CN115205599A - Multi-age-range child chest radiography pneumonia classification system based on domain generalization model - Google Patents
Multi-age-range child chest radiography pneumonia classification system based on domain generalization model Download PDFInfo
- Publication number
- CN115205599A CN115205599A CN202210879158.5A CN202210879158A CN115205599A CN 115205599 A CN115205599 A CN 115205599A CN 202210879158 A CN202210879158 A CN 202210879158A CN 115205599 A CN115205599 A CN 115205599A
- Authority
- CN
- China
- Prior art keywords
- model
- filter
- classification
- image
- layer
- 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.)
- Granted
Links
- 206010035664 Pneumonia Diseases 0.000 title claims abstract description 38
- 238000002601 radiography Methods 0.000 title claims abstract description 25
- 238000013145 classification model Methods 0.000 claims abstract description 29
- 238000000034 method Methods 0.000 claims abstract description 20
- 238000004590 computer program Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 39
- 210000004072 lung Anatomy 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 13
- 238000012795 verification Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 10
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000011156 evaluation Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 2
- 238000003745 diagnosis Methods 0.000 description 8
- 238000005070 sampling Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000011218 segmentation Effects 0.000 description 5
- 239000013598 vector Substances 0.000 description 4
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 201000001178 Bacterial Pneumonia Diseases 0.000 description 2
- 201000004813 Bronchopneumonia Diseases 0.000 description 2
- 206010035737 Pneumonia viral Diseases 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000002538 fungal effect Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000002604 ultrasonography Methods 0.000 description 2
- 208000009421 viral pneumonia Diseases 0.000 description 2
- 206010003805 Autism Diseases 0.000 description 1
- 208000020706 Autistic disease Diseases 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000002599 functional magnetic resonance imaging Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 208000015181 infectious disease Diseases 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002610 neuroimaging Methods 0.000 description 1
- 230000001717 pathogenic effect Effects 0.000 description 1
- 230000002685 pulmonary effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
Images
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
- G06V10/70—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
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/50—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment specially adapted for specific body parts; specially adapted for specific clinical applications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
- A61B6/5211—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data
-
- 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
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- 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
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
- G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
-
- 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
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B2503/00—Evaluating a particular growth phase or type of persons or animals
- A61B2503/06—Children, e.g. for attention deficit diagnosis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- High Energy & Nuclear Physics (AREA)
- Veterinary Medicine (AREA)
- Pathology (AREA)
- Optics & Photonics (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Heart & Thoracic Surgery (AREA)
- Radiology & Medical Imaging (AREA)
- Public Health (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Dentistry (AREA)
- Probability & Statistics with Applications (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention discloses a multi-age children chest radiography classification system based on a domain generalization model, which comprises a computer memory, a computer processor and a computer program which is stored in the computer memory and can be executed on the computer processor, wherein the trained domain generalization classification model is stored in the computer memory; the domain generalization classification model adopts an improved F-conv network model and comprises a weight layer, a filter layer and a sharing layer; after the image to be detected is input into the domain generalization classification model, point multiplication weighting is carried out on the image to be detected and a corresponding filter bank in the filter layer through a weight group, and all results are added to obtain the output of the filter layer; and the output of the filter layer is input to the shared layer, and the classification result of the model is finally obtained through the full-connection layer through multiple 2D convolutions. By using the method and the system, the children of multiple ages can realize high-performance classification of pneumonia on the same model.
Description
Technical Field
The invention relates to the field of medical artificial intelligence, in particular to a multi-age-range child chest radiography image pneumonia classification system based on a domain generalization model.
Background
The children pneumonia is a disease of the first three infection rates of children, and at present, many researchers analyze and diagnose children pneumonia images through an intelligent algorithm, wherein X-ray chest radiography images are main objects of research.
For example, chinese patent publication No. CN114566276A discloses a training method and a training device for a lung color Doppler ultrasound-based children pneumonia auxiliary diagnosis model, wherein the training method includes: acquiring a diagnosis and treatment information database comprising color Doppler ultrasound images of a plurality of children pneumonia patients; acquiring a plurality of sample sets according to a database; training the training set in each sample set to obtain a training model; testing the training model according to the test set of each sample set to obtain the accuracy of the training model; determining a diagnostic model based on the accuracy of the plurality of training models; each sample set comprises a training set and a corresponding test set.
