic image recognition device based on deep learning further includes: an endoscopic detector;

[0028] The endoscopic detector is configured to obtain the original gastroscope image of the target user and feed the original gastroscope image back to the processor.

[0029] The endoscopic image recognition system based on deep learning proposed by the present invention includes a preliminary screening unit for substituting the original gastroscope image into a preset convolutional neural network model to obtain a qualified gastric image; a secondary screening unit for substituting the qualified gastric image into a preset deep learning classification model to obtain an abnormal image; a recognition unit for substituting the abnormal image into an infection feature recognition model to obtain the infection features of Helicobacter pylori, and determining the current infection status according to the infection features; it can improve the accuracy and effectiveness of Helicobacter pylori infection judgment, provide a strong diagnostic basis for endoscopic physicians to diagnose HP infection, and at the same time assist endoscopic physicians to analyze the disease risk more effectively and accurately. Description of the Drawings

[0030] Figure 1Schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present invention;

[0031] Figure 2 Schematic flowchart of the first embodiment of the endoscopic image recognition system based on deep learning of the present invention;

[0032] Figure 3 Schematic flowchart of the second embodiment of the endoscopic image recognition system based on deep learning of the present invention;

[0033] Figure 4 Training diagram of the convolutional neural network model in the endoscopic image recognition system based on deep learning.

[0034] The realization, functional features and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0035] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0036] The solution of the embodiment of the present invention is mainly: through a preliminary screening unit, which is used to substitute the original gastroscope image into a preset convolutional neural network model to obtain a qualified gastric image; a secondary screening unit, which is used to substitute the qualified gastric image into a preset deep learning classification model to obtain abnormal images; an identification unit, which is used to substitute the abnormal images into an infection feature recognition model to obtain the infection features of Helicobacter pylori, and determine the current infection status according to the infection features; it can improve the accuracy and effectiveness of Helicobacter pylori infection judgment, provide a strong diagnostic basis for endoscopic physicians to diagnose HP infection, and at the same time assist endoscopic physicians to analyze the disease risk more effectively and accurately, solving the technical problem that the accuracy and effectiveness of HP infection judgment are relatively low due to the different effects of the sensitivity, specificity and accuracy of different features of HP infection under endoscopy in the prior art.

[0037] Refer to Figure 1 , Figure 1 Schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present invention.

[0038] Such as Figure 1As shown in the figure, the device may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface 1003 may further include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM memory or a stable memory (Non-Volatile Memory), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0039] Those skilled in the art can understand that Figure 1 the device structure shown in the figure does not constitute a limitation on the device, and it may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0040] As Figure 1 shown, the memory 1005, as a storage medium, may include an operating system, a network communication module, a user interface module, and an endoscope image recognition program based on deep learning.

[0041] The device of the present invention calls the endoscope image recognition program based on deep learning stored in the memory 1005 through the processor 1001 and performs the following operations:

[0042] A preliminary screening unit for substituting the original gastroscope image into a preset convolutional neural network model to obtain a qualified gastric image;

[0043] A secondary screening unit for substituting the qualified gastric image into a preset deep learning classification model to obtain an abnormal image;

[0044] An identification unit for substituting the abnormal image into an infection feature recognition model to obtain the infection features of Helicobacter pylori, and determining the current infection status according to the infection features.

[0045] The device of the present invention calls the endoscope image recognition program based on deep learning stored in the memory 1005 through the processor 1001 and also performs the following operations:

[0046] The model construction unit is used to obtain an input item dataset of Helicobacter pylori infection features, and construct a preset convolutional neural network model, the preset deep learning classification model, and the infection feature recognition model based on the residual network according to the input item dataset.

[0047] The device of the present invention calls the deep learning-based endoscopic image recognition program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0048] The preliminary screening unit is further configured to construct a preset convolutional neural network model based on the residual network, substitute the original gastroscope image into the preset convolutional neural network model, and obtain a qualified gastric image.

[0049] The device of the present invention calls the deep learning-based endoscopic image recognition program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0050] The secondary screening unit is further configured to substitute the qualified gastric image into a preset deep learning classification model, filter out normal gastric images, and obtain abnormal images of the stomach.

[0051] The device of the present invention calls the deep learning-based endoscopic image recognition program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0052] The recognition unit is further configured to substitute the abnormal image into an infection feature recognition model, obtain the features that match the preset Helicobacter pylori symptom set as the infection features of Helicobacter pylori, and determine the current infection status according to the infection features.

