CN113989551B - Alzheimer's disease classification method based on improved ResNet network - Google Patents
Alzheimer's disease classification method based on improved ResNet network Download PDFInfo
- Publication number
- CN113989551B CN113989551B CN202111237755.XA CN202111237755A CN113989551B CN 113989551 B CN113989551 B CN 113989551B CN 202111237755 A CN202111237755 A CN 202111237755A CN 113989551 B CN113989551 B CN 113989551B
- Authority
- CN
- China
- Prior art keywords
- network
- training
- slices
- resnet
- alzheimer
- 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
- 208000024827 Alzheimer disease Diseases 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 43
- 230000006870 function Effects 0.000 claims abstract description 19
- 238000011176 pooling Methods 0.000 claims abstract description 17
- 238000012360 testing method Methods 0.000 claims abstract description 17
- 238000001514 detection method Methods 0.000 claims abstract description 13
- 210000004556 brain Anatomy 0.000 claims description 19
- 238000002595 magnetic resonance imaging Methods 0.000 claims description 14
- 238000000605 extraction Methods 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 7
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 5
- 230000000877 morphologic effect Effects 0.000 claims description 5
- 210000004884 grey matter Anatomy 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 4
- 238000005481 NMR spectroscopy Methods 0.000 claims description 3
- IXKSXJFAGXLQOQ-XISFHERQSA-N WHWLQLKPGQPMY Chemical compound C([C@@H](C(=O)N[C@@H](CC=1C2=CC=CC=C2NC=1)C(=O)N[C@@H](CC(C)C)C(=O)N[C@@H](CCC(N)=O)C(=O)N[C@@H](CC(C)C)C(=O)N1CCC[C@H]1C(=O)NCC(=O)N[C@@H](CCC(N)=O)C(=O)N[C@@H](CC(O)=O)C(=O)N1CCC[C@H]1C(=O)N[C@@H](CCSC)C(=O)N[C@@H](CC=1C=CC(O)=CC=1)C(O)=O)NC(=O)[C@@H](N)CC=1C2=CC=CC=C2NC=1)C1=CNC=N1 IXKSXJFAGXLQOQ-XISFHERQSA-N 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 208000010877 cognitive disease Diseases 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 2
- 238000013145 classification model Methods 0.000 claims description 2
- 208000027061 mild cognitive impairment Diseases 0.000 claims description 2
- 210000005013 brain tissue Anatomy 0.000 claims 1
- HGBLNBBNRORJKI-WCABBAIRSA-N cyclacillin Chemical compound N([C@H]1[C@H]2SC([C@@H](N2C1=O)C(O)=O)(C)C)C(=O)C1(N)CCCCC1 HGBLNBBNRORJKI-WCABBAIRSA-N 0.000 claims 1
- 238000011156 evaluation Methods 0.000 claims 1
- 238000004422 calculation algorithm Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 3
- 230000003213 activating effect Effects 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 230000000926 neurological effect Effects 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 241000282412 Homo Species 0.000 description 1
- 210000003484 anatomy Anatomy 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 231100000871 behavioral problem Toxicity 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000001175 cerebrospinal fluid Anatomy 0.000 description 1
- 230000003930 cognitive ability Effects 0.000 description 1
- 238000004195 computer-aided diagnosis Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 208000011977 language disease Diseases 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000013421 nuclear magnetic resonance imaging Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 210000004872 soft tissue Anatomy 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
- 210000004885 white matter Anatomy 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/047—Probabilistic or stochastic networks
-
- 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
- G06N3/048—Activation functions
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses a classification method of Alzheimer's disease based on an improved ResNet network, which mainly solves the problems of data imbalance and feature loss in pooling in the existing classification detection algorithm process. The scheme is as follows: (1) Improving ResNet network, adding an effective channel attention ECA module, and building ResNet classification detection network for Alzheimer disease; (2) To reduce information loss during pooling operations, the maximum pooling Maxpool is modified to softpool; (3) Replacing the cross entropy loss function by using the focus loss function to solve the problems of data imbalance and difficult model learning; (4) Selecting 2D-form slices of each MRI 3D data at three Axial positions Axial-transversal, sagittal-sagittal, coronal-coronal; (5) training and testing the improved ResNet network. The invention improves the classification detection accuracy of the Alzheimer disease, and compared with the existing model, the model has better performance.
