CN113705680A - Cancer associated depression identification method based on multi-modal magnetic resonance data - Google Patents
Cancer associated depression identification method based on multi-modal magnetic resonance data Download PDFInfo
- Publication number
- CN113705680A CN113705680A CN202110999956.7A CN202110999956A CN113705680A CN 113705680 A CN113705680 A CN 113705680A CN 202110999956 A CN202110999956 A CN 202110999956A CN 113705680 A CN113705680 A CN 113705680A
- Authority
- CN
- China
- Prior art keywords
- network
- node
- nodes
- magnetic resonance
- brain
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 63
- 206010028980 Neoplasm Diseases 0.000 title claims abstract description 42
- 201000011510 cancer Diseases 0.000 title claims abstract description 38
- 210000004556 brain Anatomy 0.000 claims abstract description 75
- 239000010410 layer Substances 0.000 claims abstract description 42
- 239000011229 interlayer Substances 0.000 claims abstract description 32
- 238000002599 functional magnetic resonance imaging Methods 0.000 claims abstract description 28
- 238000002598 diffusion tensor imaging Methods 0.000 claims abstract description 24
- 238000002595 magnetic resonance imaging Methods 0.000 claims abstract description 24
- 230000000877 morphologic effect Effects 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 20
- 230000002739 subcortical effect Effects 0.000 claims abstract description 14
- 230000000284 resting effect Effects 0.000 claims abstract description 11
- 238000012937 correction Methods 0.000 claims description 15
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000007781 pre-processing Methods 0.000 claims description 13
- 238000002790 cross-validation Methods 0.000 claims description 12
- 238000000354 decomposition reaction Methods 0.000 claims description 11
- 210000003128 head Anatomy 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 6
- 238000009499 grossing Methods 0.000 claims description 5
- 210000004761 scalp Anatomy 0.000 claims description 5
- 210000003625 skull Anatomy 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 239000000126 substance Substances 0.000 claims description 3
- 230000002123 temporal effect Effects 0.000 claims description 3
- 208000020016 psychiatric disease Diseases 0.000 abstract description 10
- 238000013145 classification model Methods 0.000 abstract description 2
- 238000011160 research Methods 0.000 description 12
- 230000003993 interaction Effects 0.000 description 10
- 238000005516 engineering process Methods 0.000 description 6
- 210000001320 hippocampus Anatomy 0.000 description 6
- 210000004727 amygdala Anatomy 0.000 description 5
- 238000003745 diagnosis Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000002610 neuroimaging Methods 0.000 description 5
- 238000012706 support-vector machine Methods 0.000 description 5
- 230000004927 fusion Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 208000025174 PANDAS Diseases 0.000 description 3
- 208000021155 Paediatric autoimmune neuropsychiatric disorders associated with streptococcal infection Diseases 0.000 description 3
- 240000004718 Panda Species 0.000 description 3
- 235000016496 Panda oleosa Nutrition 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 239000000090 biomarker Substances 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000013399 early diagnosis Methods 0.000 description 2
- 230000002996 emotional effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000004886 head movement Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000007500 overflow downdraw method Methods 0.000 description 2
- 230000008506 pathogenesis Effects 0.000 description 2
- 230000001575 pathological effect Effects 0.000 description 2
- 238000004393 prognosis Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 244000144725 Amygdalus communis Species 0.000 description 1
- 235000011437 Amygdalus communis Nutrition 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 235000020224 almond Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 210000001175 cerebrospinal fluid Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 208000024714 major depressive disease Diseases 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 210000004885 white matter Anatomy 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
- G06F18/256—Fusion techniques of classification results, e.g. of results related to same input data of results relating to different input data, e.g. multimodal recognition
-
- 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
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- 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
-
- 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
- G16H50/20—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)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The application provides a cancer associated depression identification method based on multi-modal magnetic resonance data, solves the problem of cancer associated depression identification of the multi-modal magnetic resonance data, and improves the identification accuracy. The method comprises the steps of constructing a functional network of the brain by using average time sequence correlation of a resting state functional magnetic resonance imaging (fMRI) brain image data set, constructing a structural network of the brain by using brain image data of Diffusion Tensor Imaging (DTI), and further constructing a multi-layer network, wherein the multi-layer network is a network for establishing interlayer connection of the same nodes in the functional network and the structural network. Further, morphological features of sub-cortical structures are extracted using structural magnetic resonance imaging (srmri), while topological features of the multi-layer network are extracted for classification. And finally, fusing the classification models obtained by training different characteristics. The application provides that a multi-layer network is applied to mental disease identification, hidden relations among brain networks in different modes are mined, and classification accuracy is improved.
Description
Technical Field
The application relates to the technical field of medical auxiliary diagnosis, in particular to a cancer associated depression identification method based on multi-modal magnetic resonance data.
