An attention deficit hyperactivity disorder recognition system based on the developmental norm of brain functional connectome

Abstract

The present invention provides an attention deficit hyperactivity disorder (ADHD) recognition system based on a brain functional connectome development norm. It classifies and recognizes ADHD by constructing a brain functional connectome development norm and using the obtained individual development deviation index of brain functional connections. The present invention provides an ADHD recognition system based on a brain functional connectome development norm. By constructing a norm for the variation of brain functional connection features with age, the degree of individual deviation from the normal range is characterized, and the recognition of ADHD patients is achieved based on the obtained individual deviation index, thereby providing a quantitative and individual-level new brain imaging biomarker for ADHD patients. In norm construction, the Bayesian spline regression method used in the present invention can more precisely model non-linear distributions by combining B-spline transformations, and at the same time can also shorten the calculation time. Therefore, it can be efficiently extended to large brain imaging data cohorts and flexibly applied to new sites not included in the training samples.

Description

Technical Field

[0001] The present invention relates to a system for identifying attention deficit hyperactivity disorder based on a brain functional connectome development norm. Background Art

[0002] Attention deficit hyperactivity disorder (ADHD) is one of the most common neurodevelopmental disorders in children and adolescents. Its core symptoms are age-inappropriate inattention, impulsivity, and hyperactive behaviors. Currently, the diagnosis of ADHD mainly relies on clinical assessment and behavioral observation, which is highly subjective. Therefore, there is an urgent need to reveal objective and reliable biomarkers for early identification and precise diagnosis and treatment of ADHD.

[0003] CN202310853785.6, Invention Title: A Method and System for Assisting in the Diagnosis of Attention Deficit Hyperactivity Disorder, provides a method and system for assisting in the diagnosis of attention deficit hyperactivity disorder, which relates to the fields of medical image processing and auxiliary diagnosis. The method for assisting in the diagnosis of attention deficit hyperactivity disorder includes: dividing the acquired brain time series data into at least one continuous and non-overlapping sub-time window; extracting discriminative features of each sub-time window and obtaining a high-order functional connection network through the extracted discriminative features; capturing the dynamic features of the high-order functional connection network of each acquired sub-time window through a TCN module to depict the relationship between brain functional connections over time; obtaining the value of the correlation between each sub-time window and the disease according to the dynamic change law of brain diseases, and using the value of the correlation between each sub-time window and the disease as a weight to perform feature fusion on the high-order functional connection network of each sub-time window. This invention uses dynamic functional connection for the auxiliary diagnosis of ADHD. In order to capture the evolving characteristics of functional connections over time, dynamic functional connection usually requires longer scanning time and higher time resolution to obtain more accurate and representative dynamic connection patterns, which is highly dependent on the scanning time and increases the difficulty of data acquisition. In addition, dynamic functional connection is more sensitive to noise. Given that dynamic functional connection is easily affected by scanning time, time resolution, and noise, and the noise levels and scanning time parameter settings are different due to the equipment and operating environment of different scanning centers, the generalization ability of the model may be reduced.

[0004] CN202311304937.3, Invention Title: An Auxiliary Diagnosis System for Attention Deficit Hyperactivity Disorder Syndrome Based on an Omnidirectional Cognitive Experiment Set and a Multi-Stream Dual-Line Decision Neural Network Model, provides a reaction test system for attention deficit hyperactivity disorder syndrome. The reaction test system consists of multiple cognitive reaction experiments, which are all experiments designed and optimized based on cognitive psychology and cognitive neurology data. It achieves an all-round assessment of the subject. During the experiment, the electroencephalogram (EEG) data of the subject is collected. Then, based on the EEG data of the experiment, the frequency bands of interest in the reaction experiment are processed, and data conversion and feature extraction are sent into a multi-stream dual-line decision neural network (MSDD) model. This model selects different learning schemes according to different data types, learns the EEG data features, and compares their similarity with the healthy norm, so as to achieve the purpose of auxiliary diagnosis. This invention uses EEG data to identify ADHD. The spatial resolution of EEG is much lower than that of functional magnetic resonance imaging (fMRI), making it difficult to accurately locate the activity source inside the brain. In addition, EEG signals are easily affected by electromyographic interference (such as blinking, facial muscle activities, etc.) and external electromagnetic interference, and may require complex signal processing techniques to purify the data.

