CN113298038A - Construction method for differentiating sub-network pairs of multi-frequency brain network for auxiliary diagnosis of AD - Google Patents
Construction method for differentiating sub-network pairs of multi-frequency brain network for auxiliary diagnosis of AD Download PDFInfo
- Publication number
- CN113298038A CN113298038A CN202110684023.9A CN202110684023A CN113298038A CN 113298038 A CN113298038 A CN 113298038A CN 202110684023 A CN202110684023 A CN 202110684023A CN 113298038 A CN113298038 A CN 113298038A
- Authority
- CN
- China
- Prior art keywords
- network
- sub
- frequency
- networks
- 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.)
- Granted
Links
- 210000004556 brain Anatomy 0.000 title claims abstract description 116
- 238000003745 diagnosis Methods 0.000 title claims abstract description 22
- 238000010276 construction Methods 0.000 title claims abstract description 11
- 238000002599 functional magnetic resonance imaging Methods 0.000 claims abstract description 31
- 238000005065 mining Methods 0.000 claims abstract description 10
- 230000003925 brain function Effects 0.000 claims description 60
- 208000024827 Alzheimer disease Diseases 0.000 claims description 56
- 238000000034 method Methods 0.000 claims description 44
- 230000004069 differentiation Effects 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 7
- 238000012163 sequencing technique Methods 0.000 claims description 4
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000012217 deletion Methods 0.000 claims description 2
- 238000004195 computer-aided diagnosis Methods 0.000 abstract description 2
- 230000007177 brain activity Effects 0.000 description 6
- 238000012937 correction Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000007781 pre-processing Methods 0.000 description 4
- 230000000284 resting effect Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 3
- 239000008280 blood Substances 0.000 description 3
- 210000004369 blood Anatomy 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 3
- 238000002595 magnetic resonance imaging Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 229910052760 oxygen Inorganic materials 0.000 description 3
- 239000001301 oxygen Substances 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 230000001419 dependent effect Effects 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 239000010410 layer Substances 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 208000024891 symptom Diseases 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000001638 cerebellum Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000003920 cognitive function Effects 0.000 description 1
- 230000001054 cortical effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004886 head movement Effects 0.000 description 1
- 239000011229 interlayer Substances 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002610 neuroimaging Methods 0.000 description 1
- 238000010827 pathological analysis Methods 0.000 description 1
- 230000035479 physiological effects, processes and functions Effects 0.000 description 1
- 230000002285 radioactive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000002269 spontaneous effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000002618 waking effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Signal Processing (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a construction method of a multi-frequency brain network distinguishing sub-network pair for assisting in AD diagnosis, and belongs to the technical field of computer-aided diagnosis. Dividing time sequence signals of all interested areas of the brain in an fMRI data set containing AD and NC into a plurality of frequency bands; calculating the correlation coefficient of any two interested areas in the same frequency band; thresholding is carried out on the correlation coefficient to construct each tested multi-frequency brain network, two multi-frequency brain network data sets AD and NC are obtained, and frequent sub-network mining is carried out to obtain two multi-frequency frequent sub-network sets; respectively calculating the distinguishing capability of each sub-network in the two multi-frequency frequent sub-networks, and respectively taking a plurality of sub-networks with the strongest capability from the two sub-networks to combine and construct a sub-network pair of each frequency band, namely a distinguishing sub-network pair; and calculating the difference degree of the distinguishing sub-network pairs, sorting the first k distinguishing sub-network pairs with the maximum frequency band difference degree according to the descending order of the difference degree, and selecting the front k' sub-network pairs as the distinguishing sub-network pairs for assisting in diagnosing AD.
Description
Technical Field
The invention designs a construction method of a multi-frequency brain network distinguishing sub-network pair for assisting in AD diagnosis, and belongs to the technical field of computer-aided diagnosis.
Background
Functional Magnetic Resonance Imaging (fMRI) technology is an emerging neuroimaging method based on blood oxygen level dependent signals. The fMRI technique has the advantages of being non-invasive and non-radioactive, providing superior spatial resolution and coverage, being able to effectively locate the active areas of brain function, unlike magnetic signals that are sampled into discrete data points, the blood oxygen level dependent signals can generate digitized MRI signals, being able to further perform spatial calibration, normalization, smoothing, etc., being an ideal tool for studying the activity patterns and relationships of the whole brain.
The brain function magnetic resonance imaging in the resting state is used for researching spontaneous brain activities of a person in the waking state without a specific task, and the resting state brain function network artificially constructed based on the resting state brain function magnetic resonance imaging can well describe the activity state of the brain and the interaction among all neurons or brain areas, research and analyze some characteristics of a brain default network, systematically represent the functional connection of the whole brain, and reveal the brain network basis of the normal cognitive function of the human brain. On the basis of the constructed resting brain function network, the functional connectivity analysis and graph theory method can also be used for detecting the change of the topology of the brain function network in partial brain areas or whole brain areas caused by certain diseases such as Alzheimer's Disease (AD).