Background
Cancer-associated depression (CRD) refers to the emotional and pathological response of a patient to loss of personal normality, such as cancer diagnosis, treatment, and complications thereof. In recent years, researches show that almost all cancers have pathogenesis factors such as depression, and the cancer-associated depression is closely related to the occurrence, development and prognosis of tumors, but the occurrence mechanism is still not quite clear at present, and reliable and effective biomarkers are not available to help early diagnosis. In the existing domestic and foreign research, the research on cancer-associated depression groups is less, the research field of multi-modal brain imaging is rarely related, and the application of a machine learning method in medical auxiliary diagnosis is not mature.
The multi-modality Magnetic Resonance Imaging (MRI) technology is a flexible combination of various functional MRI technologies based on conventional MRI. Currently, multi-modality MRI used for neurosurgery mainly includes conventional MRI, resting-state functional magnetic resonance imaging (fMRI), Diffusion Tensor Imaging (DTI), magnetic resonance imaging (srmri), perfusion-weighted imaging (PWI), and the like, and the multi-modality MRI technology combined with neuronavigation has become one of important auxiliary tools for neurosurgery.
Currently, the number of studies using neuroimaging for cancer-associated depression populations is small and mostly dependent on a single modality.
Disclosure of Invention
The application provides a cancer associated depression identification method based on multi-modal magnetic resonance data, solves the problem of cancer associated depression identification of the multi-modal magnetic resonance data, and improves the identification accuracy.
The application provides a cancer associated depression identification method based on multi-modal magnetic resonance data, which comprises the following steps: constructing a multilayer network, wherein the multilayer network is stacked by at least a functional network constructed by brain image data of a resting state functional magnetic resonance imaging modality and a structural network constructed by brain image data of a diffusion tensor imaging DTI modality; extracting a plurality of features of the multi-layer network; respectively training to obtain a plurality of models respectively corresponding to the plurality of characteristics according to the plurality of characteristics; obtaining a prediction result of the sample according to the plurality of models; and fusing the plurality of models according to the prediction result.
Therefore, the functional network of the brain is constructed by using the average time sequence correlation of the resting state functional magnetic resonance imaging fMRI brain image data set, the structural network of the brain is constructed by using the brain image data of the diffusion tensor imaging DTI, and then a multilayer network is constructed, wherein the multilayer network is a network for establishing interlayer connection of the same nodes in the functional network and the structural network. A multi-layer network may expose complex, highly interdependent structures between multiple subsystems in a complex system or within a subsystem. In the study of brain images, such complex relationships not only appear as interactions between brain regions, but also appear as interactions and interdependencies between different modalities (structural networks and functional networks). Therefore, the multi-layer network can more accurately model the brain network. According to the method and the system, the multilayer network is applied to a framework of mental disease identification, hidden interaction and interdependence relation among different modal brain networks are mined, and then classification accuracy is improved.
In some embodiments, the method for cancer-associated depression identification based on multi-modality magnetic resonance data further comprises: acquiring morphological characteristics of a subcortical structure in structural magnetic resonance imaging; and training to obtain a morphological feature model corresponding to the morphological feature according to the morphological feature.
In this way, morphological features of sub-cortical structures are extracted by structural magnetic resonance imaging srmri, obtaining a large number of characteristics of hippocampus and amygdala. According to the method and the device, the morphological characteristics are selected, meanwhile, the topological characteristics of the multilayer network are extracted for classification, the information of the MRI in three modes is fused, the characteristics of the MRI in the cortex are considered, and hidden information can be mined to the maximum extent.
In some embodiments, in the step of extracting a plurality of features of the multilayer network, the plurality of features includes a node degree, an inter-layer connection, and a node entropy.
In some embodiments, the multi-layer network G in the method for identifying cancer associated with depression based on multi-modal magnetic resonance data is defined as follows:
wherein A is1Expressed as m by m of the functional network, A2Expressed as m by m of the structural network, H12And H21Representing an interlayer connection matrix, wherein interlayer connection exists between the same nodes, and the interlayer connection value between different nodes is set to be 0;
the interlayer connections between the same nodes are defined as follows:
wherein the content of the first and second substances,
representing the degree of the node j in the matrix corresponding to the ith mode, when i is 1
Is A1Degree of middle node j, i is 2
Is A2The degrees, α and β, of the middle node j are two adjustable parameters.
In some embodiments, the constructing of the multi-layered network in the method for identifying cancer associated with depression based on multi-modal magnetic resonance data further comprises: a singular value decomposition step of performing singular value decomposition on the structural network and the functional network; and an efficiency cost optimization step, wherein efficiency cost optimization is carried out on the structural network and the functional network according to the singular value decomposition result.
In this way, Singular Value Decomposition (SVD) and Efficiency Cost Optimization (ECO) are performed on the structural and functional networks respectively to achieve the purpose of normalization and sparsification, and the ECO enables the densities of the edges of the two networks to be consistent for constructing a multi-layer network.