Description
Technical Field
The invention belongs to the field of intersection of brain images and computer science, relates to the technical field of image classification, and particularly relates to an Alzheimer disease classification method based on an improved ResNet network.
Background
As the population ages, alzheimer's Disease (AD) is the most common cognitive disorder in the elderly and is one of the greatest health threats facing humans in the 21 st century. Early symptoms include decline of memory and other cognitive abilities, and as the disease progresses, behavioral problems such as language disorders, disorientation (e.g., lost walking), altered sexuality, etc. may also occur. Eventually, the patient loses self-care ability and physical function. The incidence rate of Alzheimer's disease of old people over 85 years old can reach 25%, and early identification and early treatment are particularly critical. Therefore, it is significant to develop and improve such prediction methods.
The development of medical artificial intelligence is significant for the computer-aided diagnosis of AD, which is a field most closely related to human life, and in addition, the nuclear magnetic resonance imaging (Magnetic Resonance Imaging, MRI) is widely applied to the diagnosis of clinical AD, has the characteristics of noninvasive, high resolution, high contrast, multi-azimuth imaging and strong resolution to soft tissues, and provides effective information for the diagnosis of AD.
The traditional Alzheimer disease classification method is mainly based on machine learning and mainly comprises two parts of feature extraction and classification: (1) extracting features by manual means; (2) The extracted features are sent to a classifier, such as a Support Vector Machine (SVM), a Random Forest (RF) and the like, for classification, but the classification method needs manual extraction in the process of extracting the features, can be interfered by human factors to generate careless mistakes, and has the problem of high subjectivity.
Disclosure of Invention
In order to solve the problems, the invention provides an Alzheimer's disease classification method based on an improved residual error network (ResNet), which utilizes an effective channel attention module (ECA) to improve the accuracy of a network model on Alzheimer's disease detection; aiming at the problem of pooling feature loss, softpool pooling is adopted to reserve fine granularity features, so that the classification precision of the network model is improved.
The invention is realized by the following steps:
In order to achieve the above object, the invention provides a method for classifying Alzheimer's disease based on an improved ResNet network, which is realized by the following steps:
(1) Constructing an Alzheimer disease MRI brain data set, wherein the Alzheimer disease MRI brain data set comprises three data sets of Alzheimer disease patients (AD), mild cognitive impairment patients (MCI) and normal patients (CN);
(1-1) sequentially performing data preprocessing and data set division on the MRI image;
(1-2) slice selection: three axial image slices of multiple view angles X, Y, Z under 3D MRI scanning are adopted in training, and key indexes alpha of view positions are selected respectively, so that the positions can present a brain outline with clearer brain because each MRI in a standard image space morphological structure has the same structure under the same index. And randomly taking an integer beta in the range of the segment [ -9,9], respectively selecting three slices of alpha+beta-1, alpha+beta and alpha+beta-2, and then respectively inputting the three slices at an angle after splicing, so that the synthesized image contains more spatial information and has certain robustness.
(2) Improving ResNet the network to complete detection classification of Alzheimer's disease;
(2-1) ResNet the network model uses a residual network of 18 layers, by embedding ECA modules in each layer of blocks, a depth feature extraction network ECA-ResNet is designed, the ECA modules omit the dimension reduction part and use 1-dimensional convolution to replace full connection, and ECA operation is expressed as: In the formula, sigma represents a Sigmoid function,/> Representing 1-dimensional 1 x1 convolution, GAP representing global average pooling operation, x being the input tensor;
(2-2) to reduce information loss during pooling operations, the pooling layer uses softpool, which can retain more fine feature information by an exponential function combined with a softmax weighting method. Dividing the region R into l characteristic regions averagely, wherein the activating factor of the nth region is a n, softpool, and each a n is assigned a corresponding weight W n, and the weight W n is the ratio of the activated natural index to the sum of the activated natural indexes of the l characteristic regions: The output value of softpool is obtained by weighted summation of all the active factors in the kernel neighborhood R: /(I)
(2-3) Replacing the cross entropy loss function with the focus loss function Focalloss to suppress the positive and negative sample imbalance, and difficult sample learning;
(3) Training and testing the improved ResNet network:
(3-1) data set training: initializing parameters of a network by using an Adam optimization method in a Pytorch framework to obtain initial parameters of the network, and setting relevant training super parameters for parameter optimization of a network model; randomly disturbing the nuclear magnetic resonance image training set, and sequentially sending the 2D slices with three angles into a network for training;
(3-2) calculation by Focus loss function Wherein y represents the picture category, p represents the probability of model prediction sample category, and the gamma index factor is used for reducing the loss of simple samples, so that the network pays more attention to difficult samples; the excessive weight of the positive samples when the alpha factor is used for balancing the gamma index factor is excessive causes the problem of unbalance between the positive and negative samples. Calculating the error between the label obtained by classifying the training data through the network and the real label, obtaining a loss value, carrying out back propagation on the network by utilizing the obtained loss value, carrying out parameter adjustment on the network, analyzing the loss obtained by training the whole network, continuously and iteratively updating network parameters if the loss is not converged, training the network again, obtaining a network model if the whole network tends to be converged, testing the obtained network model, determining a predictive label by majority voting of three 2D slice images, and if at least two support labels AD exist in the three pictures, predicting that the patient has Alzheimer disease, obtaining a classification result by comparing and evaluating, realizing the classification detection of the Alzheimer disease, and finally carrying out analysis and summarization.
