CN116665906B - Resting state functional magnetic resonance brain age prediction method based on similarity twin network - Google Patents

Resting state functional magnetic resonance brain age prediction method based on similarity twin network Download PDF

Info

Publication number
CN116665906B
CN116665906B CN202310601158.3A CN202310601158A CN116665906B CN 116665906 B CN116665906 B CN 116665906B CN 202310601158 A CN202310601158 A CN 202310601158A CN 116665906 B CN116665906 B CN 116665906B
Authority
CN
China
Prior art keywords
similarity
network
age
brain
neural network
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
Application number
CN202310601158.3A
Other languages
Chinese (zh)
Other versions
CN116665906A (en
Inventor
丁卫平
周天奕
沈鑫杰
黄嘉爽
鞠恒荣
姜舒
程纯
陈悦鹏
孙颖
尹涛
侯涛
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.)
Nantong University
Original Assignee
Nantong 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 Nantong University filed Critical Nantong University
Priority to CN202310601158.3A priority Critical patent/CN116665906B/en
Publication of CN116665906A publication Critical patent/CN116665906A/en
Application granted granted Critical
Publication of CN116665906B publication Critical patent/CN116665906B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves 
    • A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
    • A61B5/004—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part
    • A61B5/0042—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part for the brain
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7271—Specific aspects of physiological measurement analysis
    • A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • 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
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2576/00—Medical imaging apparatus involving image processing or analysis
    • A61B2576/02—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part
    • A61B2576/026—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part for the brain
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • General Physics & Mathematics (AREA)
  • Public Health (AREA)
  • Pathology (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Surgery (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Radiology & Medical Imaging (AREA)
  • Databases & Information Systems (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Computational Linguistics (AREA)
  • Psychiatry (AREA)
  • Physiology (AREA)
  • Mathematical Analysis (AREA)
  • Signal Processing (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)

Abstract

The invention provides a resting-state functional magnetic resonance brain age prediction method based on a similarity twin network, belongs to the technical field of intelligent diagnosis of medical images, and solves the technical problems of insufficient accuracy and stability in the traditional brain age prediction method. The technical proposal is as follows: the method comprises the following steps: s1: collecting tested functional magnetic resonance imaging rs-fMRI data; s2: constructing a twin neural network; s3: designing a feature similarity and tag similarity measurement module; s4: defining a confidence evaluation brain age prediction module; s5: the brain image data in the test data set is input into the model for analysis, thereby obtaining the predicted brain age of each test data sample. The beneficial effects of the invention are as follows: the prediction accuracy is high, accurate prediction is carried out on brain image data, and doctors are helped to evaluate the brain age of patients more accurately.