Chinese patent publication No. CN113241184A discloses a children pneumonia auxiliary diagnosis model and a training method thereof, wherein the training method comprises: acquiring a medical image and a corresponding medical diagnosis sentence of a child pneumonia patient, wherein the medical image is used as a training image set, and the medical diagnosis sentence is used as a training sentence; extracting image depth characteristic vectors from the image training set data through a CNN neural network to obtain a depth characteristic map set, and performing word vector training on the training sentences through a word2vec model to obtain a depth characteristic vector word set; and performing feature fusion on the depth feature map set and the depth feature vector word set, and then training through an LSTM neural network to obtain a trained auxiliary diagnosis model of the child pneumonia.
The children pneumonia is divided into viral pneumonia, bronchopneumonia, bacterial pneumonia, fungal pneumonia and the like according to pathogenic reasons, and the cure rate can be effectively improved by adopting a special treatment scheme aiming at subdivided diseases. However, since there is a certain similarity in the manifestations of various pneumonia and other lung related diseases, and since children are in the developmental stage, pneumonia varies in its distribution and onset of symptoms at different ages. Therefore, the X-ray chest radiography images of the children pneumonia have the same and different characteristics in different age groups, and the method brings challenges to the uniform classification model of the whole age group.
Aiming at the problem, many researchers adopt a domain generalization method to classify similar data sets, such as a domain generalization algorithm MatchDG, a multi-view space-time graph convolutional network (MSTGCN) with a domain generalization function, a domain generalization meta-learning model, and the like. The domain generalized meta-learning model can be used for diagnosis and classification of multi-site functional magnetic resonance imaging, the network is verified on a public Autism Brain Imaging Data Exchange (ABIDE) data set, and the diagnosis and classification performance of unseen parts and visible parts is remarkably improved. But the method focuses on extracting domain-invariant information across sources to generalize the target domain, and ignores useful domain-specific information closely related to a single domain label.
Disclosure of Invention
The invention provides a chest image pneumonia classification system for children in multiple ages based on a domain generalization model, which can realize high-performance pneumonia classification for children in multiple ages on the same model.
A domain generalization model-based multi-age children chest radiography classification system comprising a computer memory having stored therein a trained domain generalization classification model, a computer processor, and a computer program stored in and executable on said computer memory; the domain generalization classification model adopts an improved F-conv network model and comprises a weight layer, a filter layer and a sharing layer; the training process of the domain generalization classification model is as follows:
(1) Collecting X-ray chest radiography images of children of multiple age groups, and preprocessing image data;
(2) Extracting lung regions in the X-ray chest radiography images of the children by using a trained ResUnet model;
(3) Based on the characteristics of lung regions, combining clinical information, clustering the images by using an unsupervised Kmeans + + model, and determining the number n of groups and the center C = { C of each group according to the number of the clustered classes 1 ,C 2 ,…,C n }；
(4) Randomly dividing all image data into a training set, a verification set and a test set;
(5) Sending the training set into a constructed domain generalization classification model for training, evaluating the performance of the domain generalization classification model by using a verification set, adjusting the hyper-parameters of the model according to the evaluation effect, and finally obtaining the domain generalization classification model with the performance reaching the standard through repeated training and verification;
the computer processor when executing the computer program implements the steps of:
preprocessing an X-ray chest radiography image to be classified, inputting the preprocessed X-ray chest radiography image into a trained domain generalization classification model, and calculating a characteristic group F and the center C = { C of each group of the input image in a weight group 1 ,C 2 ,…,C n Distance D = { D = } distance D = { D i ＝||F-C i || 2 ,i∈[1,n]Then normalize the distance D to [0,1 ] using Softmax]In order to obtain the weight of the image in different groups
Performing dot product weighting on the weight groups and corresponding filter groups in the filter layer, and adding all results to obtain the output of the filter layer; and the output of the filter layer is input to the shared layer, and the classification result of the model is finally obtained through the full-connection layer through multiple 2D convolutions.
