A new-born brain function evaluation and neurodevelopment prognosis prediction system

Abstract

The application discloses a new-born brain function evaluation and neural development prognosis prediction system, relates to the technical field of new-born brain function evaluation, and comprises an EEG electrode module, an fNIRS optical electrode module and a prediction module. The EEG electrode module is used for collecting EEG signals of a new-born in real time within a set time period. The fNIRS optical electrode module is used for collecting rSO2 signals of the new-born in real time within the set time period. The prediction module is used for performing signal preprocessing on EEG signals of the whole brain of the new-born and the rSO2 signals of the new-born, obtaining preprocessed EEG signals and preprocessed rSO2 signals, generating a dynamic brain function connection network sequence capable of representing the time evolution characteristics of brain function connection structures by using the preprocessed EEG signals and the preprocessed rSO2 signals, obtaining a time-varying feature vector by using the dynamic brain function connection network sequence, and obtaining a prediction result of new-born brain function evaluation and neural development prognosis based on the time-varying feature vector and a prediction model established in advance.

Description

Technical Field

[0001] This invention relates to the field of neonatal brain function assessment technology, and in particular to a neonatal brain function assessment and neurodevelopment prognosis prediction system. Background Technology

[0002] The neonatal period is a critical stage for brain development. Abnormalities during this stage can lead to a variety of clinical diseases and affect the newborn's subsequent learning abilities and social cognition. It is worth noting that early brain functional abnormalities are often reversible, while by the time structural brain damage occurs, the condition has often progressed to an irreversible state.

[0003] Currently, the main methods commonly used in clinical practice to assess neonatal brain function include electroencephalography (EEG), amplitude-integrated electroencephalography (aEEG), near-infrared spectroscopy (NIRS), and functional near-infrared spectroscopy (fNIRS), but each has its own limitations. EEG has high temporal resolution and can directly reflect the dynamic changes in neonatal brain electrical activity, but it contains complex spatiotemporal information, requiring interpretation by electrophysiologists, and it is difficult to reflect cerebral blood flow and metabolic status. aEEG is a simplified version of EEG, easier to apply and interpret, but its spectral and spatial information is greatly simplified, making it difficult to finely characterize the details of brain function. NIRS can non-invasively and continuously monitor neonatal brain tissue oxygenation and blood perfusion, reflecting brain metabolism and circulation, but its temporal resolution is low, and it cannot directly characterize neural electrical activity. fNIRS builds upon NIRS to achieve multi-channel and dynamic functional imaging, revealing brain function-related changes in blood oxygenation and functional connectivity between brain regions, but it is not sensitive to rapid neural activity.

[0004] In 2017, Chalak and Tian et al. developed a neurovascular wavelet transform method to differentiate between newborns with anencephaly and those with hypoxic-ischemic encephalopathy (HIE). This method first monitors the EEG activity at the C3-C4 sites in newborns using EEG and converts the EEG signals into aEEG signals; simultaneously, NIRS is used to monitor cerebral oxygen saturation (rSO2) in the prefrontal cortex. Both signals need to be continuously acquired simultaneously for more than 20 hours. Wavelet transforms are then performed on the simultaneously acquired aEEG and rSO2 signals (over 20 hours) to obtain the wavelet coherence between the two signals. Finally, the average wavelet coherence is calculated within a specific frequency band, and this average value is used to differentiate between newborns with anencephaly and those with HIE.

[0005] However, this technology has the following drawbacks:

[0006] (1) It is not possible to comprehensively assess the brain function and brain connectivity networks of newborns. This is because aEEG signals are only collected at the C3-C4 sites, and rSO2 signals are only collected in the prefrontal cortex (NIRS can only collect signals in the prefrontal cortex). Analyzing signals collected only from these two locations cannot comprehensively and accurately assess brain function. In addition, since both aEEG and rSO2 signals are collected from only a single location, it is also impossible to assess the brain connectivity networks throughout the entire brain.

[0007] (2) It cannot reflect the dynamic changes in the brain function of newborns. This is because the above method can only obtain a static value by analyzing aEEG and rSO2 signals collected synchronously for more than 20 hours. This value can only reflect the average state of brain function in these 20 hours and cannot reflect the dynamic changes in brain function.

[0008] (3) The neurodevelopmental prognosis of newborns cannot be predicted. Current technology has not conducted research on the prediction of neurodevelopmental prognosis.

