CN117351218B - Method for identifying inflammatory bowel disease pathological morphological feature crypt stretching image - Google Patents
Method for identifying inflammatory bowel disease pathological morphological feature crypt stretching image Download PDFInfo
- Publication number
- CN117351218B CN117351218B CN202311646973.8A CN202311646973A CN117351218B CN 117351218 B CN117351218 B CN 117351218B CN 202311646973 A CN202311646973 A CN 202311646973A CN 117351218 B CN117351218 B CN 117351218B
- Authority
- CN
- China
- Prior art keywords
- image
- crypt
- feature
- convolution
- stretching
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 208000022559 Inflammatory bowel disease Diseases 0.000 title claims abstract description 24
- 230000001575 pathological effect Effects 0.000 title claims abstract description 14
- 230000000877 morphologic effect Effects 0.000 title claims abstract description 11
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 18
- 239000011159 matrix material Substances 0.000 claims description 26
- 238000011176 pooling Methods 0.000 claims description 19
- 239000013598 vector Substances 0.000 claims description 19
- 238000000605 extraction Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 abstract description 5
- 230000002757 inflammatory effect Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 8
- 230000000968 intestinal effect Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 208000011231 Crohn disease Diseases 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 206010009900 Colitis ulcerative Diseases 0.000 description 2
- 201000006704 Ulcerative Colitis Diseases 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 208000027138 indeterminate colitis Diseases 0.000 description 2
- 208000028774 intestinal disease Diseases 0.000 description 2
- NLMDJJTUQPXZFG-UHFFFAOYSA-N 1,4,10,13-tetraoxa-7,16-diazacyclooctadecane Chemical compound C1COCCOCCNCCOCCOCCN1 NLMDJJTUQPXZFG-UHFFFAOYSA-N 0.000 description 1
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 235000008495 Chrysanthemum leucanthemum Nutrition 0.000 description 1
- 235000000604 Chrysanthemum parthenium Nutrition 0.000 description 1
- 208000017189 Gastrointestinal inflammatory disease Diseases 0.000 description 1
- 241000764238 Isis Species 0.000 description 1
- 240000004460 Tanacetum coccineum Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001684 chronic effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 235000008384 feverfew Nutrition 0.000 description 1
- 210000001035 gastrointestinal tract Anatomy 0.000 description 1
- 210000004907 gland Anatomy 0.000 description 1
- 238000002347 injection Methods 0.000 description 1
- 239000007924 injection Substances 0.000 description 1
- 210000004347 intestinal mucosa Anatomy 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000004400 mucous membrane Anatomy 0.000 description 1
- 210000000813 small intestine Anatomy 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/34—Smoothing or thinning of the pattern; Morphological operations; Skeletonisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The application relates to the field of image processing, in particular to a method for identifying a crypt stretching image of pathological morphological characteristics of inflammatory bowel disease, which comprises the following steps: setting a tag on a historical microscopic image of the crypt, wherein the tag comprises a crypt stretching image and a non-crypt stretching image, so as to obtain a data set; building a convolutional neural network model; and setting a loss function, training the convolutional neural network model by using the data set to obtain an optimal model, responding to the newly shot microscopic image, and generating an image classification result of the stretching of the crypt according to the optimal model. According to the method and the device, the crypt stretching image can be identified rapidly, a doctor is assisted to accurately judge the inflammatory bowel crypt stretching phenomenon, and the judging accuracy and efficiency are improved.
Description
Technical Field
The application relates to the field of image processing, in particular to a method for identifying inflammatory bowel disease pathomorphology feature crypt stretching images.
Background
In recent years, with rapid development of image processing and artificial intelligence technology, artificial intelligence technology is applied in more and more fields. IBD (inflammatory bowel disease ) refers to a group of nonspecific chronic gastrointestinal inflammatory diseases of unknown cause, including UC (ulcerative colitis ), CD (Crohn disease) and IC (indeterminate colitis ). In recent years, IBD, particularly CD, has been diagnosed in the early-late-stage and feverfew stages, and about 20% to 30% of IBD is diagnosed in childhood, and the clinical manifestations of childhood inflammatory bowel disease are mainly primary, with the symptoms being more severe as the onset age is smaller.
