CN111916162A - Multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint - Google Patents

Multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint Download PDF

Info

Publication number
CN111916162A
CN111916162A CN202010754793.1A CN202010754793A CN111916162A CN 111916162 A CN111916162 A CN 111916162A CN 202010754793 A CN202010754793 A CN 202010754793A CN 111916162 A CN111916162 A CN 111916162A
Authority
CN
China
Prior art keywords
sample
mode
modal
feature
feature selection
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.)
Pending
Application number
CN202010754793.1A
Other languages
Chinese (zh)
Inventor
陈伟斌
张笑钦
钱乐旦
赵丽
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wenzhou University
Original Assignee
Wenzhou University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Wenzhou University filed Critical Wenzhou University
Priority to CN202010754793.1A priority Critical patent/CN111916162A/en
Publication of CN111916162A publication Critical patent/CN111916162A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/20—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for electronic clinical trials or questionnaires
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/211—Selection of the most significant subset of features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • 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
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Public Health (AREA)
  • Theoretical Computer Science (AREA)
  • Medical Informatics (AREA)
  • Primary Health Care (AREA)
  • General Health & Medical Sciences (AREA)
  • Epidemiology (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint, which comprises the following steps of: (1) collecting multi-mode brain image data; (2) after data of a plurality of modes are obtained, feature selection is carried out in a multi-mode feature collaborative analysis mode, a regression model from each mode data to a classification class target is established, group sparse constraint is carried out on regression vectors, and therefore a public feature subset relevant to all tasks is obtained; (3) modeling multi-modal data features; (4) writing the target function into an augmented Lagrange form, wherein the target function is changed into a convex function; (5) obtaining a weight matrix of each mode and a weight of each feature of each sample; (6) and fusing the multi-modal characteristics of the sample by using a method of a multi-core support vector machine for classification. According to the technical scheme, the importance of the samples among the modes is considered, the correlation among the modes is also considered, more meaningful features are found out, and the accuracy of classification and prediction is improved.