Further, in the step (1), the preprocessing includes performing data normalization on the image, and filtering noise by using a gaussian filter.
The specific process of the step (3) is as follows:
extracting size and shape characteristics according to the segmented lung region, combining the characteristics of age, sex, height and weight in clinical information to form a characteristic group F of each image, and clustering the images based on an unsupervised Kmeans + + model; determining the number n of groups and the center C = { C of each group according to the clustering result 1 ,C 2 ,…,C n }。
In the step (4), all image data are randomly divided into a training set, a verification set and a test set according to the ratio of 7.
And (5) training the domain generalization classification model by adopting a supervision training method.
At the Filter level, the model defines n +1 Filter banks Filter = { Filter 0 ,Filter 1 ,…,Filter n Filter of Filter bank with 0 0 Is a shared Filter bank used for learning common characteristics on all age data sets, and the Filter bank is numbered from 1 to n 1 ,Filter 2 ,…,Filter n Are used to learn the characteristics of different groups, respectively; each filter is a standard convolution filter, which contains 4-dimensional data corresponding to the number of input channels, the number of output channels, the number of convolution kernels, and the width of the kernels.
The output of the filter layer is:
wherein, the first and the second end of the pipe are connected with each other,
shown is a feature map, the output of the filter layer and the input of the share layer, W, fitter, x representing the set of weights, the filter bank and the input image, respectively.
The domain generalization classification model also constructs a Loss function Loss which is used for the back propagation learning of the model;
the Loss function Loss comprises cross entropy based on pneumonia labels, a Loss function based on similarity in a class and cross entropy based on filter weight set; wherein the cross entropy of the pneumonia label is used for learning the overall classification performance, the loss function based on the similarity in the classes is used for enabling the classification of data of each age group to be closer, and the cross entropy based on the filter weight set is used for learning common features of the age groups.
Compared with the prior art, the invention has the following beneficial effects:
the invention innovatively provides a domain generalization method for multi-classification of pneumonia X-ray chest images of children of multiple age groups. Pulmonary features and clinical features segmented based on ResUnnet can form a feature group of images, unsupervised clustering is performed through a Kmeans + + method, grouping according to age is avoided, grouping which is most consistent with actual conditions is achieved, then a domain-generalized improved F-conv model is established, different filters are weighted according to the distance between the feature group of the images and the center of each group, and therefore most core general features, namely features of different pneumonia, and differences of pneumonia expressions of different age groups can be learned from X-ray chest radiographs, and high-performance pneumonia classification of children of multiple ages on the same model can be achieved.
Drawings
FIG. 1 is a classification flow chart of a multi-age-segment chest radiography pneumonia classification system based on a domain generalization model according to the present invention;
FIG. 2 is a diagram of the ResUnet model for lung segmentation according to the present invention;
FIG. 3 is a view of the structure of the Kmeans + + model for image clustering according to the present invention;
FIG. 4 is a block diagram of the domain-generalized classification model according to the present invention;
FIG. 5 is a diagram illustrating a weighting layer in the domain-generalized classification model according to the present invention
FIG. 6 is a diagram illustrating a filter layer in a domain generalized classification model according to the present invention;
FIG. 7 is a diagram illustrating a sharing layer in the domain-generalized classification model according to the present invention.
Detailed Description
The invention will be described in further detail below with reference to the drawings and examples, which are intended to facilitate the understanding of the invention without limiting it in any way.
A multi-age-segment child chest radiography classification system based on a domain generalization model comprises a computer memory, a computer processor and a computer program which is stored in the computer memory and can be executed on the computer processor, wherein the trained domain generalization classification model is stored in the computer memory.
As shown in FIG. 1, the process of classifying pneumonia in chest images of children of multiple ages by using the system of the present invention is as follows:
1. image pre-processing
Collecting X-ray chest radiography image data of children from multiple age groups, including four types of viral pneumonia, bronchopneumonia, bacterial pneumonia, fungal pneumonia and non-pneumonia. In order to reduce the influence of the difference between the images on the model, the data normalization operation is firstly carried out on the images, and for the X-ray film with the input size of (X, y), the mean value of each channel is firstly obtained
And standard deviation Std, then for different channels, the mean value is subtracted separately
After dividing by the standard deviation Std, the final data will be distributed in [ -0.5]In between. And finally, filtering noise by adopting a Gaussian filter.
