Gastric atrophy typing detection method and system for digestive endoscopy

Abstract

The invention relates to the technical field of medical detection, and particularly discloses a stomach atrophy type detection method and system for a digestive endoscopy, and the method comprises the steps: 1, obtaining a stomach image generated through the digestive endoscopy, and then carrying out the preprocessing of the stomach image, and eliminating irrelevant information in the image; 2, segmenting the stomach image based on a color space through a fuzzy C mean value, and extracting an abnormal image region; 3, inputting the abnormal image region into a trained gastric atrophy typing model for processing, and determining a gastric atrophy typing result; maximum pooling Dropout and weight attenuation are combined with an SSD network, the overfitting problem in the model training process is effectively avoided, the generalization ability of the model is improved, positive samples and negative samples in an image sample data set are processed to achieve the balance of the number of the positive samples and the number of the negative samples, and the image quality is improved. When the image sample data set is applied to model training and recognition, the model recognition precision can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of medical detection technology, and in particular relates to a gastric atrophy typing detection method and system for digestive endoscopy. Background Art

[0002] Atrophy of the gastric mucosa is often accompanied by gastric diseases such as chronic gastritis, gastric ulcer, and gastric cancer. By detecting the atrophic area, these diseases can be detected and diagnosed early. And the degree and range of atrophy are usually related to the degree and prognosis of the gastric disease. Therefore, for patients with gastric diseases, during gastroscopy, if the location and range of the atrophic area are determined in advance, the scope of gastroscopy can be more accurately guided to avoid omissions or excessive examinations. At the same time, detecting the atrophic area can evaluate the degree and prognosis of the disease and guide the formulation of treatment plans. By sampling and pathological examination of the atrophic area, the type and degree of the lesion can be determined, providing a basis for the diagnosis and treatment of the disease.

[0003] There are deficiencies in the existing technology for detecting atrophic areas, especially the accuracy still needs to be improved. During the training process of the backbone network CNN model used in the detection, the data set is too small, contains a lot of noise, or the model is too complex, has too many training parameters, and is over-trained, which are all very likely to cause the model to fall into overfitting.

[0004] Therefore, it is necessary to propose a method and system for detecting gastric atrophy typing by digestive endoscopy to at least partially solve the problems existing in the prior art.

[0005] The above information disclosed in this background technology is only used to increase the understanding of the background technology of the present invention and therefore, it may include information that does not constitute the prior art known to ordinary technicians in this field. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for detecting gastric atrophy typing by digestive endoscopy, so as to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for detecting gastric atrophy typing by digestive endoscopy, comprising:

[0009] Step 1: obtaining a stomach image generated by digestive endoscopy, and then preprocessing the image to eliminate irrelevant information in the image;

[0010] Step 2: Segment the stomach image based on the color space by fuzzy C-means to extract the abnormal image area;

[0011] Step 3, inputting the abnormal image area into the trained gastric atrophy classification model for processing to determine the gastric atrophy classification result;

[0012] Step 4: Comprehensively consider the classification results and the patient's clinical symptoms and medical history to conduct a comprehensive assessment of the patient's gastric lesions.

[0013] Preferably, in step 1, the stomach image is preprocessed: Gaussian filtering is used to denoise the image, and a Gaussian function is used to calculate the weighted average of the neighborhood around each pixel in the image, where the weight is determined by a Gaussian distribution, so that noise can be reduced while retaining edge information of the image;

[0014] Then, contrast enhancement, brightness adjustment, sharpening enhancement and wavelet transform are used to improve the visual effect of the image. Histogram equalization is used to adjust the image contrast, the image brightness is adjusted by a linear function, and the Laplace operator is applied to enhance the edges and details of the image.

[0015] Preferably, in step 2, the stomach image is segmented using fuzzy C-means combined with local space and local grayscale information to maintain robustness and insensitivity to noise, and the original stomach image is used to avoid loss of details during preprocessing in step 1. Then, based on the result of this segmentation, the stomach image that has completed preprocessing in step 1 is segmented to obtain a segmented image for model processing.

[0016] Preferably, in step three, a gastric atrophy classification model is constructed by introducing maximum pooling Dropout and weight attenuation in the SSD model, the model is trained using the prepared training set, and the high-level semantic information is fused with the underlying detail information.