[0005] CN201910987452.6, Invention Title: An Auxiliary Diagnosis Device for Attention Deficit Hyperactivity Disorder and Its Using Method, proposes an auxiliary diagnosis device for attention deficit hyperactivity disorder and its using method, including: a display screen for interacting with the tested person and displaying the test scenario; a wearable acceleration sensor for collecting the limb acceleration of the tested person in this test scenario; a data processing module for preprocessing the limb acceleration to obtain the data to be tested, inputting the data to be tested into an attention deficit hyperactivity disorder detection model based on a deep neural network for identification to obtain an auxiliary diagnosis result, and sending it to the display screen. This invention uses a wearable acceleration sensor to measure limb acceleration to identify ADHD. However, the motion data of the acceleration sensor may be affected by various factors, such as the state of the subject (such as anxiety, fatigue, excitement, etc.), motion errors. In addition, some other neuropsychiatric diseases may also show abnormal movements, such as bipolar disorder, schizophrenia, tic disorder, etc. Therefore, its specificity for auxiliary diagnosis of ADHD may be insufficient.

[0006] CN202010922095.8, Invention Title: ADHD Disease Diagnosis Aided Decision-making System Based on Deep Convolutional Spiking Neural Network, discloses an attention deficit hyperactivity disorder (ADHD) disease diagnosis aided decision-making system based on a deep convolutional spiking neural network, including: a preprocessing device for preprocessing functional magnetic resonance imaging data using a signal-to-noise ratio feature selection method; an encoding device for encoding the preprocessed functional magnetic resonance imaging data using a forward algorithm to generate a spiking sequence; a diagnostic prediction output device for using a deep convolutional spiking neural network to perform diagnostic prediction based on the spiking sequence and output a diagnostic prediction result. This invention uses the spiking sequence generated from the original functional magnetic resonance imaging data for ADHD diagnosis aided decision-making. The data dimension is very high, the processing and analysis are extremely complex, and the interpretability of the indicators is poor and the physiological significance is unclear.

[0007] In addition to the different observation indicators between the present invention and several other inventions as described above, previous inventions all constructed machine learning classification models based on observation data and judged ADHD according to the diagnostic results output by the classification models. However, machine learning models, especially deep learning models, are often regarded as "black boxes" and it is difficult to explain their decision-making processes. In the medical field, especially in mental health diagnosis, clinical practitioners need to understand the decision-making basis of the models, and this non-explainability hinders their application. The brain functional connectome analysis technology based on resting-state functional magnetic resonance imaging is considered the most effective means to reveal the brain functional mechanisms of various mental diseases and has been widely applied to ADHD, such as Sripada, C., et al. Disrupted network architecture of the resting brain in attention-deficit / hyperactivity disorder. Human brain mapping 35, 4693-4705 (2014); Norman, L.J., Sudre, G., Price, J., Shastri, G.G. & Shaw, P. Evidence from "big data" for the default-mode hypothesis of ADHD: a mega-analysis of multiple large samples. Neuropsychopharmacology 48, 281-289 (2023); Norman, L.J., Sudre, G., Price, J. & Shaw, P. Subcortico-Cortical Dysconnectivity in ADHD: A Voxel-Wise Mega-Analysis Across Multiple Cohorts. The American journal of psychiatry 181, 553-562 (2024).

[0008] The above studies have all explored the differences in the characteristics of the brain functional connectome between the ADHD patient group and the normal development control group, and found that there are abnormal functional connections in the default mode network, fronto-parietal network, salience network, etc. in ADHD patients. However, the results among the studies are highly inconsistent and cannot be clinically translated and applied. Since ADHD is a highly heterogeneous disease, there are significant differences among different patients in terms of clinical symptom manifestations and potential biological characteristics of the disease. The above studies all adopted the traditional case-control comparison method, regarding patients with the same diagnosis as a homogeneous group. Therefore, the results obtained reflect the differences in group averages, ignoring the heterogeneity among individual patients and also ignoring the impact of development on brain function. Such differences at the group level are not generalizable, and different results will be obtained when comparing and analyzing a different group of patients. Therefore, it cannot be applied to the precise identification of individual ADHD patients.

[0009] The emergence of the normative modeling method provides a new idea for solving this problem. This method constructs a norm of brain structure or functional characteristics based on the brain imaging data of a large sample of healthy people. The obtained norm can reflect the typical patterns and variation reference ranges of brain structure or function at a specific age, gender, and cognitive state, so that it can be used to evaluate the degree of deviation of an individual from the normal pattern, "microscopically" distinguish the differences of each patient at the individual level, which is of great significance for the early diagnosis and precise treatment of patients with mental disorders.

[0010] Currently, there is no developmental norm of the brain functional connectome for children and adolescents, and even less is there a use of the constructed developmental norm of the brain functional connectome for the identification of patients with attention deficit hyperactivity disorder. Summary of the Invention

[0011] The present invention proposes an attention deficit hyperactivity disorder identification system based on the developmental norm of the brain functional connectome. By constructing a reference model of the changes in brain functional connectome characteristics with age in the normally developing population, the degree of deviation of each ADHD patient from the normal range can be obtained according to this model, and an individualized brain functional connection deviation pattern can be obtained. Based on this deviation pattern, classification and discrimination between ADHD patients and normally developing controls are carried out, so as to realize the identification of ADHD patients.