Description

Resting state functional magnetic resonance brain age prediction method based on similarity twin network
Technical Field
The invention relates to the technical field of intelligent diagnosis of medical images, in particular to a resting state functional magnetic resonance brain age prediction method based on a similarity twin network.
Background
With age, the cognitive ability of human beings is reduced, and different parts of the brain show nonlinear relations in the aging process, so that the degree of brain aging of a patient suffering from wounds is more uncertain. With the advent of the concept of brain age, research into brain age has also increased exponentially, and in addition to predicting brain age, researchers have explored different brain aging patterns, and predictive applications related to cognitive impairment, mortality, cardiovascular disease, etc., for clinical diagnosis. As humans age, the risk of morbidity and mortality increases. On a macroscopic level, this can lead to physiological changes such as ventricle enlargement. At the microscopic level, mitochondrial changes and the like can occur. Thus, brain age studies are important for developing clinically useful biological models that can be used to predict brain health status in aging or disease.
The traditional brain age prediction method based on feature engineering has some technical defects. Conventional methods typically use feature engineering-based methods to extract features of the input image, which require human selection and design of feature extractors, are difficult to process high-dimensional complex data, and are susceptible to noise and interference, which limits the performance of conventional methods to be unstable and consistent results cannot be obtained across a variety of different data sets. Secondly, the influence of a neural network structure is ignored by the traditional method, in the neural network, the weight and deviation of each layer are key factors of network learning, and different neural network structures have different learning capacities and expression capacities, so that the selection of a proper neural network structure is crucial to brain age prediction tasks.
Disclosure of Invention
The invention aims to provide a resting state functional magnetic resonance brain age prediction method based on a similarity twin network, which aims to improve the accuracy and generalization capability of brain age prediction and has wide application prospects in brain science research and clinical medicine. The twin neural network can automatically learn the characteristics of the input images and can realize the brain age prediction task by comparing the similarity of the input images. The similarity twin neural network can also process multi-mode input data and has good generalization performance.
In order to achieve the above object, the present invention adopts the following technical scheme: a resting state functional magnetic resonance brain age prediction method based on a similarity twin network comprises the following steps:
s1: acquiring functional magnetic resonance imaging data to be tested to form an original sample set, wherein a sample of the original sample set comprises resting state functional magnetic resonance imaging to be tested and corresponding actual age information thereof, preprocessing, performing operations such as slicing timing, head movement correlation and the like on the resting state functional magnetic resonance imaging, converting the preprocessed data into three-dimensional image data, and then, performing one-to-one correspondence on the preprocessed image data and the corresponding actual ages to be tested to form the sample set, wherein the sample set comprises a plurality of groups of image data and corresponding actual age information thereof, and finally, dividing the sample set into a training sample set and a testing sample set;
s2: constructing a twin neural network, using a double convolutional neural network based on a full convolutional neural network as a branch network, and modifying the branch network so that the network of two paths shares parameters, wherein the network is composed of two parts: convolutional neural network trunks for extracting depth features from a pair of input images and trunks for fusing the extracted depth features to similarity metrics
S3: the design feature similarity and label similarity measurement module is recorded asThe known rs-fMRI is entered in the first pass,the characteristic information set Z is obtained through convolution and collection x Inputting unknown rs-fMRI in the second path, and obtaining the unknown characteristic vector set Z through the same operation y Wherein rs-fMRI represents resting-state functional magnetic resonance imaging, then the obtained feature pairs describe similarity by adopting a cosine distance CS (), a transformation matrix is learned through training data, finally the contrast loss is used as a loss function to train similarity learning, and the similarity loss function is optimized;