The intestinal crypt (intestinal crypt) is a tubular gland formed by the subsidence of the small intestinal epithelium into the lamina propria at the root of the villus, opening between adjacent villi. Intestinal crypt stretching refers to a morphological feature of the small intestine that causes pathological stretching of the crypt relative to the crypt in its normal state.
A doctor can judge the stretching phenomenon of the intestinal crypt according to the image of the intestinal canal, but the artificial recognition workload is large, the subjectivity is strong, the judgment accuracy and efficiency are low, and the prior art lacks a method for assisting the doctor in judging the stretching of the inflammatory intestinal crypt.
Disclosure of Invention
In order to assist doctors in accurately judging inflammatory bowel crypt stretching, the application provides a method for identifying inflammatory bowel disease pathological morphological feature crypt stretching images.
The application provides a method for identifying a crypt stretching image of pathological morphological characteristics of inflammatory bowel disease, which adopts the following technical scheme: the method comprises the following steps: setting a tag on a historical microscopic image of the crypt, wherein the tag comprises a crypt stretching image and a non-crypt stretching image, so as to obtain a data set; building a convolutional neural network model; setting a loss function and training the convolutional neural network model by using the data set to obtain an optimal model, wherein the expression of the loss function is as follows:，
wherein,representing a loss function->Indicate->Zhang Yinwo prediction of image, < >>Indicate->Zhang Yinwo true value of the image; and responding to the newly shot microscopic image, and generating an image classification result of the crypt stretching according to the optimal model.
Optionally, the convolutional neural network model includes: the device comprises a first convolution layer, a first extraction layer, a second convolution layer and a full connection layer which are sequentially connected, wherein the first extraction layer and the second extraction layer comprise a residual error module and an attention mechanism module which are sequentially connected; the input historical microscopic image of the hidden pit sequentially passes through the first convolution layer, the first extraction layer, the second convolution layer and the full connection layer and then outputs a classification result.
Optionally, the establishing of the attention mechanism module includes the steps of: passing a historical microscopic image of a recess through a first convolution layer to generate an input feature tensor, wherein the input feature tensor consists of feature graphs of a plurality of channels; calculating the characteristic difference value of the characteristic graph to obtain a vector with the length being the input characteristic tensor channel number; normalizing the vector to obtain the weight of the feature map of each channel; and multiplying the input characteristic tensor with the weight of each channel characteristic graph to obtain a new characteristic tensor, and obtaining the attention mechanism module for representing the row-column difference.
Optionally, the calculating the feature difference value of the feature map includes the steps of: carrying out row convolution operation on the input characteristic tensor according to the row convolution kernel of 1 multiplied by 3 to obtain the row convolution of the characteristic graph, and carrying out column convolution operation according to the column convolution kernel of 3 multiplied by 1 to obtain the column convolution of the characteristic graph; the method comprises the steps of performing row convolution according to a row using global average pooling method and global maximum pooling method, and splicing to obtain a first feature matrix; a global average pooling method and a global maximum pooling method are used for the column convolution according to the columns, and a second feature matrix is obtained through splicing; calculating L2 norms of the first feature matrix and the second feature matrix to obtain a row-column difference matrix; and using a global maximum pooling method for the row-column difference matrix to obtain a difference characteristic value of the characteristic map.
Optionally, in the training of the convolutional neural network model, the convolutional neural network is trained according to the error back propagation between the classification result and the label.
Optionally, in the training of the convolutional neural network model: the training termination condition is that the training times reach a preset time threshold or the loss function value is less than a preset loss threshold.
