Space-time dynamics-based schizophrenia auxiliary diagnosis method

Abstract

The invention discloses a schizophrenia auxiliary diagnosis method based on spatio-temporal dynamics, and belongs to the technical field of biomedical image mode recognition. A functional brain network image is obtained by f-MRI, and a space-time dynamic model is constructed by means of a Markov chain, so that an individual functional brain network convergence mode and a maximum information flow transmission mode between brain networks can be extracted. Three corresponding index quantization function brain network spatio-temporal dynamic modes are provided, and the three corresponding index quantization function brain network spatio-temporal dynamic modes are a network stationary probability vector, a network optimal step number mean value matrix and a network direct connection strength matrix. According to the method, an effective auxiliary diagnosis model is constructed for three proposed functional brain network spatio-temporal dynamic indexes by means of a statistical test method and a logistic regression algorithm, so that accurate quantitative calculation of the probability of suffering from schizophrenia of an individual is realized, and a powerful technical support is provided for early diagnosis and intervention of schizophrenia; the kit fills the blank of the existing diagnosis technology in the field, and has remarkable clinical application value and scientific significance.

Description

Technical Field

[0001] The present invention belongs to the technical field of biomedical image pattern recognition, and in particular relates to a schizophrenia auxiliary diagnosis method based on spatiotemporal dynamics. Background Art

[0002] The maturity of brain imaging technology has greatly provided technical support and convenience for the study of diseases, and also provided many excellent methods for the diagnosis of mental illness. Among them, f-MRI (Functional Magnetic Resonance Imaging) technology reflects the activity of neurons over time by measuring changes in brain blood flow, and can be used to detect brain activity. Through f-MRI technology, doctors can determine the location of important functional areas of the brain, and can also find functional abnormalities in specific areas of the brain, thereby providing a reference for the diagnosis and treatment of diseases.

[0003] Schizophrenia (SZ) is a serious mental disorder that affects a person's thinking, emotions, behavior and perception. However, the patient's brain does not have obvious organic damage, so s-MRI (Structural Magnetic Resonance Imaging) technology, which focuses on reflecting tissue structure, is difficult to provide auxiliary diagnosis and treatment plans for it. Although traditional methods reflect real brain physiological activities through f-MRI technology, they often only model a simple space or time domain, and the extracted information is incomplete. Summary of the invention

[0004] The purpose of the present invention is to provide a method for auxiliary diagnosis of schizophrenia based on spatiotemporal dynamics. From the perspective of spatiotemporal dynamics, three indicators are proposed to describe the spatiotemporal information flow pattern of individual functional networks, namely, network stationary probability vector, network optimal step mean matrix and network direct connection strength matrix. The f-MRI image data is passed through a Markov chain to obtain a probability model of information flow transmission between brain regions or networks. The intergroup differences of the above indicators of the mental illness patient group and the healthy control (health control, HC) group are compared by statistical test methods, and then an effective diagnostic model is constructed by means of logistic regression to achieve accurate quantitative calculation of the probability of an individual suffering from schizophrenia, and provide strong technical support for the early diagnosis and intervention of schizophrenia. In order to solve the technical problem that the existing technology uses f-MRI technology to reflect real brain physiological activities, but often only models a simple space or time domain, and the extracted information is incomplete.

[0005] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0006] A method for auxiliary diagnosis of schizophrenia based on spatiotemporal dynamics, comprising the following steps:

[0007] Step S11: collecting f-MRI brain image data of individual subjects, and preprocessing the collected f-MRI brain image data to obtain preprocessed f-MRI brain image data.

[0008] Step S12: Using the cortical brain region partitioning atlas of N ROIs based on the Yeo 7 network, the preprocessed f-MRI brain image data is mapped to the N ROIs to obtain the time signal of the brain region.

[0009] Step S13: Divide the temporal signal of the brain region into sliding windows, and calculate the functional connectivity matrix in each sliding window to obtain a dynamic functional connectivity matrix.

[0010] Step S14: Calculate the brain region stationary probability vector and map it to the network to calculate the network stationary probability vector.

[0011] Step S15: Calculate the network optimal step mean matrix.

[0012] Step S16: Calculate the direct connection strength matrix.

[0013] Step S17: Construct an auxiliary diagnosis model through statistical analysis of inter-group differences and logistic regression.

[0014] Step S18: f-MRI images of the subjects to be diagnosed are collected, and after preprocessing, the three indicators of the network stationary probability vector, the network optimal step mean matrix, and the network direct connection strength matrix are calculated, and the elements with significant differences in each indicator are input into the corresponding auxiliary diagnosis model to calculate and predict the probability of suffering from schizophrenia.