Most of brain function networks constructed by the traditional method are constructed based on time domain signals, detail differences of brain activity information at different frequencies are covered, namely different conclusions can be obtained by analyzing the networks constructed under different frequency division scales. As at low frequencies, bilaterally homologous brain regions tend to be tightly connected and functional connectivity is generally greater, and the correlation of cortical networks is generally concentrated in the ultra-low frequencies (0.01-0.06Hz), while the connections of marginal networks are distributed over a wider frequency range (0.01-0.14 Hz). These all indicate that the fMRI signal is frequency specific. Therefore, the analysis method of the brain function network constructed based on a single frequency cannot describe the frequency domain information of the brain activity signal more accurately. On the other hand, the traditional AD classification auxiliary diagnosis method based on the brain function network is mainly based on the inter-group analysis of AD and normal contrast, a series of features such as node degree, clustering coefficient, path length, centrality and the like are selected from network topology attributes, and the features are connected in series to form feature vectors, and then the feature vectors are combined with a machine learning method to be used for subsequent classification. However, such feature selection may lose some more detailed topology information in the network, such as the topology of the network itself and the common topology among networks, and then affect the accuracy of the subsequent classification.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a construction method of a multi-frequency brain network distinguishing sub-network pair for assisting in diagnosing AD (AD), aiming at fully revealing the detail difference of brain activity information at different frequencies through the constructed multi-frequency brain network distinguishing sub-network pair, describing the frequency domain information of brain activity signals more accurately, extracting more detailed topology structure information of the brain network, being applicable to assisting in diagnosing AD and other symptoms and helping to improve the diagnosis accuracy of AD and other symptoms.
The technical scheme of the invention is as follows:
a construction method of a multi-frequency brain network distinguishing sub-network pair for assisting in AD diagnosis comprises the following steps:
step 1: acquiring fMRI original data, and processing the fMRI original data to obtain a time sequence signal of each interested area of each tested brain; the fMRI raw data comprise Alzheimer disease AD data and normal control NC data;
step 2: performing frequency division processing on the time sequence signal of each interested area, and dividing each time sequence signal into a plurality of frequency bands to obtain frequency division time sequence signals of the plurality of frequency bands;
and step 3: for each frequency band, respectively calculating correlation coefficients between any two interested areas, and representing the relevance between two interested brain area nodes in the same frequency band through the correlation coefficients;
and 4, step 4: performing thresholding treatment on the correlation coefficient, constructing a brain function network of each tested frequency band to obtain each tested multi-frequency brain network, and further obtaining an AD multi-frequency brain network data set and an NC multi-frequency brain network data set;
and 5: performing frequent sub-network mining on the brain function networks of all frequency bands in the AD multi-frequency brain network data set and the NC multi-frequency brain network data set to obtain frequent sub-networks of all frequency bands in each data set, and further obtaining an AD multi-frequency frequent sub-network set and an NC multi-frequency frequent sub-network set;
step 6: respectively calculating respective 2 classification distinguishing capabilities of all sub-networks in the AD multi-frequency frequent sub-network set and the NC multi-frequency frequent sub-network set, namely the capacity of distinguishing AD and NC brain function networks, aiming at each frequency band, and respectively combining a plurality of sub-networks with the strongest distinguishing capabilities in the two frequent sub-network sets to construct a sub-network pair of each frequency band, namely a distinguishing sub-network pair;
and 7: calculating the difference degree of two sub-networks in each distinguishing sub-network pair, respectively sequencing the distinguishing sub-network pairs of each frequency band according to the sequence of the difference degrees from large to small, and selecting the first k distinguishing sub-network pairs from the sequencing sequence of the distinguishing sub-network pairs of each frequency band;
and 8: and sorting the difference degrees of the first k distinguishing sub-network pairs of all the frequency bands according to a descending order, and selecting the front k' sub-network pairs as final distinguishing sub-network pairs for carrying out class prediction on the given brain network.
Further, according to the construction method of the multi-frequency brain network distinguishing sub-network pair for assisting in diagnosing AD, frequency division processing is performed on the time series signals of each region of interest by using a multi-scale wavelet transformation method.
Further, according to the construction method of the multi-frequency brain network for assisting in diagnosing AD, the correlation coefficient is a Pearson correlation coefficient.
Further, according to the method for constructing the pair of sub-network distinguishing networks of the multi-frequency brain network for assisting in diagnosing AD, the method for mining the frequent sub-networks for all the brain function networks of each frequency band j in the data set of the multi-frequency brain network comprises the following steps:
(a) counting the frequency of all edges in all brain function networks of each frequency band j in a multi-frequency brain network data set, selecting the edges with the frequency greater than the preset support degree, taking each edge as a candidate sub-network, and obtaining a candidate sub-network set and recording the candidate sub-network set as Sj 1；
(b) For set Sj 1Each candidate subnetwork in (1) is expanded at most K times in a mode of adding one edge at a time, and a formed set is marked as Sj 2，Sj 3，…，Sj KAnd in the process of expanding, the set Sj 2，Sj 3，…，Sj KDeletion of duplicate subnets;
(c) computing a set Sj 2，Sj 3，……，Sj KThe frequency of each sub-network in the multi-frequency brain network data set is selected, the sub-network with the frequency of occurrence larger than the preset support degree is selected and is respectively marked as Sj 2’，Sj 3’，…，Sj K’Then get the frequent sub-network set S ═ { S) for each frequency bandj 1,Sj 2’,Sj 3’,…,Sj K’}。
Further, according to the construction method for differentiating the sub-network pairs of the multi-brain network for the auxiliary diagnosis of AD, the support degree is defined as: for a given set of brain function networks P and children thereinNetwork gsWill sub-network gsSupport (g) in a network set PsP) is defined as a subnetwork gsThe number of occurrences in the network set P is a percentage of the total number N of networks in the network set P, that is:
wherein ,
is a sub-network gsSupport set of (1), GiRepresenting the networks in the network set P and N representing the number of networks contained in P.