[0012] The present invention provides an attention deficit hyperactivity disorder identification system based on the developmental norm of the brain functional connectome, which classifies and identifies attention deficit hyperactivity disorder by constructing the developmental norm of the brain functional connectome and using the obtained individual developmental deviation index of the brain functional connection.

[0013] Among them, the developmental norm of the brain functional connectome is constructed by using Bayesian spline regression.

[0014] The attention deficit hyperactivity disorder recognition system based on the brain functional connectome development norm includes the following functional modules:

[0015] a. Data acquisition module: used to collect the demographic and magnetic resonance imaging data of the subjects;

[0016] b. Data preprocessing module: connected to the data acquisition module, used to perform preprocessing such as correction, registration, noise reduction, smoothing, and filtering on the collected resting-state functional magnetic resonance imaging data;

[0017] c. Brain functional connectome calculation module: connected to the data preprocessing module, generates a functional connection matrix from the preprocessed resting-state functional magnetic resonance imaging data, and calculates the functional connections of brain regions and brain networks;

[0018] d. Brain functional connectome development norm construction module: connected to the brain functional connectome calculation module, in the normal development control group, uses Bayesian spline regression to construct the norm of the functional connection characteristics of each brain region and brain network changing with age. Based on this development norm, calculates the degree of deviation of each functional connection characteristic of each attention deficit hyperactivity disorder patient from the norm, and obtains the individual development deviation score;

[0019] e. Attention deficit hyperactivity disorder recognition module: connected to the brain functional connectome development norm construction module, according to the obtained brain functional connection development deviation scores, uses the support vector machine algorithm to construct a classification model for attention deficit hyperactivity disorder patients and normal development controls, and combines the brain functional connection patterns of individual extreme positive and negative deviations to determine whether it is an attention deficit hyperactivity disorder patient.

[0020] Among them, the data acquisition module collects the demographic data, T1 structural magnetic resonance, and resting-state functional magnetic resonance imaging data of multi-center attention deficit hyperactivity disorder patients and normal development controls;

[0021] The data preprocessing module is the preprocessing of image data. The preprocessing of the functional magnetic resonance imaging data of all data sets uses the brain image data processing and analysis toolbox based on the MATLAB platform for preprocessing;

[0022] Brain functional connectome calculation; used to analyze and understand the functional collaborative activities or connection patterns between different regions of the brain, calculate the correlation of time series between different brain regions and networks, and obtain the brain functional connectome characteristics;

[0023] The construction of the brain functional connectome development norm is to construct the brain functional connectome norm in a normal development control group. The stratified 10-fold cross-validation method is adopted. The dataset is divided into 10 parts, and the proportion of data in each center in each part is the same as that of the whole dataset. Each time, one part is selected as the validation set, and the remaining 9 parts are used as the training set, and this is repeated 10 times; Modeling is carried out separately for men and women to reduce the influence of gender differences on the model; Based on the training set data, Bayesian Spline Regression is used to characterize the change pattern of the functional connection features of each brain region and brain network with age; Bayesian Spline Regression introduces a cubic B-spline transformation into the general Bayesian linear regression model to capture non-linear effects, and the Powell method is used to fit and optimize the hyperparameters; The explained variance (EV) and mean standardized log loss (MSLL) are used to evaluate the norm;

[0024] The individual deviation score calculation method is as follows: Based on the established brain functional connection norm, the degree to which each brain functional connection feature d of each subject n deviates from the norm is statistically estimated, that is, the individual deviation score Z nd , and its calculation formula is:

[0025]

[0026] where, ŷ nd is the predicted functional connection value, y nd is the true functional connection value, and the noise σ 2 d variance (reflecting the variability in the data) and uncertainty variance (σ 2 * ) d are used for normalization.

[0027] Furthermore, the image data preprocessing method includes: removing the first 5 time points; time slice correction; head motion correction; image registration to the T1 structural image and then to the MNI standard space; regressing the Friston-24 head motion parameters, linear drift, white matter signal, cerebrospinal fluid signal, and whole brain average signal; smoothing with a 6mm Gaussian smoothing kernel; 0.01 - 0.1Hz band-pass filtering.

[0028] The brain functional connectome calculation method includes the following steps:

[0029] S1: Using the Schaefer brain functional atlas to divide the cerebral cortex into 400 regions of interest (ROIs);

[0030] S2: Calculate the Pearson correlation coefficients between the time series of every two of these 400 ROIs, and perform Fisher-z transformation to obtain a 400×400 functional connectivity matrix;

[0031] S3: Calculate the average functional connectivity of each ROI [the sum of the functional connectivities of this ROI with other ROIs / (the total number of ROIs - 1)] to obtain 400 functional connectivity values at the brain region level;

[0032] S4: According to the Yeo brain functional network atlas, assign these 400 ROIs to 7 brain networks, calculate the within-network and between-network functional connectivities, a total of 28 [7 + 7×(7 - 1) / 2] functional connectivity values; the within-network functional connectivity of each brain network is the average of the functional connectivities between the ROIs within the network, and the between-network functional connectivity is the average of the functional connectivities between the ROIs of two brain networks.