Preferably: in the step 1, the preprocessing operation for the magnetic resonance image data of a given tested is specifically as follows: performing AC-PC origin correction on all data, dividing the corrected image, normalizing the image into a standard template space, extracting gray matter information of the image, performing smoothing and uniform size operation on the gray matter information, performing random sample division according to tested sample labels, and dividing the sample data corresponding to the tested sample into a training set and a testing set in proportion.
Preferably: in the step (1-3), three axial image slices of multiple view angles X, Y, Z under 3D MRI scanning are adopted in training, and the key indexes of view positions are respectively 57, 79 and 60, and because each MRI in the standard image space morphological structure has the same structure under the same index, the positions can present the brain outline with clearer brain. And randomly taking an integer beta in the range of the segment [ -9,9], respectively selecting three slices of alpha+beta-1, alpha+beta and alpha+beta-2, and splicing the three slices at an angle to ensure that the synthesized image contains more spatial information and has certain robustness. Network inputs are detected to accommodate classification and the slices are uniformly sized 224 x 224.
Preferably: in the step 2, in the feature extraction network ResNet, input (7, 64) layers are connected to SoftPool layers, connected to the basic block A, B, C, D, and shorted between layers, wherein A representsModule, B represents/>Module, C represents/>Module, D represents/>And (3) a module, namely connecting an average pooling AvgPool layer and a complete connection layer FC.
Preferably: and 3, adopting an Adam optimization method to dynamically adjust training super parameters in a Pytorch framework, optimizing the training, extracting features of the improved ResNet on the 2D slice image of the MRI, carrying out parameter adjustment on the model by utilizing the back propagation of a network, calculating the accuracy of model classification by a test sample, and reserving optimal model parameters.
The beneficial effects of the invention are as follows:
The Alzheimer's disease classification method based on the improved ResNet network is different from the traditional detection method. On one hand, through the multi-view image slice under the 3D MRI scanning adopted, three axial slices including sagittal X, coronal Y and transversal Z are sequentially input into a group of network models, more space information is reserved to a certain extent, the network training difficulty is reduced, and the detection speed is improved; on the other hand, the focus loss function Focalloss is used for replacing the cross entropy loss function to solve the problems of unbalanced data and difficult sample learning, and finally, a better classification result can be obtained.
Drawings
In order to more clearly illustrate the technical solutions of the present invention, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present invention, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of the overall algorithm of the present invention;
FIG. 2 is a diagram of a dataset image preprocessing process of the present invention;
FIG. 3 is a diagram of the improved ResNet network architecture of the present invention;
FIG. 4 is a diagram of a test procedure embodying the present invention;
Detailed Description
Exemplary embodiments of the present invention will be described hereinafter with reference to the accompanying drawings. It should be noted that, in order to avoid obscuring the present invention due to unnecessary details, only the portions closely related to the solution according to the present invention are shown in the drawings, and other details not greatly related to the present invention are omitted.
The first embodiment is as follows:
a method for classifying alzheimer's disease based on a modified ResNet network of this embodiment, in combination with fig. 1, comprises the steps of:
step one, acquiring, preprocessing and dividing MRI data of Alzheimer's disease;
step two, improving ResNet networks to finish detection classification of Alzheimer's disease;
and thirdly, training and testing the improved ResNet network.