[0053] The device of the present invention calls the deep learning-based endoscopic image recognition program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0054] The recognition unit is further configured to determine the current infection status according to a preset intelligent diagnosis formula and the infection features.

[0055] The device of the present invention calls the deep learning-based endoscopic image recognition program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0056] The recognition unit is further configured to determine the number of infection features according to the infection features, and obtain an infection index through the following preset intelligent diagnosis formula;

[0057]

[0058] where f is the infection index, exp is the exponential function, b is the regression coefficient, x i represents the number of the i-th related feature of Helicobacter pylori infection, and lambda is the adjustment parameter;

[0059] Compare the infection index with a preset infection threshold, and when the infection index is greater than the preset infection threshold, determine the current infection status as the infected state;

[0060] When the infection index is not greater than the preset infection threshold, it is determined that the current infection status is a non-infected status.

[0061] The device of the present invention calls the deep learning-based endoscopic image recognition program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0062] The recognition unit is also used to perform regression processing on the correlation of each single feature to obtain the regression coefficient of each single feature;

[0063] The recognition unit is also used to fit and cross-validate each single feature through logistic regression, select the adjustment parameter with the smallest cross-validation error, and add the adjustment parameter to the regression coefficient to obtain the log odds;

[0064] And obtain the preset infection threshold according to the log odds and the following formula:

[0065]

[0066] Where Probablity is the preset infection threshold, exp is the exponential function, and Log Odds is the log odds.

[0067] Further, the deep learning-based endoscopic image recognition device further includes: an endoscopic detector;

[0068] The endoscopic detector is used to obtain the original gastroscope image of the target user and feed the original gastroscope image back to the processor.

[0069] It can be understood that the endoscopic detector is used to obtain the original gastroscope image of the target user. The endoscopic detector generally includes a guide core and a probe, and its material can be a soft material, so as to avoid the harm of the probe to the person to be detected and reduce the discomfort of the person to be detected.

[0070] In this embodiment, through the above solution, the preliminary screening unit is used to substitute the original gastroscope image into the preset convolutional neural network model to obtain a qualified gastric image; the secondary screening unit is used to substitute the qualified gastric image into the preset deep learning classification model to obtain an abnormal image; the recognition unit is used to substitute the abnormal image into the infection feature recognition model to obtain the infection feature of Helicobacter pylori, and determine the current infection status according to the infection feature; it can improve the accuracy and effectiveness of Helicobacter pylori infection judgment, provide a strong diagnostic basis for endoscopic physicians to diagnose HP infection, and at the same time assist endoscopic physicians to analyze the disease risk more effectively and accurately.

[0071] Based on the above hardware structure, an embodiment of the deep learning-based endoscopic image recognition system of the present invention is proposed.

[0072] Reference Figure 2 , Figure 2 is a schematic flowchart of the first embodiment of the endoscopic image recognition system based on deep learning of the present invention.

[0073] In the first embodiment, the endoscopic image recognition system based on deep learning includes the following steps:

[0074] The preliminary screening unit 10 is used to substitute the original gastroscope image into a preset convolutional neural network model to obtain a qualified gastric image.

[0075] It should be noted that the original gastroscope image is the original detection image generated by the target person to be detected during the detection of the stomach through an endoscopic detection device, and the preset convolutional neural network model is a pre-set convolutional neural network model for preliminarily screening and identifying the original gastric image.

[0076] The secondary screening unit 20 is used to substitute the qualified gastric image into a preset deep learning classification model to obtain abnormal images.

[0077] It can be understood that the preset deep learning classification model is a pre-set deep learning classification model for judging normal and abnormal images in the qualified gastric images. By substituting the qualified gastric image into the preset deep learning classification model, abnormal images can be obtained.

[0078] The recognition unit 30 is used to substitute the abnormal image into an infection feature recognition model to obtain the infection features of Helicobacter pylori, and determine the current infection status according to the infection features.

[0079] It should be understood that the infection feature recognition model is a pre-set recognition model for identifying whether there are infection features of Helicobacter pylori in the abnormal image, so as to determine different current infection statuses according to different infection features.

[0080] In this embodiment, through the above solution, the preliminary screening unit is used to substitute the original gastroscope image into a preset convolutional neural network model to obtain a qualified gastric image; the secondary screening unit is used to substitute the qualified gastric image into a preset deep learning classification model to obtain abnormal images; the recognition unit is used to substitute the abnormal image into an infection feature recognition model to obtain the infection features of Helicobacter pylori, and determine the current infection status according to the infection features; it can improve the accuracy and effectiveness of Helicobacter pylori infection judgment, provide a strong diagnostic basis for endoscopic physicians to diagnose HP infection, and at the same time assist endoscopic physicians to analyze the disease risk more effectively and accurately.