[0033] The steps for constructing the norm are as follows:

[0034] (1) Divide the normal control data set into 10 parts, with the proportion of data at each center in each part being the same as that in the entire data set. Each time, select one of them as the validation set, and the remaining 9 parts as the training set, and repeat this 10 times. Based on the training set data, use cubic B-spline basis functions to expand the independent variable age, denoted as φ(x), where Φ is an N x K matrix containing the basis function expansions of all subjects; the single functional connectivity value y is assumed to be a linear combination of the B-spline basis function transformation plus a noise term:

[0035] y = w T φ(x) + ∈ s

[0036] where w is the estimated weight vector, and ε s is the Gaussian noise distribution at site s, with a mean of 0 and a precision of β s ;

[0037] All the noise precisions from different sites will be combined in a vector β, allowing the noise precision to vary between sites to accommodate the between-site variability and site-specific intercepts; add the dummy variable site regression variable to the final basis expansion matrix of the B-spline to make it a random intercept model;

[0038] (2) Use Bayesian linear regression to set Gaussian priors for the model parameters where α is a hyperparameter dependent on the weights; assume that the mean of the Gaussian prior is zero and the precision matrix is Λ α , and use an isotropic precision matrix Λ α = αI; the choice of the Gaussian prior allows the posterior distribution of w to be calculated in closed form:

[0039]

[0040] The posterior for each subject was obtained using standard posterior derivation:

[0041]

[0042] The Powell method was used to fit the optimal hyperparameters α and β; Powell is a derivative-free method that can fit the hyperparameters faster, and the resulting predictive distribution is:

[0043]

[0044] This predictive distribution reflects the changing pattern of brain functional connectivity features with age, and thus the developmental norms of functional connectivity for each brain region and network are obtained. The model was evaluated using the above validation set data, and evaluation metrics such as explained variance (EV) and mean standardized log loss (MSLL) were calculated to evaluate the fitting effect of the norms.

[0045] Among them, the attention deficit hyperactivity disorder identification method described in the e step specifically includes the following steps:

[0046] S1: Using patients with attention deficit hyperactivity disorder as the test set, predicting the brain functional connectivity values of patients with attention deficit hyperactivity disorder based on the above constructed brain functional connectivity norms, calculating the individual deviation scores of each brain functional connectivity feature for each ADHD patient, and defining the absolute value of z greater than 2 as an extreme deviation to identify brain regions or networks showing extreme positive and negative deviations;

[0047] S2: Analyzing the overall deviation pattern of ADHD patients by combining the deviations of all brain regions and network functional connectivities, and at the same time using extreme value statistics to analyze the most extreme deviated brain regions and networks of individuals;

[0048] S3: Using the individual deviation scores of brain functional connectivity of patients with attention deficit hyperactivity disorder and normal development controls as input features to construct a support vector classification model to discriminate patients with attention deficit hyperactivity disorder.

[0049] The feature adopted by the present invention is a static functional connectome calculated based on resting-state functional magnetic resonance imaging (fMRI) data. The measurement environment of the adopted resting-state functional magnetic resonance imaging is more controllable. The subject only needs to keep still, with a higher degree of cooperation. The data acquisition is relatively more stable. The obtained brain functional connectome features can provide potentially more specific neural activity patterns, which helps to improve the diagnostic accuracy. The adopted static functional connectivity analyzes the average connectivity pattern during the entire scanning process, requires relatively less scanning time, is relatively less sensitive to noise, and will be more stable. Therefore, it has better generalization across different scanning centers. The present invention calculates brain functional connectome metrics based on fMRI data. These metrics reflect the synchrony of functional activities between different brain regions or brain networks, integrate complex information into relatively easy-to-understand connectivity patterns, reduce data redundancy, and improve information extraction efficiency. As a biomarker, it is more feasible and more practical in clinical applications.

[0050] The present invention constructs a norm, calculates the degree to which each individual's brain functional characteristics deviate from the norm, so as to obtain the brain regions or networks with abnormal functions of each individual, and issues a brain functional imaging report similar to the traditional diagnostic report of the radiology department to clinicians, providing a more interpretable diagnostic tool for clinicians, thereby improving the diagnostic accuracy. By establishing a norm for the development of the brain functional connectome of a healthy population, the degree of deviation of each individual from the norm can be calculated. According to whether the deviation degree of the individual exceeds the normal range, abnormal individuals can be identified, and further assisted diagnosis of attention deficit hyperactivity disorder can be carried out according to the deviation pattern of the individual, thus realizing the technical improvement from group average comparison to individual precise assessment and solving the bottleneck problem of the transformation and application of brain imaging indicators to clinical practice.