Unlike the first embodiment, the method for classifying alzheimer's disease based on the improved ResNet network according to the present embodiment, in conjunction with fig. 2, the specific flow of the data set processing in the first step is as follows:
(1) Constructing an Alzheimer disease MRI brain data set, wherein the Alzheimer disease MRI brain data set comprises three data sets including AD illness, MCI mild and CN normal;
(2) Sequentially preprocessing data of the MRI image: firstly, carrying out AC-PC (Anterior Commissure, pre-AC connection; posterior Commissure, post-PC connection) origin correction on all data, then dividing corrected MRI data into a brain gray image, a brain white matter image and a cerebrospinal fluid image, adding the three images to obtain a whole brain image, normalizing the divided images into a Montreal neurological institute (Montreal Neurological Institute, MNI) standard template space for registration, and smoothing the images to inhibit noise of functional images, improve signal-to-noise ratio, reduce the residual anatomical structure or functional difference among the images, wherein a Gaussian kernel function is adopted for smoothing treatment; the gray scale normalization is to extend the gray scale distribution in the original image to the image with the whole gray scale by using a gray scale stretching method, and the formula is as follows: Wherein I (I, j) and N (I, j) respectively represent the gray value of the original image and the gray value of the converted image, and min and max respectively represent the minimum gray value and the maximum gray value of the original image; unified size operation; and randomly dividing samples according to the tested sample labels, and dividing the sample data corresponding to the tested samples into a training set and a testing set in proportion.
(3) Dividing the data set: the data are classified into a training set and a testing set according to proportion by reading clinical information, and the data set comprises a Magnetic Resonance Image (MRI) and a tested label; the training data samples are denoted as M = { (a i,bi) |i e [1, Z ] }, where a represents MRI image ID, b represents the corresponding label to be tested, i represents the sample subscript, and test set N = { (a i,bi) |i e [1, W ] }, where Z and W represent the number of samples after division, respectively.
(4) Slice selection: three axial image slices of multiple view angles X, Y, Z under 3D MRI scanning are adopted in training, and the key indexes alpha of view positions are respectively 57, 79 and 60, so that each MRI in the standard image space morphological structure has the same structure under the same index, and the positions can present the brain outline with clearer brain. And randomly taking an integer beta in the range of the segment [ -9,9], respectively selecting three slices of alpha+beta-1, alpha+beta and alpha+beta-2, and then respectively inputting the three slices at an angle after splicing, so that the synthesized image contains more spatial information and has certain robustness. To accommodate the classification detection network input, the slices are uniformly sized 224×224.
And a third specific embodiment:
Unlike the first or second embodiment, the method for classifying alzheimer's disease based on the improved ResNet network according to the present embodiment, in combination with fig. 3, the specific flow of improving the ResNet network in the second step is as follows:
(1) The ResNet network model uses a residual network of 18 layers, and a depth feature extraction network ECA-ResNet is designed by embedding an ECA module in each layer of blocks, wherein the ECA module omits a dimension reduction part and uses 1-dimensional convolution to replace full connection, and ECA operation is expressed as follows: In the formula, sigma represents a Sigmoid function,/> Representing 1-dimensional 1 x1 convolution, GAP representing global average pooling operation, x being the input tensor;
(2) To reduce information loss during the pooling operation, the pooling layer uses softpool, which can retain more fine feature information by combining an exponential function with a softmax weighting method. Dividing the region R into l characteristic regions averagely, wherein the activating factor of the nth region is a n, softpool, and each a n is assigned a corresponding weight W n, and the weight W n is the ratio of the activated natural index to the sum of the activated natural indexes of the l characteristic regions: The output value of softpool is obtained by weighted summation of all the active factors in the kernel neighborhood R: /(I)
The specific structure of the network in the second step is as follows: in the feature extraction network ResNet, input (7, 64) layers are connected to SoftPool layers, to basic block A, B, C, D, and then layers are shorted together, where A representsLayer module, B represents/>Module, C represents/>Module, D representsAnd (3) a module, namely connecting an average pooling AvgPool layer and a complete connection layer FC.
