CN109859839B - Alzheimer disease classification method based on layered ensemble learning - Google Patents
Alzheimer disease classification method based on layered ensemble learning Download PDFInfo
- Publication number
- CN109859839B CN109859839B CN201910065074.6A CN201910065074A CN109859839B CN 109859839 B CN109859839 B CN 109859839B CN 201910065074 A CN201910065074 A CN 201910065074A CN 109859839 B CN109859839 B CN 109859839B
- Authority
- CN
- China
- Prior art keywords
- classification
- alzheimer
- classifier
- disease
- ensemble learning
- 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
Images
Landscapes
- Magnetic Resonance Imaging Apparatus (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an Alzheimer's disease classification method based on layered ensemble learning, which comprises the following steps: s1, acquiring a nuclear magnetic resonance image of the Alzheimer disease; s2, preprocessing the nuclear magnetic resonance image obtained in the step S1; s3, inputting the preprocessed image into a model based on hierarchical ensemble learning by taking a slice as a unit, extracting a feature matrix by adopting a pre-trained Deep Neural Network (DNN), and inputting the extracted feature matrix into a classifier for classification to obtain a coarse prediction result of the slice level; s4, integrating the coarse prediction results obtained in the step S3, and classifying the coarse prediction results by the classifier again to obtain fine classification prediction results of the slices; and S5, integrating the fine classification prediction results obtained in the step S4, and performing classification prediction by using a classifier again to obtain the final classification result of the Alzheimer' S disease of the patient grade. The method has the advantages of stable classification, high classification efficiency and good universality and flexibility of the classification method.
Description
Technical Field
The invention relates to the field of medical image processing, in particular to an Alzheimer's disease classification method based on hierarchical ensemble learning.
Background
Alzheimer's Disease (AD) is a serious irreversible degenerative disease of the brain, and it is predicted that the number of alzheimer's patients will exceed 1.068 billion worldwide by 2050. The burden of patients and society is reduced, and the method has important significance for early diagnosis and treatment of the Alzheimer disease. In recent years, neuroimaging-based diagnosis of alzheimer's disease, such as Magnetic Resonance Imaging (MRI), has attracted considerable attention because neuroimaging can provide a visual indication of the pathology and its progress. Therefore, extensive research on computer-aided diagnosis of alzheimer's disease using neuroimages has been conducted with remarkable results.
In general, the traditional approach to this topic can be generalized to two key steps, i.e., feature extraction and modeling (classification). The former focuses on designing effective neural image representations, such as HOG; the latter attempts to classify the extracted features as accurately as possible, and typical classifiers used for this purpose include random forests, Support Vector Machines (SVMs). However, since conventional features are designed manually, it is difficult to optimize them with classifiers, which necessarily leads to performance gaps in real-world applications. One potential solution to this problem is a Deep Learning (DL) based approach that can unify the two steps into a complete Deep Neural Network (DNN) and train the Network in an end-to-end manner. Recent studies have shown that the DL-based method has superior performance in the related tasks compared to the conventional method.
However, applying DL techniques to the Alzheimer's disease classification is not so straightforward due to the fact that, first, most deep learning networks are proposed for processing natural images, usually containing rich and sharp textures; secondly, the volume of the neural image is much larger, for example, the MRI image of the ADNI dataset 1 may contain 192 slices at most, while the natural image has only RGB channels; last but not least, training a deep learning network requires a large number of neural images, which are also expensive to compute.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides the Alzheimer's disease classification method based on hierarchical ensemble learning, and the method has the advantages of stable classification, high classification efficiency and good universality and flexibility.
The technical scheme for realizing the purpose of the invention is as follows:
an Alzheimer's disease classification method based on layered ensemble learning comprises the following steps:
s1, acquiring a nuclear magnetic resonance image of the Alzheimer disease;
s2, preprocessing the nuclear magnetic resonance image obtained in the step S1;
s3, inputting the preprocessed image into a model based on hierarchical ensemble learning by taking a slice as a unit, extracting a feature matrix by adopting a pre-trained Deep Neural Network (DNN), and inputting the extracted feature matrix into a classifier for classification to obtain a coarse prediction result of the slice level;
s4, integrating the coarse prediction results obtained in the step S3, and classifying the coarse prediction results by the classifier again to obtain fine classification prediction results of the slices;
and S5, integrating the fine classification prediction results obtained in the step S4, and performing classification prediction by using a classifier again to obtain the final classification result of the Alzheimer' S disease of the patient grade.
In step S2, the preprocessing is performed by preprocessing a slice of a magnetic resonance image of an alzheimer patient using SPM12, performing head correction, temporal layer correction, registration, and segmentation on an image in nii format, and finally obtaining separation of a gray brain matter, a white brain matter, a cerebrospinal fluid, and a skull by segmentation.