[0017] Preferably, the maximum pooling Dropout is introduced in the SSD model: the retention probability of each pooling area in the image to be pooled is set to p, generally set p = 0.5, then the suppression probability of each unit in the pooling area is q = 1-p, and at the same time, assuming that the unit value in each pooling area m of the lth layer Rearrange in ascending order, the cell values ​​after arrangement are (Note: The semi-linear activation function ReLU is used here to make the activation values ​​of all units non-negative, so the minimum ),So The output after pooling is selected as the maximum value of the entire pooling area, which is greater than The unit value of

[0018] Suppressed, only less than or equal to The value of is retained. Among these retained values, since the maximum value pooling (taking the maximum value) is adopted, the final output value of the pooling is The probability of its occurrence is p j ,Right now

[0019]

[0020] When performing maximum pooling Dropout in the pooling area, the jth activation value in the pooling area that is arranged in ascending order is selected by polynomial arrangement. As the output value of the pooling area,

[0021]

[0022] Preferably, the weight decay calculation is: Let the loss function L of the model be 0 As follows, when performing weight decay calculation, in the original loss function L 0 Add a penalty term in

[0023]

[0024] Among them, L 0 is the original loss function, w is the network weight, that is, the connection coefficient of the neuron, and λ (λ>0) is the penalty coefficient, which is used to measure the penalty term and L 0 The penalty term is the square sum of the network weights w. Taking the derivative we get:

[0025]

[0026] Among them, b is the bias of the network neuron, which is included in L 0 o n In, that is, o n =w·x n-1 +b, it can be found that adding the penalty term has no effect on the update of bias b. Let η be the learning rate, and the formula for the weight value w is as follows:

[0027]

[0028] Among them, η and λ are both positive numbers less than 1, so 1-ηλ<1.

[0029] Preferably, in step three, a training set is generated by bidirectional sampling of a sample correlation unbalanced data set, and high-level features rich in semantic information in the SSD network structure are mapped to a low-level network through deconvolution.

[0030] Preferably, in step four, the patient's clinical symptoms and medical history are used to test the typing results, confirm the correctness of the typing results, conduct a comprehensive assessment of the patient's gastric lesions, generate a comprehensive assessment report and send it to the doctor.

[0031] A gastric atrophy typing detection system for digestive endoscopy, comprising:

[0032] The image acquisition module acquires images and video data from digestive endoscopy and pre-processes the acquired image data to improve the accuracy and efficiency of subsequent analysis;

[0033] The image analysis module segments the image data through the fuzzy C-means clustering segmentation technology to segment the image data of the abnormal area;

[0034] An image recognition module, which uses a gastric atrophy classification model to identify abnormal image areas and determines a gastric atrophy classification result based on the recognition result;

[0035] Comprehensive evaluation module, which combines information from different modules to comprehensively evaluate the patient's gastric lesions and generate a corresponding diagnosis report;

[0036] The gastric atrophy typing detection system for digestive endoscopy is used to execute the gastric atrophy typing detection method for digestive endoscopy as described in any one of the above items.

[0037] Preferably, the detection system may also include a pathology evaluation module for performing pathological analysis on biopsy tissue samples, extracting lesion characteristics, and providing a basis for evaluating the type and extent of lesions; and a remote diagnosis module for supporting doctors in remote diagnosis and consultation, and providing auxiliary decision-making for patients' gastric lesions.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention combines maximum pooling Dropout with weight decay and SSD network to effectively avoid the overfitting problem in the model training process, improve the generalization ability of the model, and process the positive and negative samples in the image sample data set to achieve a balance between the number of positive and negative samples. The image sample data set is applied to model training and recognition, which can improve the model recognition accuracy and avoid the problem of high misclassification rate caused by poor training effect.

[0040] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the detection method of the present invention;

[0042] Figure 2 It is a framework diagram of the detection system of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Example:

[0045] See also Figure 1 As shown, a method for detecting gastric atrophy typing by digestive endoscopy comprises:

[0046] Step 1: obtaining a stomach image generated by digestive endoscopy, and then preprocessing the image to eliminate irrelevant information in the image;

[0047] Step 2: Segment the stomach image based on the color space by fuzzy C-means to extract the abnormal image area;

[0048] Step 3, inputting the abnormal image area into the trained gastric atrophy classification model for processing to determine the gastric atrophy classification result;

[0049] Step 4: Comprehensively consider the classification results and the patient's clinical symptoms and medical history to conduct a comprehensive assessment of the patient's gastric lesions.