The specific embodiment IV is as follows:
unlike the third embodiment, the method for classifying alzheimer's disease based on the improved ResNet network according to the present embodiment, in combination with fig. 4, the specific flow of training and testing the network in the third step is as follows:
(1) The data set is trained: initializing parameters of a network by using an Adam optimization method in a Pytorch framework to obtain initial parameters of the network, and setting relevant training super parameters for parameter optimization of a network model; randomly disturbing the MRI training set through a shuffle statement in a Python programming language, and sequentially sending the 2D slices with three angles into a network for training;
(2) And adopting an Adam optimization method to dynamically adjust training super parameters in Pytorch frames, optimizing the training, extracting features of the improved ResNet-based MRI 2D slice images, carrying out parameter adjustment on the models by utilizing the back propagation of a network, calculating the accuracy of model classification by test samples, evaluating the trained classification models, and reserving optimal model parameters.
(3) Calculation by means of a focal loss functionWherein y represents the picture category, p represents the probability of model prediction sample category, and the gamma index factor is used for reducing the loss of simple samples, so that the network pays more attention to difficult samples; the excessive weight of the positive samples when the alpha factor is used for balancing the gamma index factor is excessive causes the problem of unbalance between the positive and negative samples.
Calculating the error between the label obtained by classifying the training data through the network and the training real label, obtaining a loss value, carrying out back propagation by utilizing the obtained loss value, carrying out parameter adjustment on the network, analyzing a loss function of the whole network training, continuously and iteratively updating a network parameter adjustment network structure if the loss function is not converged, carrying out training on the network again, obtaining a network model if the whole network tends to be converged, testing the obtained optimal network model, determining a predictive label by majority voting of three 2D slice images, if at least two support labels AD exist in the three images, predicting the patient to have the Alzheimer disease, comparing and evaluating to obtain a classification result, realizing the classified detection of the Alzheimer disease, and finally analyzing and summarizing.
Claims (5)
1. A method for classifying alzheimer's disease based on an improved ResNet network, characterized by: the method is realized by the following steps:
(1) Constructing an Alzheimer disease MRI brain data set, wherein the Alzheimer disease MRI brain data set comprises three data sets of AD patients, mild cognitive impairment patients (MCI) and normal patients (CN);
(1-1) sequentially performing data preprocessing and data set division on the MRI image;
(1-2) slice selection: in training, three axial image slices of multiple view angles X, Y, Z under 3D MRI scanning are adopted, and key indexes alpha of view positions are selected respectively, so that the positions can present brain contours with clearer brains because each MRI in a standard image space morphological structure has the same structure under the same index; randomly taking an integer beta in the range of the segment [ -9,9], respectively selecting three slices of alpha+beta-1, alpha+beta and alpha+beta-2, and then respectively inputting the three slices at an angle after splicing, so that the synthesized image contains more spatial information and has certain robustness;
(2) Improving ResNet the network to complete detection classification of Alzheimer's disease;
(2-1) ResNet the network model uses a residual network of 18 layers, by embedding an ECA model in each layer of blocks, by embedding an ECA module in each layer of blocks, a depth feature extraction network ECA-ResNet is designed, the ECA module omits a dimension reduction part and uses 1-dimensional convolution to replace full connection, the feature extraction capability is further improved and the robustness of the model is enhanced, and ECA operation is expressed as: In the formula, sigma represents a Sigmoid function,/> Representing 1-dimensional 1 x1 convolution, GAP representing global average pooling operation, x being the input tensor;
(2-2) to reduce information loss during pooling operations, the pooling layer adopts softpool, which can retain more fine feature information by an exponential function combined with a softmax weighting method;
(2-3) replacing the cross entropy loss function with the focus loss function Focalloss to suppress the positive and negative sample imbalance, and difficult sample learning;
(3) Training and testing the improved ResNet network:
(3-1) data set training: initializing parameters of a network by using an Adam optimization method in a Pytorch framework to obtain initial parameters of the network, and setting relevant training super parameters for parameter optimization of a network model; randomly disturbing the nuclear magnetic resonance image training set, and sequentially sending the 2D slices with three angles into a network for training;
(3-2) calculation by Focus loss function Wherein y represents the picture category, p represents the probability of model prediction sample category, and the gamma index factor is used for reducing the loss of simple samples, so that the network pays more attention to difficult samples; the alpha factor is used for balancing the problem that unbalance between positive and negative samples is caused by excessive weight of the positive samples when the gamma index factor is excessive;
Calculating the error between the label obtained by classifying the training data through the network and the real label, obtaining a loss value, carrying out back propagation on the obtained loss value, carrying out parameter adjustment on the network, analyzing the loss obtained by training the whole network, continuously and iteratively updating network parameters if the loss is not converged, training the network again, obtaining a network model if the whole network tends to be converged, testing the obtained network model, determining the prediction label by majority vote of three 2D slice images, and if at least two support labels AD exist in the three pictures, predicting that the patient has Alzheimer disease, obtaining a classification result through comparison and evaluation, realizing the classification detection of the Alzheimer disease, and finally carrying out analysis and summarization.