In step S3, the model based on layered ensemble learning includes three layers, the first layer adopts M characteristics, trains a classifier for each characteristic, gives a slice, and generates a middle probability characteristic vector y ∈ [0,1 ]]1×|C|×MElement y thereofc,mRepresents the probability that the slice belongs to class c predicted by classifier m; the second layer carries out intelligent classifier integration, takes the rough prediction y as input, outputs a refined prediction which is recorded as Z belonging to [0,1 ]]1×C(ii) a The third layer connects all the N fine prediction results to obtain Z1,,...,ZNAgain using it as a feature to predict the final result.
In step S3, the extraction of the feature matrix using a Deep Neural Network (DNN) trained in advance is specifically to input the segmented gray brain slice into VGG19 and ALEXNET networks trained in advance to perform feature extraction, and each slice obtains a feature matrix of 1x4096 dimensions.
In step S3, the classifiers are Support Vector Machine (SVM) and Extreme Learning Machine (ELM) classifiers.
Has the advantages that: the invention provides an Alzheimer's disease classification method based on layered ensemble learning, which has the following advantages that:
(1) the internal output of the deep learning network trained in advance is directly adopted as the characteristic, so that the calculation cost for training the deep learning network can be eliminated;
(2) considering that the pre-trained neural network is not optimal for processing the MRI images, combining results obtained by processing the MRI images by a plurality of pre-trained neural networks to improve the robustness of feature representation; the results of all the slices are combined to obtain a final prediction result of the HEL, and the optimal slice does not need to be manually selected, so that the classification method has robustness on abnormal values;
(3) a classifier is trained on each feature type to reduce the effects of "cursing" dimension, and ensemble learning is done at the semantic level (classification result), giving flexibility in selecting various features and classifiers.
Drawings
FIG. 1 is a flow chart of a method for classifying Alzheimer's disease based on hierarchical ensemble learning according to the present invention;
FIG. 2 is a schematic diagram of an operational interface of the image pre-processing SPM 12;
FIG. 3 is a slice of an MRI image before preprocessing in an embodiment, which is a slice image of the same MRI image (MRI) in three orientations, namely coronal, sagittal and transverse, from left to right;
fig. 4 is a grey brain matter slice image of a magnetic resonance image corresponding to fig. 3 after preprocessing.
Detailed Description
The invention is further illustrated but not limited by the following figures and examples.
Example (b):
as shown in fig. 1, a method for classifying alzheimer's disease based on hierarchical ensemble learning includes the following steps:
s1, acquiring a nuclear magnetic resonance image of the Alzheimer disease, specifically selecting 155 groups of data sets, wherein AD of Alzheimer disease patients is 50 groups, MCI of mild cognitive impairment patients is 55 groups, and NC of healthy people is 50 groups;
s2, preprocessing the magnetic resonance image obtained in step S1 by using the SPM12, where an operation interface of the SPM12 is shown in fig. 2, and an operation flow is as follows: performing head correction, time layer correction, registration and segmentation on the nii format image, and finally obtaining separation of gray brain matter, white brain matter, cerebrospinal fluid and skull after segmentation; the obtained pretreatment effect graph is a slice of a nuclear magnetic resonance image before pretreatment, as shown in fig. 3, slice images of the same nuclear Magnetic Resonance Image (MRI) in three orientations of coronal position, sagittal position and cross section are sequentially arranged from left to right, fig. 4 is a gray matter slice image of the nuclear magnetic resonance image corresponding to fig. 3 after pretreatment, the number N of slices of each patient after pretreatment is 28, and the total number L of slices of the data set is 4340;
s3, inputting the preprocessed gray matter slice images into a model based on hierarchical ensemble learning by taking slices as units, extracting feature matrices by adopting the images inputted into pre-trained VGG19 and alexanet networks, obtaining a feature matrix of 1x4096 dimensions for each slice, taking the 'fc 7' layer in the two networks as output as the extracted feature matrix of each slice, where M in fig. 1 is 2, that is, each slice obtains two different feature matrices of 2x4096 dimensions; inputting the extracted feature matrix into a Support Vector Machine (SVM) and an Extreme Learning Machine (ELM) classifier for classification to obtain a coarse prediction result of a slice level;
s4, integrating the coarse prediction results obtained in the step S3, and classifying the coarse prediction results by the classifier again to obtain fine classification prediction results of the slices;
and S5, integrating the fine classification prediction results obtained in the step S4, and performing classification prediction by using a classifier again to obtain the final classification result of the Alzheimer' S disease of the patient grade.
In step S3, the framework based on hierarchical ensemble learning includes three layers, the first layer adopts M characteristics, trains a classifier for each characteristic, gives a slice, and generates a middle probability characteristic vector y E [0,1 ]]1×|C|×MElement y thereofc,mRepresents the probability that the slice belongs to class c predicted by classifier m; the second layer carries out intelligent classifier integration, takes the rough prediction y as input, outputs a refined prediction which is recorded as Z belonging to [0,1 ]]1×|C|(ii) a The third layer connects all the N fine prediction results to obtain Z1,,...,ZNAgain using it as a feature to predict the final result.