[0050] In step 1, the stomach image is preprocessed: Gaussian filtering is used to denoise the image, and then contrast enhancement, brightness adjustment, sharpening enhancement and wavelet transform image enhancement methods are used to improve the visual effect of the image.

[0051] In step 2, the stomach image is segmented using fuzzy C-means combined with local spatial and local grayscale information to maintain robustness and insensitivity to noise, and the original stomach image is used to avoid loss of details during preprocessing in step 1.

[0052] Fuzzy factor G ij Defined as:

[0053]

[0054] Among them, i is the center pixel of the local sample window, j is the neighbor set of pixel j falling into the window of pixel i, k is the reference cluster, and d ijrepresents the Euclidean distance between pixels, u jk is the membership of pixel j in the kth cluster, v k is the center of the reference cluster k, m is the weighted index of each fuzzy membership, N j is the number of data items in the neighbor set of all pixel j, x i is the pixel gray value of the center pixel i;

[0055] The expression of the objective function is as follows:

[0056]

[0057] Among them, u ij is the fuzzy membership relationship of pixel j to the center pixel i, c (2≤c≤n) is the number of clusters, and n represents the number of data items.

[0058] In step three, a gastric atrophy classification model is constructed by introducing maximum pooling Dropout and weight decay in the SSD model, the model is trained using the prepared training set, and the high-level semantic information is fused with the underlying detail information; the training set is generated by bidirectional sampling of the sample correlation unbalanced data set, and the high-level features rich in semantic information in the SSD network structure are mapped to the low-level network through deconvolution.

[0059] For processing of image datasets:

[0060] Assume that in the image sample data set, the number of positive image samples is z, the number of negative images is f, and the average gray value g of each image is:

[0061]

[0062] Among them, g s ∈[0,1] is the gray value of pixel s in the image, N is the number of pixels in the image, (x s ,y s ) is the average gray value in the image The position coordinates of the nearest pixel s, (x r ,y r ) is the position coordinate of pixel r, β is the parameter factor when the pixel spatial relationship is digitized, H sr is the spatial relationship value of pixel points s and r. The size of β is determined by the pixel coordinate position. The purpose is to control H sr In a reasonable range of [0,1], if the maximum distance value among all the distances between pixels and pixel s can be taken as β, then the spatial relationship H between pixel r and pixel s in the image is sr Defined as:

[0063]

[0064] After the spatial relationship between all pixels r and the pixel point s where the average grayscale is located is obtained, the average spatial correlation value between all pixels in the entire image and the pixel point s is:

[0065]

[0066] The new features after image feature synthesis are:

[0067]

[0068] Among them, ε∈[0,1] is a weight factor, which is set by the ratio of the pixel gray value to the pixel spatial relationship value in the image; after determining the sample value of spatial correlation, the negative samples are undersampled using fuzzy C-means, and the number of clusters is set to After clustering is completed, The sample closest to each cluster center is the newly generated negative sample.

[0069] Let the sample feature pixel of an image in the positive sample be x t , d at Represents pixel x a With the feature pixel x t The Euclidean distance between at =||x t -x a ||,a∈[1,n t ], and d at <γ, γ is a certain threshold, n t Indicates that d is satisfied at <γ range x t The number of surrounding pixels is expressed as x t The pixel spatial density centered at is defined as:

[0070]

[0071] The total density d of all feature pixels in the image sample N for:

[0072]

[0073] The formula for the number of copies of an image sample is as follows:

[0074]

[0075] in, is the number of samples to be copied when a certain image sample τ is oversampled, is the total number of positive samples to be obtained, is the total number of samples to be replicated, d τNis the total density of sample τ.

[0076] In step four, the patient's clinical symptoms and medical history are used to test the classification results, confirm the correctness of the classification results, conduct a comprehensive assessment of the patient's gastric lesions, generate a comprehensive assessment report and send it to the doctor.

[0077] See also Figure 2 As shown, a gastric atrophy typing detection system for digestive endoscopy includes:

[0078] Image acquisition module, which acquires images and video data from digestive endoscopy and pre-processes the acquired image data to improve the accuracy and efficiency of subsequent analysis;

[0079] The image analysis module segments the image data through the fuzzy C-means clustering segmentation technology to segment the image data of the abnormal area;

[0080] An image recognition module, which uses a gastric atrophy classification model to identify abnormal image areas and determines a gastric atrophy classification result based on the recognition result;

[0081] Comprehensive evaluation module, which combines information from different modules to comprehensively evaluate the patient's gastric lesions and generate a corresponding diagnosis report;

[0082] A gastric atrophy typing detection system for digestive endoscopy is used to execute a gastric atrophy typing detection method for digestive endoscopy as described in any one of the above items.