s4: defining a confidence evaluation brain age prediction module, selecting three groups of known brain age labels with highest similarity, and predicting brain ages p=mean (p 1 ,p 2 ,p 3 ) Is the average of three groups of known brain ages with highest similarity, where p is the predicted brain age, p 1 ,p 2 ,p 3 The mean () is the average value of the known brain age tags with the highest three groups of similarity;
s5: the brain image data in the test data set is input into the model for analysis, thereby obtaining the predicted brain age of each test data sample.
As a further optimization scheme of the resting-state functional magnetic resonance brain age prediction method based on the similarity twin network, the specific steps of the step S2 are as follows:
step S2.1: respectively sending the rs-fMRI image 1 and the rs-fMRI image 2 into a twin convolution neural network simultaneously to respectively obtain first characteristic information Z x Second channel characteristic information Z y ；
Step S2.2: the input image is subjected to operations such as convolution and pooling, the dual-path convolution fusion network architecture adopts parameter sharing, two convolution neural networks are built, the channel number of each convolution neural network is C,2C,4C,8C,
4C, wherein C is the initial number of channels;
step S2.3: the network extraction characteristic part structure consists of six modules, wherein the first five modules are identical and sequentially comprise a convolution layer, a standardization layer, a convolution layer and a nonlinear activation function layer, a dropout layer is added into the fifth module and the sixth module to solve the fitting problem, and the sixth module comprises a convolution layer with a convolution kernel size of 1;
step S2.3: results E (x) i ,y i )
Where phi () is the feature extraction network,for similarity measurement module, b is bias, x i As the first line characteristic tensor, y i For the second channel of feature tensor, i is the ith feature vector;
step S2.4: using contrast loss function L (Y, x i ,y i ) Evaluating the effect of a twin neural network to distinguish a given plurality of images, if there is dissimilarity between pairs of samples entered, their distance in the feature space will become smaller, which will lead to an increase in the loss value:
wherein,d is the similarity distance between the twin neural network outputs, P is the feature dimension of the input samples, Y is the label of whether the input samples are similar, y=1 represents that the input samples are relatively similar, y=0 represents that the input samples are not similar, m is a set threshold, and N is the number of data.
As a further optimization scheme of the resting-state functional magnetic resonance brain age prediction method based on the similarity twin network, the specific steps of the step S3 are as follows:
step S3.1: given a 4D first pass feature tensor Z extracted from an input image pair x And a second trace feature tensor Z y Wherein Z is x ＝{x 1 ,x 2 ,...,x i ,...,x n Sum Z y ＝{y 1 ,y 2 ,...,y i ,...,y n The number of the rs-fMRI images in the training data set is n, and the first age set of the rs-MRI images in the training sample is L x ＝{x' 1 ,x' 2 ,...x' n Second age set L y ＝{y' 1 ,y' 2 ,...y' n -x 'where' i ,y' i Respectively the first-pass characteristic tensor x i Corresponding age and second road characteristic tensor y i Age corresponding to the age;
step S3.2: the resulting features are divided into two groups, i.e. feature vectors of the training set, as follows, where (x 1 ,y 1 ) Representing a sample set:
step S3.3: the similarity of each group of images is described by cosine distance CS (x i ,y i A) is given by:
wherein the superscript T denotes the transpose of the matrix, A is a linear transformation matrix that computes cosine similarity in the transformed subspace, while similarity cosine distances CS (x 'for each group of ages' i ,y' i A) is given by:
the cosine similarity f (a) between vectors is defined as follows:
where α and β are given parameters for A, α is used to balance the contribution of positive and negative samples to the edge, β controls the maximization threshold, A 0 Is a predefined matrix, ||A-A 0 I represents A and A 0 The distance between the two is N, and N is the number of data;
step S3.4: dividing the objective function into two terms g (A) and h (A):
h(A)＝β||A-A 0 || 2 (8)
g (A) is a simple voting scheme for each sample, the value of alpha is determined by cross-validation, h (A) is regularizing the matrix A as close as possible to the predefined matrix A 0 ；
Step S3.5: the similarity CS (x i ,y i A), and the similarity CS (x 'of the module sample labels at the similarity measure' i ,y' i And A) fusing, namely calculating and back-propagating the loss function according to the similarity between the output characteristics and the similarity f (A) between the sample labels so as to optimize the network, and obtaining the optimal network model.
Compared with the prior art, the invention has the beneficial effects that:
1. automatic learning characteristics: the similarity twin neural network can automatically learn the characteristics of the input image through training, and a manual selection and design of a characteristic extraction module is not needed, so that the automatic characteristic learning can more accurately capture the information in the image, and the prediction accuracy is improved; in addition, the similarity twin neural network can process high-dimensional data, capture more detail information, improve the expression capacity of the neural network on complex data, and is more suitable for tasks needing complex modeling on the data.
2. Processing multi-modal data: the similarity twin neural network can process multi-mode input data simultaneously, multiple information is fused to improve prediction accuracy, and for brain age prediction tasks, the similarity twin neural network can use structural and functional magnetic resonance imaging data simultaneously, and multiple information is fused to improve prediction accuracy.
3. The prediction accuracy is improved: the similarity twin neural network realizes the brain age prediction task by comparing the similarity of the input images, and the method has more accurate processing mode of the image data and can better predict the brain age of the tested person; compared with the traditional method, the similarity twin neural network does not need to manually design a feature extractor, so that uncertainty caused by manually selecting parameters and a feature extraction method is avoided, and the prediction effect is more accurate.
4. The generalization capability is strong: the similarity twin neural network has good generalization performance, can process high-dimensional complex data, and has good robustness to noise and interference; the similarity twin neural network can learn a general rule from the sample, so that the similarity twin neural network has better generalization capability for the unseen sample, and the overfitting phenomenon is effectively avoided; this allows the similarity twin neural network to be adapted to process a wider range of data.
Drawings
The accompanying drawings are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate the invention and do not limit it.
Fig. 1 is an overall framework diagram of a resting state functional magnetic resonance brain age prediction method based on a similarity twin network.
Fig. 2 is a diagram of a model structure of a similarity twin convolutional neural network based on a resting state functional magnetic resonance brain age prediction method of the similarity twin network.
Fig. 3 is a block diagram of brain age prediction for a resting state functional magnetic resonance brain age prediction method based on a similarity twin network according to the present invention.
Fig. 4 is a block diagram of a twin neural network based on a resting state functional magnetic resonance brain age prediction method of a similarity twin network according to the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. Of course, the specific embodiments described herein are for purposes of illustration only and are not intended to limit the invention.
Example 1
Referring to fig. 1 to 4, the present embodiment provides a method for predicting brain age by resting state functional magnetic resonance based on a similarity twin network, wherein a first pair of pictures x in a dataset 1 ，y 1 For example, the classification result is obtained from the loading of data, and the method comprises the following steps:
s1: acquiring functional magnetic resonance imaging data to be tested to form an original sample set, wherein a sample of the original sample set comprises functional magnetic resonance image data to be tested and corresponding actual age information thereof, preprocessing, performing operations such as slicing timing, head movement correlation and the like on a resting state magnetic resonance image, converting the preprocessed data into three-dimensional image data, and then, performing one-to-one correspondence on the preprocessed image data and corresponding actual ages to be tested to form the sample set, wherein the sample set comprises a plurality of groups of image data and corresponding actual age information thereof, and finally dividing the sample set into a training sample set and a testing sample set;
s2: constructing a twin neural network, using a double convolutional neural network based on a full convolutional neural network as a branch network, and modifying the branch network so that the network of two paths shares parameters, wherein the network is composed of two parts: convolutional neural network trunks for extracting depth features from a pair of input images and trunks for fusing the extracted depth features to similarity metrics
S3: design feature similarity and label similarity measurement moduleInputting a group of known rs-fMRI in the first path, and obtaining characteristic information Z through convolution and collection x Inputting a group of unknowns in the second pathrs-fMRI, the unknown characteristic vector Z is obtained through the same operation y Wherein rs-fMRI represents resting-state functional magnetic resonance imaging, then the obtained feature pairs describe similarity by adopting a cosine distance CS (), a transformation matrix is learned through training data, finally the contrast loss is used as a loss function to train similarity learning, and the similarity loss function is optimized;