Description

Multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint
Technical Field
The invention relates to the technical field of neural image analysis and processing, in particular to a multi-modal neural image feature selection method based on sample weight and low-rank constraint.
Background
The rapid development of neuroimaging technology has made possible the measurement and visualization of structural information of the human brain and functional information of mental activities, thereby leading to an explosive growth of knowledge about brain cognition in recent years. However, whatever the imaging mode, it only provides an imaging means, how to effectively and quantitatively analyze and visualize the data obtained by neuroimaging, and how to integrate the scattered multi-modal multi-map data information into a panoramic image reflecting the physiological and pathological mechanisms of brain tissue structure and brain function, which is the key of the subsequent clinical application.
However, for the characteristic representation of multi-modal multi-map data isomerism, the traditional characteristic selection method, namely the multi-modal brain image data fusion diagnosis, adopts the simple splicing processing of characteristic vectors, and realizes the improvement of diagnosis precision in a simple information quantity mode. Multiple sets of features are taken from multiple atlases for each sample and then cascaded for subsequent classification tasks. In recent years, a multi-kernel learning (multi-kernel learning) method has been applied to multimodal data fusion analysis, and has achieved a favorable effect. However, a large amount of redundant features irrelevant to a diagnosis and prediction task exist in multi-modal data, so that some researchers expand on the basis of a single-modal or single-map data embedded sparse feature selection technology, and adopt a multi-task learning framework to realize multi-modal and multi-map group sparse feature selection, namely, a consistent brain image marker is selected as a diagnostic discriminant feature.
Therefore, the obvious disadvantages of the existing feature selection method are as follows: only single-mode data or simple cascade multi-mode data are used, complementary information among the multi-mode data cannot be fully utilized, the feature representation method is single, and the learning performance of the model is limited to a certain extent.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint.
In order to achieve the purpose, the invention provides the following technical scheme: a multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint comprises the following steps:
(1) collecting multi-mode brain image data;
(2) after data of a plurality of modes are obtained, feature selection is carried out in a multi-mode feature collaborative analysis mode, a regression model from each mode data to a classification class target is established, group sparse constraint is carried out on regression vectors, and therefore a public feature subset relevant to all tasks is obtained;
(3) modeling multi-modal data features;
(4) writing the target function into an augmented Lagrange form, wherein the target function is changed into a convex function;
(5) solving each variable of the formula in the step (4) by using an alternating direction multiplier algorithm by using a gradient descent method to obtain a weight matrix W of each mode of each sample and a weight beta of each characteristicj。
(6) And fusing the multi-modal characteristics of the sample by using a method of a multi-core support vector machine for classification.
Preferably, step (1) is performed by performing multiple imaging in multiple modes or multiple imaging in the same mode on the same patient, and obtaining information from several images to perform comprehensive analysis.
Preferably, in step (3), the data of each modality is used
Figure BDA0002611181550000021
Wherein the superscript j denotes the jth mode and the subscript i denotes the ith modeThen, the process is carried out; the category to which each sample belongs is used as yiIn this representation, the index i represents the ith sample; inputting the data of each sample and the belonged category into an objective function for feature selection, wherein the objective function for feature selection is as follows:
Figure BDA0002611181550000031
Figure BDA0002611181550000032
wherein n is the number of samples, m is the number of modes, betajA vector is selected for the features of the jth modality,
Figure BDA0002611181550000033
what weight is occupied by the j mode of the ith sample, W is the weight matrix of all modes of the training sample, wherein each row represents a different sample, each column represents a different mode, and lambdaLIs a regularization parameter that constrains the feature sparsity, λRIs a regularization parameter that constrains the sample multi-modal association.
Preferably, the characteristic selection target function is changed into a convex function form by constructing an augmented lagrange equation:
Figure BDA0002611181550000034
wherein c is a fixed step length parameter of gradient rise, and the step (3) is a constraint function
Figure BDA0002611181550000035
Is equivalent to
Figure BDA0002611181550000036
viIs an equivalent transform factor. λ is a penalty parameter.
Preferably, the augmented lagrange equation is optimized by the ADMM method, the optimization problem being decomposed into βj、
Figure BDA0002611181550000037
Lambda three subproblems, and continuously updating Lagrange multipliers in an iterative mode in the following mode to obtain component values:
Figure BDA0002611181550000038
Figure BDA0002611181550000039
Figure BDA00026111815500000310
wherein argmin is the value of the variable when the following expression reaches the minimum value, and the parameter C is more than or equal to 0.
Preferably, in step (5), each variable in the formula is solved by using an alternating direction multiplier algorithm ADMM to obtain a weight matrix of each mode and a weight of each feature of each sample, and then the multi-mode features of the samples are fused by using a multi-core support vector machine method to classify.
The invention has the advantages that: compared with the prior art, the method constructs a learning model with high classification performance, improves the diagnosis accuracy of the brain diseases, assists in searching biomarkers which are possibly sensitive to the diseases, solves the problems that only single-mode data or only simple cascade multi-mode data is used and the multi-mode data cannot be fully utilized by fully mining and utilizing the prior information of the neuroimage data, mines complementary information among different-mode data, fully considers the difference generated in the acquisition process of the sample and the individual difference of the sample and even the outlier contained in the sample, and selects the characteristic with strong discriminability.
According to the invention, by considering the weight of the samples, the weight of the modes and the weight of the features, not only the importance of the samples among the modes but also the correlation among the modes are considered, more meaningful features are found out, and the accuracy of classification and prediction is improved.
The invention is further described with reference to the drawings and the specific embodiments in the following description.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention.
Detailed Description
Referring to fig. 1, the invention discloses a multi-modal neuroimaging feature selection method based on sample weight and low rank constraint, comprising the following steps:
(1) collecting multi-mode brain image data;
(2) after data of a plurality of modes are obtained, feature selection is carried out in a multi-mode feature collaborative analysis mode, a regression model from each mode data to a classification class target is established, group sparse constraint is carried out on regression vectors, and therefore a public feature subset relevant to all tasks is obtained;
(3) modeling multi-modal data features;
(4) writing the target function into an augmented Lagrange form, wherein the target function is changed into a convex function;
(5) solving each variable of the formula in the step (4) by using an alternating direction multiplier algorithm by using a gradient descent method to obtain a weight matrix W of each mode of each sample and a weight beta of each characteristicj。
(6) And fusing the multi-modal characteristics of the sample by using a method of a multi-core support vector machine for classification.
Preferably, step (1) is performed by performing multiple imaging in multiple modes or multiple imaging in the same mode on the same patient, and obtaining information from several images to perform comprehensive analysis.
It is often clinically desirable to perform multiple modalities or multiple imaging of the same modality on the same patient. I.e. information is obtained from several images at the same time for comprehensive analysis. The single mode imaging uses only one imaging device and can be used for observing lesion growth, comparing treatment effects before and after an operation, and the like. In the invention, multiple imaging modes (MRI, fMRI, PET and the like) are adopted to simultaneously express multiple information of human bodies, and multiple information integration is realized.
Preferably, in step (3), the data of each modality is used
Figure BDA0002611181550000051
Where the superscript j denotes the jth modality, and the subscript i denotes the ith sample; the category to which each sample belongs is used as yiIn this representation, the index i represents the ith sample; inputting the data of each sample and the belonged category into an objective function for feature selection, wherein the objective function for feature selection is as follows:
Figure BDA0002611181550000052
Figure BDA0002611181550000053
wherein n is the number of samples, m is the number of modes, betajA vector is selected for the features of the jth modality,
Figure BDA0002611181550000054
what weight is occupied by the j mode of the ith sample, W is the weight matrix of all modes of the training sample, wherein each row represents a different sample, each column represents a different mode, and lambdaLIs a regularization parameter that constrains the feature sparsity, λRIs a regularization parameter that constrains the sample multi-modal association.
Preferably, the characteristic selection target function is changed into a convex function form by constructing an augmented lagrange equation:
Figure BDA0002611181550000061
wherein c is a fixed step length parameter of gradient rise, and the step (3) is a constraint function
Figure BDA0002611181550000062
Is equivalent to
Figure BDA0002611181550000063
viIs an equivalent transform factor. λ is a penalty parameter.
Preferably, the augmented lagrange equation is optimized by the ADMM method, the optimization problem being decomposed into βj、
Figure BDA0002611181550000064
Lambda three subproblems, and continuously updating Lagrange multipliers in an iterative mode in the following mode to obtain component values:
Figure BDA0002611181550000065
Figure BDA0002611181550000066
Figure BDA0002611181550000067
wherein argmin is the value of the variable when the following expression reaches the minimum value, and the parameter C is more than or equal to 0.
Preferably, in step (5), each variable in the formula is solved by using an alternating direction multiplier algorithm ADMM to obtain a weight matrix of each mode and a weight of each feature of each sample, and then the multi-mode features of the samples are fused by using a multi-core support vector machine method to classify.
Aiming at the characteristics that typical brain diseases have multi-modal data such as biology, physiology and the like, the invention explores a method for efficiently processing and understanding the abundant, various and complex data. On the basis of meaningful measurement in a brain feeling region or a brain network through feature extraction, efficient feature selection is performed, irrelevant and redundant features are filtered out to reduce the dimension of the features so as to facilitate subsequent classification, biomarkers which are possibly sensitive to diseases are searched for, automatic diagnosis and classification of individuals are completed, pathology is better explained, and key support is performed for realizing typical intelligent learning.
The method solves the problem that individual differences and characteristics contain noise, considers the weight of the samples, the weight of the modes and the weight of the characteristics, considers the importance of the samples among the modes and the correlation among the modes, finds out more meaningful characteristics, and improves the accuracy of classification and prediction.
The above embodiments are described in detail for the purpose of further illustrating the present invention and should not be construed as limiting the scope of the present invention, and the skilled engineer can make insubstantial modifications and variations of the present invention based on the above disclosure.