In some embodiments, the method for cancer-associated depression identification based on multi-modality magnetic resonance data further comprises: calculating a topology attribute of the multilayer network, wherein the topology attribute comprises the node degree and the node entropy;
the node degree is defined as:
wherein f is the number of layers of the network,
representing the node degree of node i in the f-th level, diiRepresents the interlayer connection of the node i;
the entropy of each layer of nodes in G is defined as the entropy of a subgraph formed by all layers of given nodes, adjacent nodes and edges connected with the nodes, and the total entropy of the nodes in G is defined as the sum of the entropy of each layer of nodes of the corresponding nodes;
the node entropy is defined as:
where k, m are any two nodes in the subgraph, and qk，m≠0。
In some embodiments, before the step of segmenting the M, further comprising: pre-processing the brain image data, the pre-processing including removing the first ten time points, temporal layer correction, scalp and skull stripping, head motion correction, spatial normalization, and gaussian smoothing.
In some embodiments, the weights of the edges in the functional network in the method for identifying cancer associated with depression based on multi-modal magnetic resonance data are taken as absolute values, and the edges refer to functional connections of brain regions. The application takes the absolute value of the weight of each side in the network to reserve the negative connection.
In some embodiments, the method for cancer-associated depression identification based on multi-modality magnetic resonance data further comprises: preprocessing the N, the preprocessing including correction of eddy currents and FA calculations.
In some embodiments, the method for cancer-associated depression identification based on multi-modality magnetic resonance data further comprises: performing cross validation on all samples by using external ten folds, and dividing the samples into a training set and a test set; the training set is cross-validated by an inner ten-fold. The present application uses nested ten-fold cross validation. The role of the outer ten-fold cross validation is to divide all samples into training and test sets. The ten-fold cross validation is performed on the training set, the internal ten-fold cross validation has the effect of finding out an optimal set of parameters (such as alpha and beta in the multi-layer network construction and relevant parameters of an SVM classifier) on the training set, and the arrangement ensures that the information of the test set cannot be leaked.
Drawings
Fig. 1 is a schematic flow chart diagram of an embodiment of a method for identifying cancer associated depression based on multi-modal magnetic resonance data provided herein;
fig. 2 is a schematic flow chart of an embodiment of constructing a multi-layered network in the method for identifying cancer associated with depression based on multi-modal magnetic resonance data provided in the present application;
fig. 3 is a schematic flow chart of an example of brain image data preprocessing of the fMRI modality provided herein;
fig. 4 is a schematic flowchart of an example of brain image data preprocessing in the DTI modality provided in the present application.
Detailed Description
Cancer-associated depression (CRD) refers to the emotional and pathological response of a patient to loss of personal normality, such as cancer diagnosis, treatment, and complications thereof. In recent years, researches show that almost all cancers have pathogenesis factors such as depression, and the cancer-associated depression is closely related to the occurrence, development and prognosis of tumors, but the occurrence mechanism is still not quite clear at present, and reliable and effective biomarkers are not available to help early diagnosis. In the existing domestic and foreign research, the research on cancer-associated depression groups is less, the research field of multi-modal brain imaging is rarely related, and the application of a machine learning method in medical auxiliary diagnosis is not mature.
The multi-modality Magnetic Resonance Imaging (MRI) technology is a flexible combination of various functional MRI technologies based on conventional MRI. Currently, multi-modality MRI used for neurosurgery mainly includes conventional MRI, resting state functional magnetic resonance imaging (fMRI), Diffusion Tensor Imaging (DTI), magnetic resonance imaging (srmri), perfusion-weighted imaging (PWI), and the like, and the multi-modality MRI technology combined with neuronavigation has become one of important auxiliary tools for neurosurgery. The multi-modal brain image is brain image data using two modalities, fMRI and DTI.
Currently, the number of studies using neuroimaging for cancer-associated depression populations is small and mostly dependent on a single modality. In recent years, the development of neuroscience has been greatly promoted by multi-modal neuroimaging research. On one hand, hidden information can be mined to the maximum extent by combining data of different modes, and on the other hand, cross validation can be performed between data of different modes, so that the reliability of results is improved.
The method combines multi-modal magnetic resonance image data, quantifies the characteristics of the cancer associated depression brain images, classifies the cancer associated depression patients and normal control groups by using a machine learning method, excavates the relation between hidden interaction and dependence between brain networks in different modes, and improves the classification accuracy.
The method for identifying cancer associated with depression based on multi-modal magnetic resonance data provided by the present application is described below with reference to the accompanying drawings.
Fig. 1 is a schematic flow chart diagram of an embodiment of a method for identifying cancer associated depression based on multi-modal magnetic resonance data provided herein. As shown in figure 1 of the drawings, in which,
s1, constructing a multilayer network, wherein the multilayer network is formed by stacking at least a functional network constructed by brain image data of a resting state functional magnetic resonance imaging modality and a structural network constructed by brain image data of a diffusion tensor imaging DTI modality;
in one implementation, constructing a multi-layer network includes the steps of: building a functional network, building a structural network, and stacking the functional network and the structural network.
Specific steps for constructing the functional network, constructing the structural network, and stacking the functional network and the structural structure are described below with reference to the drawings.