2. Lung segmentation
For the pre-processed images, segmentation of the lung regions was performed using a trained ResUnet, which achieved a Dice score of 98% on the test set. The structure of ResUnet model is shown in FIG. 2, resUnet is that Res part is added on the basis of Unet network, the input image firstly passes through four down-sampling parts to respectively obtain the characteristics F of the image in different dimensions down_1 ,F down_2 ,F down_3 ,F down_4 After each down-sampling, the down-sampled feature F down_i With features F not sampled down_i-1 F down_i-1 The merging is performed to provide richer features for the next layer down-sampling while avoiding gradient vanishing and overfitting. In the last down-sampling layer, two convolutions are performed first, and then four up-sampling F is performed up_1 ,F up_2 ,F up_3 ,F up_4 Feature F obtained after the end of each upsampling up_i Same as feature F not subjected to upsampling up_i-1 Are combined and then compared with the corresponding upsampled image feature F of the same size down_i And splicing is carried out to be used as the input of the next up-sampling layer. Finally, the image is up-sampled for four times to obtain a segmentation result which is consistent with the size of the original image.
3. Clustering groups
The development conditions of children at different ages are different, but the development conditions are not in direct proportion to the ages, and children can develop faster at the low age and slowly at the high age, so that the children cannot be directly grouped at the ages. According to the method, the characteristics of the lung such as size and shape are obtained through the result of lung segmentation, the characteristics of age, sex, height, weight and the like in clinical information are combined to form a characteristic group F of each image, and then each image is clustered based on an unsupervised Kmeans + + model. The model structure of the Kmeans + + model is shown in FIG. 3, and the clustering method can realize clustering of indefinite groups and complete classification according to a set threshold. Determining scores according to the clustering resultsThe number of groups n and the center of each group C = { C 1 ,C 2 ,…,C n }。
4. Data packet
For the full volume data set, according to the random principle, 70% of the data set is used as a training set, 10% of the data set is used as a verification set, and 20% of the data set is used as a test set.
5. Construction of domain generalization classification model
The invention selects an improved F-conv model to realize the domain generalization classification learning, the structure of the domain generalization classification model is shown in figure 4, and the specific steps of the model establishment are as follows:
(1) As shown in fig. 5, first, in the weighting layer, for an input image, the feature group F and the center C = { C } of each group are calculated 1 ,C 2 ,…,C n Distance D = { D = } i ＝||F-C i || 2 ,i∈[1,n]H, then normalize the distance D to [0,1 ] using Softmax]In order to obtain the weight of this image in the different groups
(2) As shown in fig. 6, at the Filter level, the model defines n +1 Filter banks Filter = { Filter 0 ,Filter 1 ,…,Filter n Filter of Filter bank with 0 0 Is a shared Filter bank used for learning common characteristics on all age data sets, and the Filter bank (Filter) is numbered from 1 to n 1 ,Filter 2 ,…,Filter n Are used to learn the characteristics of different packets, respectively. Each filter is a standard convolution filter that contains 4-dimensional data corresponding to the number of input channels, the number of output channels, the number of convolution kernels, and the width of the kernels.
(3) Weighting the corresponding filter bank by 18 weight groups, wherein the weight of the shared filter bank is 1, and then summing the weights as the input of the shared layer, the formula is as follows:
wherein the content of the first and second substances,
a feature map is shown, which is used as input to the shared layer, and W, fiter, x represent the weight set, filter bank, and input image, respectively.
As shown in fig. 7, in the shared layer, mainly some 2D convolution modules finally pass through the linear layer to obtain the final classification result as output.
(4) The Loss function Loss of the model is calculated. In the F-conv model training process, the difference between the predicted pneumonia types and the real pneumonia labels of the model in all data sets needs to be calculated, and the specific formula is as follows:
L CrossDomain ＝∑ types ∑ x CE(p x ,y x )
wherein, CE represents adopts cross entropy algorithm, p x ,y x The prediction type of the model for the image x and the actual pneumonia category label corresponding to the image x are respectively represented, and types represents the classification type of pneumonia.