s4: defining a confidence evaluation brain age prediction module, selecting three groups of known brain age labels with highest similarity, and predicting brain ages p=mean (p 1 ,p 2 ,p 3 ) Is the average of three groups of known brain ages with highest similarity, where p is the predicted brain age, p 1 ,p 2 ,p 3 The mean () is the average value of the known brain age tags with the highest three groups of similarity;
s5: the brain image data in the test data set is input into the model for analysis, thereby obtaining the predicted brain age of each test data sample.
Specifically, the specific steps of the step S2 are as follows:
step S2.1: respectively sending the rs-fMRI image 1 and the rs-fMRI image 2 into a twin convolution neural network simultaneously to respectively obtain first characteristic information Z x Second channel characteristic information Z y ；
Step S2.2: the input image is subjected to operations such as convolution and pooling, the two-way convolution fusion network architecture adopts parameter sharing, two convolution neural networks are built, and the channel number of each convolution neural network is 16, 32, 64, 128 and 128;
step S2.3: the network extraction characteristic part structure consists of six modules, wherein the first five modules are identical and sequentially comprise a convolution layer, a standardization layer, a convolution layer and a nonlinear activation function layer, a dropout layer is added into the fifth module and the sixth module to solve the fitting problem, and the sixth module comprises a convolution layer with a convolution kernel size of 1;
step S2.3: results E (x) 1 ,y 1 ) For calculating instance x 1 And y 1 Distance between:
where phi () is the feature extraction network,for similarity measurement module, b is bias, x 1 The 1 st eigenvector, y, being the first path eigenvector 1 The 1 st eigenvector, x, being the second channel eigenvector 1 And y 1 The middle part of (2) is as follows:
step S2.4: using contrast loss function L (Y, x 1 ,y 1 ) Evaluating the effect of a twin neural network to distinguish a given plurality of images, if there is dissimilarity between pairs of samples entered, their distance in the feature space will become smaller, which will lead to an increase in the loss value:
wherein,d is the similarity distance between the twin neural network outputs, P is the feature dimension of the input samples, where p=4, Y is the label of whether the input samples are similar, where the input samples are relatively similar, so y=1, i.e. the method comprises:
where m is a set threshold, here set to 0.5d, and n is the number of data.
As the resting state functional magnetic resonance brain age prediction method based on the similarity twin network provided in this embodiment, the specific steps of step S3 are as follows:
step S3.1: given a 4D first pass feature tensor Z extracted from an input image pair x And a second trace feature tensor Z y Wherein Z is x ＝{x 1 ,x 2 ,...,x i ,...,x n Sum Z y ＝{y 1 ,y 2 ,...,y i ,...,y n The number of the rs-fMRI images in the training data set is n, and the first age set of the rs-MRI images in the training sample is L x ＝{x' 1 ,x' 2 ,...x' n Second age set L y ＝{y' 1 ,y' 2 ,...y' n -x 'where' 1 ,y' 1 The first and second ages, here 45 and 48, respectively;
step S3.2: the obtained features are divided into two groups, namely, the feature vectors of the training set are as follows:
step S3.3: the similarity of each group of images is described by cosine distance CS (x 1 ,y 1 A) is given by:
wherein, the superscript T represents the transposition of the matrix, A is a linear transformation matrix for calculating cosine similarity in the subspace after transformation, and the matrix is introduced and subjected to matrix operation to obtain a result:
at the same time, the similarity cosine distance CS (x' i ,y' i A) is given by:
the final result CS (x' 1 ,y' 1 A) =0.786, and cosine similarity f (a) between vectors is defined as follows:
where α and β are given parameters for A, α is used to balance the contribution of positive and negative samples to the edge, β controls the maximization threshold, set to 0.1, A 0 Is a predefined matrix, ||A-A 0 I represents A and A 0 The distance between the two functions is calculated by matrix summation of a plurality of rounds, and the similarity f (A) =0.805 is finally obtained;
step S3.4: dividing the objective function into two terms g (A) and h (A):
h(A)＝β||A-A 0 || 2 (20)
g (A) is a simple voting scheme for each sample, the value of alpha is determined by cross-validation, the value of alpha is obtained by the first round of calculation to be 0.2, and h (A) is obtained by regularizing the matrix A to be as close to the predefined matrix A as possible 0 ；
Step S3.5: the similarity CS (x 1 ,y 1 A), and the similarity CS (x 'of the module sample labels at the similarity measure' 1 ,y' 1 A) fusion, followed by computation and back-propagation of the loss function from the similarity between the features of the output and the similarity f (a) between the sample tagsAnd broadcasting to optimize the network to obtain an optimal network model.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.