[0051] The beneficial effects of the present invention are:

[0052] The present invention provides an attention deficit hyperactivity disorder recognition system based on a norm for the development of the brain functional connectome. By constructing a norm for the change of brain functional connectivity characteristics with age, the degree of deviation of an individual from the normal range is characterized, and the recognition of attention deficit hyperactivity disorder patients is realized based on the obtained individualized deviation index, thereby providing a quantitative and individual-level new brain imaging biomarker for attention deficit hyperactivity disorder patients. In the construction of the norm, the Bayesian spline regression method used in the present invention can more accurately model non-linear distributions by combining B-spline transformations, and can also shorten the calculation time. Therefore, it can be efficiently extended to large brain imaging data cohorts and flexibly applied to new sites not included in the training samples. Brief Description of the Drawings

[0053] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0054] Figure 2Schematic diagram of the developmental norm constructed based on brain functional connectome features in the embodiments of the present invention;

[0055] Figure 3 Brain region map with extreme deviations of ADHD patients and normal development controls in the embodiments of the present invention. Detailed implementation manners

[0056] Flow schematic diagram of the construction of the attention deficit hyperactivity disorder recognition system based on the brain functional connectome developmental norm of the present invention is as Figure 1 shown.

[0057] 1. Data collection

[0058] Collect demographic information and T1 structural magnetic resonance and resting-state functional magnetic resonance imaging data of the ADHD-200 public dataset and the SCU dataset.

[0059] 2. Preprocessing of imaging data

[0060] For the preprocessing of the functional magnetic resonance imaging data of all datasets, the brain imaging data processing and analysis toolbox based on the MATLAB platform (a toolbox for Data Processing & Analysis for Brain Imaging, DPABI, version v4.2) is used, and a unified preprocessing process is adopted, including: removing the first 5 time points; time slice correction; head motion correction; registering the functional image to the T1 structural image and then to the standard MNI space; regressing the Friston-24 head motion parameters, linear drift, white matter signal, cerebrospinal fluid signal, and whole-brain average signal; smoothing with a 6mm Gaussian smoothing kernel; and band-pass filtering from 0.01 to 0.1 Hz.

[0061] 3. Calculation of brain functional connectome

[0062] The Schaefer brain functional atlas is used to divide the cerebral cortex into 400 regions of interest (ROIs). Then, the Pearson correlation coefficients between the pairwise time series of these 400 ROIs are calculated and Fisher-z transformed to obtain a 400×400 functional connectivity matrix. Calculate the average functional connectivity of each ROI [the sum of the functional connectivities of this ROI with other ROIs / (the total number of ROIs - 1)] to obtain 400 functional connectivity values at the brain region level. According to the Yeo brain functional network atlas, these 400 ROIs are assigned to 7 brain networks, and the intra-network and inter-network functional connectivities are calculated, with a total of 28 [7 + 7×(7 - 1) / 2] functional connectivity values; the intra-network functional connectivity of each brain network is the average of the functional connectivities between the ROIs within the network, and the inter-network functional connectivity is the average of the functional connectivities between the ROIs of two brain networks.

[0063] 4. Construction of Developmental Norm

[0064] In the normal development control group, Bayesian spline regression was used to characterize the changing patterns of the functional connectivity features of the above 400 brain regions and 28 brain network levels with age, respectively, so as to construct the developmental norm of the brain functional connectome. Modeling was carried out separately for males and females to reduce the influence of gender differences on the model. The model was validated using the stratified 10-fold cross-validation method. The specific steps are as follows:

[0065] (1) The normal control data set was divided into 10 parts, and the proportion of data in each center in each part was the same as that of the whole data set. Each time, one of them was selected as the validation set, and the remaining 9 parts were used as the training set, and this was repeated 10 times. Based on the training set data, the independent variable age was extended using the cubic B-spline basis function, and φ(x) represents this extension, where Φ is an N x K matrix containing the basis function extensions of all subjects. The single functional connectivity value y was assumed to be a linear combination of the B-spline basis function transformation plus a noise term:

[0066] y = w T φ(x)+∈ s

[0067] where w is the estimated weight vector, and ε s is the Gaussian noise distribution at site s, with a mean of 0 and a precision of β s . All noise precisions from different sites were combined in a vector β, and we allowed the noise precision to vary between different sites to accommodate the variation between sites, as well as the site-specific intercepts (i.e., the dummy variable site regressors in the design matrix). The dummy variable site regressors were added to the final basis extension matrix of the B-spline to make it a random intercept model. This method provides an effective way to accommodate site effects in the normalization modeling.