The application has the following technical effects:
1. the method for identifying the crypt stretching image of the inflammatory bowel disease by utilizing the image processing or deep learning technology provides a method for identifying the crypt stretching image of the pathological morphological characteristics of the inflammatory bowel disease so as to assist doctors in identifying the crypt stretching image and reduce the phenomena of low efficiency, strong subjectivity and the like of manually identifying the crypt stretching image of the inflammatory bowel disease. The workload of doctors can be reduced, and the accuracy and the efficiency of judgment are improved.
2. And constructing an attention mechanism module according to the row-column difference of the hidden-recess stretching image, so that model training is more focused on the feature of the hidden-recess stretching of the image, and the recognition accuracy of the model is improved. By the size ofPerforming row convolution operation on the convolution check feature images of (1) to obtain row convolution of images, performing row convolution operation on the convolution check feature images of (3×1) to obtain column convolution of images, giving a weight to the feature images according to differences of the feature images of the row and the column features, multiplying the weight of each channel feature image with the input feature image to obtain output feature images, and performing the above operation after each residual module to enable the trained model to correct inflammatory bowel disease cryptThe stretching attention is higher, and the recognition accuracy of the model is improved.
3. The method can be combined with other intestinal disease identification technologies to realize more comprehensive automatic intestinal disease monitoring.
Drawings
The above, as well as additional purposes, features, and advantages of exemplary embodiments of the present application will become readily apparent from the following detailed description when read in conjunction with the accompanying drawings. In the drawings, several embodiments of the present application are shown by way of example and not by way of limitation, and identical or corresponding reference numerals indicate identical or corresponding parts.
FIG. 1 is a flow chart of a method of step S1-S4 in a method of identifying a crypt stretch image of a pathomorphological feature of inflammatory bowel disease according to an embodiment of the present application.
Fig. 2 is a layer structure diagram of a convolutional neural network model in a method for identifying a recess stretching image of a pathological morphological feature of inflammatory bowel disease according to an embodiment of the present application.
FIG. 3 is a flow chart of the method of steps S20-S23 in a method for identifying a crypt stretch image of a pathomorphological feature of inflammatory bowel disease according to an embodiment of the present application.
Fig. 4 is a flowchart of a method of steps S210-S214 in a method of identifying a crypt stretch image of a pathomorphological feature of inflammatory bowel disease according to an embodiment of the present application.
Fig. 5 is a logic block diagram illustrating an example in step S23 in a method for identifying a crypt stretching image of a pathomorphological feature of inflammatory bowel disease according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all, of the embodiments of the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments herein without making any inventive effort, are intended to be within the scope of the present application.
It should be understood that when the terms "first," "second," and the like are used in the claims, specification, and drawings of this application, they are used merely for distinguishing between different objects and not for describing a particular sequential order. The terms "comprises" and "comprising," when used in the specification and claims of this application, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
The embodiment of the application discloses a method for identifying a crypt stretching image of pathological morphological characteristics of inflammatory bowel disease, referring to fig. 1, comprising the following steps:
s1: setting a tag on the historical microscopic image of the crypt, wherein the tag comprises a crypt stretching image and a crypt stretching image, and obtaining a data set.
In general, the pathological image occupies a large memory, and about 1G, and the computational effort required for directly inputting the pathological image into the subsequent convolutional neural network model is large, so that the case image needs to be sliced, and a plurality of microscopic images with the pixel size of 512 x 512 are obtained.
And (3) improving the definition of the microscopic image by using image definition processing technology such as histogram equalization, self-adaptive histogram equalization, contrast enhancement and the like on the microscopic image after slicing. Then, denoising treatment is carried out, and methods such as Gaussian filtering, median filtering and the like are adopted to reduce noise in the microscopic image so as to improve the image quality. Because the acquisition of the pathological image is difficult and the image data volume is small, the microscopic image is subjected to data enhancement, such as rotation, overturning, scaling and the like, so that a new image sample is generated, and the size of the data set is increased.
After the case images are processed, a large number of historical microscopic images of intestinal tracts are obtained. The historical microscopic images include crypt stretch images and non-crypt stretch images. Historical microscopic images may be acquired by medical devices such as endoscopes.