[0081] Furthermore, Figure 3 is a schematic flowchart of the second embodiment of the endoscopic image recognition system based on deep learning of the present invention, asFigure 3 As shown in the figure, based on the first embodiment, the second embodiment of the endoscopic image recognition system based on deep learning of the present invention is proposed. In this embodiment, the endoscopic image recognition system based on deep learning further includes: a model construction unit 40;

[0082] The model construction unit 40 is used to obtain an input item dataset of Helicobacter pylori infection characteristics, and based on the input item dataset, construct a preset convolutional neural network model, the preset deep learning classification model, and the infection characteristic recognition model based on the residual network.

[0083] It should be noted that after obtaining the input item dataset of Helicobacter pylori infection characteristics, several models can be constructed based on the residual network, namely the preset convolutional neural network model, the preset deep learning classification model, and the infection characteristic recognition model, which are used to assist in the recognition of endoscopic images.

[0084] In specific implementation, an intelligent diagnosis formula for judging Helicobacter pylori HP infection in endoscopic images based on deep learning can be established:

[0085] M hp = f(X, Y, Z)

[0086] where M hp is the judgment result of whether there is HP infection, f is the intelligent diagnosis formula, X is the input item dataset of HP infection characteristics, Y is the automatically optimized item dataset of HP infection parameters, and Z is the comprehensive judgment item dataset of HP infection.

[0087] Correspondingly, after obtaining the input item dataset of HP infection characteristics, a deep learning model can be constructed based on the residual network, such as Resnet50, namely the gastric image recognition model A (preset convolutional neural network model), the positive and abnormal image judgment model B (preset deep learning classification model), and the HP infection characteristic recognition model C (infection characteristic recognition model); generally, it is necessary to train and adjust the parameters of models A, B, and C to make the model accuracy meet the requirements.

[0088] In specific implementation, as Figure 4 shown, Figure 4 is the training diagram of the convolutional neural network model in the endoscopic image recognition system based on deep learning. Refer to Figure 4, after the gastroscope images are input into the sample database, they can be manually marked according to the picture attributes, marked as esophagus, duodenum or other features, as well as HP infection-related features. Through the HP infection intelligent diagnosis formula, it can be determined whether there is HP infection. At the same time, the sample database and the gastroscope images can be used for machine training and learning, and the accuracy rate can be verified. When the accuracy rate passes, a convolutional neural network model is generated. When the verification accuracy rate fails, the reasons are analyzed, and then the model is optimized, and the above steps are carried out again until the verification accuracy rate passes; the generated preset convolutional neural network model can be used to identify the esophagus, duodenum, HP infection-related features (goose skin-like changes, atrophy, intestinal metaplasia, xanthoma, mucosal swelling, serpentine swelling of folds, white fluid turbidity, diffuse redness, punctate redness and hyperplastic polyps) and other features.

[0089] Through the above solution in this embodiment, by obtaining the input item dataset of Helicobacter pylori infection characteristics, based on the input item dataset, a preset convolutional neural network model, the preset deep learning classification model and the infection characteristic recognition model are constructed based on the residual network, which can improve the speed and efficiency, and then improve the accuracy and effectiveness of Helicobacter pylori infection judgment.

[0090] Further, continue to refer to Figure 2 , Figure 2 The preliminary screening unit in it is also used to construct a preset convolutional neural network model based on the residual network, and substitute the original gastroscope image into the preset convolutional neural network model to obtain qualified gastric images.

[0091] It can be understood that a preset convolutional neural network model can be constructed based on the residual network. Substituting the original gastroscope image into the preset convolutional neural network model, that is, inputting the original white light gastroscope image to be classified, and using the preset convolutional neural network model for deep learning classification, the esophagus, duodenum, blurred and qualified gastric images can be identified from the original gastroscope image.

[0092] Further, continue to refer to Figure 2 , Figure 2 The secondary screening unit in it is also used to substitute the qualified gastric images into the preset deep learning classification model to filter out normal gastric images and obtain abnormal images of the stomach.

[0093] It should be noted that substituting the qualified gastric images into the preset deep learning classification model can perform secondary screening, that is, identifying normal images and abnormal images from the qualified gastric images.