[0068] (2) Use Bayesian linear regression to set Gaussian priors for the model parameters where α is a hyperparameter dependent on the weights. Assume that the mean of the Gaussian prior is zero and the precision matrix is Λ α , and here we use a simple isotropic precision matrix Λ α = αI. The choice of the Gaussian prior allows the posterior distribution of w to be calculated in closed form:

[0069]

[0070] The posterior of each subject was obtained using the standard posterior derivation:

[0071]

[0072] Use the Powell method to fit the optimal hyperparameters α and β. Powell is a derivative-free method that can fit hyperparameters more quickly. The resulting predictive distribution is:

[0073]

[0074] This predictive distribution reflects the pattern of changes in brain functional connectivity characteristics with age, and from this, the developmental norms of functional connectivity for each brain region and network are obtained. The quantile model form is selected, and each model fits seven curves of the 5th, 10th, 25th, 50th, 75th, 90th, and 95th percentiles to reflect the range of changes in brain functional connectivity. Figure 2 Figure showing the developmental norms of functional connectivity for two of these brain regions.

[0075] (3) Use the above validation set data to evaluate the model, and calculate evaluation metrics such as explained variance (EV) and mean standardized log loss (MSLL) to evaluate the fitting effect of the norms.

[0076] (4) To quantify the degree of deviation of each subject from the normal range of the norms, calculate the deviation z-score of each subject n for each brain functional connectivity characteristic d, with the formula as follows:

[0077]

[0078] where, ŷ nd is the predicted functional connectivity value, y nd is the true functional connectivity value, and use the noise variance σ 2 d (reflecting the variability in the data) and the uncertainty variance (σ 2 * ) d for normalization. The z-score quantifies the degree of deviation between the estimated functional connectivity characteristics of an individual and the model prediction given the model uncertainty.

[0079] 5. Identification of ADHD patients

[0080] Using the ADHD data as the test set, the brain functional connectivity values of ADHD patients were predicted based on the brain functional connectome development norm constructed above. From this, the individual deviation score z of each brain functional connectivity feature of each ADHD patient was calculated, thereby obtaining the degree to which the functional connectivity of each brain region or network in ADHD patients deviated from the normal range. Defining the absolute value of z greater than 2 as an extreme deviation to identify the brain regions or networks showing extreme positive and negative deviations. Combining the deviations of the functional connectivity of all brain regions and networks to analyze the overall deviation pattern of ADHD patients, and at the same time using the extreme value statistical method to analyze the most extreme deviated brain regions and networks of individuals. The extreme deviation brain region map of ADHD patients and normal development controls is as Figure 3 shown. It can be seen that there are almost no extremely deviated brain regions in the normal development control, while ADHD patients have extensive positive deviations, that is, the functional connectivity of brain regions is higher than the normal range.

[0081] Finally, taking the individual deviation scores of the brain functional connectivity of all subjects as input features, a classification model of ADHD patients and normal development controls was constructed using the support vector machine algorithm. The nested 10-fold cross-validation method was used to tune and test the model. The area under the curve (AUC), accuracy, sensitivity, and specificity of the Receiver Operating Characteristic (ROC) curve were used to evaluate the classification performance of the model. The results showed that compared with the original functional connectivity values, using the individual deviation scores of brain functional connectivity as features significantly improved the classification performance. The average classification AUC, accuracy, sensitivity, and specificity and their standard deviation values of the classification model in the 10-fold cross-validation are shown in Table 1.

[0082] Table 1. Performance indicators of the ADHD diagnosis classification model

[0083]

[0084] Based on the brain functional connectome norm constructed above, the degree to which the individual brain functional connectivity features of each subject deviate from the normal range can be obtained. If the deviation degree of the functional connectivity features of the subject exceeds the threshold, it indicates an abnormality. Analyze the deviation pattern of the subject's brain functional connectivity features to clarify which brain regions or networks have abnormal functional changes. Further input the individual deviation scores of the subject's brain functional connectivity into the classification model constructed above, and the classification model will return a result (ADHD or normal). Combining the deviation pattern of the subject's individual brain functional connectivity features and the result determined by the classification model, the final judgment of whether the subject is an ADHD patient is obtained, thereby realizing the identification of ADHD patients.

[0085] The following alternative solutions can achieve the same invention purpose without departing from the scope and spirit of the present invention:

[0086] 1) Alternative data preprocessing method: The preprocessing of imaging data based on DPABI can be replaced by using software such as SPM, FSL, fMRIPrep, etc., so as to achieve the preprocessing of functional magnetic resonance imaging data in the same way.

[0087] 2) Alternative method for constructing brain functional connectivity matrix: The Schaefer brain functional atlas can be replaced by other high-resolution and high-confidence templates, such as Brainnetome, Glasser, Power templates, etc.; the calculation method of functional connectivity can be replaced by wavelet coherence, etc., so as to achieve the construction of the functional connectivity matrix in the same way.