After the historical microscopic images of the crypts are obtained, denoising is carried out on the historical microscopic images of each crypt by using Gaussian filtering, after denoising is finished, a label is set for the historical microscopic images of each crypt according to the form of the crypt, the label is classified into 1 (not a stretching image of the crypt) or 0 (a stretching image of the crypt), namely, the label of the stretching image of the crypt is 0, and the labels of intestinal images of other conditions are all 1. And establishing a data set according to the historical microscopic image of the crypt and the label.
S2: and establishing a convolutional neural network model.
Referring to fig. 2, the convolutional neural network model includes: the first convolution layer, the first extraction layer, the second convolution layer and the full-connection layer are connected in sequence, and the first extraction layer and the second extraction layer comprise a residual error module and an attention mechanism module which are connected in sequence.
The input historical microscopic image of the hidden pit sequentially passes through the first convolution layer, the first extraction layer, the second convolution layer and the full connection layer and then outputs a classification result.
The convolutional neural network module is divided into a feature extraction part and a classification part, wherein the feature extraction part uses a residual module as a backbone network, and attention mechanism modules which pay attention to the row-column difference are added between the residual modules.
Referring to fig. 3, the establishment of the attention mechanism module includes steps S20-S23:
s20: the historical microscopic image of the crypt is passed through a first convolution layer to generate an input feature tensor, wherein the input feature tensor is composed of feature maps of a plurality of channels.
If the stretching phenomenon of the hidden pits occurs, and meanwhile, the bifurcation phenomenon of the hidden pits is also accompanied, the characteristic distribution of the lines and the columns of the hidden pit areas in the hidden pit image is changed due to the bifurcation connection phenomenon of the hidden pits, the hidden pit arrangement is approximately the whole arrangement in the normal hidden pit image, and in the hidden pit stretching image, the hidden pit arrangement is complex and changeable, irregular and large in line and column characteristic difference. Therefore, when the attention mechanism module is used, the feature map with larger line-row difference is more important for identifying the recess stretching, the feature map with larger line-row difference has larger weight when the deep features of the image are extracted later, the features after the feature passes through the second convolution layer are deep features, and the features before the first convolution layer are shallow features.
S21: and calculating the characteristic difference value of the characteristic graph to obtain a vector with the length being the input characteristic tensor channel number. Referring to fig. 4, steps S210-S214 are included:
s210: the input characteristic tensor is subjected to row convolution operation according to a convolution kernel of 1 multiplied by 3 to obtain row convolution of the characteristic diagram, and simultaneously subjected to column convolution operation according to a convolution kernel of 3 multiplied by 1 to obtain column convolution of the characteristic diagram.
Wherein the step size of the convolution is 1.
S211: and the line convolution is spliced according to the use of a global average pooling method and a global maximum pooling method to obtain a first feature matrix.
The global average pooling method is used for the row convolution according to the rows, and the average value of all elements in the row convolution is calculated and output.
The global maximum pooling method is used for the row convolution according to the row, and the maximum value of all elements in the row convolution is calculated and output.
S212: and (3) a global average pooling method and a global maximum pooling method are used for the column convolution according to the columns, and a second feature matrix is obtained through splicing.
The global average pooling method is used for column convolution by column to calculate the average value of all elements in the column convolution and output.
The global max pooling method is used for column convolution by column to calculate the maximum value of all elements in the column convolution and output.
S213: and calculating L2 norms of the first feature matrix and the second feature matrix to obtain a row-column difference matrix.
Specifically, for each feature map, two feature vectors with the same length are spliced together, e.g., two feature vectors with the same sizeThe feature vector of (2) becomes +.>Is a matrix of (a) in the matrix. After calculating L2 norms by the matrix A and the matrix B, an industry difference matrix D is obtained, and the calculation formula is as follows:
wherein,representing the values of the elements in the ith row and jth column of the row-column difference matrix in the figure, < >>Vector representing row i in matrix a, +.>Representing the vector of the j-th column in matrix B. />Representing the L2 norm.