[0094] Further, continue to refer to Figure 2 , Figure 2The recognition unit in it is further configured to substitute the abnormal image into an infection feature recognition model, obtain features that match the preset Helicobacter pylori symptoms as the infection features of Helicobacter pylori, and determine the current infection status according to the infection features.

[0095] It should be understood that after substituting the abnormal image into the infection feature recognition model, features that match the preset Helicobacter pylori symptoms can be further recognized, and then these features can be used as the infection features of Helicobacter pylori. The infection features can be chicken-skin-like changes, atrophy, intestinal metaplasia, xanthoma, mucosal swelling, serpentine swelling of folds, white fluid turbidity, diffuse redness, punctate redness, and hyperplastic polyps and other HP infection-related features. This embodiment does not limit this; after obtaining the infection features, different current infection statuses can be determined according to different infection features.

[0096] Further, continue to refer to Figure 2 , Figure 2 The recognition unit in it is further configured to determine the current infection status according to a preset intelligent diagnosis formula and the infection features.

[0097] It should be noted that the preset intelligent diagnosis formula is a formula for the infection index corresponding to the infection features of Helicobacter pylori set in advance. The current infection status can be determined through the infection features and the preset intelligent diagnosis formula.

[0098] Further, continue to refer to Figure 2 , Figure 2 The recognition unit in it is further configured to determine the number of infection features according to the infection features, and obtain the infection index through the following preset intelligent diagnosis formula;

[0099]

[0100] Where f is the infection index, exp is the exponential function, b is the regression coefficient, x i represents the number of the i-th related feature of Helicobacter pylori infection, and lambda is the adjustment parameter;

[0101] Compare the infection index with a preset infection threshold. When the infection index is greater than the preset infection threshold, determine the current infection status as the infected state;

[0102] When the infection index is not greater than the preset infection threshold, determine the current infection status as the non-infected state.

[0103] It should be understood that the preset infection threshold is a threshold set in advance to determine whether the current infection status is a definite infected state. By comparing the infection index with the preset infection threshold, the current infection status can be quickly determined.

[0104] In a specific implementation, when the infection index f > the preset infection threshold α, it is considered that the HP infection is positive, that is, the output result is positive, otherwise it is negative.

[0105] Further, continue to refer to Figure 2 , Figure 2 the recognition unit described in, and is further configured to obtain the correlation between each single feature in the infection characteristics and Helicobacter pylori infection;

[0106] the recognition unit is further configured to perform regression processing on the correlation of each single feature to obtain the regression coefficient of each single feature;

[0107] the recognition unit is further configured to perform fitting and cross-validation on each single feature through logistic regression, select the adjustment parameter with the smallest cross-validation error, and add the adjustment parameter and the regression coefficient to obtain the log odds;

[0108] and obtain the preset infection threshold according to the log odds and the following formula:

[0109]

[0110] where Probablity is the preset infection threshold, exp is the exponential function, and Log Odds is the log odds.

[0111] In a specific implementation, automatic optimization of the HP infection model parameters can be performed, that is, a deep learning classification model is used, and a single feature is used to judge whether there is HP infection to obtain the correlation (sensitivity, specificity, negative predictive value, positive predictive value) between each feature and HP infection; based on the correlation between the single feature and HP infection, the "glmnet" software package in the R statistical software can be used for regression processing to obtain the regression coefficient of each feature; of course, the regression coefficients of each single feature can also be obtained by other means, such as obtaining the regression coefficient by adjusting the model parameters, and this embodiment does not limit this;

[0112] It should be understood that through the obtained regression coefficients, logistic regression is used to fit different features to obtain a model. At the same time, to avoid overfitting of the model, a cross-validation method is adopted, and the adjustment parameter lambda value with the smallest cross-validation error is selected; furthermore, comprehensive judgment of H infection can be performed, including:

[0113] Based on the correlation coefficient of each obtained feature and the adjustment parameter lambda value, model D is used to refit the model with all the data for optimization to obtain model E.

[0114] 1) In model E, the regression coefficient of each obtained feature and the adjustment parameter are added to calculate the Log Odds, and according to: Convert to probability.

[0115] 2) Analyze the receiver operating characteristic (ROC) curve based on the sensitivity and specificity of HP infection judged according to the obtained single feature, select the optimal sensitivity and specificity, set the threshold α of model E, and finally, form the intelligent diagnosis formula f for HP infection.