[0083] The detection system can also include a pathology assessment module, which is used to perform pathological analysis on biopsy tissue samples, extract lesion characteristics, and provide a basis for evaluating the type and extent of lesions; a remote diagnosis module, which is used to support doctors in remote diagnosis and consultation, and provide auxiliary decision-making for patients' gastric lesions.

[0084] From the above, it can be seen that the present invention combines maximum pooling Dropout with weight decay and SSD network to effectively avoid the overfitting problem in the model training process, improve the generalization ability of the model, and process the positive and negative samples in the image sample data set to achieve a balance between the number of positive and negative samples. Applying the image sample data set to model training and recognition can improve the model recognition accuracy and avoid the problem of high misclassification rate caused by poor training effect.

[0085] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0086] In the drawings of the embodiments disclosed in the present invention, only the structures involved in the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0087] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting gastric atrophy typing by digestive endoscopy, characterized in that: include: Step 1: obtaining a stomach image generated by digestive endoscopy, and then preprocessing the image to eliminate irrelevant information in the image; Step 2: Segment the stomach image based on the color space by fuzzy C-means to extract the abnormal image area; Step 3, inputting the abnormal image area into the trained gastric atrophy classification model for processing to determine the gastric atrophy classification result; Step 4: Comprehensively consider the classification results and the patient's clinical symptoms and medical history to conduct a comprehensive assessment of the patient's gastric lesions.

2. A method for detecting gastric atrophy typing by digestive endoscopy according to claim 1, characterized in that: In step 1, the stomach image is preprocessed: Gaussian filtering is used to denoise the image, and then contrast enhancement, brightness adjustment, sharpening enhancement and wavelet transform image enhancement methods are used to improve the visual effect of the image.

3. A method for detecting gastric atrophy typing by digestive endoscopy according to claim 2, characterized in that: In step 2, the stomach image is segmented using fuzzy C-means combined with local space and local grayscale information to maintain robustness and noise insensitivity, and the original stomach image is used to avoid loss of details during preprocessing in step 1.

4. A method for detecting gastric atrophy typing by digestive endoscopy according to claim 3, characterized in that: In step three, a gastric atrophy classification model is constructed by introducing maximum pooling Dropout and weight attenuation in the SSD model, the model is trained using the prepared training set, and the high-level semantic information is fused with the underlying detail information.

5. A method for detecting gastric atrophy typing by digestive endoscopy according to claim 4, characterized in that: In step three, a training set is generated by bidirectional sampling of a sample correlation imbalanced data set, and high-level features rich in semantic information in the SSD network structure are mapped to a low-level network through deconvolution.

6. A method for detecting gastric atrophy typing by digestive endoscopy according to claim 5, characterized in that: In step 4, the patient's clinical symptoms and medical history are used to test the typing results, confirm the correctness of the typing results, conduct a comprehensive assessment of the patient's gastric lesions, generate a comprehensive assessment report and send it to the doctor.

7. A gastric atrophy typing detection system for digestive endoscopy, characterized in that: include: Image acquisition module, which acquires images and video data from digestive endoscopy and pre-processes the acquired image data to improve the accuracy and efficiency of subsequent analysis; The image analysis module segments the image data through the fuzzy C-means clustering segmentation technology to segment the image data of the abnormal area; An image recognition module, which uses a gastric atrophy classification model to identify abnormal image areas and determines a gastric atrophy classification result based on the recognition result; The comprehensive evaluation module combines the information from different modules to comprehensively evaluate the patient's gastric lesions and generate a corresponding diagnosis report.

8. A gastric atrophy typing detection system for digestive endoscopy according to claim 7, characterized in that: The detection system may also include a pathology evaluation module for performing pathological analysis on biopsy tissue samples, extracting lesion characteristics, and providing a basis for evaluating the type and extent of lesions; a remote diagnosis module for supporting doctors in remote diagnosis and consultation, and providing auxiliary decision-making for patients' gastric lesions.

Images