Further, according to the method for constructing the pair of sub-network differentiation sub-networks for assisting in diagnosing AD, the method for calculating the respective 2 classification differentiation capabilities of all sub-networks in the AD multi-frequency frequent sub-network set and all sub-networks in the NC multi-frequency frequent sub-network set for each frequency band comprises: suppose G+A brain function network set G of a certain frequency band in AD multi-frequency brain network data set-A brain function network set of a certain frequency band in the NC multi-frequency brain network data set; from G+Wherein the excavated certain frequent sub-network is g+From G-Wherein the excavated certain frequent sub-network is g-(ii) a If subnetwork g+(g-) At G+(G-) Is much more frequent than it is in G-(G+) The sub-network is a distinguishing sub-network with 2-class distinguishing capability, and the distinguishing capability of the sub-network on the AD and NC brain function networks is calculated according to the formula (5):
similarly, subnetwork g-The discriminative power for two types of brain function networks is calculated according to equation (6):
further, according to the method for constructing the pair of sub-networks of the multi-frequency brain network for assisting in diagnosing AD, the calculation formula of the difference degree of the two sub-networks in each pair of sub-networks is as follows:
diff(g1,g2)＝1-corr(g1,g2) (7)
wherein corr (g)1,g2) Is a sub-network pair (g)1,g2) In (1) sub-network g1，g2The correlation between them is calculated according to equation (8):
corr(g1,g2)＝μsimN(g1,g2)+(1-μ)simS(g1,g2) (8)
in the above formula, simN (g)1,g2) For two sub-networks g1，g2The structural similarity of (a) is calculated according to the formula (9); SimS (g)1,g2) For two sub-networks g1，g2The support set similarity of (2) is calculated according to the formula (10);
wherein ,E(g1) Shown is a sub-network g1The edge set of (1); e (g)2) Shown is a sub-network g2The edge set of (1);
in the above formula, cov (g)1，G+) As a subnetwork g1In network set G+The support degree in (1); cov (g)2，G-) As a subnetwork g2In network set G-The support degree in (1).
Further, according to the multi-frequency brain network for assisting AD diagnosis, sub-network pairs are distinguishedThe construction method of (1) utilizes a distinguishing subnetwork pair, and the method for predicting the category of the given brain network comprises the following steps: using each of the k' discriminating subnetwork pairs according to a decision rule
For a given tested brain function network tiThe class prediction is carried out, and the result is recorded as mrWhere r is 1,2, …, k', by calculating the prediction mrThe ratio of the number of sub-network pairs of "AD" or "NC" to the total number of sub-network pairs k' is obtainedi(iv) final predicted outcome of;
the judgment rule is as follows:
(1) if it is not
And is
Then m isr＝‘AD’；
(2) If it is not
And is
Then m isr＝‘NC’；
(3) If it is not
And is
Then m isr＝null；
(4) If it is not
And is
Then m isr＝null。
Compared with the prior art, the invention has the following beneficial effects: the multi-frequency brain function network is established after frequency division is carried out on the time series signals, the topological attribute difference of the brain function network under different frequency bands is fully considered, and the information of brain activities can be better reserved; by mining frequent sub-networks to construct a distinguishing sub-network pair, the intra-group commonalities and inter-group specificities of the tested brain function network set are fully represented, and the medical auxiliary diagnosis service can be better provided.
Drawings
FIG. 1 is a schematic flow chart of a method for constructing a pair of sub-network discrimination of a multi-frequency brain network for assisting diagnosis of AD according to the present invention;
FIG. 2 is a flow chart of a method for constructing a pair of sub-network discrimination of a multi-frequency brain network for assisting diagnosis of AD according to the present invention;
FIG. 3 is a diagram illustrating a method for mining frequent subnets according to an embodiment of the present invention;
FIG. 4 is a flow chart of a method for constructing a pair of distinct subnetworks using frequent subnetworks according to the present invention.
Detailed Description
To facilitate an understanding of the present application, the present application will now be described more fully with reference to the accompanying drawings. Preferred embodiments of the present application are given in the accompanying drawings. This application may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
Fig. 1 is a schematic flow chart of a method for constructing a pair of sub-network discrimination networks of a multi-frequency brain network for assisting in diagnosing AD according to the present invention, and fig. 2 is a detailed flow chart of a method for constructing a pair of sub-network discrimination networks of a multi-frequency brain network for assisting in diagnosing AD according to the present invention. The method for constructing the multi-frequency distinguishing sub-network pair of the brain network comprises the following steps:
step 1: acquiring an ADNI public data set and processing fMRI original data in the ADNI data set to obtain time sequence signals of each interested area of each tested brain; the fMRI raw data comprises AD data and Normal Control (NC) data;
the specific content comprises the following steps: firstly, preprocessing fMRI original data in an ADNI data set to obtain a required fMRI sample set, wherein the fMRI sample set comprises an AD sub-sample set consisting of AD type fMRI samples and an NC sub-sample set consisting of NC type fMRI samples; then, for a sample, using a brain region division template to obtain fMRI data of different brain regions, thus obtaining the fMRI data of different brain regions of each sample; and finally, taking an average value of data of each time point of an interested area of a sample as a value of the time point of the area, and forming a time series signal of the interested area of the brain by the values of a plurality of time points of the same area, so that the time series signal of the interested areas of the brain of the sample can be obtained, and the time series signals of the interested areas of the brain of all samples can be obtained according to the same method.