In step S2, L represents a set of magnetic resonance images, and C ═ C1,c2,c3The setting of class labels corresponds to NC, MCI and AD groups respectively, learning a function f L → C, similar latest methods, and the implementation of f can be further divided into learning a binary classifier of | C | ═ 3 "1-vs-rest", i.e. f (L) ═ max (f)1(L),...,f|C|(L)), where fc(L) represents the possibility that the image L belongs to class C, C1., | C |, the rest of the text will focus on the implementation of f without loss of generality, and we will omit the index C for clarity of expression.
In steps S2 and S3, in the model of the hierarchical ensemble learning, no restriction is imposed on the classifiers, and the model is provided with the flexibility of using any available classifier, and in fact, not only the classifier but also both ensemble parts can be realized by the same method, in order to prove this, only linear models (logistic regression) are considered, and f (x) is w · x, and f (x) is understood by minimizing the following loss function:
where T represents the number of training images from the data set, riIs an auxiliary label, assuming that the c-th binary classifier is being used for training, if the i-th training slice belongs to class c, ri=1;Otherwise ri-1; obviously, if w · xiSymbol of (a) and riThe same, the left-hand term of the above equation will be small. The correct terminology is regularization to prevent overfitting, the above-mentioned loss function is a common optimization problem, we solve directly using the liblear package, and by replacing x in the above equation with y and z, two ensemble learning steps can be achieved. Thus, the whole training process of the hierarchical ensemble learning is easily completed.
The working process of the embodiment comprises the following steps:
the image dataset was acquired from an ADNI dataset containing 50/55/50 NC/MCI/AD subjects, with the number of slices N per image set to 28, and the image data was aggregated for 4340 slices, with 10 experiments on randomly scrambled images to avoid bias in the evaluation, and the average results reported. The ratio of training to test images (subjects) was 3: 7, average accuracy was used as an evaluation index.
Claims (5)
1. A method for classifying Alzheimer's disease based on hierarchical ensemble learning is characterized by comprising the following steps:
s1, acquiring a nuclear magnetic resonance image of the Alzheimer disease;
s2, preprocessing the nuclear magnetic resonance image obtained in the step S1;
s3, inputting the preprocessed image into a model based on layered ensemble learning by taking a slice as a unit, extracting a feature matrix by adopting a pre-trained deep neural network, and inputting the extracted feature matrix into a classifier for classification to obtain a coarse prediction result of the slice level;
s4, integrating the coarse prediction results obtained in the step S3, and classifying the coarse prediction results by the classifier again to obtain fine classification prediction results of the slices;
and S5, integrating the fine classification prediction results obtained in the step S4, and performing classification prediction by using a classifier again to obtain the final classification result of the Alzheimer' S disease of the patient grade.
2. The method for classifying Alzheimer 'S disease based on hierarchical ensemble learning according to claim 1, wherein in step S2, the preprocessing comprises preprocessing slices of magnetic resonance images of Alzheimer' S disease patients by SPM12, performing head correction, temporal correction, registration and segmentation on nii-format images, and finally obtaining separation of gray brain matter, white brain matter, cerebrospinal fluid and skull by segmentation.
3. The method for classifying Alzheimer' S disease based on hierarchical ensemble learning as claimed in claim 1, wherein in step S3, the model based on hierarchical ensemble learning comprises three layers, the first layer employs M features, and trains a classifier for each feature, and gives a slice to generate an intermediate probability feature vector y e [0,1 ∈ for each feature]1×|C|×MElement y thereofc,mRepresents the probability that the slice belongs to class c predicted by classifier m; the second layer carries out intelligent classifier integration, takes the rough prediction y as input, outputs a refined prediction which is recorded as Z belonging to [0,1 ]]1×|C|(ii) a The third layer connects all the N fine prediction results to obtain Z1,...,ZNAgain using it as a feature to predict the final result.
4. The method for classifying Alzheimer' S disease based on hierarchical ensemble learning according to claim 1, wherein in step S3, the feature matrix is extracted by using a deep neural network trained in advance, specifically, the gray brain slice obtained by segmentation is input into a VGG19 and an ALEXNET network trained in advance for feature extraction, and each slice obtains a feature matrix with 1x4096 dimensions.