Fig. 2 is a schematic flow chart of an embodiment of constructing a multi-layer network in the method for identifying cancer associated with depression based on multi-modal magnetic resonance data provided in the present application. As shown in figure 2 of the drawings, in which,
s101, obtaining a brain image data set M of a resting state functional magnetic resonance imaging (fMRI) mode, wherein the resting state functional magnetic resonance imaging (fMRI) mode corresponds to M brain areas in a one-to-one mode, and samplesi(i is more than or equal to 1 and less than or equal to m) and a brain image data set N in diffusion tensor imaging DTI modei(1≤i≤m)。
Brain image data of a resting state functional magnetic resonance imaging (fMRI) mode and brain image data of a Diffusion Tensor Imaging (DTI) mode of a sample are collected, the brain image data of the fMRI mode and the brain image data of the DTI mode have the same collection parameters, a set of the brain image data of the fMRI mode is recorded as a set M, and a set of the brain image data of the DTI mode is recorded as a set N.
In one implementation, the brain is segmented into 90 brain regions, i.e., m 90, according to AAL90 (atomic Automatic Labeling template issued by montreal nerve) template. Dividing the M to obtain a subset M of different brain regions corresponding to the Mi(i is more than or equal to 1 and less than or equal to 90); dividing the N to obtain the N corresponding to differentSubset N of brain regionsi(1≤i≤m)。
Fig. 3 is a schematic flow chart of an example of brain image data preprocessing of the fMRI modality provided in the present application. As shown in fig. 3, in an example, before the step of dividing M, the method further includes: pre-processing the M, including removing the first ten time points, temporal layer correction, scalping scalp and skull, head motion correction, spatial normalization, and gaussian smoothing.
Preferably, the DPARSF tool kit based on SPM12 preprocesses the brain image data in M, and during the actual image acquisition, the initial equipment may be in an unstable state, so in order to avoid the error caused by the instability of the machine to the maximum extent, the data of the first 10 time points of each sample are removed. In addition, because the magnetic resonance image scanning adopts a layered scanning method, an interval scanning method is generally adopted, and the specific method is to firstly acquire 1, 3 and 5 … layers and then acquire 2, 4 and 6 … layers. Since the image acquisition time of each layer is different by using the interlayer scanning, the acquisition time of one volume is usually defined as one TR time. The interval between adjacent layers is half of the repetition Time (TR), namely TR/2, and the interval has a large influence on the analysis of data of the RS-fMRI in the Resting State (RS), so we perform slice correction by setting the number of slices, the order of slices, and the reference slices, so that all volumes have the same time point. The process of head motion correction is to use the first image as a reference image, then to perform position matching on other images and the reference image through rigid transformation such as rotation and translation, and then to perform resampling on all images by using an interpolation algorithm. In addition, because of individual differences among the tested objects and the sizes and shapes of human brains, the purpose of spatial standardization is to unify all tested images into the same dimension for subsequent statistical analysis. The method adopts a DARTEL model for space standardization. After the above process is completed, the data in M is preferably smoothed by a gaussian kernel with a full width at half maximum (FWHM) of 8 mm, which mainly fuses the information of the high frequency signal with the signal of the surrounding area to weaken the image of the high frequency signal. Meanwhile, the activity of the neuron is most remarkably expressed in [ 0.01-0.08 Hz ], so that band-pass filtering is needed to remove physiological noise in data. Finally, 6 brachiocephalic parameters, whole brain, white matter and cerebrospinal fluid mean signals were regressed away. To control the effect of head motion, the average inter-frame head movement distance (mean FD) is calculated. Samples with head motion >2mm displacement or >2 ° rotation or >0.3mm mean FD will be excluded.
Fig. 4 is a schematic flowchart of an example of brain image data preprocessing in the DTI modality provided in the present application. As shown in fig. 4, in an example, before the step of dividing N, the method further includes: the N is preprocessed, including first stripping the scalp, skull, and then performing head motion correction, eddy current correction, spatial normalization, gaussian smoothing, and Fractional Anisotropy (FA) FA calculations. Preferably, the raw DTI data is preprocessed using the PANDA tool kit.
The pretreatment includes first removing structures other than the brain, such as the scalp and skull, and then performing head movement correction and eddy current correction. Eddy current correction can be adjusted to correct for distortion of diffusion weighted images due to different coil gradients and to remove simple head motion artifacts. Finally, the AAL template is applied to the DTI image and the Fractional Anisotropy (FA) is calculated. Preferably, the raw DTI data is preprocessed using the PANDA tool kit.
S102, calculating any two brain image data sets MiThe average time series correlation of (a), the average time series corresponding to the brain region one to one.
Extracting the MiThe average time series of voxels in (b) is in one-to-one correspondence with the brain regions.
Preferably, the correlation between any two brain region time series is calculated using pearson correlation, resulting in a functional network. Since the range of the correlation coefficient of pearson is [ -1, 1], the absolute value of the weight of each side in the network is taken to keep the negative connection, which can provide more information and possibly help to improve the classification performance.
S103, establishing a functional network according to the average time sequence, wherein nodes of the functional network correspond to the brain areas one by one, and edges of the functional network refer to functional connection of the brain areas.