[0116] It can be understood that features related to HP infection can be effectively identified, and the correlation between different features and the diagnosis of HP infection can be obtained. That is, by further verifying the correlation between different features and the diagnosis of HP infection, then performing regression processing on the correlation, obtaining the regression coefficient of each feature, and using the cross-validation method to select the model with the smallest error, obtaining the adjusted parameter lambda, and finally forming an intelligent diagnosis formula for judging HP infection, which can provide a strong diagnostic basis for endoscopists to diagnose HP infection, and can assist endoscopists to more effectively and accurately analyze the risk of early gastric cancer, and at the same time prompt patients to perform corresponding treatments, thereby helping patients detect and eradicate HP early to prevent the progression of HP infection-related gastritis to gastric cancer.

[0117] Through the above solution, this embodiment can further improve the comprehensiveness of Helicobacter pylori infection judgment, improve the accuracy and effectiveness of Helicobacter pylori infection judgment, provide a strong diagnostic basis for endoscopists to diagnose HP infection, and at the same time assist endoscopists to more effectively and accurately analyze the disease risk.

[0118] It should be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or system. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article or system including that element.

[0119] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0120] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be included in the patent protection scope of the present invention by the same token.

Claims

1. An endoscopic image recognition system based on deep learning, characterized in that, the endoscopic image recognition system based on deep learning includes: A preliminary screening unit for substituting the original gastroscope image into a preset convolutional neural network model to obtain a qualified gastric image; A secondary screening unit for substituting the qualified gastric image into a preset deep learning classification model to obtain abnormal images; An identification unit for substituting the abnormal image into an infection feature recognition model to obtain the infection features of Helicobacter pylori, and determining the current infection status according to the infection features; wherein, the identification unit is further configured to substitute the abnormal image into the infection feature recognition model, obtain features that match the preset Helicobacter pylori symptom set as the infection features of Helicobacter pylori, and determine the current infection status according to the infection features; wherein, the identification unit is further configured to determine the current infection status according to a preset intelligent diagnosis formula and the infection features; wherein, the identification unit is further configured to determine the number of infection features according to the infection features, and obtain an infection index through the following preset intelligent diagnosis formula; where f is the infection index, exp is the exponential function, b is the regression coefficient, x i represents the number of the i-th relevant feature of Helicobacter pylori infection, and lambda is the adjustment parameter; Compare the infection index with a preset infection threshold, and when the infection index is greater than the preset infection threshold, determine that the current infection status is the infected state; When the infection index is not greater than the preset infection threshold, determine that the current infection status is the non-infected state.

2. The endoscopic image recognition system based on deep learning according to claim 1, characterized in that, the endoscopic image recognition system based on deep learning further includes: a model construction unit, The model construction unit is configured to obtain an input item dataset of Helicobacter pylori infection features, and construct a preset convolutional neural network model, the preset deep learning classification model, and the infection feature recognition model based on the residual network according to the input item dataset.

3. The endoscopic image recognition system based on deep learning according to claim 1, characterized in that, The preliminary screening unit is further configured to construct a preset convolutional neural network model based on the residual network, and substitute the original gastroscope image into the preset convolutional neural network model to obtain a qualified gastric image.

4. The endoscopic image recognition system based on deep learning according to claim 3, characterized in that, The secondary screening unit is further configured to substitute the qualified gastric image into a preset deep learning classification model to filter out normal gastric images and obtain abnormal images of the stomach.

5. The endoscopic image recognition system based on deep learning according to claim 1, characterized in that, The identification unit is further configured to obtain the correlation between each single feature in the infection features and Helicobacter pylori infection; The identification unit is further configured to perform regression processing on the correlation of each single feature to obtain the regression coefficient of each single feature; The identification unit is further configured to perform fitting and cross-validation on each single feature through logistic regression, select the adjustment parameter with the smallest cross-validation error, and add the adjustment parameter to the regression coefficient to obtain the log odds; And obtain a preset infection threshold according to the log odds and the following formula: wherein, is a preset infection threshold, exp is an exponential function, is the logit.

6. An endoscopic image recognition device based on deep learning, characterized in that, The endoscopic image recognition device based on deep learning includes: a memory, a processor, and an endoscopic image recognition program based on deep learning stored on the memory and executable on the processor, and the endoscopic image recognition program based on deep learning is configured to implement the functions of the endoscopic image recognition system based on deep learning according to any one of claims 1 to 5.

7. The endoscopic image recognition device based on deep learning according to claim 6, wherein, the endoscopic image recognition device based on deep learning further includes: an endoscopic detector; the endoscopic detector is configured to acquire an original gastroscope image of a target user and feed the original gastroscope image back to the processor.

Images