[0015] Further, step S14 includes the following steps:

[0016] Step S141: Calculate the corrected brain region functional connection matrix, and correct it by setting the diagonal and negative correlation value elements in the functional connection matrix to 0 and performing false discovery rate correction on the positive correlation value, so that the brain region functional connection strength value is not less than 0.

[0017] Step S142: Calculate the brain region single-step transfer probability matrix and the 10-step transfer probability matrix. The single-step brain region transfer probability is calculated by normalizing the brain region functional connection matrix L1-norm. Before calculating the 10-step transfer probability matrix, a time window is set and the matrix is ​​calculated by continuous multiplication of 10 matrices within the time window.

[0018] Step S143: Calculate the 10-step transfer probability mean matrix, and obtain the brain region stationary probability vector through time arithmetic mean and column mean, and then calculate the network stationary probability vector through the brain region network mapping relationship.

[0019] Further, step S15 includes the following steps:

[0020] Step S151: Calculate the network functional connection matrix, which is obtained by mapping the brain region functional connection matrix.

[0021] Step S152: Calculate the network single-step transition probability matrix, which is calculated by standardizing the network functional connection matrix L1-norm.

[0022] Step S153: Calculate the network optimal step number matrix. Set a time window before calculation, and calculate the matrix elements by comparing the maximum values ​​of the elements of the multi-step transition probability matrix within the time window.

[0023] Step S154: Calculate the network optimal step number mean matrix, which is obtained by time arithmetic averaging of the network optimal step number matrix.

[0024] Further, step S16 includes the following steps:

[0025] Step S161: Calculate the network direct connection marking matrix. If there is a direct connection between networks, the corresponding position is marked as 1, otherwise it is marked as 0.

[0026] Step S162: Calculate the network direct connection strength matrix, where the network direct connection strength matrix is ​​the time arithmetic mean of the network direct connection mark values.

[0027] Further, step S17 includes the following steps:

[0028] Step S171: Perform t-test on the three indicator elements of the network stationary probability vector of the modeling sample group, the network optimal step mean matrix, and the network direct connection strength matrix to analyze the differences between groups and find out the significant difference elements of each indicator.

[0029] Step S172: construct a logistic regression hypothesis function and estimate the regression coefficient using the maximum likelihood estimation method.

[0030] Step S173: construct the prediction probability of the network stable probability vector-assisted diagnosis model, the prediction probability of the network optimal step mean matrix-assisted diagnosis model, and the prediction probability of the network stable probability vector-assisted diagnosis model.

[0031] Compared with the prior art, the present invention has the following beneficial technical effects: the present invention uses the principle of information dynamics and proposes to use the network stable probability vector, the network optimal step mean matrix and the network direct connection strength matrix as indicators to measure the information flow characteristics of brain network functions. The present invention uses statistical test methods and logistic regression algorithms to construct an effective auxiliary diagnosis model for the three network spatiotemporal dynamics indicators of network stable probability vector, network optimal step mean matrix and network direct connection strength matrix, so as to achieve accurate quantitative calculation of the probability of an individual suffering from schizophrenia, provide strong technical support for the early diagnosis and intervention of schizophrenia, fill the gap of existing diagnostic technology in this field, and has significant clinical application value and scientific significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0033] Figure 1 It is a schematic diagram of network stable vector calculation of the present invention.

[0034] Figure 2 It is a schematic diagram for calculating the network optimal step mean matrix and the network direct connection strength matrix of the present invention.

[0035] Figure 3 A schematic diagram of constructing the auxiliary diagnosis model of the present invention.

[0036] Figure 4 Schematic diagram of the construction and use of auxiliary diagnosis model for a single indicator. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] The present invention uses Markov chain to describe the transition between different states at a certain moment with probability. The state in the present invention is a brain region of interest (ROI, referred to as brain region) or a brain network (brain network, referred to as network).

[0039] The present invention proposes a method for auxiliary diagnosis of schizophrenia based on spatiotemporal dynamics, comprising the following steps:

[0040] Step S11: collecting f-MRI brain image data of individual subjects, and preprocessing the collected f-MRI brain image data to obtain preprocessed f-MRI brain image data.

[0041] Furthermore, preprocessing was performed in the following way: the data of the first four time points of the acquired f-MRI brain imaging data were eliminated, and the time length after eliminating the data of the first four time points was T, so as to eliminate the interference of unstable factors that may exist in the initial stage; then, slice time correction was implemented to ensure the accurate synchronization of the acquisition time of each slice; then, head motion correction operation was carried out to accurately compensate for the slight movement of the subject's head and reduce head motion artifacts; further, the image edge was aligned to the structural image; covariate regression was used to remove the potential influence of physiological covariates such as white matter signals, ventricular signals, and head movement on brain imaging data; finally, bandpass filtering technology was used to screen out effective signals in the frequency range of 0.01Hz-0.08Hz.