It will be readily appreciated by those of ordinary skill in the art that preprocessing the raw data of the fMRI, which is typically performed as a conventional processing step, is generally intended to eliminate extraneous information from the image, recover useful real information, enhance the detectability of the pertinent information, and simplify the data to the greatest extent possible, thereby preparing qualified or otherwise optimal fMRI image data for subsequent steps of the method. In a preferred embodiment, the preprocessing of the fMRI raw data includes a spatio-temporal correction process, a normalization process, and a smoothing filter process. Specifically, firstly, an fMRI image corresponding to an unstable time point in an fMRI original data set is removed; then, the fMRI images of all layers obtained by adopting the interlayer scanning method are corrected to the same time point, so that the time phase difference caused by different acquisition times between the layers during scanning is eliminated; then, performing head-motion correction on the fMRI image subjected to time correction so as to eliminate the fatigue of the tested object and the influence of the physiology of the tested object caused by long acquisition time and ensure the availability of data; then, uniformly standardizing all tested brain fMRI images subjected to the head movement correction, and matching the images to the same space so as to facilitate subsequent statistics and space positioning; and finally, carrying out smooth filtering processing on the image subjected to the standardization processing to eliminate image noise.
In a preferred embodiment, the brain region division template is an aal (atomic Automatic labeling) template. Specifically, the preprocessed fMRI sample is matched with an AAL template, 26 regions related to the cerebellum in the matching result are removed, the remaining 90 brain regions are used as 90 nodes of a brain function network, and an average value of signal intensities of all voxels BOLD (Blood oxygen Level dependency) in the brain regions is used as a representative of neuron activity conditions in the brain regions, that is, a value of the node.
Step 2: performing frequency division processing on the time sequence signal of each interested area, and dividing each time sequence signal into a plurality of frequency bands to obtain frequency division time sequence signals of the plurality of frequency bands;
in a preferred embodiment, the time series signals of each region of the brain are subjected to frequency division processing by using a multi-scale wavelet transform method, each time series signal is divided into a plurality of frequency bands, and frequency division time series signals of the plurality of frequency bands are obtained.
The wavelet multi-scale transformation is to gradually carry out multi-scale refinement on signals through telescopic transformation and translation transformation, so that frequency subdivision at low frequency and time subdivision at high frequency are achieved. Assuming that the fMRI time series signal is x (n), and x (n) is respectively passed through a high pass filter and a low pass filter, then x (n) is decomposed by wavelet transform, and the result of one frequency band, for example, the e-th frequency band, is:
in the above formula, n represents the number of time points included in one fMRI scan after pretreatment; z represents the number of divided frequency bands; z is a count variable; x is the number ofe,H(n) represents the high frequency portion of the signal decomposition of the e-th frequency band, which is also the result of frequency division of this frequency band; x is the number ofe,L(n) represents the low frequency part of the signal decomposition of the e-th frequency band, which is also the input signal for the next cascade decomposition; g (-) and h (-) are a high pass filter and a low pass filter, respectively.
And step 3: for each frequency band, respectively calculating correlation coefficients between any two interested areas, and representing the relevance between two interested brain area nodes in the same frequency band through the correlation coefficients;
the connection of brain function networks refers to the dynamic coordination of activities between neurons in different brain regions. The weight of the connection represents the strength of the interaction relation between two brain regions, and the higher the weight is, the stronger the coupling between the brain regions is. After the frequency-divided time series signals are obtained, the correlation between two brain area nodes in the same frequency band can be represented through a correlation coefficient.
In the preferred embodiment, the correlation between two brain region nodes in the same frequency band is characterized by a pearson correlation coefficient. The pearson correlation coefficient, which is a linear correlation coefficient, can be used to reflect the degree of linear correlation between two variables. The preferred embodiment calculates the pearson correlation coefficient between any two brain region time series signals in the same frequency band according to equation (3).
wherein ,rijRepresenting a pearson correlation coefficient between the time-series signal of the ith brain region and the time-series signal of the jth brain region; x is the number ofi(i＝1,2,…,n)，yj(j ═ 1,2, …, n) represents the time-series signal of the i-th brain region and the time-series signal of the j-th brain region, respectively;
representing the mean value of the ith brain region time series signal and the mean value of the jth brain region time series signal; n represents the number of time points included in one scan after preprocessing.