6. Model training and classification testing
When the classification model is trained, the training set is sent into a domain generalization classification model; the verification set adjusts the hyper-parameters of the model, an optimizer is used for updating the parameters, the network is optimized, the learning rate is automatically adjusted, the trained classification network is obtained, and the parameter updating process is as follows:
wherein, theta Y ,
Respectively representing the network weight during training and the updated network weight, and delta represents the size of the update.
And after the model obtains better performance on the training set and the verification set, classifying the model in the test set, and evaluating the generalization capability of the model through the classification performance of the model on the test set.
7. Evaluation phase
On the test set, the effect of the model was evaluated: for the evaluation of the multi-classification task, five indexes need to be calculated: average classification Accuracy (ACC), average classification accuracy (Precision), average classification Recall (Recall), average classification harmonic mean F1 value (F1 score), and area under various ROC curves AUC. For each class, the model correctly classifies images belonging to this class into TP (True Positive), not FP (fast Positive), not TN (True Negative), and incorrectly classifies images not belonging to this class into FN (False Negative). For each class, the first four index formulas used for evaluation are as follows:
the area AUC under the ROC curve is an area under the ROC (Receiver Operating characterization) curve having the false positive rate (FP _ rate) and the false negative rate (TP _ rate) as axes.
The final model will be jointly evaluated according to the average accuracy, precision, recall (Recall), F1 value, AUC for the four categories.
The technical solutions and advantages of the present invention have been described in detail with reference to the above embodiments, it should be understood that the above embodiments are only specific examples of the present invention and should not be construed as limiting the present invention, and any modifications, additions and equivalents made within the scope of the principles of the present invention should be included in the scope of the present invention.
Claims (8)
1. A system for classifying chest radiography (pneumonia) in a multi-age child based on a domain generalization model, comprising a computer memory, a computer processor and a computer program stored in said computer memory and executable on said computer processor, wherein said computer memory stores a trained domain generalization model; the domain generalization classification model adopts an improved F-conv network model and comprises a weight layer, a filter layer and a sharing layer; the training process of the domain generalization classification model is as follows:
(1) Collecting X-ray chest radiography images of children of multiple age groups, and preprocessing image data;
(2) Extracting lung regions in the X-ray chest radiography images of the children by using a trained ResUnet model;
(3) Based on the characteristics of the lung region, combining clinical information, clustering the images by using an unsupervised Kmeans + + model, and determining the number n of groups and the center C = { C } of each group according to the number of the clustered categories 1 ,C 2 ,…,C n }；
(4) Randomly dividing all image data into a training set, a verification set and a test set;
(5) Sending the training set into a constructed domain generalization classification model for training, evaluating the performance of the domain generalization classification model by using a verification set, adjusting the hyper-parameters of the model according to the evaluation effect, and finally obtaining the domain generalization classification model with the performance reaching the standard through repeated training and verification;
the computer processor when executing the computer program implements the steps of:
preprocessing an X-ray chest radiography image to be classified, inputting the preprocessed X-ray chest radiography image into a trained domain generalization classification model, and calculating a characteristic group F and the center C = { C of each group of the input image in a weight group 1 ,C 2 ,…,C n Distance D = { D = } i ＝||F-C i || 2 ,i∈[1,n]Then normalize the distance D to [0,1 ] using Softmax]In order to obtain the weight of this image in the different groups
Performing point multiplication weighting on the weighted sum and a corresponding filter bank in the filter layer, and adding all results to obtain the output of the filter layer; and the output of the filter layer is input to the shared layer, and the classification result of the model is finally obtained through the full-connection layer through multiple 2D convolutions.
2. The system for classifying chest radiography image pneumonia in multi-age children according to claim 1 wherein in step (1), said preprocessing comprises normalizing the image data and filtering out noise by using gaussian filter.
3. The system for classifying chest radiography of children of multiple age groups based on domain generalization model according to claim 1, wherein the detailed process of step (3) is as follows:
extracting size and shape characteristics according to the segmented lung region, combining the characteristics of age, sex, height and weight in clinical information to form a characteristic group F of each image, and clustering the images based on an unsupervised Kmeans + + model; determining the number n of groups and the center C = { C of each group according to the clustering result 1 ,C 2 ,…,C n }。
4. The system for classifying chest radiography of multi-age children according to claim 1, wherein in step (4), all image data are randomly divided into a training set, a verification set and a test set according to the ratio of 7.