2. The method for classifying alzheimer's disease based on a modified ResNet network according to claim 1, wherein in step 1, the preprocessing operation for the given subject magnetic resonance image data is specifically: performing AC-PC origin correction on all data, dividing the corrected image, normalizing the image into a standard template space, extracting gray matter information of the image, performing smoothing and uniform size operation on the gray matter information, performing random sample division according to a sample label to be tested, and dividing the sample data corresponding to the tested sample into a training set and a testing set in proportion; the data set comprises nuclear magnetic resonance images and a tested label; the training data samples are denoted as M = { (a i,bi) |i e [1, Z ] }, where a represents MRI image ID, b represents the corresponding label to be tested, i represents the sample subscript, and test set N = { (a i,bi) |i e [1, W ] }, where Z and W represent the number of samples after division, respectively.
3. The method of claim 1, wherein in the step (1-2), three axial image slices of multi-view X, Y, Z under 3D MRI scan are taken, the ranges of the slices on the three axes are respectively [1,121], [1,145], [1,121], 121D slices can be obtained by slicing from the three axes, and among the slices in each direction, some slices near both ends of the axes have little brain tissue, and the slices contain less effective information and have low training value, so that the slices are omitted; the view position key indexes alpha are respectively 57, 79 and 60, and because each MRI in the standard image space morphological structure has the same structure under the same index, the positions can present the brain outline with clearer brain; randomly taking an integer beta in the range of the segment [ -9,9], respectively selecting three slices of alpha+beta-1, alpha+beta and alpha+beta-2, and then splicing and respectively inputting the three slices at an angle, so that the synthesized image contains more spatial information and has certain robustness; to ensure that the input network size pictures remain consistent, the slices are uniformly sized 224 x 224.
4. The method of claim 1, wherein in step 2, in the feature extraction network ResNet, input (7, 64) layers are connected to SoftPool layers, foundation blocks A, B, C, D are connected, and layers are shorted together, wherein a representsModule, B representsModule, C represents/>Module, D represents/>And (3) a module, namely connecting an average pooling AvgPool layer and a complete connection layer FC.
5. The method for classifying alzheimer's disease based on an improved ResNet network according to claim 1, wherein in step 3, an optimization method is adopted to dynamically adjust and optimize training super parameters, the improved ResNet is used to extract features of MRI 2D slice images, the model is subjected to parameter adjustment by using the back propagation of the network, the accuracy of model classification is calculated by testing samples, the trained classification model is evaluated, and the optimal model parameters are reserved.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111237755.XA CN113989551B (en) | 2021-10-25 | 2021-10-25 | Alzheimer's disease classification method based on improved ResNet network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111237755.XA CN113989551B (en) | 2021-10-25 | 2021-10-25 | Alzheimer's disease classification method based on improved ResNet network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113989551A CN113989551A (en) | 2022-01-28 |
| CN113989551B true CN113989551B (en) | 2024-05-03 |
Family
ID=79740702
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111237755.XA Active CN113989551B (en) | 2021-10-25 | 2021-10-25 | Alzheimer's disease classification method based on improved ResNet network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113989551B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114999629B (en) * | 2022-04-28 | 2024-03-01 | 安徽大学 | AD early prediction method, system and device based on multi-feature fusion |
| CN114937171B (en) * | 2022-05-11 | 2023-06-09 | 复旦大学 | Deep learning-based Alzheimer's classification system |
| CN115553787A (en) * | 2022-09-30 | 2023-01-03 | 哈尔滨理工大学 | Fetal electrocardiosignal extraction method based on multi-scale residual shrinkage U-Net |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111242233A (en) * | 2020-01-18 | 2020-06-05 | 温州大学 | Alzheimer disease classification method based on fusion network |