[0042] Step S12: Using the cortical brain region partitioning atlas of N ROIs based on the Yeo 7 network, the preprocessed f-MRI brain image data is mapped to the N ROIs to obtain the time signal of the brain region.

[0043] In order to distinguish each other, each brain region is represented by a serial number, and the i-th brain region is recorded as ROI. i The ROI time signal is represented by a matrix, denoted as Y, with a dimension of N×T, where N is the number of ROIs in the partitioned atlas and T is the time length of the time signal.

[0044] The cortical brain region division map is based on Yeo's 7-network division, which is a more refined division. Each ROI can be mapped to one of Yeo's 7 networks. 7 networks means there are 7 networks in total, and the kth network is represented by net k , where k∈{1,2,…,7}.

[0045] Step S13: Divide the temporal signal of the brain region into sliding windows, and calculate the functional connectivity matrix in each sliding window to obtain a dynamic functional connectivity matrix.

[0046] With L as the window length and 1 as the step length, the time signal of the ROI is divided into overlapping matrices over the time length T, called sliding windows. The tth sliding window is recorded as W t , corresponding to time t, where t∈{1,2,…,T-L+1}, the dimension of each sliding window is N×L. The t-th sliding window W t The i-th row vector (i.e., brain region ROI iIn the sliding window W t The time signal in The tth sliding window W t The j-th row vector (i.e., brain region ROI j In the sliding window W t The time signal in

[0047] Calculate the functional connectivity (FC) matrix of the brain region at time t, denoted as FC ROI (t). Where t∈{1,2,…,T-(L-1)}, the matrix elements Indicates the brain area ROI at time t i and brain region ROI j The functional connection strength between them, where i,j∈{1,2,…,N}, Brain region ROI i ROI j Pearson correlation coefficient at time t Right now Pearson correlation coefficient Depend on and The calculation results are:

[0048]

[0049] in express The sample mean of ; express The sample mean of ; represents the tth sliding window W t The kth item of the i-th row vector of ; The tth sliding window W t The kth entry of the jth row vector of .

[0050] Brain region functional connectivity matrix FC at time t ROI (t) dimension is N×N, FC ROI (t) is expressed as:

[0051]

[0052] By calculating the functional connectivity matrices at all times, a total of T-(L-1) functional connectivity matrices can be obtained. These matrices are called dynamic functional connectivity (dynamic FC, d-FC).

[0053] Step S14: Calculate the brain region stationary probability vector and map it to the network to calculate the network stationary probability vector.

[0054] Step S141: Calculate the corrected brain region functional connectivity matrix.

[0055] The brain region functional connectivity matrix FC at time t ROI The diagonal and negative correlation value elements in (t) are set to 0, and the positive correlation values ​​are further corrected for false discovery rate (FDR) so that the functional connection strength value of the brain region is not less than 0. The corrected brain region functional connection matrix at time t is obtained, which is recorded as FC ROI′ (t), t∈{1,2,…,T-(L-1)}, matrix elements Represents the corrected brain region ROI i and brain region ROI j The functional connection strength of , where i∈[1,N],j∈[1,N], and Matrix FC ROI′ (t) in the form of:

[0056]

[0057] Step S142: Calculate the brain region single-step transition probability matrix and the 10-step transition probability matrix.

[0058] Calculate the brain region single-step transition probability matrix A at time t (t) , where t∈{1,2,…,T-(L-1)}, the matrix elements Indicates the brain area ROI at time t i To the brain region ROI j The single-step transition probability can be simply written as i,j∈{1,2,…,N}. Its elements are calculated by L1-norm normalization of brain region functional connection matrix. The calculation formula of L1-norm normalization of brain region functional connection matrix is ​​as follows:

[0059]

[0060] Note that since the corrected brain region functional connectivity matrix FC is non-negative ROI′ (t)Diagonal elements so The brain region single-step transition probability matrix A at time t (t) The dimension is N×N, which realizes the normalization of probability and does not consider self-loops. The matrix A (t) It is expressed as:

[0061]

[0062] Assume A (τ,n) is the n-step transition probability matrix of the brain region at time τ, where τ∈{1,2,…T-(L-1)-(n-1)}, and the matrix elements Indicates that it is in ROI at time τ i Under the condition of j The probability of the ROI at time τ i To ROI j The n-step transition probability, i,j∈{1,2,…,N}, The calculation method is:

[0063]

[0064] in, Indicates that at time τ, i To ROI k1 The single-step transition probability, Indicates the time τ+1 from ROI k1 To ROI k2 The single-step transition probability of Indicates the time τ+n-2 from arrive The single-step transition probability, Indicates the time τ+n-1 from To ROI j The single-step transition probability.