And 4, step 4: performing thresholding on the correlation coefficient, constructing a brain function network of each tested frequency band to obtain each tested multi-frequency brain network, and further obtaining an AD multi-frequency brain network data set consisting of all AD tested multi-frequency brain networks in the ADNI data set and an NC multi-frequency brain network data set consisting of all NC tested multi-frequency brain networks in the ADNI data set;
those skilled in the art can easily know that the range of the correlation coefficient value is (-1,1), when thresholding is performed, assuming that a threshold value is set as T, comparing the calculated absolute value of the correlation coefficient corresponding to each brain region in the same frequency band with the threshold value T, and when the absolute value of the correlation coefficient between nodes is greater than the threshold value T, considering that the connection between nodes is valid, and setting the value as 1; conversely, when the absolute value of the correlation coefficient between nodes is smaller than the threshold, the connection between nodes is considered invalid and set to 0. Thus, brain function networks of various frequency bands can be constructed, and a multi-frequency brain network is obtained.
And 5: for the AD multi-frequency brain network data set and the NC multi-frequency brain network data set, performing frequent sub-network mining on brain function networks of all frequency bands, excavating frequently-appearing sub-networks, obtaining the frequent sub-networks of all frequency bands in each data set, and further obtaining an AD multi-frequency frequent sub-network set formed by the frequent sub-networks of all frequency bands in the AD multi-frequency brain network data set and an NC multi-frequency frequent sub-network set formed by the frequent sub-networks of all frequency bands in the NC multi-frequency brain network data set;
the goal of frequent sub-network mining of brain function networks is to mine frequently occurring sub-network patterns in a given brain function network dataset to characterize the topology common to these brain function networks. Specifically, in the brain function network set under each frequency band, all the minimum support degrees S meeting the preset requirement are searchedminFrequent subnetworks. For a given set of brain function networks P and sub-networks g thereinsWill sub-network gsSupport (g) in a network set PsP) is defined as a subnetwork gsThe number of occurrences in the network set P is a percentage of the total number N of networks in the network set P, that is:
wherein ,
is a sub-network gsSupport set of (1), GiRepresenting the networks in the network set P and N representing the number of networks contained in P.
In this embodiment, a specific method for mining a frequent subnetwork for all brain function networks of one frequency band, for example, the jth frequency band, in a multi-frequency brain network data set includes:
(a) counting the frequency of all edges in all brain function networks of the jth frequency band in a data set of the multi-frequency brain network, selecting the edges with the frequency greater than the preset support degree, taking each edge as a candidate sub-network, and recording a set formed by the candidate sub-networks as Sj 1；
(b) For set Sj 1Each candidate subnetwork in (1) is expanded at most K times in a mode of adding one edge at a time, and a formed set is marked as Sj 2，Sj 3，…，Sj K. In the process of expansion, set Sj 2，Sj 3，…，Sj KDuplicate sub-networks in (1) are deleted.
(c) Computing a set Sj 2，Sj 3，……，Sj KThe frequency of each sub-network in the multi-frequency brain network data set is selected, the sub-network with the frequency of occurrence larger than the preset support degree is selected and is respectively marked as Sj 2’，Sj 3’，…，Sj K’Then, a frequent subnetwork set S ═ S of the jth band can be obtainedj 1,Sj 2’,Sj 3’,…,Sj K’}。
Therefore, frequent sub-networks can be mined from all brain function networks of each frequency band in the AD multi-frequency brain network data set and the NC multi-frequency brain network data set respectively to obtain the frequent sub-networks of each frequency band in each data set, the AD multi-frequency frequent sub-network set is formed by the frequent sub-networks of each frequency band in the AD multi-frequency brain network data set, and the NC multi-frequency frequent sub-network set is formed by the frequent sub-networks of each frequency band in the NC multi-frequency brain network data set.
In a preferred embodiment, as shown in fig. 3, the candidate subnetworks are DFS-encoded first, and then an edge is added to the L-edge frequent subnetwork by using the rightmost path extension method, so as to obtain an L + 1-edge frequent subnetwork. In the process of establishing the depth-first search space tree, a plurality of different DFS codes are possible for the same candidate subnetwork, but only the minimum DFS code can uniquely identify one brain network, namely, other non-minimum DFS codes of the candidate subnetwork are redundant. Redundant DFS coding and all child nodes thereof are clipped by applying Apriori property. And calculating the occurrence frequency of the generated candidate sub-networks in the brain function network data set, and deleting the candidate sub-networks which do not meet the preset support degree condition, so that the frequent sub-networks of each frequency band in each data set can be obtained.
Step 6: as shown in fig. 4, for each frequency band, the respective 2-classification distinguishing capabilities of all sub-networks in the AD multi-frequency frequent sub-network set and all sub-networks in the NC multi-frequency frequent sub-network set, that is, the capabilities of distinguishing the AD brain function networks from the NC brain function networks, are calculated, and the previous sub-networks with the strongest distinguishing capabilities in the two frequent sub-network sets are taken to be combined to construct a sub-network pair, that is, a distinguishing sub-network pair, of each frequency band.