Claims (6)

1. A multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint is characterized in that: the method comprises the following steps:
(1) collecting multi-mode brain image data;
(2) after data of a plurality of modes are obtained, feature selection is carried out in a multi-mode feature collaborative analysis mode, a regression model from each mode data to a classification class target is established, group sparse constraint is carried out on regression vectors, and therefore a public feature subset relevant to all tasks is obtained;
(3) modeling multi-modal data features;
(4) writing the target function into an augmented Lagrange form, wherein the target function is changed into a convex function;
(5) solving each variable of the formula in the step (4) by using an alternating direction multiplier algorithm by using a gradient descent method to obtain a weight matrix of each mode and a weight of each characteristic of each sample;
(6) and fusing the multi-modal characteristics of the sample by using a method of a multi-core support vector machine for classification.
2. The method of claim 1, wherein the method for multi-modal neuroimaging feature selection based on sample weights and low rank constraints comprises: step (1), multiple modes or multiple times of imaging of the same mode are carried out on the same patient, and information is obtained from several images at the same time for comprehensive analysis.
3. The method of claim 2, wherein the method for multi-modal neuroimaging feature selection based on sample weights and low rank constraints comprises: step (3) of using data of each modality
Figure FDA0002611181540000014
Where the superscript j denotes the jth modality, and the subscript i denotes the ith sample; the category to which each sample belongs is used as yiIn this representation, the index i represents the ith sample; inputting the data of each sample and the belonged category into an objective function for feature selection, wherein the objective function for feature selection is as follows:
Figure FDA0002611181540000011
Figure FDA0002611181540000012
wherein n is the number of samples, m is the number of modes, betajA vector is selected for the features of the jth modality,
Figure FDA0002611181540000013
what weight is occupied by the j mode of the ith sample, W is the weight matrix of all modes of the training sample, wherein each row represents a different sample, each column represents a different mode, and lambdaLIs a regularization parameter that constrains the feature sparsity, λRIs a regularization parameter that constrains the sample multi-modal association.
4. The method of claim 3, wherein the method for multi-modal neuroimaging feature selection based on sample weights and low rank constraints comprises: the characteristic selection target function is changed into a convex function form by constructing an augmented Lagrange equation:
Figure FDA0002611181540000021
wherein c is a fixed step length parameter of gradient rise, and the step (3) is a constraint function
Figure FDA0002611181540000022
Is equivalent to
Figure FDA0002611181540000023
viIs an equivalent transform factor. λ is a penalty parameter.
5. The method of claim 4, wherein the method for multi-modal neuroimaging feature selection based on sample weights and low rank constraints comprises: the augmented Lagrange equation is optimized by the ADMM method, and the optimization problem is decomposed into betaj、
Figure FDA0002611181540000024
Lambda three subproblems, and continuously updating Lagrange multipliers in an iterative mode in the following mode to obtain component values:
Figure FDA0002611181540000025
Figure FDA0002611181540000026
Figure FDA0002611181540000027
wherein argmin is the value of the variable when the following expression reaches the minimum value, and the parameter C is more than or equal to 0.
6. The method of claim 5, wherein the method for multi-modal neuroimaging feature selection based on sample weights and low rank constraints comprises: and (5) solving each variable in the formula by using an alternative direction multiplier Algorithm (ADMM) to obtain a weight matrix of each mode and a weight of each feature of each sample, and then fusing multi-mode features of the samples for classification by using a multi-core support vector machine method.
CN202010754793.1A 2020-07-31 2020-07-31 Multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint Pending CN111916162A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010754793.1A CN111916162A (en) 2020-07-31 2020-07-31 Multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010754793.1A CN111916162A (en) 2020-07-31 2020-07-31 Multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint

Publications (1)

Publication Number Publication Date
CN111916162A true CN111916162A (en) 2020-11-10

Family

ID=73286477

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010754793.1A Pending CN111916162A (en) 2020-07-31 2020-07-31 Multi-modal neuroimaging feature selection method based on sample weight and low-rank constraint

Country Status (1)

Country Link
CN (1) CN111916162A (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104299216A (en) * 2014-10-22 2015-01-21 北京航空航天大学 Multimodality medical image fusion method based on multiscale anisotropic decomposition and low rank analysis
CN109770932A (en) * 2019-02-21 2019-05-21 河北工业大学 The processing method of multi-modal brain neuroblastoma image feature

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104299216A (en) * 2014-10-22 2015-01-21 北京航空航天大学 Multimodality medical image fusion method based on multiscale anisotropic decomposition and low rank analysis
CN109770932A (en) * 2019-02-21 2019-05-21 河北工业大学 The processing method of multi-modal brain neuroblastoma image feature

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
程波;朱丙丽;熊江;: "基于多模态多标记迁移学习的早期阿尔茨海默病诊断", 计算机应用, no. 08, pages 2283 - 2286 *

Similar Documents

Publication Publication Date Title
Mahmud et al. Deep learning in mining biological data
CN113040715B (en) Human brain function network classification method based on convolutional neural network
Liu et al. EEG emotion recognition based on the attention mechanism and pre-trained convolution capsule network
Khojaste-Sarakhsi et al. Deep learning for Alzheimer's disease diagnosis: A survey
Feng et al. MRI radiomics classification and prediction in Alzheimer’s disease and mild cognitive impairment: a review
CN109528197B (en) Individual prediction method and system for mental diseases based on brain function map
Cao et al. Sparse shared structure based multi-task learning for MRI based cognitive performance prediction of Alzheimer’s disease
CN113421652A (en) Method for analyzing medical data, method for training model and analyzer
Luo et al. Retinal image classification by self-supervised fuzzy clustering network
CN115131642B (en) Multi-modal medical data fusion system based on multi-view subspace clustering
Bedel et al. BolT: Fused window transformers for fMRI time series analysis
CN116503680B (en) Brain image structured analysis and brain disease classification system based on brain atlas
CN112674720A (en) Alzheimer disease pre-diagnosis method based on 3D convolutional neural network
CN111540467B (en) Schizophrenia classification identification method, operation control device and medical equipment
Bakkouri et al. BG-3DM2F: Bidirectional gated 3D multi-scale feature fusion for Alzheimer’s disease diagnosis
Laghari et al. How to collect and interpret medical pictures captured in highly challenging environments that range from nanoscale to hyperspectral imaging
Hacking et al. Deep learning for the classification of medical kidney disease: a pilot study for electron microscopy
Yang et al. Data augmentation for depression detection using skeleton-based gait information
Duan et al. Cross task neural architecture search for EEG signal recognition
Wu et al. A state-of-the-art survey of U-Net in microscopic image analysis: From simple usage to structure mortification
CN113255734A (en) Depression classification method based on self-supervision learning and transfer learning
Boorboor et al. NeuRegenerate: A framework for visualizing neurodegeneration
Bi et al. Explainable and programmable hypergraph convolutional network for imaging genetics data fusion
Holderrieth et al. Transfer Learning for Neuroimaging via Re-use of Deep Neural Network Features
Goutham et al. Brain tumor classification using Efficientnet-B0 model

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