5. The system for classifying chest radiography of children of multiple age groups according to claim 1, wherein in step (5), the domain generalization classification model is trained by supervised training.
6. The system for classifying chest radiography image pneumonia in children of multiple age groups based on domain generalization model according to claim 1, wherein at Filter level, the model defines n +1 Filter banks Filter = { Filter 0 ,Filter 1 ,…,Filter n Filter of Filter bank with 0 0 Is a shared Filter bank used for learning common characteristics on all age data sets, and the Filter bank is numbered from 1 to n 1 ,Filter 2 ,…,Filter n Are used to learn the characteristics of different groups, respectively; each filter is a standard convolution filter that contains 4-dimensional data corresponding to the number of input channels, the number of output channels, the number of convolution kernels, and the width of the kernels.
7. The system for classifying chest radiography image pneumonia in multi-age children according to claim 6 based on domain generalization model, wherein the output of the filter layer is:
wherein the content of the first and second substances,
shown is a feature map, as the output of the filter layer and the input of the shared layer, W, filter, x representing the set of weights, the filter bank, and the input image, respectively.
8. The system for classifying chest radiography of children of multiple age groups according to claim 1, wherein said domain generalization model further constructs a Loss function Loss for back propagation learning of the model;
the Loss function Loss comprises cross entropy based on pneumonia labels, a Loss function based on similarity in the similarity, and cross entropy based on filter weight set; the cross entropy of the pneumonia label is used for learning the overall classification performance, the loss function based on the similarity in the class is used for enabling the classification of data of each age group to be closer, and the cross entropy based on the filter weight group is used for learning common features of multiple age groups.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210879158.5A CN115205599B (en) | 2022-07-25 | 2022-07-25 | Multi-age-segment child chest image pneumonia classification system based on domain generalization model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210879158.5A CN115205599B (en) | 2022-07-25 | 2022-07-25 | Multi-age-segment child chest image pneumonia classification system based on domain generalization model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115205599A true CN115205599A (en) | 2022-10-18 |
| CN115205599B CN115205599B (en) | 2023-02-10 |
Family
ID=83583144
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210879158.5A Active CN115205599B (en) | 2022-07-25 | 2022-07-25 | Multi-age-segment child chest image pneumonia classification system based on domain generalization model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115205599B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116363438A (en) * | 2023-04-14 | 2023-06-30 | 浙江大学 | Child encephalitis classification system based on multiple modes |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108171232A (en) * | 2017-11-15 | 2018-06-15 | 中山大学 | The sorting technique of bacillary and viral children Streptococcus based on deep learning algorithm |
| CN113327258A (en) * | 2021-07-15 | 2021-08-31 | 重庆邮电大学 | Lung CT image identification method based on deep learning |
| US20220036564A1 (en) * | 2020-08-03 | 2022-02-03 | Korea Advanced Institute Of Science And Technology | Method of classifying lesion of chest x-ray radiograph based on data normalization and local patch and apparatus thereof |
| CN114359629A (en) * | 2021-12-20 | 2022-04-15 | 桂林理工大学 | Pneumonia X chest radiography classification and identification method based on deep migration learning |
| CN114511737A (en) * | 2022-01-24 | 2022-05-17 | 北京建筑大学 | Training method of image recognition domain generalization model |
| CN114694150A (en) * | 2022-05-31 | 2022-07-01 | 成都考拉悠然科技有限公司 | Method and system for improving generalization capability of digital image classification model |
| CN114708255A (en) * | 2022-04-29 | 2022-07-05 | 浙江大学 | Multi-center children X-ray chest image lung segmentation method based on TransUNet model |
-
2022
- 2022-07-25 CN CN202210879158.5A patent/CN115205599B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108171232A (en) * | 2017-11-15 | 2018-06-15 | 中山大学 | The sorting technique of bacillary and viral children Streptococcus based on deep learning algorithm |