Claims (1)

1. The resting state functional magnetic resonance brain age prediction method based on the similarity twin network is characterized by comprising the following steps of:
s1: acquiring functional magnetic resonance imaging data to be tested to form an original sample set, wherein a sample of the original sample set comprises a tested resting state functional magnetic resonance imaging rs-fMRI and corresponding actual age information thereof, then preprocessing, performing slicing timing and head movement correlation operation on the resting state functional magnetic resonance imaging rs-fMRI, converting the preprocessed data into three-dimensional image data, then, performing one-to-one correspondence on the preprocessed image data and corresponding tested actual ages to form the sample set, wherein the sample set comprises a plurality of groups of image data and corresponding actual age information thereof, and finally dividing the sample set into a training sample set and a testing sample set;
s2: constructing a twin neural network model, using a double convolutional neural network based on a full convolutional neural network as a branch network, and modifying the branch network so that the network of two paths shares parameters, wherein the network consists of two parts: convolutional neural network trunks for extracting depth features from a pair of input images and trunks for fusing the extracted depth features to similarity metrics
S3: the design feature similarity and label similarity measurement module is recorded asInputting known rs-fMRI in the first path, and obtaining a characteristic information set Z through convolution and collection x Inputting unknown rs-fMRI in the second path, and obtaining the unknown characteristic vector set Z through the same operation y Wherein rs-fMRI represents resting-state functional magnetic resonance imaging, then the obtained feature pairs describe similarity by adopting a cosine distance CS (), a transformation matrix is learned through training data, finally the contrast loss is used as a loss function to train similarity learning, and the similarity loss function is optimized;
s4: defining a confidence evaluation brain age prediction module, selecting three groups of known brain age labels with highest similarity, and predicting brain ages p=mean (p 1 ,p 2 ,p 3 ) Is the average of three groups of known brain ages with highest similarity, where p is the predicted brain age, p 1 ,p 2 ,p 3 The mean () is the average value of the known brain age tags with the highest three groups of similarity;
s5: inputting brain image data in the test data set into a twin neural network model for analysis, so as to obtain the predicted brain age of each test data sample;
the specific steps of the step S2 are as follows:
step S2.1: respectively sending the rs-fMRI image 1 and the rs-fMRI image 2 into a twin convolution neural network simultaneously to respectively obtain first characteristic information Z x Second channel characteristic information Z y ；
Step S2.2: the input image is subjected to operations such as convolution and pooling, the double-path convolution fusion network architecture adopts parameter sharing, two convolution neural networks are built, and the channel number of each convolution neural network is C,2C,4C,8C and 4C, wherein C is the initial channel number;
step S2.3: the network extraction characteristic part structure consists of six modules, wherein the first five modules are identical and sequentially comprise a convolution layer, a standardization layer, a convolution layer and a nonlinear activation function layer, a dropout layer is added into the fifth module and the sixth module to solve the fitting problem, and the sixth module comprises a convolution layer with a convolution kernel size of 1;
step S2.3: results E (x) i ,y i )
Where phi () is the feature extraction network,for similarity measurement module, b is bias, x i As the first line characteristic tensor, y i For the second channel of feature tensor, i is the ith feature vector;
step S2.4: using contrast loss function L (Y, x i ,y i ) Evaluating the effect of a twin neural network to distinguish a given plurality of images, if there is dissimilarity between pairs of samples entered, their distance in the feature space will become smaller, which will lead to an increase in the loss value:
wherein,d is the similarity distance between the twin neural network outputs, P is the feature dimension of the input samples, Y is the label of whether the input samples are similar, y=1 represents that the input samples are similar, y=0 represents that the input samples are dissimilar, m is a set threshold value, and N is the number of data;
the specific steps of the step S3 are as follows:
step S3.1: given a 4D first pass feature tensor Z extracted from an input image pair x And a second trace feature tensor Z y Wherein Z is x ＝{x 1 ,x 2 ,...,x i ,...,x n Sum Z y ＝{y 1 ,y 2 ,...,y i ,...,y n The number of the rs-fMRI images in the training data set is n, and the first year of the rs-fMRI images in the training sampleAge set L x ＝{x' 1 ,x' 2 ,...x' n Second age set L y ＝{y' 1 ,y' 2 ,...y' n -x 'where' i ,y' i Respectively the first-pass characteristic tensor x i Corresponding age and second road characteristic tensor y i Age corresponding to the age;
step S3.2: the resulting features are divided into two groups, i.e. feature vectors of the training set, as follows, where (x 1 ,y 1 ) Representing a sample set:
step S3.3: the similarity of each group of images is described by cosine distance CS (x i ,y i A) is given by:
wherein the superscript T denotes the transpose of the matrix, A is a linear transformation matrix that computes cosine similarity in the transformed subspace, while similarity cosine distances CS (x 'for each group of ages' i ,y' i A) is given by:
the cosine similarity f (a) between vectors is defined as follows:
where α and β are given parameters for A, α is used to balance the contribution of positive and negative samples to the edge, β controls the maximization threshold, A 0 Is a predefined matrix, ||A-A 0 Representation ofA and A 0 The distance between the two is N, and N is the number of data;
step S3.4: dividing the objective function into two terms g (A) and h (A):
h(A)＝β||A-A 0 || 2 (8)
g (A) is a simple voting scheme for each sample, the value of alpha is determined by cross-validation, h (A) is regularizing the matrix A as close as possible to the predefined matrix A 0 ；
Step S3.5: the similarity CS (x i ,y i A), and the similarity CS (x 'of the module sample labels at the similarity measure' i ,y' i And A) fusing, namely calculating and back-propagating the loss function according to the similarity between the output characteristics and the similarity f (A) between the sample labels so as to optimize the network, and obtaining the optimal network model.
CN202310601158.3A 2023-05-25 2023-05-25 Resting state functional magnetic resonance brain age prediction method based on similarity twin network Active CN116665906B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310601158.3A CN116665906B (en) 2023-05-25 2023-05-25 Resting state functional magnetic resonance brain age prediction method based on similarity twin network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310601158.3A CN116665906B (en) 2023-05-25 2023-05-25 Resting state functional magnetic resonance brain age prediction method based on similarity twin network