In this embodiment, G is assumed+A brain function network set G of a certain frequency band in AD multi-frequency brain network data set-The method is a brain function network set of a certain frequency band in an NC multi-frequency brain network data set. From G+Wherein the excavated certain frequent sub-network is g+From G-Wherein the excavated certain frequent sub-network is g-. If subnetwork g+(g-) At G+(G-) Is much more frequent than it is in G-(G+) The sub-network is a distinguishing sub-network with 2-class distinguishing capability, and the distinguishing capability of the sub-network on two types of brain function networks of AD and NC can be calculated according to the formula (5):
similarly, subnetwork g can be calculated according to equation (6)-Discriminative power on two types of brain function networks:
and 7: calculating the difference degree of two sub-networks in each distinguishing sub-network pair, sorting the two sub-networks in the sequence from large to small according to the difference degree, and selecting the first k distinguishing sub-network pairs from the sorting sequence;
in this embodiment, for the pair of subnetworks (g)1,g2) Computing two subnetworks g in a subnetwork pair1，g2The expression of the degree of difference of (a) is as follows:
diff(g1,g2)＝1-corr(g1,g2) (7)
wherein corr (g)1,g2) For two sub-networks g1，g2The correlation between them is calculated according to equation (8):
corr(g1,g2)＝μsimN(g1,g2)+(1-μ)simS(g1,g2) (8)
in the above formula, simN (g)1,g2) For two sub-networks g1，g2The structural similarity of (a) is calculated according to the formula (9); SimS (g)1,g2) For two sub-networks g1，g2The support set similarity of (2) is calculated according to equation (10).
Where E (g) represents the edge set of subnetwork g.
In the above formula, cov (g)1，G+) As a subnetwork g1In network set G+The support degree in (1); cov (g)2，G-) As a subnetwork g2In network set G-The support degree in (1).
And 8: and sorting the difference degrees of the first k distinguishing sub-network pairs of all frequency bands according to a sequence from large to small, and selecting the first k' sub-network pairs as final distinguishing sub-network pairs for carrying out 2-class distinguishing on the given brain network to distinguish whether the given brain network is an AD brain function network or an NC brain function network, thereby realizing the auxiliary diagnosis of the AD.
In this embodiment, an arbitrary brain function network t to be tested is giveniUsing the obtained k' pairs of discriminating subnetworks tiAnd (4) performing category prediction, wherein the prediction result is used for auxiliary diagnosis of AD.
Using each of the k' dividing sub-network pairs obtained in step 6 according to the following determination rules (1) to (4)
For a given tested brain function network tiThe class prediction is carried out, and the result is recorded as mrWherein r is 1,2, …, k':
(1) if it is not
And is
Then m isr＝‘AD’；
(2) If it is not
And is
Then m isr＝‘NC’；
(3) If it is not
And is
Then m isr＝null；
(4) If it is not
And is
Then m isr＝null；
According to the above rule, t is respectively paired with k' sub-network pairsiPerforming prediction to calculate prediction result mrThe number of pairs of subnetworks, which are "AD" or "NC", is proportional to the total number of pairs k', giving the pair tiIf for tiAnd the final prediction result of the test and the pathological diagnosis label liIf the two networks are consistent, the method of the invention is proved to be applied to the given tested brain function network tiThe prediction is effective, and can provide effective service for the auxiliary diagnosis of AD.
It should be understood that various changes and modifications can be made by those skilled in the art without departing from the spirit and scope of the invention.
Claims (8)
1. A construction method of a multi-frequency brain network distinguishing sub-network pair for assisting in AD diagnosis is characterized by comprising the following steps:
step 1: acquiring fMRI original data, and processing the fMRI original data to obtain a time sequence signal of each interested area of each tested brain; the fMRI raw data comprise Alzheimer disease AD data and normal control NC data;
step 2: performing frequency division processing on the time sequence signal of each interested area, and dividing each time sequence signal into a plurality of frequency bands to obtain frequency division time sequence signals of the plurality of frequency bands;
and step 3: for each frequency band, respectively calculating correlation coefficients between any two interested areas, and representing the relevance between two interested brain area nodes in the same frequency band through the correlation coefficients;
and 4, step 4: performing thresholding treatment on the correlation coefficient, constructing a brain function network of each tested frequency band to obtain each tested multi-frequency brain network, and further obtaining an AD multi-frequency brain network data set and an NC multi-frequency brain network data set;
and 5: performing frequent sub-network mining on the brain function networks of all frequency bands in the AD multi-frequency brain network data set and the NC multi-frequency brain network data set to obtain frequent sub-networks of all frequency bands in each data set, and further obtaining an AD multi-frequency frequent sub-network set and an NC multi-frequency frequent sub-network set;
step 6: respectively calculating the respective abilities of distinguishing AD brain function networks and NC brain function networks of all sub-networks in the AD multi-frequency frequent sub-network set and the NC multi-frequency frequent sub-network set aiming at each frequency band, and respectively combining a plurality of sub-networks with the strongest distinguishing abilities in the AD multi-frequency frequent sub-network set and the NC multi-frequency frequent sub-network set to construct a distinguishing sub-network pair of each frequency band;
and 7: calculating the difference degree of two sub-networks in each distinguishing sub-network pair, respectively sequencing the distinguishing sub-network pairs of each frequency band according to the sequence of the difference degrees from large to small, and selecting the first k distinguishing sub-network pairs from the sequencing sequence of the distinguishing sub-network pairs of each frequency band;
and 8: and sorting the difference degrees of the first k distinguishing sub-network pairs of all the frequency bands according to a descending order, and selecting the front k' sub-network pairs as final distinguishing sub-network pairs for carrying out class prediction on the given brain network.