[0065] In fact, according to the definition and rules of matrix multiplication, It can be calculated by a more convenient matrix multiplication operation. Multiply all the single-step probability matrices from time τ to time τ+n-1 in sequence to obtain the brain region n-step probability transfer matrix A at time τ (τ,n) , the element at position (i,j) is The calculation formula is as follows:

[0066]

[0067] In order to calculate the 10-step transition probability matrix of the brain region, the single-step transition probability matrix A of the brain region at 10 consecutive moments is (t) Form a sequence, which is called a time window. Then the τth time window Ω τ =[A (τ) ,A (τ+1) ,…,A (τ+9) ], where τ∈{1,2,…,(TL-8)}, because at most [T-(L-1)-(10-1)] time windows can be obtained, that is, (TL-8) time windows. τ Calculate the corresponding brain area 10-step transition probability matrix A (τ,10) , with a dimension of N×N, by τ All single-step transition probability matrices in are multiplied in sequence to obtain the result. The calculation formula is:

[0068]

[0069] in, That is, from ROI at time τ i To ROI j The 10-step transition probability.

[0070] Step S143: Calculate the 10-step transfer probability mean matrix, and obtain the brain region stationary probability vector through time arithmetic mean and column mean, and then calculate the network stationary probability vector through the brain region network mapping relationship.

[0071] Further calculate the 10-step transition probability mean matrix Matrix Elements Indicates that from ROI i To ROI j The time arithmetic mean of the 10-step transition probabilities, where i,j∈{1,2,…,N}, The calculation formula is:

[0072]

[0073] 10-step transition probability mean matrix The dimension is N×N, matrix The calculation formula is:

[0074]

[0075] Calculate the brain region stationary probability vector u=(u1,…,u j ,…,u N ), where j∈{1,2,…,N}. j Brain Region of Interest j The stationary probability, that is, to calculate the mean matrix of the 10-step transition probability The arithmetic mean of the elements in column j, u j The calculation method is:

[0076]

[0077] Brain Region ROI j The stationary probability u j When the brain region reaches a stable state, it converges from any brain region to the ROI j The transition probability of all brain regions is 1, which satisfies

[0078] The steady state means that after a large number of steps of transfer, the state distribution of the Markov chain will gradually stabilize, and the similarity of the elements in each row (corresponding to different initial states) in the transition probability matrix will become higher and higher. The final probability distribution almost no longer depends on the specific state at the start, but only on the target state to be reached.

[0079] The stable state here means that, under a sufficiently long time, the information flow model will show that the probability distribution of information transmission between brain regions (networks) has nothing to do with the starting brain region (network), but is only related to the target brain region (network) of the information flow. In other words, when the number of steps n is large enough (according to research results, it is often n ≥ 7), the n-step transfer probability matrix will show that the similarity of the row vectors increases sharply or even approaches consistency. Therefore, the stable probability vector u reflects the probability distribution of each brain region as the target brain region of the information flow in a stable state.

[0080] The kth network is net k , where k∈{1,2,…,7}, now calculate the network stability probability vector v=(v1,…,v k ,…,v7),v k That is, network k The probability of stability.

[0081] Any ROI can be mapped to a network. A network contains several ROIs. Let P k Indicates that it belongs to the network net k ROI number set, if ROI i Belong to the networknet k , denoted as i∈P k .

[0082] The network k The stationary probabilities of all brain regions included are added together, and the result is the network net k The stability probability of the network k The stationary probability v k The calculation formula is:

[0083]

[0084] j∈P k Representing ROI j Belong to the networknet k ;Networknet k The stationary probability v k When the network reaches a stable state, start from any network to net k The transition probability satisfies

[0085] The network stability probability vector v=(v1,…,vk ,…,v N ) reflects the probability of each network being the target network of information flow in a stable state.

[0086] Step S15: Calculate the network optimal step mean matrix.

[0087] Step S151: Calculate the network functional connection matrix at time t, where the network functional connection matrix is ​​obtained by mapping the brain region functional connection matrix.

[0088] The network functional connectivity matrix at time t is represented as FC net (t), where t∈{1,2,…,T-(L-1)}, the matrix elements Indicates that at time t, the network net k and network m The functional connection strength between them, where k,m∈{1,2,…,7}. And the functional connection strength between the same network is set to 0. When k≠m, Through the network k Midbrain ROI i (i∈P k ) and network m Midbrain ROI j (j∈P m ) after correction of the functional connectivity strength Double summation is performed, and considering that different brain networks contain different numbers of ROIs, the above sum needs to be divided by the product of the number of ROIs of the two brain networks to achieve normalization, which is The calculation formula is as follows:

[0089]

[0090] Where P k and P m Respectively represent the network net k and network m The set of ROI numbers contained in |P k | and |P m | respectively represent the network net k and network m The number of ROIs contained in the matrix FC net (t) has a dimension of 7×7 and is expressed as:

[0091]

[0092] Step S152: Calculate the network single-step transition probability matrix at time t, which is calculated by normalizing the network functional connection matrix L1-norm.