[0088] 3) Alternative modeling method: Hierarchical Bayesian regression can be used as an alternative to the norm modeling method. The Bayesian statistical method is also used to distinguish the sources of variation, and it is applicable to multi-center brain imaging data.

[0089] 4) Alternative classification and discrimination algorithm: SVM classification can be replaced by machine learning algorithms such as XGBoost, random forest, decision tree, etc., so as to achieve the classification and discrimination of ADHD patients in the same way.

[0090] In summary, the present invention provides an attention deficit hyperactivity disorder recognition system based on the developmental norm of the brain functional connectome. For the first time, a norm of the characteristics of the brain functional connectome changing with age is constructed for children and adolescents. The Bayesian spline regression method adopted can more accurately model the non-linear distribution, and at the same time shorten the calculation time. Therefore, it can be efficiently extended to a large brain imaging data cohort and flexibly applied to new sites not included in the training samples. Based on the constructed developmental norm, the degree to which ADHD patients deviate from the normal range can be characterized at the individual level. The obtained individualized brain functional connectivity development deviation index contains brain functional development information, which improves the classification effect of ADHD patients and controls compared with the original functional connectivity values. The present invention provides a quantitative and individual-level novel brain imaging biomarker for ADHD patients, thereby further revealing the mechanism of ADHD neural heterogeneity and promoting the precise diagnosis and treatment of ADHD.

Claims

1. An attention deficit hyperactivity disorder recognition system based on the developmental norm of brain functional connectome, characterized in that: It classifies and identifies attention deficit hyperactivity disorder (ADHD) by constructing a normative model of brain functional connectome development and using the obtained individual developmental deviation index of brain functional connectivity; It includes the following functional modules: a. Data acquisition module: used to collect demographic and magnetic resonance imaging data of subjects; b. Data preprocessing module: connected to the data acquisition module, used to preprocess the collected resting-state functional magnetic resonance imaging data, including correction, registration, noise reduction, smoothing and filtering; c. Brain functional connectome calculation module: connected to the data preprocessing module, generates a functional connectivity matrix from the preprocessed resting-state functional magnetic resonance imaging data, and calculates the functional connectivity of brain regions and brain networks; d. Brain functional connectome development norm construction module: connected to the brain functional connectome calculation module, in the normal development control group, uses Bayesian spline regression to construct the norm of the functional connectivity characteristics of each brain region and brain network changing with age. Based on this development norm, calculate the degree of deviation of each functional connectivity characteristic of each ADHD patient from the norm to obtain the individual developmental deviation score; e. ADHD identification module: connected to the brain functional connectome development norm construction module, according to the obtained brain functional connectivity development deviation score, uses the support vector machine algorithm to construct a classification model for ADHD patients and normal development controls, and combines the brain functional connectivity patterns of individual extreme positive and negative deviations to determine whether it is an ADHD patient; The data acquisition module collects demographic information, T1 structural magnetic resonance and resting-state functional magnetic resonance imaging data of multi-center ADHD patients and normal development controls; The data preprocessing module is image data preprocessing, and the preprocessing of functional magnetic resonance imaging data of all data sets uses the brain image data processing and analysis toolbox based on the MATLAB platform; Brain functional connectome calculation; used to analyze and understand the functional collaborative activities or connection patterns between different regions of the brain, calculate the correlation of time series between different brain regions and networks, and obtain the characteristics of the brain functional connectome; The construction of the brain functional connectome development norm is to construct the brain functional connectome norm in a normal development control group. The hierarchical 10-fold cross-validation method is adopted. The data set is divided into 10 parts, and the proportion of data in each center in each part is the same as that in the whole data set. Each time, one part is selected as the validation set, and the remaining 9 parts are used as the training set, and this cycle is repeated 10 times; Modeling is carried out separately for men and women to reduce the influence of gender differences on the model; Based on the training set data, Bayesian Spline Regression is used to characterize the change patterns of functional connectivity features of each brain region and brain network with age; Cubic B-spline transformation is introduced into the general Bayesian linear regression model in Bayesian Spline Regression to capture non-linear effects, and the Powell method is used to fit and optimize hyperparameters; The norm is evaluated using the explained variance (EV) and the mean standardized log loss (MSLL). The individual deviation score calculation method is as follows: Based on the established brain functional connectivity norm, statistically estimate the degree to which each brain functional connectivity feature d of each subject n deviates from the norm, that is, the individual deviation score Z nd , and its calculation formula is: ； Among them, ŷ nd is the predicted functional connectivity value, and y nd is the true functional connectivity value. The noise σ 2 d variance (reflecting the variability in the data) and the uncertainty variance (σ 2 * ) d are used for normalization.