S104, constructing the N according to a deterministic bundle marking methodiAnd the structure network corresponds to the brain areas, the nodes of the structure network correspond to the brain areas one by one, and the edges of the structure network refer to the structure connection of the brain areas.
Preferably, the structural network is constructed by deterministic bundle tagging using the PANDA toolkit.
And S105, constructing a multilayer network according to the functional network and the structural network, wherein the multilayer network is a network for establishing interlayer connection of the same node in the functional network and the structural network.
The network nodes in the functional network and the structural network obtained through the above steps S3 and S4 are brain regions in the AAL90 template, and edges in the functional network and the structural network represent functional connections and structural connections between brain regions, respectively. It will be appreciated that the functional network may be represented as an m adjacency matrix and the structural network may be represented as an m adjacency matrix, so that two m adjacency matrices are obtained for each sample.
In one implementation, according to the AAL90 template, the network nodes in the functional network and the structural network obtained through the above steps S3 and S4 are brain areas in the AAL90 template, and the edges in the functional network and the structural network represent the functional connection and the structural connection between the brain areas, respectively. It will be appreciated that the functional network may be represented as a 90 x 90 adjacency matrix and the structural network may be represented as a 90 x 90 adjacency matrix, so that two 90 x 90 adjacency matrices are obtained for each sample.
The definition of the multilayer network G is as follows:
wherein A is1Expressed as m by m of the functional network, A2Expressed as m by m of the structural network, H12And H21Representing an interlayer connection matrix, wherein interlayer connection exists between the same nodes, and the interlayer connection value between different nodes is set to be 0;
the interlayer connections between the same nodes are defined as follows:
wherein the content of the first and second substances,
representing the degree of the node j in the matrix corresponding to the ith mode, when i is 1
Is A1Degree of middle node j, i is 2
Is A2The degrees, α and β, of the middle node j are two adjustable parameters.
In one implementation, in accordance with AAL90 templates, where
Representing the degree of the node j in the matrix corresponding to the ith mode. Alpha and beta are two adjustable parameters, and interlayer connection with different interlayer dependence degrees can be obtained by setting different alpha and beta. Because the interlayer connection value between different nodes is 0, in the classification process, the interlayer connection matrix H can be directly extracted12Or H21Is characterized by a diagonal line of (a). Thus, the input feature size is N × 90, N being the number of samples.
In a certain implementation manner, because the weight size ranges of the edges in the structural network and the functional network are different from the edge density, a multi-layer network cannot be directly constructed through the definition of the multi-layer network G. Constructing a multilayer network comprises a singular value decomposition step, wherein singular value decomposition is carried out on the structural network and the functional network; effect of (1)And a rate cost optimization step, namely performing efficiency cost optimization on the structural network and the functional network according to the singular value decomposition result. Therefore, the purposes of normalization and sparsification are achieved, the weight value range of the edges in the structural network and the functional network is consistent with the edge density, and a multilayer network is established. Wherein, singular value decomposition is carried out on the matrix A, so that A is UAVTWhere Λ represents the singular value of the matrix A, λ1Represents the largest singular value; the matrix after normalization is defined as
It will be appreciated that the ECO keeps the edge densities ρ of both networks consistent, i.e., the density ρ is 3/(n-1), where n is the number of vertices.
S2, extracting a plurality of characteristics of the multilayer network.
In a certain implementation manner, in the step of extracting a plurality of features of the multilayer network, the plurality of features include node degrees, interlayer connections, and node entropies.
The node degree is defined as:
wherein f is the number of layers of the network,
representing the node degree of node i in the f-th level, diiRepresents the interlayer connection of the node i;
the entropy of each layer of nodes in G is defined as the entropy of a subgraph formed by all layers of given nodes, adjacent nodes and edges connected with the nodes, and the total entropy of the nodes in G is defined as the sum of the entropy of each layer of nodes of the corresponding nodes;
the node entropy is defined as:
where k, m are any two nodes in the subgraph, and qk，m≠0。
And S3, respectively training and obtaining a plurality of models respectively corresponding to the plurality of characteristics according to the plurality of characteristics.
The training of the model refers to training different models by using different characteristics (node degrees, interlayer connections and node entropies).
In one implementation, the node degree characteristic is trained to form a node degree model, the interlayer connection characteristic is trained to form an interlayer connection model, and the node entropy characteristic is trained to form a node entropy model.
And S4, obtaining a prediction result of the sample according to the plurality of models.
The result of a tested data is predicted by using different models, some models can predict the result, and some models can predict the result incorrectly. And inputting the data into different models to obtain the prediction results of the models.
And S5, fusing the multiple models according to the prediction result.
Model fusion is to give a classification result, such as voting, by comprehensively considering the prediction results of the models. The model fusion method is a stacking method, model fusion is effectively completed, and the advantages of multiple models are integrated to produce a better result.