[0093] The network single-step transition probability matrix at time t is expressed as B (t) , where t∈{1,2,…,T-(L-1)}, the matrix elements Indicates that at time t, the network net k to network m The single-step transition probability is obtained by the network functional connection matrix L1-norm, where k,m∈{1,2,…,7}. The network single-step transition probability matrix calculation formula is:

[0094]

[0095] Since the network functional connectivity matrix FC net (t)Diagonal elements so The network single-step transition probability matrix B at time t (t) The dimension is 7×7, which realizes the normalization of probability and does not consider self-loops. The network single-step transition probability matrix B at time t (t) It is expressed as:

[0096]

[0097] Step S153: Calculate the network optimal step matrix. Before calculation, set a time window and calculate the matrix elements by comparing the maximum values ​​of the elements of the multi-step transition probability matrix within the time window.

[0098] In order to calculate the optimal step number matrix of the network, the network single-step transition probability matrix B at 10 consecutive moments is (t) Form a time window, let the τth time window be φ τ =[B (τ) ,…,B (τ+i) ,…,B (τ+9) ].

[0099] Time window φ τ The optimal step matrix of the network is denoted as S * (τ), where τ∈{1,2,…,TL-8}, and the matrix elements s km (τ) represents the time window τ from the network net k to network m The optimal number of steps, where k,m∈{1,2,…,7}.

[0100] The optimal number of steps refers to determining a specific number of steps within a given time window so that the probability sum of all reachable paths from the starting network to the target network reaches the maximum value and takes the minimum value, and the optimal number of steps between the same network is set to 0, so s km (τ)∈{0,1,2,…,10},s km(τ) is calculated as:

[0101]

[0102] in is the network net at time τ k to network m The s-step transition probability, that is, the matrix B (τ,s) The (k,m) position element of the matrix B (τ,s) is the s-step network transition probability matrix, Among them B (τ+i) is the network single-step transition probability matrix at time τ+i; The purpose is to find The set of s when the maximum value is obtained. The min function here is used if When there are at least two elements in the obtained s value set, take the smallest s value.

[0103] Then the time window φ τ The optimal step matrix S * (τ) dimension is 7×7, and the expression is:

[0104]

[0105] Step S154: Calculate the network optimal step number mean matrix, which is obtained by time arithmetic averaging of the network optimal step number matrix.

[0106] Calculate the network optimal step mean matrix Matrix Elements Indicates network net k to network m The optimal step time arithmetic mean, k,m∈{1,2,…,7}, The calculation method is as follows:

[0107]

[0108] The network optimal step mean matrix The characteristics of the spatiotemporal information path distance between individual networks are quantified from the perspective of spatiotemporal dynamics. The dimension is 7×7, and the expression is:

[0109]

[0110] Step S16: Calculate the direct connection strength matrix.

[0111] Step S161: Calculate the network direct connection marking matrix. If there is a direct connection between networks, the corresponding position is marked as 1, otherwise it is marked as 0.

[0112] Calculate the time window φ τ The network in is directly connected to the label matrix, denoted as I direct (τ), where τ∈{1,2,…,TL-8}, and the matrix elements are If in the time window φ τ China Network k to network m There is a direct connection, otherwise Right now where k,m∈{1,2,…,7},

[0113] Among them, if the network net in the time window τ k to network m The optimal number of steps s km (τ)=1, then it is said that in the time window φ τ China Network k and network m There is a direct connection between them, where k,m∈{1,2,…,7}. Calculation method:

[0114]

[0115] Matrix I direct The form of (τ) is:

[0116]

[0117] Step S162: Calculate the network direct connection strength matrix, where the network direct connection strength matrix is ​​the time arithmetic mean of the network direct connection mark values.

[0118] Calculate the direct connection strength matrix S direct , the matrix elements Indicates network net k net m The direct connection strength of the network k net m The time arithmetic mean of the direct connection mark values, and The calculation method is:

[0119]

[0120] Direct connection strength matrix S direct The form is:

[0121]

[0122] The network direct connection strength Sdirect Quantifies the average degree to which direct connections exist between individual networks in the temporal dimension.

[0123] Step S17: Construct an auxiliary diagnosis model through statistical analysis of inter-group differences and logistic regression.

[0124] Step S171: Perform t-test on the three indicator elements of the network stationary probability vector of the modeling sample group, the network optimal step mean matrix, and the network direct connection strength matrix to analyze the differences between groups and find out the significant difference elements of each indicator.