| US20220036564A1 (en) * | 2020-08-03 | 2022-02-03 | Korea Advanced Institute Of Science And Technology | Method of classifying lesion of chest x-ray radiograph based on data normalization and local patch and apparatus thereof |
| CN113327258A (en) * | 2021-07-15 | 2021-08-31 | 重庆邮电大学 | Lung CT image identification method based on deep learning |
| CN114359629A (en) * | 2021-12-20 | 2022-04-15 | 桂林理工大学 | Pneumonia X chest radiography classification and identification method based on deep migration learning |
| CN114511737A (en) * | 2022-01-24 | 2022-05-17 | 北京建筑大学 | Training method of image recognition domain generalization model |
| CN114708255A (en) * | 2022-04-29 | 2022-07-05 | 浙江大学 | Multi-center children X-ray chest image lung segmentation method based on TransUNet model |
| CN114694150A (en) * | 2022-05-31 | 2022-07-01 | 成都考拉悠然科技有限公司 | Method and system for improving generalization capability of digital image classification model |
Non-Patent Citations (2)
| Title |
|---|
| KAREN SANCHEZ ET AL.: "CX-DaGAN: Domain Adaptation for Pneumonia Diagnosis on a Small Chest X-ray Dataset", 《IEEE TRANSACTIONS ON MEDICAL IMAGING》 * |
| 徐海 等: "视觉域泛化技术及研究进展", 《广州大学学报(自然科学版)》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116363438A (en) * | 2023-04-14 | 2023-06-30 | 浙江大学 | Child encephalitis classification system based on multiple modes |
| CN116363438B (en) * | 2023-04-14 | 2023-09-15 | 浙江大学 | Child encephalitis classification system based on multiple modes |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115205599B (en) | 2023-02-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tseng et al. | DNetUnet: a semi-supervised CNN of medical image segmentation for super-computing AI service | |
| CN113693563B (en) | Brain function network classification method based on hypergraph attention network | |
| CN110889852A (en) | Liver segmentation method based on residual error-attention deep neural network | |
| CN110889853A (en) | Tumor segmentation method based on residual error-attention deep neural network | |
| CN110570432A (en) | CT image liver tumor segmentation method based on deep learning | |
| CN111009321A (en) | Application method of machine learning classification model in juvenile autism auxiliary diagnosis | |
| CN110084790B (en) | Bionic mode identification improvement method for judging pneumonitis in imaging | |
| CN112766376A (en) | Multi-label eye fundus image identification method based on GACNN | |
| CN112085742B (en) | NAFLD ultrasonic video diagnosis method based on context attention | |
| CN115205599B (en) | Multi-age-segment child chest image pneumonia classification system based on domain generalization model | |
| CN108596228B (en) | Brain function magnetic resonance image classification method based on unsupervised fuzzy system | |
| CN111524140B (en) | Medical image semantic segmentation method based on CNN and random forest method | |
| Shrivastava et al. | Control or autism-classification using convolutional neural networks on functional MRI | |
| CN115496720A (en) | Gastrointestinal cancer pathological image segmentation method based on ViT mechanism model and related equipment | |
| CN112863664A (en) | Alzheimer disease classification method based on multi-modal hypergraph convolutional neural network | |
| CN115474939A (en) | Autism spectrum disorder recognition model based on deep expansion neural network | |
| CN115880523A (en) | Image classification model, model training method and application thereof | |
| CN110032973A (en) | A kind of unsupervised helminth classification method and system based on artificial intelligence | |
| CN116030063B (en) | Classification diagnosis system, method, electronic device and medium for MRI image | |
| CN114821176B (en) | Viral encephalitis classification system for MR (magnetic resonance) images of children brain | |
| Riyadi et al. | Covid-19 Detection Based-On CT-Scan Images Using Inception Deep Learning | |
| CN115035346A (en) | Classification method for Alzheimer disease based on cooperative learning method enhancement | |
| Swarup et al. | Biologically inspired CNN network for brain tumor abnormalities detection and features extraction from MRI images | |
| Ahmed et al. | 3D U-Net: fully convolutional neural network for automatic brain tumor segmentation | |
| CN116012903A (en) | Automatic labeling method and system for facial expressions |
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 |