The method comprises the steps of constructing a functional network of a brain by using average time sequence correlation of a resting state functional magnetic resonance imaging (fMRI) brain image data set, constructing a structural network of the brain by using brain image data of two modes of Diffusion Tensor Imaging (DTI), and further constructing a multilayer network, wherein the multilayer network is a network for establishing interlayer connection of the same nodes in the functional network and the structural network. A multi-layer network may expose complex, highly interdependent structures between multiple subsystems in a complex system or within a subsystem. In the study of brain images, such complex relationships not only appear as interactions between brain regions, but also appear as interactions and interdependencies between different modalities (structural networks and functional networks). Therefore, the multi-layer network can more accurately model the brain network. According to the method and the system, the multilayer network is applied to a framework of mental disease identification, hidden interaction and interdependence relation among different modal brain networks are mined, and then classification accuracy is improved.
The mental diseases such as major depressive disorder and the like are not only abnormally changed in the structure and functional network, but also remarkably changed in the form and structure of the subcortical structure including the hippocampus, amygdala and the like.
The present application provides another embodiment of a method for cancer-associated depression identification based on multi-modality magnetic resonance data. The method comprises the steps of extracting the characteristics of the multilayer network, wherein the characteristics of the multilayer network are topological attributes of the multilayer network, the topological attributes comprise interlayer connection, node degree and node entropy, further, collecting morphological characteristics of a subcortical structure in the structure magnetic resonance imaging (sMRI), and training according to the morphological characteristics to obtain a morphological characteristic model corresponding to the morphological characteristics. By fusing the information of the three modalities of MRI, the method and the system have the characteristics of cortex and cortex, and can furthest mine hidden information.
Further, a method of acquiring morphological features of sub-cortical structures in structural magnetic resonance imaging srmri is described.
In one implementation, the morphological feature of the subcortical structure is collected by measuring the subcortical hippocampus, amygdala morphology. Firstly, extracting a sub-cortical structure from original sMRI image data by using an FSL tool, and obtaining a binary image by threshold processing. A marching cubes algorithm is used to generate a triangular surface mesh, and a 'progressive mesh' and mesh refinement are applied to the mesh to eliminate obtuse angles and smoothing noise generated in the process of generating the mesh. All smooth meshes are normalized into a standard space by affine transformation, 9 parameter matrixes (comprising three parameters of translation, rotation and scaling) are calculated by FIRST (a module of FSL), then the inverse consistent surface fluid registration is used, and the problem of surface registration is converted into image configuration by conformal mapping, so that the working difficulty is reduced. Finally, the 'Log-Euclidean metric' is considered according to the idea of the multivariate tensor to analyze the change of the shape instead of the eigenvalue based on the deformation tensor, the calculation of the tensor by the Log-Euclidean metric is easier, and any deviation can not be introduced by using the method. By the treatment of the method, a large number of characteristics of the hippocampus and the almond can be obtained.
For the above features extracted from the multi-layer network and the sub-cortical structure, preferably, the method of F-score is used for dimension reduction and feature selection. In one example, the morphological feature and the topological attribute are input to a Support Vector Machine (SVM), which selects the morphological feature and the topological attribute that meet a threshold as a fused feature.
To avoid the chance of sample grouping, the present application uses nested ten-fold cross-validation. The role of the outer ten-fold cross validation is to divide all samples into training and test sets. The ten-fold cross validation is performed on the training set, the internal ten-fold cross validation has the effect of finding out an optimal set of parameters (such as alpha and beta in the multi-layer network construction and relevant parameters of an SVM classifier) on the training set, and the arrangement ensures that the information of the test set cannot be leaked. And respectively training and classifying different features, and finally performing model fusion by adopting a stacking method. The ten-fold cross validation is repeated for ten times to obtain 100 accuracy rates, the average value of the accuracy rates is calculated to measure the performance of the whole classification model, the obtained classification result is up to more than 90%, and compared with a common method, the accuracy rate is obviously improved.
The embodiment integrates the information of MRI in three modes, gives consideration to the characteristics of cortex and cortex, can furthest mine hidden information, and improves the identification accuracy. A large number of researches prove that different types of mental diseases cause morphological structure changes with remarkable distinguishing characteristics on structures such as hippocampus, amygdala and the like, so that the application is not only limited in the identification of cancer-associated depression, but also can be used for carrying out adaptive characteristic modification on different mental diseases so as to realize the identification of other mental diseases.
The method applies a multi-modal MRI feature fusion method to cancer-associated depression identification. On one hand, a functional network and a structural network are respectively constructed based on fMRI and DTI, and then a multi-layer network is constructed. And extracting multiple topological attributes (such as interlayer connection, node degree, node entropy and the like) of the multilayer network as feature input SVM for classification, and selecting the best-performing features for final model fusion. In recent years, multilayer networks have been gaining attention in many research fields, and have many emerging applications in complex systems such as key infrastructure, transportation, ecosystem, human brain connectors, and the like. The multi-layer network may expose complex, highly interdependent structures between or within multiple subsystems in these complex systems. In the study of brain images, such complex relationships not only appear as interactions between brain regions, but also as interactions and interdependencies between different modalities (structures and functions). Therefore, the multi-layer network can help us model the brain network more accurately. The method firstly proposes a framework for applying a multi-layer network to mental disease identification, and excavates the hidden interaction and interdependence relationship between different modal brain networks, thereby improving the classification accuracy.