[0125] The pre-collected imaging data of the schizophrenia group (SZ group) and the healthy control group (HC group) were converted into three indicators: network stationary probability vector, network optimal step mean matrix, and network direct connection strength matrix.

[0126] For the convenience of explanation, the three indicators, namely the network stability probability vector, the network optimal step mean matrix, and the network direct connection strength matrix, are all regarded as vectors (if they are originally matrices, the elements are sorted by columns to form a vector).

[0127] The two-sample t-test was used to analyze the intergroup differences of all indicator elements, and FDR was used to correct for multiple hypothesis testing. If the p-value of an element after FDR correction was still no greater than 0.05, the element was considered to be significantly different.

[0128] The significant difference elements selected by the network stationary probability vector are recorded as x1,…,x i1 ,…,x n1 , n1 represents the number of significant difference elements in the network stationary probability vector; x i1 represents the significant difference element of the i1th network stationary probability vector, i1∈{1,…,n1}, and n1≤7.

[0129] The significant difference elements selected by the mean matrix of the optimal number of steps of the network are recorded as x1′,…,x i ′2,…,x′ n2 , n2 represents the number of significant difference elements in the mean matrix of the network optimal step number; x i ′2 represents the significant difference element of the mean matrix of the optimal number of steps of the i2th network, i2∈{1,…,n2}, and n2≤49.

[0130] The significant difference elements selected by the network direct connection strength matrix are recorded as x1″,…,x i ′3′,…,x′ n ′3, n3 represents the number of significant difference elements in the network direct connection strength matrix; x i '3' represents the significant difference element of the i3th network direct connection strength matrix, i3∈{1,…,n3}, and n3≤49.

[0131] Step S172: construct a logistic regression hypothesis function and estimate the regression coefficient using the maximum likelihood estimation method.

[0132] Use z, z′, z″ to represent the linear prediction value.

[0133] Construct a linear combination of network stationary probability vectors, and its calculation formula is:

[0134] z=β0+β1x1+…+β i1 x i1 +…+β n1 x n1

[0135] Among them, β0, β1…, β i1 ,…β n1 is the network stationary probability vector regression coefficient, and β0 is the intercept term.

[0136] Construct the linear combination of the network optimal step mean matrix, and its calculation formula is:

[0137] z′=β0′+β1′x1′+…+β i ′2x i ′2+…+β′ n2 x′ n2

[0138] Among them, β0′, β1′…, β i ′2,…,β′ n2 is the network optimal step mean matrix regression coefficient, β0′ is the intercept term.

[0139] Construct the linear combination of the network direct connection strength matrix, and its calculation formula is:

[0140] z″=β0″+β1″x1″+…+β i '3'x i '3'+...+β' n '3x' n '3

[0141] Among them, β0″,β0″…,β i ′3′,…,β′ n ′3 is the regression coefficient of the network direct connection matrix, and β0″ is the intercept term.

[0142] Step S173: Construct the prediction probability of the network stable probability vector auxiliary diagnosis model, the prediction probability of the network optimal step mean matrix auxiliary diagnosis model, and the prediction probability of the network stable probability vector auxiliary diagnosis model. Convert z, z′, z″ through the Sigmoid function to obtain the probability that the auxiliary diagnosis model predicts that the subject to be diagnosed is a patient. Its value is between 0 and 1, which can be interpreted as the probability of being a schizophrenia patient; if the probability is greater than 0.75, it can be judged that the subject suffers from schizophrenia.

[0143] The prediction probability of the network stationary probability vector-assisted diagnosis model is:

[0144]

[0145] Among them, e is a natural constant.

[0146] The prediction probability of the network optimal step mean matrix auxiliary diagnosis model is:

[0147]

[0148] The prediction probability of the network stationary probability vector-assisted diagnosis model is:

[0149]

[0150] The regression coefficients of each indicator are estimated using the maximum likelihood estimation method, so that the auxiliary diagnosis model can predict the known patient groups and healthy groups as accurately as possible. The independent variables of the training logistic regression model are the significant difference element values ​​of each indicator of the subjects, and the dependent variable is the label value y i , where y i is the label value of the ith subject. If the ith subject is a patient, let y i =1, otherwise record y i =0.

[0151] Step S18: Collect f-MRI images of the subjects to be diagnosed, and after preprocessing, calculate the three indicators of the network stationary probability vector, the network optimal step mean matrix, and the network direct connection strength matrix, input the elements with significant differences in each indicator into the corresponding auxiliary diagnosis model, and calculate and predict the probability of suffering from schizophrenia.

[0152] For the network stable probability vector, elements with significant differences are extracted, and the extracted elements showing differences are input into the stable probability vector auxiliary diagnosis model to calculate the predicted probability of the network stable probability vector.

[0153] For the network optimal step mean matrix, elements with significant differences are extracted, and the extracted elements showing differences are input into the network optimal step mean matrix auxiliary diagnosis model to calculate the prediction probability of the network optimal step mean matrix.

[0154] For the network direct connection strength matrix, the elements with significant differences are extracted, and the extracted elements showing differences are input into the network stationary probability vector auxiliary diagnosis model to calculate the predicted probability of the network direct connection strength matrix.

[0155] Experts further analyzed the predicted probabilities of the three indicators and finally obtained a comprehensive diagnosis result.

[0156] The spatiotemporal dynamics-based schizophrenia auxiliary diagnosis method of the present invention is described below with a specific implementation.

[0157] The data used in the embodiments of the present invention are all static data collected by Siemens machine (magnetic field strength 3.0T). The scanning parameters are echo time (TE) 0.03s, repetition time (TR) 2.5s, flip angle 90°, layer thickness 3mm, spatial resolution 3.75mm×3.75mm×4mm, and 212 time points are collected.

[0158] Preprocessing of f-MRI images was based on the CBIG pipeline tool (based on FSL and FreeSurfer), and the processing steps included removal of the first four time points, slice time correction, motion correction, registration of image edges to the structural image, regression of covariates (white matter signal, ventricular signal, head motion), and bandpass filtering (0.01–0.08 Hz).

[0159] Among them, subjects with the following characteristics in the original data were excluded: severe visual segmentation errors, intra-individual registration cost exceeding 0.7, abnormal FD (framewise displacement) index, too high a proportion of the reviewed volume data, and unsatisfactory registration of functional and anatomical images.

[0160] The network stationary probability vector is calculated as follows:

[0161] like Figure 1 For each subject, the preprocessed f-MRI brain image data were mapped to the brain region (ROI) level using a 400-ROI partitioning atlas based on the Yeo 7 network (see Figure 1 In a), the f-MRI image data is mapped to functional image signals at the brain region level, and the signal matrix dimension is 400×208.

[0162] With a window length of 50 and a step length of 1, 159 sliding windows were divided, each with a dimension of 50×208, corresponding to time 1 to time 159. In each sliding window, the Pearson correlation value between ROIs was calculated to obtain the brain region dynamic functional connectivity matrix (see Figure 1b), and finally 159 functional connectivity matrices were obtained, all with a dimension of 400 × 400. For each functional connectivity matrix, negative values ​​were set to zero, and positive correlation values ​​were further corrected for false discovery rate (FDR) (see Figure 1 c), and then the corrected functional connectivity matrix was standardized to obtain the single-step transition probability matrix of 159 brain regions (see Figure 1 d), the dimensions are all 400×400.

[0163] The single-step probability transfer matrices of 10 consecutive moments are combined into a time window, and a total of 150 time windows are obtained. All single-step probability matrices in the time window are multiplied in turn to obtain a 10-step transfer probability matrix (see Figure 1 e), calculate the 10-step transfer probability mean matrix, and calculate the brain region stationary probability vector by averaging it column by column (see Figure 1 f), and then mapped to the network stability probability vector based on the cortical partitioning map based on the Yeo7 network (see Figure 1 g) in.

[0164] The network optimal step mean matrix and network direct connection strength matrix are calculated as follows:

[0165] like Figure 2 As shown, the corrected functional connection matrix of the brain region is mapped to the network functional connection matrix (see Figure 2 a), the number is 159, the dimension is 7×7, and the network single-step transition probability matrix is ​​calculated by L1-norm (see Figure 2 b), the number is 159 and the dimension is 7×7. It is divided into 150 time windows in the same way. According to the direct connection marking rule, the direct connection marking matrix of each time window is calculated (see Figure 2 d), calculate the optimal step matrix mean and direct connection strength matrix (see Figure 2 e), the dimensions are all 7×7.

[0166] The auxiliary diagnosis model is constructed in the following way:

[0167] like Figure 3 As shown in Figure 1, indicator A represents the network stability probability vector, indicator B represents the network optimal step mean matrix, and indicator C represents the network direct connection strength matrix. A t-test was performed on each element of the three indicators of the subjects in the existing sample set to analyze the differences between groups, and the significant difference elements of each indicator after FDR correction were screened (see Figure 3 Step a). Use the existing sample set to estimate the corresponding regression coefficient using the maximum likelihood estimation method (see Figure 3 Step b) uses the conversion formula to convert the predicted value into a predicted probability (see Figure 3 Step c).

[0168] Specifically, Figure 4 The prediction model construction of indicator A is used as an example to explain in detail. The prediction probability formula is obtained through t-test, FDR correction, and maximum likelihood estimation. The indicator is calculated for the imaging data of the subjects to be diagnosed, and the significant difference elements are selected. The prediction probability is calculated through the estimated regression coefficient (see Figure 4 Step d). The predicted probabilities of the three indicators are calculated respectively, and the experts make a comprehensive analysis to determine whether the diagnosis is confirmed.

[0169] Compared with other auxiliary diagnostic methods for schizophrenia, this method proposes three indicators, namely, network stationary probability vector, network optimal step mean matrix and network direct connection strength, from the perspective of spatiotemporal dynamics. It uses statistical tests and logistic regression methods to derive a probability calculation formula for evaluating an individual's diagnosis of schizophrenia.

[0170] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.

Claims

1. A method for auxiliary diagnosis of schizophrenia based on spatiotemporal dynamics, characterized in that: The steps include: Step S11: collecting f-MRI brain image data of individual subjects, and preprocessing the collected f-MRI brain image data to obtain preprocessed f-MRI brain image data; Step S12: using the cortical brain region partitioning atlas of N ROIs based on the Yeo 7 network, mapping the preprocessed f-MRI brain image data to the N ROIs to obtain the time signal of the brain region; Step S13: Divide the time signal of the brain area into sliding windows, and calculate the functional connection matrix in each sliding window to obtain a dynamic functional connection matrix; Step S14: Calculate the brain region stationary probability vector and map it to the network to calculate the network stationary probability vector; Step S15: Calculate the network optimal step mean matrix; Step S16: Calculate the direct connection strength matrix; Step S17: constructing an auxiliary diagnosis model through statistical analysis of inter-group differences and logistic regression; Step S18: f-MRI images of the subjects to be diagnosed are collected, and after preprocessing, the three indicators of the network stationary probability vector, the network optimal step mean matrix, and the network direct connection strength matrix are calculated, and the elements with significant differences in each indicator are input into the corresponding auxiliary diagnosis model to calculate and predict the probability of suffering from schizophrenia.

2. The method for auxiliary diagnosis of schizophrenia based on spatiotemporal dynamics according to claim 1, characterized in that: Step S14 includes the following steps: Step S141: Calculate the corrected brain region functional connection matrix, and correct it by setting the diagonal and negative correlation value elements in the functional connection matrix to 0 and performing false discovery rate correction on the positive correlation value, so that the brain region functional connection strength value is not less than 0; Step S142: calculating the brain region single-step transfer probability matrix and the 10-step transfer probability matrix. The single-step brain region transfer probability is calculated by the L1-norm normalization of the brain region functional connection matrix. A time window is set before calculating the 10-step transfer probability matrix, and the matrix is ​​calculated by continuous cumulative multiplication of 10 matrices within the time window. Step S143: Calculate the 10-step transfer probability mean matrix, and obtain the brain region stationary probability vector through time arithmetic mean and column mean, and then calculate the network stationary probability vector through the brain region network mapping relationship.

3. The method for auxiliary diagnosis of schizophrenia based on spatiotemporal dynamics according to claim 2, characterized in that: Step S15 includes the following steps: Step S151: calculating a network functional connection matrix, where the network functional connection matrix is ​​obtained by mapping the brain region functional connection matrix; Step S152: Calculate the network single-step transition probability matrix, which is calculated by the network functional connection matrix L1-normal form normalization; Step S153: Calculate the network optimal step number matrix. Set a time window before calculation, and calculate the matrix elements by comparing the maximum values ​​of the elements of the multi-step transition probability matrix within the time window; Step S154: Calculate the network optimal step number mean matrix, which is obtained by time arithmetic averaging of the network optimal step number matrix.

4. The method for auxiliary diagnosis of schizophrenia based on spatiotemporal dynamics according to claim 3, characterized in that: Step S16 includes the following steps: Step S161: Calculate the network direct connection marking matrix. If there is a direct connection between networks, the corresponding position is marked as 1, otherwise it is marked as 0; Step S162: Calculate the network direct connection strength matrix, where the network direct connection strength matrix is ​​the time arithmetic mean of the network direct connection mark values.

5. The method for auxiliary diagnosis of schizophrenia based on spatiotemporal dynamics according to claim 4, characterized in that: Step S17 includes the following steps: Step S171: Perform t-tests on the three indicator elements of the network stationary probability vector, the network optimal step mean matrix, and the network direct connection strength matrix of the modeling sample group to analyze the differences between the groups and find out the significant difference elements of each indicator; Step S172: construct a logistic regression hypothesis function and estimate the regression coefficient using the maximum likelihood estimation method; Step S173: construct the prediction probability of the network stable probability vector-assisted diagnosis model, the prediction probability of the network optimal step mean matrix-assisted diagnosis model, and the prediction probability of the network stable probability vector-assisted diagnosis model.

Images