Publications (2)

Publication Number Publication Date
CN116665906A CN116665906A (en) 2023-08-29
CN116665906B true CN116665906B (en) 2023-12-26

Family

ID=87714680

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310601158.3A Active CN116665906B (en) 2023-05-25 2023-05-25 Resting state functional magnetic resonance brain age prediction method based on similarity twin network

Country Status (1)

Country Link
CN (1) CN116665906B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112784130A (en) * 2021-01-27 2021-05-11 杭州网易云音乐科技有限公司 Twin network model training and measuring method, device, medium and equipment
CN114944229A (en) * 2022-05-23 2022-08-26 中国科学院自动化研究所 Brain age prediction method based on deep learning and magnetic resonance structure brain image
CN115861716A (en) * 2023-02-16 2023-03-28 之江实验室 Glioma classification method and device based on twin neural network and iconomics

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112784130A (en) * 2021-01-27 2021-05-11 杭州网易云音乐科技有限公司 Twin network model training and measuring method, device, medium and equipment
CN114944229A (en) * 2022-05-23 2022-08-26 中国科学院自动化研究所 Brain age prediction method based on deep learning and magnetic resonance structure brain image
CN115861716A (en) * 2023-02-16 2023-03-28 之江实验室 Glioma classification method and device based on twin neural network and iconomics

Also Published As

Publication number Publication date
CN116665906A (en) 2023-08-29

Similar Documents

Publication Publication Date Title
CN113040715B (en) Human brain function network classification method based on convolutional neural network
CN111528859B (en) Child ADHD screening and evaluating system based on multi-modal deep learning technology
CN107133651B (en) The functional magnetic resonance imaging data classification method of subgraph is differentiated based on super-network
CN109544518B (en) Method and system applied to bone maturity assessment
CN111863244B (en) Functional connection mental disease classification method and system based on sparse pooling graph convolution
CN111738363B (en) Alzheimer disease classification method based on improved 3D CNN network
CN113616184A (en) Brain network modeling and individual prediction method based on multi-mode magnetic resonance image
CN111090764B (en) Image classification method and device based on multitask learning and graph convolution neural network
CN110731773B (en) Abnormal electrocardiogram screening method based on fusion of global and local depth features of electrocardiogram
CN111009321A (en) Application method of machine learning classification model in juvenile autism auxiliary diagnosis
CN110598793A (en) Brain function network feature classification method
CN110889496A (en) Human brain effect connection identification method based on confrontation generation network
CN111513717A (en) Method for extracting brain functional state
Hossin et al. Breast cancer detection: an effective comparison of different machine learning algorithms on the Wisconsin dataset
CN116579975A (en) Brain age prediction method and system of convolutional neural network
CN115272295A (en) Dynamic brain function network analysis method and system based on time domain-space domain combined state
CN115496953A (en) Brain network classification method based on space-time graph convolution
CN116612335A (en) Few-sample fine-granularity image classification method based on contrast learning
CN109935321B (en) Risk prediction system for converting depression patient into bipolar affective disorder based on functional nuclear magnetic resonance image data
Ahmad et al. Autism spectrum disorder detection using facial images: A performance comparison of pretrained convolutional neural networks
CN114334162A (en) Intelligent prognosis prediction method and device for disease patient, storage medium and equipment
CN113786185A (en) Static brain network feature extraction method and system based on convolutional neural network
CN116665906B (en) Resting state functional magnetic resonance brain age prediction method based on similarity twin network
Goutham et al. Brain tumor classification using EfficientNet-B0 model
CN115909438A (en) Pain expression recognition system based on depth time-space domain convolutional neural network

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