On the other hand, because the information of the subcortical structure is ignored by the multilayer network, the method further extracts the morphological characteristics of the subcortical hippocampus and the amygdala as the characteristics for classification. And finally, fusing by adopting a stacking method. The method integrates information of different modalities of MRI and subcortical structures, compared with other methods, more hidden information can be mined, the classification accuracy can reach 90%, and the method is remarkably improved. Meanwhile, as the structure and the function of the brain of the mental disease patient are changed to a certain extent, the method can be popularized to the related research of other mental diseases.
Claims (10)
1. A cancer-associated depression identification method based on multi-modal magnetic resonance data is characterized by comprising the following steps:
constructing a multilayer network, wherein the multilayer network is stacked by at least a functional network constructed by brain image data of a resting state functional magnetic resonance imaging modality and a structural network constructed by brain image data of a diffusion tensor imaging DTI modality;
extracting a plurality of features of the multi-layer network;
respectively training to obtain a plurality of models respectively corresponding to the plurality of characteristics according to the plurality of characteristics;
obtaining a prediction result of the sample according to the plurality of models;
and fusing the plurality of models according to the prediction result.
2. The method of claim 1, further comprising:
acquiring morphological characteristics of a subcortical structure in structural magnetic resonance imaging;
and training to obtain a morphological feature model corresponding to the morphological feature according to the morphological feature.
3. The method of claim 1, wherein the step of extracting a plurality of features of the multi-layered network comprises node degrees, inter-layer connections and node entropy.
4. The method of claim 3, wherein the multi-layered network G is defined as follows:
wherein A is1Expressed as m by m of the functional network, A2Expressed as m by m of the structural network, H12And H21Representing an interlayer connection matrix, wherein interlayer connection exists between the same nodes, and the interlayer connection value between different nodes is set to be 0;
the interlayer connections between the same nodes are defined as follows:
wherein the content of the first and second substances,
representing the degree of the node j in the matrix corresponding to the ith mode, when i is 1
Is A1Degree of middle node j, i is 2
Is A2The degrees, α and β, of the middle node j are two adjustable parameters.
5. The method of claim 4, further comprising:
a singular value decomposition step of performing singular value decomposition on the structural network and the functional network;
and an efficiency cost optimization step, wherein efficiency cost optimization is carried out on the structural network and the functional network according to the singular value decomposition result.
6. The method of claim 5, further comprising:
calculating a topology attribute of the multilayer network, wherein the topology attribute comprises the node degree and the node entropy;
the node degree is defined as:
wherein f is the number of layers of the network,
representing the node degree of node i in the f-th level, diiRepresents the interlayer connection of the node i;
the entropy of each layer of nodes in G is defined as the entropy of a subgraph Gs composed of all layers of given nodes, adjacent nodes and edges connected with the nodes, and the total entropy of the nodes in G is defined as the sum of the entropy of each layer of nodes of the corresponding nodes;
the node entropy is defined as:
where k, m are any two nodes in the subgraph, qk，mRepresents the weight of the connection between node k and node m and qk，m≠0。
7. The method for identifying cancer associated with depression according to any one of claims 1-6, further comprising:
pre-processing the brain image data, the pre-processing including removing the first ten time points, temporal layer correction, scalp and skull stripping, head motion correction, spatial normalization, and gaussian smoothing.
8. The method of claim 7, wherein the weights of the edges in the functional network are taken as absolute values, and the edges are functional connections of brain regions.
9. The method of claim 8, further comprising:
preprocessing the N, the preprocessing including correction of eddy currents and FA calculations.