2. The attention deficit hyperactivity disorder recognition system based on the developmental norm of brain functional connectome according to claim 1, characterized in that: The brain functional connectome development norm described above is constructed using Bayesian Spline Regression.

3. The attention deficit hyperactivity disorder recognition system based on the brain functional connectome development norm according to claim 1, wherein: The image data preprocessing method described above includes: removing the first 5 time points; time slice correction; head motion correction; image registration to the T1 structural image and then to the MNI standard space; regressing the Friston-24 head motion parameters, linear drift, white matter signal, cerebrospinal fluid signal and whole brain average signal; smoothing processing using a 6 mm Gaussian smoothing kernel; 0.01 - 0.1 Hz band-pass filtering.

4. The attention deficit hyperactivity disorder recognition system based on the brain functional connectome development norm according to claim 1, wherein: The brain functional connectome calculation method described above includes the following steps: S1: The cerebral cortex is divided into 400 regions of interest (ROIs) using the Schaefer brain functional atlas; S2: Calculate the Pearson correlation coefficient between the time series of each pair of the 400 ROIs and perform Fisher-z transformation to obtain a 400×400 functional connectivity matrix; S3: Calculate the average functional connectivity of each ROI [the sum of the functional connectivity of this ROI with other ROIs / (the total number of ROIs - 1)] to obtain 400 functional connectivity values at the brain region level; S4: According to the Yeo brain functional network atlas, assign these 400 ROIs to 7 brain networks, and calculate the intra-network and inter-network functional connectivity, with a total of 28 [7 + 7×(7 - 1) / 2] functional connectivity values; The intra-network functional connectivity of each brain network is the average of the functional connectivity between the ROIs within the network, and the inter-network functional connectivity is the average of the functional connectivity between the ROIs of two brain networks.

5. The attention deficit hyperactivity disorder recognition system based on the brain functional connectome development norm according to claim 1, wherein: The norm construction steps are as follows: (1) Divide the normal control data set into 10 parts. The proportion of data from each center in each part is the same as that in the whole data set. Each time, select one of them as the validation set, and the remaining 9 parts as the training set, and loop 10 times in this way; Based on the training set data, use cubic B-spline basis functions to expand the independent variable age. Let φ(x) represent this expansion, where Φ is an N x K matrix containing the basis function expansions of all subjects; The single functional connectivity value y is assumed to be a linear combination of the B-spline basis function transformation plus a noise term: ； where w is the estimated weight vector, and ε s is the Gaussian noise distribution at site s with mean 0 and precision β s ; All noise precisions from different sites will be combined in a vector β, allowing the noise precision to vary between sites to accommodate the variation between sites and site-specific intercepts; Add dummy variable site regression variables to the final basis expansion matrix of the B-spline to make it a random intercept model; (2) Use Bayesian linear regression to give the model parameter p(w|α) = 𝓝 (𝐰|0, Λ 𝛼 -1 ) Set a Gaussian prior, where α is a hyperparameter dependent on the weights; assume that the mean of the Gaussian prior is zero and the precision matrix is Λ 𝛼， Use an isotropic precision matrix Λ 𝛼 = αI; The choice of Gaussian prior allows the posterior distribution of w to be calculated in closed form: ； The posterior of each subject is obtained using standard posterior derivation: ； Use the Powell method to fit the optimal hyperparameters α and β; Powell is a derivative-free method that can fit hyperparameters faster. The resulting predictive distribution is: ； This predictive distribution reflects the change pattern of brain functional connectivity characteristics with age, and thus obtains the developmental norms of functional connectivity in each brain region and network; Use the above validation set data to evaluate the model, and calculate evaluation metrics such as explained variance (EV) and mean standardized log loss (MSLL) to evaluate the fitting effect of the norms.

6. The attention deficit hyperactivity disorder recognition system based on the developmental norms of brain functional connectomes according to claim 1, wherein: The attention deficit hyperactivity disorder recognition method described in step e specifically includes the following steps: S1: Use patients with attention deficit hyperactivity disorder as the test set. Based on the brain functional connectivity norms constructed above, predict the brain functional connectivity values of patients with attention deficit hyperactivity disorder, calculate the individual deviation scores of each brain functional connectivity feature of each patient with attention deficit hyperactivity disorder, and define the absolute value of z greater than 2 as an extreme deviation to identify brain regions and networks showing extreme positive and negative deviations; S2: Combine the deviations of all brain regions and network functional connectivities to analyze the overall deviation pattern of patients with attention deficit hyperactivity disorder, and at the same time use extreme value statistics to analyze the most extreme deviated brain regions and networks of individuals; S3: Use the individual deviation scores of brain functional connectivities of patients with attention deficit hyperactivity disorder and normal development controls as input features to construct a support vector classification model to discriminate patients with attention deficit hyperactivity disorder.

Images