| CN111461233A (en) * | 2020-04-02 | 2020-07-28 | 大连海事大学 | Automatic nuclear magnetic resonance image classification method and device based on MDC L STM-L DenseNet network |
| CN111488914A (en) * | 2020-03-17 | 2020-08-04 | 哈尔滨工业大学 | Alzheimer disease classification and prediction system based on multitask learning |
| CN111738363A (en) * | 2020-07-24 | 2020-10-02 | 温州大学 | Alzheimer disease classification method based on improved 3D CNN network |
| CN112674720A (en) * | 2020-12-24 | 2021-04-20 | 四川大学 | Alzheimer disease pre-diagnosis method based on 3D convolutional neural network |
-
2021
- 2021-10-25 CN CN202111237755.XA patent/CN113989551B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111242233A (en) * | 2020-01-18 | 2020-06-05 | 温州大学 | Alzheimer disease classification method based on fusion network |
| CN111488914A (en) * | 2020-03-17 | 2020-08-04 | 哈尔滨工业大学 | Alzheimer disease classification and prediction system based on multitask learning |
| CN111461233A (en) * | 2020-04-02 | 2020-07-28 | 大连海事大学 | Automatic nuclear magnetic resonance image classification method and device based on MDC L STM-L DenseNet network |
| CN111738363A (en) * | 2020-07-24 | 2020-10-02 | 温州大学 | Alzheimer disease classification method based on improved 3D CNN network |
| CN112674720A (en) * | 2020-12-24 | 2021-04-20 | 四川大学 | Alzheimer disease pre-diagnosis method based on 3D convolutional neural network |
Non-Patent Citations (1)
| Title |
|---|
| 一种深度置信网络和迭代量化的图像检索方法;宋立新;徐军;;信息与控制(第02期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113989551A (en) | 2022-01-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113989551B (en) | Alzheimer's disease classification method based on improved ResNet network | |
| CN110070935B (en) | Medical image synthesis method, classification method and device based on antagonistic neural network | |
| CN113616184B (en) | Brain network modeling and individual prediction method based on multi-mode magnetic resonance image | |
| Voets et al. | Replication study: Development and validation of deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs | |
| CN112365464B (en) | GAN-based medical image lesion area weak supervision positioning method | |
| CN109614991A (en) | A kind of segmentation and classification method of the multiple dimensioned dilatancy cardiac muscle based on Attention | |
| CN110930416A (en) | MRI image prostate segmentation method based on U-shaped network | |
| CN110211671B (en) | Thresholding method based on weight distribution | |
| CN115393269A (en) | Extensible multi-level graph neural network model based on multi-modal image data | |
| CN112733961A (en) | Method and system for classifying diabetic retinopathy based on attention mechanism | |
| Aslam et al. | Neurological Disorder Detection Using OCT Scan Image of Eye | |
| CN116097302A (en) | Connected machine learning model with joint training for lesion detection | |
| CN113177943A (en) | Cerebral apoplexy CT image segmentation method | |
| CN110298364B (en) | Multi-threshold-value multi-task-based feature selection method for functional brain network | |
| CN114359629B (en) | Deep migration learning-based X-chest X-ray pneumonia classification and identification method | |
| CN105184794B (en) | A kind of CSM Computer Aided Analysis Systems and method based on tensor image | |
| CN110853048A (en) | MRI image segmentation method, device and storage medium based on rough training and fine training | |
| CN115760586A (en) | Medical image enhancement method based on multi-scale attention generation countermeasure network | |
| CN117218453B (en) | Incomplete multi-mode medical image learning method | |
| CN115601535A (en) | Chest radiograph abnormal recognition domain self-adaption method and system combining Wasserstein distance and difference measurement | |
| CN115330600A (en) | Lung CT image super-resolution method based on improved SRGAN | |
| Estermann et al. | Pwml detection in 3d cranial ultrasound volumes using over-segmentation and multimodal classification with deep learning | |
| CN114723937A (en) | Method and system for classifying blood vessel surrounding gaps based on nuclear magnetic resonance image | |
| CN114331996A (en) | Medical image classification method and system based on self-coding decoder | |
| Wu et al. | Mscan: Multi-scale channel attention for fundus retinal vessel segmentation |
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 |