10. The method for identifying cancer associated with depression according to any one of claims 1-9, further comprising:
performing cross validation on all samples by using external ten folds, and dividing the samples into a training set and a test set;
the training set is cross-validated by an inner ten-fold.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110999956.7A CN113705680A (en) | 2021-08-27 | 2021-08-27 | Cancer associated depression identification method based on multi-modal magnetic resonance data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110999956.7A CN113705680A (en) | 2021-08-27 | 2021-08-27 | Cancer associated depression identification method based on multi-modal magnetic resonance data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113705680A true CN113705680A (en) | 2021-11-26 |
Family
ID=78656395
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110999956.7A Pending CN113705680A (en) | 2021-08-27 | 2021-08-27 | Cancer associated depression identification method based on multi-modal magnetic resonance data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113705680A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114418982A (en) * | 2022-01-14 | 2022-04-29 | 太原理工大学 | Method for constructing DTI multi-parameter fusion brain network |
| CN117315425A (en) * | 2023-10-12 | 2023-12-29 | 无锡市第五人民医院 | Fusion method and system of multi-mode magnetic resonance images |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103778349A (en) * | 2014-01-29 | 2014-05-07 | 思博奥科生物信息科技（北京）有限公司 | Biomolecular network analysis method based on function module |
| CN111127441A (en) * | 2019-12-25 | 2020-05-08 | 兰州大学 | Multi-modal brain image depression recognition method and system based on graph node embedding |
| CN111178431A (en) * | 2019-12-20 | 2020-05-19 | 北京邮电大学 | Network node role identification method based on neural network and multi-dimensional feature extraction |
| US20200289044A1 (en) * | 2016-11-28 | 2020-09-17 | Cornell University | Systems and methods for identifying a neurophysiological biotype of depression in the brain of a patient |
| CN113255728A (en) * | 2021-04-26 | 2021-08-13 | 浙江工业大学 | Depression classification method based on map embedding and multi-modal brain network |
-
2021
- 2021-08-27 CN CN202110999956.7A patent/CN113705680A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103778349A (en) * | 2014-01-29 | 2014-05-07 | 思博奥科生物信息科技（北京）有限公司 | Biomolecular network analysis method based on function module |
| US20200289044A1 (en) * | 2016-11-28 | 2020-09-17 | Cornell University | Systems and methods for identifying a neurophysiological biotype of depression in the brain of a patient |
| CN111178431A (en) * | 2019-12-20 | 2020-05-19 | 北京邮电大学 | Network node role identification method based on neural network and multi-dimensional feature extraction |
| CN111127441A (en) * | 2019-12-25 | 2020-05-08 | 兰州大学 | Multi-modal brain image depression recognition method and system based on graph node embedding |
| CN113255728A (en) * | 2021-04-26 | 2021-08-13 | 浙江工业大学 | Depression classification method based on map embedding and multi-modal brain network |
Non-Patent Citations (1)
| Title |
|---|
| 苏瑞;王勇;杨指挥;: "基于拓扑势的网络拓扑建模方法", 计算机工程, no. 05 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114418982A (en) * | 2022-01-14 | 2022-04-29 | 太原理工大学 | Method for constructing DTI multi-parameter fusion brain network |
| CN114418982B (en) * | 2022-01-14 | 2022-11-01 | 太原理工大学 | Method for constructing DTI multi-parameter fusion brain network |
| CN117315425A (en) * | 2023-10-12 | 2023-12-29 | 无锡市第五人民医院 | Fusion method and system of multi-mode magnetic resonance images |
| CN117315425B (en) * | 2023-10-12 | 2024-03-26 | 无锡市第五人民医院 | Fusion method and system of multi-mode magnetic resonance images |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Dalca et al. | Learning conditional deformable templates with convolutional networks | |
| US20220122250A1 (en) | Brain feature prediction using geometric deep learning on graph representations of medical image data | |
| CN113616184B (en) | Brain network modeling and individual prediction method based on multi-mode magnetic resonance image | |
| CN109741343B (en) | T1WI-fMRI image tumor collaborative segmentation method based on 3D-Unet and graph theory segmentation | |
| CN110363760B (en) | Computer system for recognizing medical images | |
| CN113705680A (en) | Cancer associated depression identification method based on multi-modal magnetic resonance data | |
| Abbas et al. | Transformed domain convolutional neural network for Alzheimer's disease diagnosis using structural MRI | |
| CN104361318A (en) | Disease diagnosis auxiliary system and disease diagnosis auxiliary method both based on diffusion tensor imaging technology | |
| Liu et al. | An enhanced multi-modal brain graph network for classifying neuropsychiatric disorders | |
| CN111161226A (en) | Method for uniformly segmenting cerebral cortex surface based on spectral clustering algorithm | |
| CN115272295A (en) | Dynamic brain function network analysis method and system based on time domain-space domain combined state | |
| Ferrarini et al. | GAMEs: growing and adaptive meshes for fully automatic shape modeling and analysis | |
| CN105184794A (en) | CSM assistant analysis system and method based on tensor image | |
| CN116503680A (en) | Brain image structured analysis and brain disease classification system based on brain atlas | |
| Qiu et al. | Multi-channel sparse graph transformer network for early alzheimer’s disease identification | |
| Ramana | Alzheimer disease detection and classification on magnetic resonance imaging (MRI) brain images using improved expectation maximization (IEM) and convolutional neural network (CNN) | |
| Xi et al. | Brain Functional Networks with Dynamic Hypergraph Manifold Regularization for Classification of End-Stage Renal Disease Associated with Mild Cognitive Impairment. | |
| Ziyan et al. | Consistency clustering: a robust algorithm for group-wise registration, segmentation and automatic atlas construction in diffusion MRI | |
| Zong et al. | Multiscale autoencoder with structural-functional attention network for Alzheimer’s disease prediction | |
| CN112515653A (en) | Brain network construction method based on nuclear magnetic resonance image | |
| Yeung et al. | Pipeline comparisons of convolutional neural networks for structural connectomes: predicting sex across 3,152 participants | |
| Ulloa et al. | Improving multiple sclerosis lesion boundaries segmentation by convolutional neural networks with focal learning | |
| CN114373095A (en) | Alzheimer disease classification system and method based on image information | |
| CN114494132A (en) | Disease classification system based on deep learning and fiber bundle spatial statistical analysis | |
| Bhasin et al. | Multiple-activation parallel convolution network in combination with t-SNE for the classification of mild cognitive impairment |
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 |