S214: and (5) using a global average pooling method for the row-column difference matrix to obtain the difference characteristic value of the characteristic map.
S22: and normalizing the vector to obtain the weight of the feature map of each channel.
And (3) carrying out the operations of the steps S210-S214 on the feature graphs of each channel in the input feature tensor to obtain a vector with the length being the number of the channels of the input feature tensor, marking the vector as an attention vector, normalizing the attention vector by a softmax function, and obtaining the weight of each channel feature graph.
S23: and multiplying the input characteristic tensor with the weight of each channel characteristic graph to obtain a new characteristic tensor so as to obtain an attention mechanism module for representing the row-column difference.
Referring to fig. 5, for example, the input size isIs obtained by performing a row convolution operation through a convolution kernel with the size of 1×3, and consists of C feature graphs with the size of W×H×1>Is then for each of the feature tensorsAfter the feature graphs are subjected to step S210-step S214, feature difference vectors composed of C feature difference values are obtained, the feature difference vectors are normalized by a softmax function to obtain attention vectors with the length of C, and the input size is +.>Is multiplied by the attention vector to obtain an output feature tensor. The input of the attention module of row-column difference is a feature tensor, so that an output feature tensor, namely a new feature tensor, is obtained, and the input feature tensor and the output feature tensor are different in that the output feature tensor comprises the importance degree of each feature map.
The input historical microscopic image of the crypt firstly passes through the first convolution layer and then passes through a residual error module to extract features, the extracted features are distinguished through an attention mechanism injection module, the features which are possibly stretched by the crypt of inflammatory bowel disease are given larger weight, then pass through another residual error module and then pass through another attention mechanism module to distinguish the feature images of different channels, the feature images which are favorable for classification tasks are given larger weight, and then pass through the second convolution layer and then classify the input image through the full connection layer.
S3: setting a loss function and training a convolutional neural network model by using a data set to obtain an optimal model, wherein the expression of the loss function is as follows:，
wherein,representing a loss function->Indicate->Zhang Yinwo prediction of image, < >>Indicate->Zhang Yinwo image.
The training termination condition is that the training times reach a preset time threshold or the loss function value is less than a preset loss threshold. The preset frequency threshold is 100 times, and the preset frequency threshold can be adjusted by an implementer according to the application scene. The loss threshold is set to 0.001, which can be adjusted by the implementer according to the application scenario.
And after the one-time training is finished, adjusting the hyper-parameters to train the convolutional neural network model for multiple times, and selecting an optimal model according to the evaluation index prediction accuracy of the model.
In the training of the convolutional neural network model, a backward propagation algorithm is used for training the convolutional neural network according to the classification result and the error between the labels.
In this application, the back propagation algorithm used is: SGD (Stochastic Gradient Descent, random gradient descent) or ADAM (Adaptive Moment Estimation ).
S4: and responding to the newly shot microscopic image, and generating an image classification result of the crypt stretching according to the optimal model.
Classifying the crypt images by using the trained convolutional neural network model, inputting the acquired microscopic images, and outputting classification results, wherein the classification results are crypt stretching images or not.
While various embodiments of the present application have been shown and described herein, it will be obvious to those skilled in the art that such embodiments are provided by way of example only. Many modifications, changes, and substitutions will now occur to those skilled in the art without departing from the spirit and spirit of the application. It should be understood that various alternatives to the embodiments of the present application described herein may be employed in practicing the application.
The foregoing are all preferred embodiments of the present application, and are not intended to limit the scope of the present application in any way, therefore: all equivalent changes in structure, shape and principle of this application should be covered in the protection scope of this application.