5. The method for classifying Alzheimer' S disease based on hierarchical ensemble learning according to claim 1, wherein in step S3, the classifiers are a support vector machine and an extreme learning machine classifier.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910065074.6A CN109859839B (en) | 2019-01-23 | 2019-01-23 | Alzheimer disease classification method based on layered ensemble learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910065074.6A CN109859839B (en) | 2019-01-23 | 2019-01-23 | Alzheimer disease classification method based on layered ensemble learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109859839A CN109859839A (en) | 2019-06-07 |
CN109859839B true CN109859839B (en) | 2022-05-03 |
Family
ID=66895758
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910065074.6A Active CN109859839B (en) | 2019-01-23 | 2019-01-23 | Alzheimer disease classification method based on layered ensemble learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109859839B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110718301B (en) * | 2019-09-26 | 2024-04-19 | 东北大学 | Alzheimer's disease auxiliary diagnosis device and method based on dynamic brain function network |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108154924A (en) * | 2017-12-07 | 2018-06-12 | 广州市本真网络科技有限公司 | Alzheimer's disease tagsort method and system based on support vector machines |
CN109087298A (en) * | 2018-08-17 | 2018-12-25 | 电子科技大学 | A kind of Alzheimer's disease MRI image classification method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2012074565A1 (en) * | 2010-01-26 | 2012-06-07 | University Of Utah Research Foundation | Imaging-based identification of a neurological disease or a neurological disorder |
-
2019
- 2019-01-23 CN CN201910065074.6A patent/CN109859839B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108154924A (en) * | 2017-12-07 | 2018-06-12 | 广州市本真网络科技有限公司 | Alzheimer's disease tagsort method and system based on support vector machines |
CN109087298A (en) * | 2018-08-17 | 2018-12-25 | 电子科技大学 | A kind of Alzheimer's disease MRI image classification method |
Non-Patent Citations (3)
Title |
---|
Learning Deep Convolutional Features for MRI Based Alzheimer’s Disease Classification;Fayao Liu 等;《https://arXiv.org/pdf/:1404.3366v1.pdf》;20140413;第1-12页 * |
基于多尺度残差神经网络的阿尔茨海默病诊断分类;刘振丙 等;《山东大学学报(工学版)》;20181231;第48卷(第6期);第1-7页 * |
基于深度学习的阿尔茨海默症分类算法研究;王如月;《中国优秀博硕士学位论文全文数据库(硕士) 医药卫生科技辑》;20220215(第02期);第E071-40页 * |
Also Published As
Publication number | Publication date |
---|---|
CN109859839A (en) | 2019-06-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2023077603A1 (en) | Prediction system, method and apparatus for abnormal brain connectivity, and readable storage medium | |
CN111461232A (en) | Nuclear magnetic resonance image classification method based on multi-strategy batch type active learning | |
Kong et al. | Multi-modal data Alzheimer’s disease detection based on 3D convolution | |
CN112465058A (en) | Multi-modal medical image classification method under improved GoogLeNet neural network | |
CN104881683A (en) | Cataract eye fundus image classification method based on combined classifier and classification apparatus | |
Wang et al. | Applications of generative adversarial networks in neuroimaging and clinical neuroscience | |
CN112785593B (en) | Brain image segmentation method based on deep learning | |
CN114359164A (en) | Method and system for automatically predicting Alzheimer disease based on deep learning | |
Li et al. | Comparison of machine learning approaches for enhancing Alzheimer’s disease classification | |
Yao et al. | Artificial intelligence-based diagnosis of Alzheimer's disease with brain MRI images | |
CN109859839B (en) | Alzheimer disease classification method based on layered ensemble learning | |
Shi et al. | Discriminative feature network based on a hierarchical attention mechanism for semantic hippocampus segmentation | |
Herzog et al. | Convolutional neural networks-based framework for early identification of dementia using MRI of brain asymmetry | |
Pan et al. | Multi-classification prediction of Alzheimer’s disease based on fusing multi-modal features | |
Thangavel et al. | EAD-DNN: Early Alzheimer's disease prediction using deep neural networks | |
CN112489048B (en) | Automatic optic nerve segmentation method based on depth network | |
Jin et al. | Unsupervised detection of individual atrophy in Alzheimer's disease | |
Park et al. | Deep joint learning of pathological region localization and Alzheimer’s disease diagnosis | |
Alharthi et al. | Do it the transformer way: a comprehensive review of brain and vision transformers for autism spectrum disorder diagnosis and classification | |
Mohapatra et al. | A machine learning approach to assist prediction of Alzheimer's disease with convolutional neural network | |
Lin et al. | Hybrid CNN-SVM for alzheimer’s disease classification from structural MRI and the alzheimer’s disease neuroimaging initiative (ADNI) | |
LR et al. | Multimodal screening for dyslexia using anatomical and functional MRI data | |
Zhang et al. | Latent multi-relation reasoning for gan-prior based image super-resolution | |
Chen et al. | Attention-based node-edge graph convolutional networks for identification of autism spectrum disorder using multi-modal mri data | |
Wang et al. | Automatic segmentation for retinal vessel using concatenate UNet++ |
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 |