2. The method for constructing a pair of multi-frequency brain network differentiation sub-networks for assisting in the diagnosis of AD as claimed in claim 1, wherein the time series signal of each region of interest is frequency-divided by using a multi-scale wavelet transform method.
3. The method of constructing a pair of multi-frequency brain network differentiation sub-networks for aiding in the diagnosis of AD as claimed in claim 1, wherein said correlation coefficient is a Pearson correlation coefficient.
4. The method for constructing a pair of multi-frequency brain network differentiation sub-networks for assisting in the diagnosis of AD according to claim 1, wherein the method for mining frequent sub-networks for all brain function networks of each frequency band j in the data set of the multi-frequency brain network comprises:
(a) counting the frequency of all edges in all brain function networks of each frequency band j in a multi-frequency brain network data set, selecting the edges with the frequency greater than the preset support degree, taking each edge as a candidate sub-network, and obtaining a candidate sub-network set and recording the candidate sub-network set as Sj 1；
(b) For set Sj 1Each candidate subnetwork in (1) is expanded at most K times in a mode of adding one edge at a time, and a formed set is marked as Sj 2，Sj 3，…，Sj KAnd in the process of expanding, the set Sj 2，Sj 3，…，Sj KDeletion of duplicate subnets;
(c) computing a set Sj 2，Sj 3，……，Sj KThe frequency of each sub-network in the multi-frequency brain network data set is selected, the sub-network with the frequency of occurrence larger than the preset support degree is selected and is respectively marked as Sj 2’，Sj 3’，…，Sj K’Then get the frequent sub-network set S ═ { S) for each frequency bandj 1,Sj 2’,Sj 3’,…,Sj K’}。
5. The method for constructing a pair of multi-frequency brain network differentiation sub-networks for assisting in the diagnosis of AD as claimed in claim 4, wherein said support degree is defined as: for a given set of brain function networks P and sub-networks g thereinsWill sub-network gsSupport (g) in a network set PsP) is defined as a subnetwork gsThe number of occurrences in the network set P is a percentage of the total number N of networks in the network set P, that is:
wherein ,
is a sub-network gsSupport set of (1), GiRepresenting the networks in the network set P and N representing the number of networks contained in P.
6. The method for constructing a pair of multi-frequency brain network differentiation sub-networks for assisting in diagnosing AD according to claim 1, wherein the method for calculating respective 2-class differentiation capabilities of all sub-networks in the AD multi-frequency frequent sub-network set and all sub-networks in the NC multi-frequency frequent sub-network set for each frequency band is as follows: suppose G+A brain function network set G of a certain frequency band in AD multi-frequency brain network data set-A brain function network set of a certain frequency band in the NC multi-frequency brain network data set; from G+Wherein the excavated certain frequent sub-network is g+From G-Wherein the excavated certain frequent sub-network is g-(ii) a If subnetwork g+(g-) At G+(G-) Is much more frequent than it is in G-(G+) The sub-network is a distinguishing sub-network with 2-class distinguishing capability, and the distinguishing capability of the sub-network on the AD and NC brain function networks is calculated according to the formula (5):
similarly, subnetwork g-The discriminative power for two types of brain function networks is calculated according to equation (6):
7. the method for constructing pairs of multi-frequency brain network differentiation sub-networks for aiding in the diagnosis of AD as claimed in claim 5, wherein the calculation formula of the difference degree of two sub-networks in each pair of differentiation sub-networks is as follows:
diff(g1,g2)＝1-corr(g1,g2) (7)
wherein corr (g)1,g2) Is a sub-network pair (g)1,g2) In (1) sub-network g1，g2The correlation between them is calculated according to equation (8):
corr(g1,g2)＝μsimN(g1,g2)+(1-μ)simS(g1,g2) (8)
in the above formula, simN (g)1,g2) For two sub-networks g1，g2The structural similarity of (a) is calculated according to the formula (9); SimS (g)1,g2) For two sub-networks g1，g2The support set similarity of (2) is calculated according to the formula (10);
wherein ,E(g1) Shown is a sub-network g1The edge set of (1); e (g)2) Shown is a sub-network g2The edge set of (1);
in the above formula, cov (g)1，G+) As a subnetwork g1In network set G+The support degree in (1); cov (g)2，G-) As a subnetwork g2On a networkCollection G-The support degree in (1).