Claims (3)
1. A method for identifying a crypt stretch image of a pathological morphological feature of inflammatory bowel disease, comprising the steps of:
setting a tag on a historical microscopic image of the crypt, wherein the tag comprises a crypt stretching image and a non-crypt stretching image, so as to obtain a data set;
building a convolutional neural network model;
setting a loss function and training the convolutional neural network model by using the data set to obtain an optimal model, wherein the expression of the loss function is as follows:，
wherein,representing a loss function->Indicate->Zhang Yinwo prediction of image, < >>Indicate->Zhang Yinwo true value of the image;
responding to the newly shot microscopic image, and generating an image classification result of the recess stretching according to the optimal model;
the convolutional neural network model includes:
the device comprises a first convolution layer, a first extraction layer, a second convolution layer and a full connection layer which are sequentially connected, wherein the first extraction layer and the second extraction layer comprise a residual error module and an attention mechanism module which are sequentially connected;
the input historical microscopic image sequentially passes through a first convolution layer, a first extraction layer, a second convolution layer and a full connection layer and then outputs a classification result;
the establishment of the attention mechanism module comprises the following steps:
passing a historical microscopic image of a recess through a first convolution layer to generate an input feature tensor, wherein the input feature tensor consists of feature graphs of a plurality of channels;
calculating the characteristic difference value of the characteristic graph to obtain a vector with the length being the input characteristic tensor channel number;
normalizing the vector to obtain the weight of the feature map of each channel;
multiplying the input characteristic tensor with the weight of each channel characteristic graph to obtain a new characteristic tensor, and obtaining an attention mechanism module for representing the row-column difference;
calculating a characteristic difference value of the characteristic graph, comprising the following steps:
carrying out row convolution operation on the input characteristic tensor according to the row convolution kernel of 1 multiplied by 3 to obtain the row convolution of the characteristic graph, and carrying out column convolution operation according to the column convolution kernel of 3 multiplied by 1 to obtain the column convolution of the characteristic graph;
the method comprises the steps of performing row convolution according to a row using global average pooling method and global maximum pooling method, and splicing to obtain a first feature matrix;
a global average pooling method and a global maximum pooling method are used for the column convolution according to the columns, and a second feature matrix is obtained through splicing;
calculating L2 norms of the first feature matrix and the second feature matrix to obtain a row-column difference matrix;
and using a global average pooling method for the row-column difference matrix to obtain a difference characteristic value of the characteristic map.
2. The method of claim 1, wherein in training the convolutional neural network model, the convolutional neural network is trained based on error back propagation between the classification result and a label.
3. The method for identifying the hidden-stretching image of the pathological morphological feature of the inflammatory bowel disease according to claim 1, wherein in the training of the convolutional neural network model, the training termination condition is that the training times reach a preset time threshold or the loss function value is less than a preset loss threshold.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311646973.8A CN117351218B (en) | 2023-12-04 | 2023-12-04 | Method for identifying inflammatory bowel disease pathological morphological feature crypt stretching image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311646973.8A CN117351218B (en) | 2023-12-04 | 2023-12-04 | Method for identifying inflammatory bowel disease pathological morphological feature crypt stretching image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117351218A CN117351218A (en) | 2024-01-05 |
| CN117351218B true CN117351218B (en) | 2024-02-27 |
Family
ID=89371483