[0009] (4) It has little clinical application value and is not very helpful for the clinical diagnosis and early intervention of neonatal brain developmental abnormalities. This is because the best results are achieved when neonates with brain developmental abnormalities are treated within 6 hours of birth, which can significantly reduce the incidence of death and neurodevelopmental disorders. The above methods require continuous signal collection for more than 20 hours to obtain information reflecting the neonatal encephalopathy. By then, treatment may be too late.

[0010] Therefore, timely assessment of brain function in the neonatal period plays an important role in the clinical diagnosis of brain developmental abnormalities, prediction of neurodevelopmental prognosis, and early intervention. Summary of the Invention

[0011] To address the aforementioned technical problems, this invention provides a neonatal brain function assessment and neurodevelopmental prognosis prediction system.

[0012] This invention provides a neonatal brain function assessment and neurodevelopmental prognosis prediction system, comprising:

[0013] The EEG electrode module has X EEG electrodes placed symmetrically on the newborn's head to collect the newborn's EEG signals in real time within a set time period. i ;

[0014] The fNIRS photoelectric module consists of Y fNIRS photoelectric electrodes placed around each EEG electrode on the newborn's head to collect the newborn's rSO2 signal OR in real time within a set time period. i,j ;

[0015] The prediction module is used to predict the EEG signal of the newborn. i and the rSO2 signal OR of the newborn i,j Signal preprocessing is performed to obtain the preprocessed EEG signal E. i and the preprocessed rSO2 signal O i,j And using the preprocessed EEG signal E i and the preprocessed rSO2 signal O i,j A dynamic brain functional connectivity network is generated that can characterize the evolution of brain functional connectivity structures over time. Using the dynamic brain functional connectivity network, a time-varying feature vector is obtained. Based on the time-varying feature vector and a pre-constructed prediction model, the prediction results of neonatal brain function assessment and neurodevelopmental prognosis are obtained.

[0016] Preferably, the prediction module includes:

[0017] A preprocessing unit is used to process the EEG signals of the newborn. i and the rSO2 signal OR of the newborn i,j Extracting the EEG signal EC within the same time period is performed along the time dimension. i and rSO2 signal OC i,j and the EEG signal EC i Converted to aEEG signal aEC i The EEG signal EC i and the aEEG signal aEC i Downsampled to the rSO2 signal OC i,j The frequency is used to obtain the preprocessed EEG signal E. i Preprocessed aEEG signal aEi and the preprocessed rSO2 signal O i,j ;

[0018] The segmentation unit is used to segment the preprocessed EEG signal of each EEG electrode within a set time period using a time-sliding window. i or preprocessed aEEG signal aE i And the preprocessed rSO2 signal O of each fNIRS photoelectric pole around it i,j The signal is segmented to obtain K sub-EEG signal segments arranged in chronological order. or sub-aEEG signal fragment and rSO2 signal fragment .

[0019] Preferably, the prediction module further includes:

[0020] The time-frequency coupling feature extraction unit is used to extract the sub-EEG signal segments from each EEG electrode. or the sub-aEEG signal segment and its subrSO2 signal segments with multiple fNIRS optical poles placed around it. Time-frequency domain analysis was performed to obtain the time-frequency coupling characteristics of each EEG electrode as each node under different time windows;

[0021] A dynamic brain functional connectivity network unit is constructed to calculate the connection weights of all node pairs under different time windows based on the time-frequency coupling characteristics of each node and other nodes under different time windows, and to obtain the brain functional connectivity matrix under different time windows based on the connection weights of all node pairs under different time windows; and to model each EEG electrode node as a node set, all EEG electrode pairs as an edge set, and the brain functional connectivity matrix under different time windows as edge weights, and to construct a dynamic brain functional connectivity network under different time windows based on the node set, the edge set, and the edge weights under different time windows.

[0022] Preferably, the prediction module further includes:

[0023] The time-varying feature extraction unit is used to extract multiple network parameters under each time window based on the dynamic brain functional connectivity network under different time windows, and obtain the time-varying features within a set time period based on the multiple network parameters under each time window.

[0024] The feature vector unit is used to form a feature vector by combining the time-varying features.

[0025] Preferably, the prediction module further includes:

[0026] A prediction model building and training unit is used to build a prediction model and train and optimize the parameters of the prediction model using training set data containing feature vectors to obtain a trained prediction model.

[0027] The prediction unit is used to obtain the prediction results of neonatal brain function assessment and neurodevelopmental prognosis by inputting the feature vector into the trained prediction model.

[0028] Preferably, the construction of the dynamic brain functional connectivity network unit is specifically used to calculate the connection weights of all node pairs under different time windows based on the time-frequency coupling characteristics of each node and other nodes under different time windows, using the difference metric method, the normalized similarity metric method, or the inverse distance method.

[0029] Preferably, the plurality of network parameters include node degree, clustering coefficient, feature path length, and global efficiency; correspondingly, the time-varying feature extraction unit is specifically used to perform statistical analysis on the node degree, clustering coefficient, feature path length, and global efficiency respectively within a set time period to obtain the mean and standard deviation of node degree, the mean and standard deviation of clustering coefficient, the mean and standard deviation of feature path length, and the mean and standard deviation of global efficiency.

[0030] Preferably, the node degree is obtained based on the edge weights in the dynamic brain functional connectivity network under different time windows.

[0031] Preferably, the clustering coefficient is obtained based on the edge weights and node degrees in the dynamic brain functional connectivity network under different time windows.

[0032] Preferably, the feature path length and the global efficiency are obtained based on the number of nodes in the node set of the dynamic brain functional connectivity network under different time windows and the shortest path length between nodes.

[0033] The beneficial effects of this invention are that by simultaneously acquiring neonatal electroencephalogram (EEG) and brain oxygenation signals, and introducing time-frequency domain analysis and functional connectivity network modeling methods, a multimodal, quantifiable, and dynamically analyzable neonatal brain function assessment and neurodevelopmental prognosis prediction technology system is constructed. First, this invention overcomes the limitations of traditional single-modal brain function assessment. By combining the high temporal resolution characterization of neural electrical activity by EEG with continuous monitoring of cerebral blood flow and metabolic status by brain oxygenation, it achieves complementary advantages between neural activity information and blood flow and metabolic information within the same framework, reflecting the neonatal brain function status more comprehensively from the physiological mechanism level. Second, this invention employs time-frequency domain analysis methods such as wavelet transform or Hilbert transform to perform multi-scale decomposition and joint characterization of multimodal signals. This not only characterizes the dynamic changes of signals in the time and frequency dimensions but also further extracts time-frequency coupling parameters reflecting neurovascular coupling characteristics, obtaining novel brain function features that are difficult to acquire using traditional time-domain or frequency-domain methods. Furthermore, this invention constructs a brain functional connectivity matrix and a brain functional connectivity network based on the aforementioned time-frequency coupling characteristics, and generates a dynamic brain functional connectivity network through a time sliding window approach. This characterizes the collaborative activities between different brain regions and their time-varying evolution at the network level, elevating brain function assessment from local signal analysis to the overall network organization and time-varying regulation level, thus improving sensitivity to brain functional abnormalities and developmental immaturity. In addition, this invention further extracts multidimensional network parameters from the dynamic brain functional connectivity network and establishes a mapping relationship between these parameters and neonatal neurodevelopmental prognosis, forming a neurodevelopmental prognostic prediction model that can be used for early risk stratification and outcome prediction. This provides an objective and quantitative technical means for the early identification and intervention of neurodevelopmental disorders. In summary, this invention achieves deep fusion and networked modeling of multimodal brain functional information without increasing invasive procedures, significantly improving the comprehensiveness, stability, and predictive value of neonatal brain function assessment, and has good clinical application prospects and promotional value. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of a module of a neonatal brain function assessment and neurodevelopment prognosis prediction system provided in Embodiment 1 of the present invention;

[0035] Figure 2 This invention provides Figure 1 A schematic diagram of the EEG electrode module;

[0036] Figure 3 This invention provides Figure 1 A schematic diagram of the fNIRS optical module;

[0037] Figure 4 This invention provides Figure 1 A schematic diagram of the prediction module;

[0038] Figure 5 This is a flowchart illustrating the neonatal brain function assessment and neurodevelopmental prognosis prediction method provided by the present invention.

[0039] Figure 6 This is a schematic diagram of the location for acquiring signals from the whole brain of a newborn, provided by the present invention.

[0040] Figure 7 This is a schematic diagram of the modules of the neonatal brain function assessment and neurodevelopment prognosis prediction system provided in Embodiment 2 of the present invention. Detailed Implementation

[0041] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In the following description, suffixes such as "module," "part," or "unit" used to denote elements are used only for the purpose of illustrative purposes and have no inherent meaning. Therefore, "module," "part," or "unit" may be used interchangeably.

[0042] Neonatal brain function is the result of a high degree of coupling between neural electrical activity and cerebral blood flow metabolism. Single-modal monitoring techniques are insufficient for a comprehensive and accurate assessment of brain function status. Electroencephalography (EEG / aEEG) has extremely high temporal resolution and can directly reflect neuronal electrical activity, but it cannot characterize cerebral blood flow and oxygen metabolism. National Intensive Care Unit (NIRS / fNIRS) can continuously reflect changes in brain tissue oxygenation and perfusion and is highly sensitive to hypoxia and metabolic abnormalities, but it is difficult to characterize rapid neural electrical activity. These two techniques are significantly complementary in terms of physiological mechanisms and information dimensions. Therefore, if EEG and NIRS signals can be monitored simultaneously and analyzed together to characterize both neural activity and metabolic status, it is hoped that a more comprehensive and accurate assessment of neonatal brain function can be achieved, and the reliability of predicting neurodevelopmental prognosis can be improved.

[0043] The present invention employs the following technical solutions: (a) EEG and rSO2 signals of the whole brain are collected at multiple locations using EEG and fNIRS; (b) Time-frequency domain analysis is used to calculate the time-frequency coupling characteristics between EEG (or aEEG) and rSO2 signals collected at adjacent locations; (c) Using the above time-frequency coupling characteristics as node feature inputs, a brain functional connectivity matrix reflecting the time-frequency coupling characteristics between multiple nodes is constructed, and a brain functional connectivity network is further generated; (d) Through time sliding windowing, the collected signals are divided into multiple time-varying sub-signal segments, and time-frequency domain analysis and brain functional connectivity network generation are repeated for all sub-signal segments to obtain a dynamic brain functional connectivity network reflecting the time-varying brain function; (e) The parameters of the dynamic brain functional connectivity network are extracted, and the mapping relationship between the above parameters and the prognosis of neonatal neurodevelopment is established through correlation analysis, regression analysis, or classification and discrimination models, thereby establishing a prediction model.

[0044] Figure 1This is a schematic diagram of a module of a neonatal brain function assessment and neurodevelopmental prognosis prediction system provided in Embodiment 1 of the present invention, as shown below. Figure 1 As shown, it includes:

[0045] The EEG electrode module has X EEG electrodes placed symmetrically according to the 10-20 system electrode placement method, such as... Figure 2 and Figure 6 As shown, this is used to collect EEG signals from newborns in real time within a set time period. i The set time period includes, but is not limited to, 6 hours; where i = 1, 2, ..., X;

[0046] The fNIRS photoelectric module consists of Y fNIRS photoelectric electrodes placed around each EEG electrode on the newborn's head, such as... Figure 3 and Figure 6 As shown, this is used to collect the rSO2 signal OR of newborns in real time within a set time period. i,j The set time period includes, but is not limited to, 6 hours; where i = 1, 2, ..., X; j = 1, 2, ..., Y;

[0047] The prediction module is used to predict the EEG signal of the newborn. i and the rSO2 signal OR of the newborn i,j Signal preprocessing is performed to obtain the preprocessed EEG signal E. i and the preprocessed rSO2 signal O i,j And using the preprocessed EEG signal E i and the preprocessed rSO2 signal O i,j A dynamic brain functional connectivity network is generated that can characterize the evolution of brain functional connectivity structures over time. Using the dynamic brain functional connectivity network, a time-varying feature vector is obtained. Based on the time-varying feature vector and a pre-constructed prediction model, the prediction results of neonatal brain function assessment and neurodevelopmental prognosis are obtained.

[0048] In one specific embodiment of the present invention, such as Figure 4 As shown, the prediction module includes:

[0049] A preprocessing unit is used to process the EEG signals of the newborn. i and the rSO2 signal OR of the newborn i,j Extracting the EEG signal EC within the same time period is performed along the time dimension. i and rSO2 signal OC i,j and the EEG signal EC i Converted to aEEG signal aEC i The EEG signal ECi and the aEEG signal aEC i Downsampled to the rSO2 signal OC i,j The frequency is used to obtain the preprocessed EEG signal E. i Preprocessed aEEG signal aE i and the preprocessed rSO2 signal O i,j ;

[0050] The segmentation unit is used to segment the preprocessed EEG signal of each EEG electrode within a set time period using a time-sliding window. i or preprocessed aEEG signal aE i And the preprocessed rSO2 signal O of each fNIRS photoelectric pole around it i,j The signal is segmented to obtain K sub-EEG signal segments arranged in chronological order. or sub-aEEG signal fragment and rSO2 signal fragment Where k = 1, 2, ..., K.

[0051] In one specific embodiment of the present invention, such as Figure 4 As shown, the prediction module further includes:

[0052] The time-frequency coupling feature extraction unit is used to extract the sub-EEG signal segments from each EEG electrode. or the sub-aEEG signal segment and its subrSO2 signal segments with multiple fNIRS optical poles placed around it. Time-frequency domain analysis was performed to obtain the time-frequency coupling characteristics of each EEG electrode as each node under different time windows;

[0053] A dynamic brain functional connectivity network unit is constructed to calculate the connection weights of all node pairs under different time windows based on the time-frequency coupling characteristics of each node and other nodes under different time windows, and to obtain the brain functional connectivity matrix under different time windows based on the connection weights of all node pairs under different time windows; and to model each EEG electrode node as a node set, all EEG electrode pairs as an edge set, and the brain functional connectivity matrix under different time windows as edge weights, and to construct a dynamic brain functional connectivity network under different time windows based on the node set, the edge set, and the edge weights under different time windows.

[0054] Furthermore, the construction of the dynamic brain functional connectivity network unit is specifically used to calculate the connection weights of all node pairs under different time windows based on the time-frequency coupling characteristics of each node and other nodes under different time windows, using the difference metric method, the normalized similarity metric method, or the inverse distance method.

[0055] In one specific embodiment of the present invention, such as Figure 4 As shown, the prediction module further includes:

[0056] The time-varying feature extraction unit is used to extract multiple network parameters under each time window based on the dynamic brain functional connectivity network under different time windows, and obtain the time-varying features within a set time period based on the multiple network parameters under each time window.

[0057] The feature vector unit is used to form a feature vector by combining the time-varying features.

[0058] Furthermore, the multiple network parameters include node degree, clustering coefficient, feature path length, and global efficiency. Correspondingly, the time-varying feature extraction unit is specifically used to perform statistical analysis on the node degree, clustering coefficient, feature path length, and global efficiency within a set time period to obtain the mean and standard deviation of the node degree, the mean and standard deviation of the clustering coefficient, the mean and standard deviation of the feature path length, and the mean and standard deviation of the global efficiency. Specifically, the node degree is obtained based on the edge weights in the dynamic brain functional connectivity network under different time windows; the clustering coefficient is obtained based on the edge weights and node degree in the dynamic brain functional connectivity network under different time windows; and the feature path length and global efficiency are obtained based on the number of nodes in the node set and the shortest path length between nodes in the dynamic brain functional connectivity network under different time windows.

[0059] In one specific embodiment of the present invention, such as Figure 4 As shown, the prediction module further includes:

[0060] A prediction model building and training unit is used to build a prediction model and train and optimize the parameters of the prediction model using training set data containing feature vectors to obtain a trained prediction model.

[0061] The prediction unit is used to obtain the prediction results of neonatal brain function assessment and neurodevelopmental prognosis by inputting the feature vector into the trained prediction model.

[0062] Figure 5 The diagram shown is a flowchart illustrating a neonatal brain function assessment and neurodevelopmental prognosis prediction method provided in an exemplary embodiment of this application, including the following steps:

[0063] Step S10: Collect EEG and rSO2 signals from the whole brain at multiple locations using EEG and fNIRS;

[0064] Using EEG technology at a high sampling frequency (not less than 200Hz), and following the 10-20 system electrode placement method, X electrodes (X≥8) were symmetrically placed on the newborn's scalp to collect EEG signals from the entire brain. i (i=1,2,…,X). Simultaneously, using fNIRS technology at a specific sampling frequency (not less than 0.5Hz), Y fNIRS electrodes (X≥1) were placed around each EEG electrode to collect rSO2 signals OR from the entire neonatal brain. i,j (i=1,2,…,X; j=1,2,…,Y). Simultaneously acquire the two types of signals for a shorter fixed time T (including but not limited to 6 hours). A schematic diagram of the whole-brain signal acquisition locations is shown below. Figure 6 As shown.

[0065] Step S20: Preprocess the acquired EEG and rSO2 signals;

[0066] First, based on the EEG signal ER i and rSO2 signal OR i,j Time information extraction: extract the common part of the two signals within the same time period (EC) i and OC i,j The original EEG and rSO2 signals from the common portion were then bandpass filtered within a specified frequency range (preferably 2-15 Hz) to strongly attenuate low-frequency physiological artifacts, muscle activity, and power line noise. The EEG signal was gradually amplified within this frequency range at a specified slope (preferably 12 dB / decade) to compensate for amplitude reduction due to attenuation from the scalp and skull. The EEG and rSO2 signals were then fitted with second-order polynomial curves to eliminate the tendency for slow drift, while threshold denoising was applied to suppress speckle noise. The EEG signal was converted to an aEEG signal (aEC) using the Washington University-Neonatal EEG Analysis Toolbox (WU-NEAT). i (i=1,2,…,X). Finally, the EEG and aEEG signals are downsampled to the frequency of the rSO2 signal. The preprocessed EEG signal E is then obtained. i (i=1,2,…,X), aEEG signal aE i (i=1,2,…,X), and rSO2 signal O i,j (i=1,2,…,X; j=1,2,…,Y).

[0067] Step 30: Calculate the time-frequency coupling characteristics between EEG / aEEG and rSO2 signals acquired at adjacent locations using time-frequency domain analysis.

[0068] For each electrode, the preprocessed EEG signal E i (It could also be the corresponding aEEG signal aE) i ), and compared with the preprocessed rSO2 signal O collected by the surrounding optical electrodes. i,j Time-frequency domain analysis is performed, using methods including but not limited to wavelet transform and Hilbert transform, to obtain time-frequency coupling characteristic values ​​F representing the N nodes (i.e., the locations of each EEG electrode) at different time and frequency scales. i (i=1,2,…,X).

[0069] Step S40: Using time-frequency coupling features as node feature input, construct a brain functional connectivity matrix that reflects the time-frequency domain coupling features between multiple nodes, and further generate a brain functional connectivity network.

[0070] The time-frequency coupling features F of each node obtained in the time-frequency coupling feature extraction step i As the node feature input, for any two spatial nodes m and n, based on their corresponding feature values ​​F m and F n Calculate the functional connection weights between node pairs. Preferably, the connection weights can be defined in the following manner:

[0071] (1) Difference measurement method:

[0072]

[0073] in, This is a scaling parameter used to control the degree of similarity decay; Let be the connection weight between any two spatial nodes m and n.

[0074] (2) Normalized similarity measurement method:

[0075]

[0076] (3) Inverse distance method:

[0077]

[0078] By using any of the above methods, the brain functional connectivity matrix can be constructed for all node pairs (m,n):

[0079]

[0080] Where X is the number of spatial nodes, and W is the matrix element. m,nIt characterizes the strength of functional associations between nodes and is used to reflect the coupling relationship between different brain regions in the multi-scale time-frequency domain.

[0081] Furthermore, a brain functional connectivity network G=(V,E,W) is constructed based on the aforementioned brain functional connectivity matrix, where the node set V consists of all spatial nodes, the edge set E consists of all node pairs, and the edge weights are determined by the brain functional connectivity matrix W of all node pairs. m,n This results in the formation of a brain functional connectivity network capable of characterizing the coordinated activity features of neonatal brain function in the time-frequency domain.

[0082] Step S50: By using a time sliding window, the acquired signal is divided into multiple time-varying sub-signal segments. Time-frequency domain analysis and brain functional connectivity network generation are repeated for all sub-signal segments to obtain a dynamic brain functional connectivity network that reflects the changes in brain function over time.

[0083] The acquired EEG and rSO2 signals are segmented using a time-sliding window method. A time window of length L and a sliding interval of step size Δ are defined. The time interval corresponding to the k-th time window is then:

[0084]

[0085] in, This refers to the time interval corresponding to the k-th time window; For the k-th time window The start time; The length of the time window; This represents the step size of the sliding interval.

[0086] By using a sliding time window, the original continuous signal is divided into K sub-signal segments arranged in chronological order:

[0087]

[0088] This represents the sub-signal segment corresponding to the i-th EEG electrode node in the k-th time window; This refers to the sub-signal segment corresponding to the j-th fNIRS photoelectrode around the i-th EEG electrode node in the k-th time window.

[0089] For each time window T k Time-frequency coupled feature extraction is performed on the signals of each node to obtain the feature value F of node i. i k Subsequently, a brain functional connectivity matrix for the k-th time window is constructed based on the feature values ​​of each node. :

[0090]

[0091]

[0092] Inter-node connection weights The difference measurement method described above can be used, or the normalized similarity measurement method or the inverse distance method can be used to calculate the connection weights between nodes. .

[0093] Further construct the corresponding brain functional connectivity network G k =(V,E,W k By combining the brain functional connectivity networks obtained under all time windows k=1,2,…,K in chronological order, a dynamic brain functional connectivity network sequence that can characterize the evolution of brain functional connectivity structures over time is formed. :

[0094]

[0095] This is a sequence of dynamic brain functional connectivity networks; This represents the brain functional connectivity network under the k-th time window.

[0096] This allows us to characterize the time-varying features and developmental trends of the brain function state in newborns.

[0097] Step S60: Extract dynamic brain functional connectivity network parameters, and establish a mapping relationship between brain functional connectivity network parameters and neurodevelopmental prognosis through correlation analysis, regression analysis, or classification and discrimination models, thereby establishing a prediction model.

[0098] Based on the dynamic brain functional connectivity network, network topology parameters are extracted for each time window. These network parameters include, but are not limited to:

[0099] (1) Node degree :

[0100]

[0101] (2) Clustering coefficient :

[0102]

[0103] Where n and h are two different neighboring nodes of node m, and the connection weights are calculated by pairwise combinations (mn, mh, nh), and finally the clustering coefficients are obtained through the above formula.

[0104] (3) Feature path length :

[0105]

[0106] Where, d m,n k This represents the shortest path length between node m and node n in the k-th time window.

[0107] (4) Global efficiency :

[0108]

[0109] Furthermore, statistical analysis is performed on the above parameters over time to extract their time-varying characteristics, including but not limited to mean and standard deviation. The time-varying statistical characteristics of the network parameters are then combined to form a feature vector. :

[0110]

[0111] in, For feature vectors; The mean of the node degree; The mean of the clustering coefficients; This represents the average length of the feature paths; This represents the average global efficiency. The standard deviation of the node degree; The standard deviation of the clustering coefficients; The standard deviation of the feature path length; The standard deviation of the global efficiency.

[0112] Subsequently, a predictive model is established based on the feature vector P and the known neonatal neurodevelopmental prognostic index ASQ (Ages and Stages Questionnaire) or Griffiths scale score Q. Preferably, a linear regression model is used:

[0113]

[0114] Where β is the model parameter and ε is the error term.

[0115] By training and optimizing the model parameters, a mapping relationship is established between the feature vector formed by the combination of time-varying statistical features of the network parameters and the results of neonatal brain function assessment and neurodevelopmental prognosis prediction. For a newborn to be assessed, inputting its corresponding feature vector P yields the prediction result Y, thereby achieving quantitative prediction and risk assessment of neurodevelopmental prognosis.

[0116] Figure 7The diagram shows a schematic of the modules of a neonatal brain function assessment and neurodevelopmental prognosis prediction system provided in an exemplary embodiment of this application, including: a whole-brain signal acquisition module M1: acquiring EEG and rSO2 signals of the whole brain at multiple locations using EEG and fNIRS; a signal preprocessing module M2: preprocessing the acquired raw EEG and rSO2 signals and aligning the acquisition time; a time-frequency coupling feature extraction module M3: calculating the time-frequency coupling features between EEG (or aEEG) and rSO2 signals acquired at adjacent locations using time-frequency domain analysis; and a brain functional connectivity network generation module M4: using the above-mentioned time-frequency coupling features as node features. The system takes input signals and constructs a brain functional connectivity matrix that reflects the time-frequency coupling characteristics between multiple nodes, further generating a brain functional connectivity network. The time-varying brain functional connectivity network construction module M5 uses a time sliding window to divide the collected signals into multiple time-varying sub-signal segments. Time-frequency domain analysis and brain functional connectivity network generation are repeated for all sub-signal segments to obtain a dynamic brain functional connectivity network reflecting the changes in brain function over time. The neurodevelopmental prognosis prediction module M6 extracts parameters from the dynamic brain functional connectivity network and establishes a mapping relationship between these parameters and neonatal neurodevelopmental prognosis through correlation analysis, regression analysis, or classification models, thereby establishing a prediction model.

[0117] The preferred embodiments of the present invention have been described above with reference to the accompanying drawings, but this does not limit the scope of the invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and spirit of the present invention should be within the scope of the present invention.

Claims

1. A neonatal brain function assessment and neurodevelopmental prognosis prediction system, characterized in that, include: The EEG electrode module has X EEG electrodes placed symmetrically on the newborn's head to collect the newborn's EEG signals in real time within a set time period. i ; The fNIRS photoelectric module consists of Y fNIRS photoelectric electrodes placed around each EEG electrode on the newborn's head to collect the newborn's rSO2 signal OR in real time within a set time period. i,j ; The prediction module is used to predict the EEG signal of the newborn. i and the rSO2 signal OR of the newborn i,j Signal preprocessing is performed to obtain the preprocessed EEG signal E. i and the preprocessed rSO2 signal O i,j And using the preprocessed EEG signal E i and the preprocessed rSO2 signal O i,j A dynamic brain functional connectivity network is generated that can characterize the evolution of brain functional connectivity structures over time. Using the dynamic brain functional connectivity network, a time-varying feature vector is obtained. Based on the time-varying feature vector and a pre-constructed prediction model, the prediction results of neonatal brain function assessment and neurodevelopmental prognosis are obtained.

2. The system according to claim 1, characterized in that, The prediction module includes: A preprocessing unit is used to process the EEG signals of the newborn. i and the rSO2 signal OR of the newborn i,j Extracting the EEG signal EC within the same time period is performed along the time dimension. i and rSO2 signal OC i,j and the EEG signal EC i Converted to aEEG signal aEC i The EEG signal EC i and the aEEG signal aEC i Downsampled to the rSO2 signal OC i,j The frequency is used to obtain the preprocessed EEG signal E. i Preprocessed aEEG signal aE i and the preprocessed rSO2 signal O i,j ; The segmentation unit is used to segment the preprocessed EEG signal of each EEG electrode within a set time period using a time-sliding window. i or preprocessed aEEG signal aE i And the preprocessed rSO2 signal O of each fNIRS photoelectric pole around it i,j The signal is segmented to obtain K sub-EEG signal segments arranged in chronological order. or sub-aEEG signal fragment and rSO2 signal fragment .

3. The system according to claim 2, characterized in that, The prediction module also includes: The time-frequency coupling feature extraction unit is used to extract the sub-EEG signal segments from each EEG electrode. or the sub-aEEG signal segment and its subrSO2 signal segments with multiple fNIRS optical poles placed around it. Time-frequency domain analysis was performed to obtain the time-frequency coupling characteristics of each EEG electrode as each node under different time windows; A dynamic brain functional connectivity network unit is constructed to calculate the connection weights of all node pairs under different time windows based on the time-frequency coupling characteristics of each node and other nodes under different time windows, and to obtain the brain functional connectivity matrix under different time windows based on the connection weights of all node pairs under different time windows; and to model each EEG electrode node as a node set, all EEG electrode pairs as an edge set, and the brain functional connectivity matrix under different time windows as edge weights, and to construct a dynamic brain functional connectivity network under different time windows based on the node set, the edge set, and the edge weights under different time windows.

4. The system according to claim 3, characterized in that, The prediction module also includes: The time-varying feature extraction unit is used to extract multiple network parameters under each time window based on the dynamic brain functional connectivity network under different time windows, and obtain the time-varying features within a set time period based on the multiple network parameters under each time window. The feature vector unit is used to form a feature vector by combining the time-varying features.

5. The system according to claim 4, characterized in that, The prediction module also includes: A prediction model building and training unit is used to build a prediction model and train and optimize the parameters of the prediction model using training set data containing feature vectors to obtain a trained prediction model. The prediction unit is used to obtain the prediction results of neonatal brain function assessment and neurodevelopmental prognosis by inputting the feature vector into the trained prediction model.

6. The system according to claim 3, characterized in that, The aforementioned dynamic brain functional connectivity network unit is specifically used to calculate the connection weights of all node pairs under different time windows based on the time-frequency coupling characteristics of each node and other nodes under different time windows, using the difference metric method, normalized similarity metric method, or inverse distance ratio method.

7. The system according to claim 4, characterized in that, The multiple network parameters include node degree, clustering coefficient, feature path length, and global efficiency. Accordingly, the time-varying feature extraction unit is specifically used to perform statistical analysis on the node degree, clustering coefficient, feature path length, and global efficiency within a set time period to obtain the mean and standard deviation of node degree, the mean and standard deviation of clustering coefficient, the mean and standard deviation of feature path length, and the mean and standard deviation of global efficiency.

8. The system according to claim 7, characterized in that, The node degree is obtained based on the edge weights in the dynamic brain functional connectivity network under different time windows.

9. The system according to claim 8, characterized in that, The clustering coefficient is obtained based on the edge weights and node degrees in the dynamic brain functional connectivity network under different time windows.

10. The system according to claim 7, characterized in that, Based on the number of nodes and the shortest path length between nodes in the dynamic brain functional connectivity network under different time windows, the feature path length and the global efficiency are obtained respectively.

Images