8. The method for constructing a pair of multi-frequency brain networks differentiation sub-networks for the auxiliary diagnosis of AD as claimed in claim 1, wherein the method for class prediction of a given brain network by using the pair of differentiation sub-networks is as follows: using each of the k' discriminating subnetwork pairs according to a decision rule
For a given tested brain function network tiThe class prediction is carried out, and the result is recorded as mrWhere r is 1,2, …, k', by calculating the prediction mrThe ratio of the number of sub-network pairs of "AD" or "NC" to the total number of sub-network pairs k' is obtainedi(iv) final predicted outcome of;
the judgment rule is as follows:
(1) if it is not
And is
Then m isr＝‘AD’；
(2) If it is not
And is
Then m isr＝‘NC’；
(3) If it is not
And is
Then m isr＝null；
(4) If it is not
And is
Then m isr＝null。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110684023.9A CN113298038B (en) | 2021-06-21 | 2021-06-21 | Construction method of multi-frequency brain network region molecular network pair for auxiliary diagnosis of AD |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110684023.9A CN113298038B (en) | 2021-06-21 | 2021-06-21 | Construction method of multi-frequency brain network region molecular network pair for auxiliary diagnosis of AD |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113298038A true CN113298038A (en) | 2021-08-24 |
| CN113298038B CN113298038B (en) | 2023-09-19 |
Family
ID=77328858
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110684023.9A Active CN113298038B (en) | 2021-06-21 | 2021-06-21 | Construction method of multi-frequency brain network region molecular network pair for auxiliary diagnosis of AD |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113298038B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108830827A (en) * | 2017-05-02 | 2018-11-16 | 通用电气公司 | Neural metwork training image generation system |
| CN109034263A (en) * | 2018-08-15 | 2018-12-18 | 东北大学 | The Alzheimer disease auxiliary diagnostic equipment and method of the brain network multi-frequency fusion kernel of graph |
| US20200396134A1 (en) * | 2019-06-11 | 2020-12-17 | At&T Intellectual Property I, L.P. | Apparatus and method for object classification based on imagery |
| WO2021026400A1 (en) * | 2019-08-06 | 2021-02-11 | Neuroenhancement Lab, LLC | System and method for communicating brain activity to an imaging device |
| EP3816853A1 (en) * | 2019-10-31 | 2021-05-05 | NVIDIA Corporation | Gaze determination using one or more neural networks |
-
2021
- 2021-06-21 CN CN202110684023.9A patent/CN113298038B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108830827A (en) * | 2017-05-02 | 2018-11-16 | 通用电气公司 | Neural metwork training image generation system |
| CN109034263A (en) * | 2018-08-15 | 2018-12-18 | 东北大学 | The Alzheimer disease auxiliary diagnostic equipment and method of the brain network multi-frequency fusion kernel of graph |
| US20200396134A1 (en) * | 2019-06-11 | 2020-12-17 | At&T Intellectual Property I, L.P. | Apparatus and method for object classification based on imagery |
| WO2021026400A1 (en) * | 2019-08-06 | 2021-02-11 | Neuroenhancement Lab, LLC | System and method for communicating brain activity to an imaging device |
| EP3816853A1 (en) * | 2019-10-31 | 2021-05-05 | NVIDIA Corporation | Gaze determination using one or more neural networks |
Non-Patent Citations (1)
| Title |
|---|
| 陈金义: "基于脑功能网络的阿尔茨海默症生物学标记物研究", 《中国优秀硕士学位论文全文数据库基础科学辑》, no. 5 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113298038B (en) | 2023-09-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113616184B (en) | Brain network modeling and individual prediction method based on multi-mode magnetic resonance image | |
| CN110188836B (en) | Brain function network classification method based on variational self-encoder | |
| CN113040715A (en) | Human brain function network classification method based on convolutional neural network | |
| CN110797123A (en) | Graph convolution neural network evolution method of dynamic brain structure | |
| CN113947157B (en) | Dynamic brain effect connection network generation method based on hierarchical clustering and structural equation model | |
| CN112348833B (en) | Dynamic connection-based brain function network variation identification method and system | |
| CN117172294B (en) | Method, system, equipment and storage medium for constructing sparse brain network | |
| CN112634214A (en) | Brain network classification method combining node attributes and multilevel topology | |
| Liao et al. | Classify autism and control based on deep learning and community structure on resting-state fMRI | |
| Huang et al. | Coherent pattern in multi-layer brain networks: Application to epilepsy identification | |
| Liu et al. | BrainTGL: A dynamic graph representation learning model for brain network analysis | |
| CN116433590A (en) | Mental disorder population whole brain function analysis method and system based on functional magnetic resonance image | |
| US20230148955A1 (en) | Method of providing diagnostic information on alzheimer's disease using brain network | |
| Ieva et al. | Component-wise outlier detection methods for robustifying multivariate functional samples | |
| CN114155952A (en) | Senile dementia illness auxiliary analysis system for elderly people | |
| Heena et al. | Machine learning based biomedical image processing for echocardiographic images | |
| Anderson et al. | Classification of spatially unaligned fMRI scans | |
| CN113298038B (en) | Construction method of multi-frequency brain network region molecular network pair for auxiliary diagnosis of AD | |
| Saneipour et al. | Improvement of MRI brain image segmentation using Fuzzy unsupervised learning | |
| CN112883905B (en) | Human behavior recognition method based on coarse-grained time-frequency features and multi-layer fusion learning | |
| CN114287910A (en) | Brain function connection classification method based on multi-stage graph convolution fusion | |
| Shekerbek et al. | APPLICATION OF MATHEMATICAL METHODS AND MACHINE LEARNING ALGORITHMS FOR CLASSIFICATION OF X-RAY IMAGES. | |
| Vallaboju et al. | Bioinformatics image based decision support system for bone cancer detection | |
| Asha et al. | Heart Block Recognition Using Image Processing and Back Propagation Neural Networks | |
| Naveen et al. | A Novel Layer Based Logical Approach (LLA) Clustering Method for Performance Analysis in Medical Images |
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 |