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311646973.8A Active CN117351218B (en) | 2023-12-04 | 2023-12-04 | Method for identifying inflammatory bowel disease pathological morphological feature crypt stretching image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117351218B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102283265B1 (en) * | 2020-12-31 | 2021-07-29 | 가천대학교 산학협력단 | Method and apparatus for colorectal disease classification using scaled dilation in convolutional neural network |
| CN114898284A (en) * | 2022-04-08 | 2022-08-12 | 西北工业大学 | Crowd counting method based on feature pyramid local difference attention mechanism |
| WO2023137433A2 (en) * | 2022-01-14 | 2023-07-20 | Board Of Regents, The University Of Texas System | Methods and systems for determining leukemia or lymphoma levels using lymphoid images |
| CN116977748A (en) * | 2023-08-30 | 2023-10-31 | 南昌大学 | Colitis image classification method and system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220164569A1 (en) * | 2020-11-26 | 2022-05-26 | POSTECH Research and Business Development Foundation | Action recognition method and apparatus based on spatio-temporal self-attention |
-
2023
- 2023-12-04 CN CN202311646973.8A patent/CN117351218B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102283265B1 (en) * | 2020-12-31 | 2021-07-29 | 가천대학교 산학협력단 | Method and apparatus for colorectal disease classification using scaled dilation in convolutional neural network |
| WO2023137433A2 (en) * | 2022-01-14 | 2023-07-20 | Board Of Regents, The University Of Texas System | Methods and systems for determining leukemia or lymphoma levels using lymphoid images |
| CN114898284A (en) * | 2022-04-08 | 2022-08-12 | 西北工业大学 | Crowd counting method based on feature pyramid local difference attention mechanism |
| CN116977748A (en) * | 2023-08-30 | 2023-10-31 | 南昌大学 | Colitis image classification method and system |
Non-Patent Citations (3)
| Title |
|---|
| Anca Loredana Udristoiu等.Deep Learning Algorithm for the Confirmation of Mucosal Healing in Crohn's Disease, Based on Confocal Laser Endomicroscopy Images.J Gastrointestin Liver Dis.2021,第30卷(第1期),59-65. * |
| 基于深度学习人工智能在结肠镜检查中应用研究;阿依木克地斯・亚力孔;庄惠军;蔡世伦;牛雪静;谭伟敏;颜波;姚礼庆;周平红;钟芸诗;;中国实用外科杂志(第03期);118-122 * |
| 焦海宁等.深度学习与智慧交通.冶金工业出版社,2022,29. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117351218A (en) | 2024-01-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ali et al. | Structural crack detection using deep convolutional neural networks | |
| CN108647585B (en) | Traffic identifier detection method based on multi-scale circulation attention network | |
| CN108133188B (en) | Behavior identification method based on motion history image and convolutional neural network | |
| CN110852316B (en) | Image tampering detection and positioning method adopting convolution network with dense structure | |
| CN109345508B (en) | Bone age evaluation method based on two-stage neural network | |
| CN112967243A (en) | Deep learning chip packaging crack defect detection method based on YOLO | |
| CN108154519A (en) | Dividing method, device and the storage medium of eye fundus image medium vessels | |
| CN113642634A (en) | Shadow detection method based on mixed attention | |
| US11435719B2 (en) | System and method for identifying manufacturing defects | |
| CN111582126A (en) | Pedestrian re-identification method based on multi-scale pedestrian contour segmentation fusion | |
| CN113449811B (en) | Low-illumination target detection method based on MS-WSDA | |
| CN112149616A (en) | Figure interaction behavior recognition method based on dynamic information | |
| CN111539456B (en) | Target identification method and device | |
| CN113313678A (en) | Automatic sperm morphology analysis method based on multi-scale feature fusion | |
| CN115994900A (en) | Unsupervised defect detection method and system based on transfer learning and storage medium | |
| CN116258632A (en) | Text image super-resolution reconstruction method based on text assistance | |
| CN113762151B (en) | Fault data processing method, system and fault prediction method | |
| CN108549899B (en) | Image identification method and device | |
| CN113159071B (en) | Cross-modal image-text association anomaly detection method | |
| CN111401209B (en) | Action recognition method based on deep learning | |
| CN111209886B (en) | Rapid pedestrian re-identification method based on deep neural network | |
| CN117830889A (en) | Video saliency detection method, system and storage medium based on optical flow graph feature fusion mechanism | |
| CN117351218B (en) | Method for identifying inflammatory bowel disease pathological morphological feature crypt stretching image | |
| CN113033371A (en) | CSP model-based multi-level feature fusion pedestrian detection method | |
| CN116977859A (en) | Weak supervision target detection method based on multi-scale image cutting and instance difficulty |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |