Brain network prediction method and system for abstinence neural recovery of methylamphetamine disorder

Abstract

According to the withdrawal stage neural recovery brain network prediction method and system for methylamphetamine disorder, time sequence function connection and withdrawal stage dynamic recovery process are combined for the first time through dynamic network modeling, the limitation of traditional static analysis is broken through, dynamic characteristics of withdrawal stage neural recovery are disclosed, and the prediction method and system for the withdrawal stage neural recovery brain network of methylamphetamine disorder are provided. The positive network is mainly based on sensory motor-cerebellar connection, and the negative network relates to default mode network interaction, provides quantifiable neural markers to reveal cortical region composition of the withdrawal network, provides neural targets for clinical intervention and strength values of the withdrawal network during different withdrawal periods, evaluates recovery conditions, and provides quantitative neural markers for clinical intervention. And making of an individual rehabilitation plan is assisted.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical image analysis and drug addiction treatment, and particularly relates to a neural recovery brain network prediction method and system for the withdrawal period of methamphetamine disorder. BACKGROUND

[0002] Methamphetamine use disorder (MUD) is a widely prevalent substance use disorder (SUD). SUDs are often accompanied by neurologic impairment, the pattern of which can be drug-specific. Repeated neuroimaging studies have confirmed that long-term methamphetamine exposure leads to structural, functional, and neurochemical abnormalities in the prefrontal cortex and other cortical and subcortical brain regions. Importantly, these changes are often accompanied by changes in cognitive function, affective symptoms, blood flow differences, and treatment outcomes, suggesting their clinical significance.

[0003] To explore the neurobiological changes in the withdrawal period of MUD, neurofunctional studies have analyzed different subgroups of MUD patients with different withdrawal durations. For example, Chen and his team compared the differences in neural responses to drug and sexual cues between MUD patients who had been abstinent for 1-3 months and those who had been abstinent for 16-40 months. The results showed that withdrawal may modulate individual responses to drug and sexual cues, thereby affecting drug use and sexual behavior. A recent longitudinal study aimed to identify markers of functional recovery by comparing neural functional activities before and after withdrawal. The study found that the neural functions of the occipital and parietal lobes changed significantly during the process of short-term (1 week to 3 months) to long-term (10-15 months) withdrawal in MUD patients. In addition, several reviews have suggested that there is a time dynamic correlation between neural recovery during the withdrawal period of addiction and the duration of withdrawal, suggesting that prefrontal function may improve with withdrawal. However, there is still a lack of evidence on the neural functional characteristics related to the time dynamic of neural recovery during the withdrawal period of MUD.

[0004] A significant feature of the aforementioned studies is their hypothesis-driven nature, i.e., exploring the clinical significance of specific neural signals by comparing pre-specified subgroups. For example, researchers concluded by comparing the neural responses to drug / sexual cues in MUD patients with different lengths of abstinence. However, current research trends are shifting towards identifying neural markers of trait-like phenomena, behaviors, clinical symptoms, and treatment responses to enable individualized predictions. Emerging data-driven machine learning methods, such as connectional predictive modeling (CPM), utilize functional connectivity data ("connectional signatures") for behavior prediction and identification of behavior-related neural networks. Previous studies have identified pre-treatment neural markers of cocaine, opioid, and cannabis addiction abstinence using CPM. These markers represent complex functional neural networks that are consistent with established neural network changes suggested by previous hypothesis-driven studies. Therefore, CPM can reveal the dynamic association between neural function changes and abstinence, providing insights into relapse mechanisms and supporting precision treatment planning for MUD. However, there is currently no study that uses whole-brain machine learning methods to identify markers of neural function changes during MUD abstinence. SUMMARY

[0005] The purpose of the present application is to provide a neural recovery brain network prediction method and system for the abstinence period of methamphetamine disorder, to solve the problems of lack of large-scale feature data fusion capability, insufficient dynamic monitoring of neural recovery in the abstinence period, and lack of quantitative evaluation of the degree of neural recovery in the abstinence period.

[0006] To achieve the above purpose, the present application provides a neural recovery brain network prediction method for the abstinence period of methamphetamine disorder, comprising the following steps:

[0007] Step S100, acquiring multi-modal brain neural image data and performing data standardization preprocessing;

[0008] Step S200, establishing a whole-brain functional connectivity network and feature extraction, extracting functional connectivity edges in the brain region functional matrix that are significantly related to the length of abstinence, establishing a positive network that predicts longer abstinence time with connection enhancement and a negative network that predicts longer abstinence with connection weakening;

[0009] Step S300, training a connection group prediction model and verification, assigning weights to the positive and negative networks respectively, and constructing a multi-source regression model to calculate the individual network strength with the length of abstinence as the target variable;

[0010] Step S400, dynamic network strength analysis and clinical correlation mapping, calculating the individual positive and negative network strengths, generating a dynamic recovery curve, and incorporating age, medication history, and other covariates into the model to verify the independence of the network prediction;

[0011] Step S500, multi-modal data fusion and model optimization, adopts high-precision head movement data correction, verifies the stability of the core network, integrates functional connection data and clinical indicators, constructs a multi-dimensional prediction model, and improves the long-term abstinence prediction accuracy.

[0012] Further, in step 100, the multi-modal brain neural image data includes high-resolution T1 weighted images covering the whole brain, and the preprocessing includes head motion and distortion correction, spatial standardization processing, noise regression analysis and time series cleaning.

[0013] Further, in step S200, first brain region segmentation and time series extraction are performed, including:

[0014] S201, brain region segmentation based on Shen-268 brain atlas, dividing the whole brain into 268 functional nodes;

[0015] S202, extracting the standardized BOLD time series of each node, calculating the Pearson correlation coefficient between nodes, and generating a 268x268 functional connection matrix.

[0016] Further, in step S200, dynamic network modeling is further performed, including:

[0017] S203, using double threshold method and permutation test to correct and screen significant functional connection edges and abstinence time;

[0018] S204, dividing the significant edges into positive network predicting longer abstinence time and negative network predicting longer abstinence time.

[0019] Further, in step S300, the model training includes:

[0020] S301, feature weighting: assigning weights to the positive network and the negative network respectively, and obtaining individual network strength by calculating the sum of significant edge weights;

[0021] S302, regression modeling: taking the logarithm of abstinence time as the target variable, constructing a multiple linear regression model, and optimizing parameters using leave-one-out cross-validation.

[0022] Further, in step S300, the model verification and generalization includes:

[0023] S303, permutation test: generating zero distribution by 5,000 random permutations to verify the significance of the model;

[0024] S304, independent sample verification: testing the model prediction ability in a new sample set, and calculating the Spearman correlation between the predicted value and the actual value.

[0025] Further, in step S400, further comprising:

[0026] S401, network strength quantification: calculating individual positive network strength and negative network strength, generating dynamic recovery curve;

[0027] S402, clinical variable control, including age, medication history and other covariates in the model, verifying the independence of network prediction.

[0028] Further, in step S500, comprising:

[0029] S501, motion artifact correction, using high-precision head motion correction, high-quality data FD<0.3mm, verifying the stability of the core network;

[0030] S502, integrating functional connectivity data and clinical indicators, constructing a multi-dimensional prediction model, and improving long-term abstinence prediction accuracy.

[0031] On the other hand, the present application also provides a neural recovery brain network prediction system for the abstinence period of methamphetamine disorder, comprising:

[0032] A data preprocessing module acquires multi-modal brain neural image data and performs data standardization preprocessing;

[0033] A whole brain functional connectivity network and feature extraction module extracts functional connectivity edges in the brain region functional matrix that are significantly related to the abstinence duration, establishes a positive network for predicting longer abstinence time by connection enhancement and a negative network for predicting longer abstinence by connection weakening;

[0034] A connection group prediction model and verification training module assigns weights to the positive network and the negative network respectively, and constructs a multiple regression model to calculate the individual network strength with the abstinence duration logarithm as the target variable;

[0035] A dynamic network strength analysis and clinical correlation mapping module calculates individual positive network strength and negative network strength, generates a dynamic recovery curve, and includes age, medication history and other covariates in the model to verify the independence of network prediction.

[0036] A multi-modal data fusion and model optimization module uses high-precision head motion data correction to verify the stability of the core network, integrates functional connectivity data and clinical indicators to construct a multi-dimensional prediction model, and improves long-term abstinence prediction accuracy.

[0037] On the other hand, the present application also provides a computer readable storage medium, the computer readable storage medium stores a computer program, the computer program includes program instructions, the program instructions when being executed by a processor make the processor execute the steps of the above method.

[0038] The application provides a methamphetamine disorder withdrawal period neural recovery brain network prediction method and system, which combines time series functional connections with withdrawal period dynamic recovery processes for the first time through dynamic network modeling, breaks through the limitations of traditional static analysis, reveals the dynamic characteristics of withdrawal period neural recovery, and provides quantifiable neural markers to reveal the cortical area composition of withdrawal networks, provides neural target points for clinical intervention and intensity values of withdrawal networks during different withdrawal periods, evaluates recovery conditions, and helps to develop individualized rehabilitation plans. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor.

[0040] Figure 1 is a flowchart of a methamphetamine disorder withdrawal period neural recovery brain network prediction method according to an embodiment of the present application.

[0041] Figure 2 is a prediction flowchart of a CPM model according to an embodiment of the present application.

[0042] Figure 3 is an architecture diagram of a methamphetamine disorder withdrawal period neural recovery brain network prediction system according to an embodiment of the present application.

[0043] Figure 4 is a structural schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0045] In the following description, "some embodiments" are described, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.

[0047] Embodiment one

[0048] Figure 1 is a flow chart of a method for predicting the neural recovery brain network in the withdrawal period of a methamphetamine disorder according to an embodiment of the present application, as shown in Figure 1 A method for predicting the neural recovery brain network in the withdrawal period of a methamphetamine disorder according to the present application includes the following steps:

[0049] Step S100, acquire multi-modal brain neural image data and perform data standardization preprocessing.

[0050] Specifically, in the present embodiment, the input data includes resting-state functional magnetic resonance (rsfMRI) data and structural image data. The resting-state functional magnetic resonance (rsfMRI) data needs to meet the following conditions: 1) whole brain coverage, no signal loss or artifact interference. 2) ii. After head motion correction, the average frame displacement (FD) is less than or equal to 0.5 mm, and the standardized DVARS is less than or equal to 1.5. 3) The time series is band filtered (0.01-0.08Hz) and de-trended.

[0051] Data standardization preprocessing includes:

[0052] a) Head motion and distortion correction: Non-linear registration (ANTs 2.3.3) is used to correct head motion and magnetic susceptibility artifacts, and a head motion parameter matrix (6 degrees of freedom) is generated.

[0053] b) Spatial standardization: Individual T1 images are registered to the MNI152NLin2009cAsym template, and the same transformation matrix is applied to standardize the rsfMRI data to the template space.

[0054] c) Noise regression: 36 covariates (including white matter, cerebrospinal fluid signals, head motion parameters and their quadratic terms, derivative terms) are extracted, and the CompCor algorithm is used to remove physiological noise.

[0055] d) Time series cleaning: Frames with abnormal head motion (FD>0.5mm or DVARS>1.5) are removed, and high-quality BOLD signals are retained for functional connectivity analysis.

[0056] Step S200, establish whole brain functional connectivity network and feature extraction, extract the functional connectivity edges in the brain region functional matrix that are significantly related to the withdrawal time, and establish the positive network that predicts longer withdrawal time by connection enhancement and the negative network that predicts longer withdrawal by connection weakening.

[0057] Specifically, step S200 also includes the following steps:

[0058] S201, Brain region segmentation and time series extraction.

[0059] Specifically, this embodiment is based on the Shen-268 brain atlas, dividing the whole brain into 268 functional nodes (covering the cortex, subcortex, and cerebellum). Individual fMRI data are mapped to the MNI152 standard space using nonlinear registration (FNIRT algorithm).

[0060] S202, Functional Connection Calculation.

[0061] Specifically, this embodiment extracts the BOLD time series of each node, calculates the Pearson correlation coefficient between nodes, and generates a 268×268 functional connectivity matrix. A Fisher z-transform is then applied to the connectivity matrix to enhance the normality of the data.

[0062] S203, Dynamic Network Modeling and Feature Extraction.

[0063] Specifically, significant edge screening is performed first. This includes permutation tests and double-threshold determination. The withdrawal duration labels are randomly permuted 1000 times to generate an empirical null distribution of correlation coefficients for each edge. The double-threshold determination requires the actual correlation coefficient to satisfy P < 0.005 (one-tailed test) and belong to intra-module connections (corrected by NBS).

[0064] Then, the network is partitioned, with significant edges divided into positive and negative networks. For the positive network, edges positively correlated with withdrawal duration are selected, and the modularity index (Q-value > 0.3) and hub node degree centrality are calculated. For the negative network, edges negatively correlated with withdrawal duration are extracted, and their small-world properties (σ > 1.5) are analyzed.

[0065] Step S300: Train and validate the connection group prediction model. Assign weights to the positive and negative networks respectively, and construct a multi-source regression model to calculate the individual network strength using the logarithm of abstinence duration as the target variable.

[0066] Specifically, such as Figure 2 As shown in this embodiment, the feature selection of the prediction model first filters for salient edges and models the correlation based on the target variable. In a specific embodiment, the feature selection is based on the logarithm of the withdrawal duration (log...). 10(T+1) is the target variable, and permutation test is used for statistical screening of whole brain functional connection edges. The zero distribution is generated by 5000 random permutations, and the empirical P value of the Pearson correlation coefficient of each edge is calculated. A double-threshold screening mechanism is applied to retain the significant correlation edges (one-sided test P<0.005, FDR correction) to ensure that the false positive rate of multiple comparisons is ≤0.05%.

[0067] Specifically, in the present embodiment, the forward network extracts the functional connection edges positively correlated with the withdrawal duration (correlation coefficient r>0), and the increase in the connection strength can predict a longer withdrawal period. The negative network extracts the functional connection edges negatively correlated with the withdrawal duration (correlation coefficient r<0), and the decrease in the connection strength is significantly associated with the prolongation of the withdrawal duration. The calculation process is as follows:

[0068] Forward network strength:

[0069] Negative network strength:

[0070] where e +,i , e -,j are the functional connection strength values of the significant edges in the positive / negative network, w +,i , w -,j are the regularization weights optimized by LOOCV.

[0071] Specifically, in the present embodiment, the prediction model uses a linear regression model, and the construction of the linear regression model is as follows: The target variable is the logarithmic value of the withdrawal duration; the regularization parameter λ∈[0.1, 1.0] is optimized by leave-one-out cross-validation (LOOCV) to prevent model overfitting. The model generalization ability is verified by independent samples, and the Spearman rank correlation coefficient ρ between the model prediction value and the actual value is ≥0.6 (95% confidence interval: 0.55-0.72).

[0072] Step S400, dynamic network strength analysis and clinical correlation mapping, calculating the individual positive network strength and negative network strength, generating a dynamic recovery curve, and incorporating age, medication history and other covariates in the model to verify the independence of network prediction.

[0073] Specifically, the calculation of the network strength includes:

[0074] Forward network strength calculation, based on the dynamic functional connection matrix, the forward network strength of the individual in the time series is calculated, defined as the sum of the weights of the significant positive correlation edges where w+,i is the regularization weight optimized by LOOCV, and e+,i is the connection strength of the positive edge filtered by permutation test (P<0.005, FDR correction).

[0075] Negative network strength calculation, the same method is used to calculate the negative network strength w-,j is the regularization weight optimized by LOOCV, e-,j is the negative edge connection strength screened by permutation test, and an individual dynamic recovery curve is generated to reflect the time sequence change characteristics of the positive / negative network strength in the withdrawal process.

[0076] Dynamic recovery curve analysis, the time series is segmented by a sliding time window (window length 30 seconds, step 10 seconds), the network strength in each time window is calculated, and the recovery trend is predicted by combining a polynomial regression model (fitting degree R 2 ≥0.75).

[0077] Specifically, in this embodiment, the dynamic recovery curve is generated by using the method of covariate stratified modeling, the covariates such as age and medication history are included in the multivariate linear regression model, and the dimensional difference is eliminated by using standardization (Z-score).

[0078] The stratified regression model can be constructed as:

[0079]

[0080] where X age is the age covariate, and X med is the medication history covariate.

[0081] After the dynamic recovery curve is generated, the independence of the network prediction can be verified by partial correlation analysis, variance inflation factor (VIF) test, and subgroup analysis.

[0082] Step S500, multi-modal data fusion and model optimization, high-precision head movement data correction is used to verify the stability of the core network, functional connection data and clinical indicators are integrated to construct a multi-dimensional prediction model, and the long-term withdrawal prediction accuracy is improved.

[0083] Specifically, in this embodiment, the head movement data correction uses Framewise Displacement (FD) as the head movement quantization index, and defines the high-quality data threshold as FD<0.3mm. The head displacement is monitored in real time by using the motion tracking technology based on the optical flow method (Optical Flow), and the high-frequency noise interference is eliminated by using the Gaussian filter (σ=1.5). Then, the motion artifacts are corrected at the pixel level by using the generative adversarial network (GAN). The corrected image without artifacts is generated by inputting the to-be-corrected image and the motion parameters (6 degrees of freedom head movement parameters, FD value). On the corrected data set, the stability of the functional connection network (core network node degree centrality variation coefficient CV<5%) is verified by Monte Carlo simulation (1000 iterations).

[0084] The multi-dimensional prediction model adopts a double-branch fusion network, a functional connection branch (3-layer fully connected network) and a clinical index branch (LSTM time sequence network) are processed in parallel, and a joint feature vector is generated through weighted splicing in the fusion layer; the regularization calculation adopts Elastic Net (α=0.5, λ=0.1), and the hyperparameters are adjusted through Bayesian optimization (50 iterations). The target variable defines long-term abstinence as the abstinence duration ≥ 6 months (binary variable), and the survival rate is evaluated by combining the Cox proportional hazards model.

[0085] Model verification and effect analysis:

[0086] 1. Prediction accuracy:

[0087] a) The correlation of the model to the abstinence duration is r=0.51, P<0.001.

[0088] 2. Independent sample verification:

[0089] a) The model is tested in a heterogeneous sample (N=48), and the combined network strength prediction validity r=0.407 (P=0.004).

[0090] b) The positive network prediction validity r=0.413 (P=0.003), and the negative network r=0.345 (P=0.016).

[0091] 3. Dynamic network strength analysis:

[0092] a) The network strength during the abstinence period increases with the duration: the strength of the 6-24 month group is significantly improved compared with the <1 month group (H 3,109 =31.314, P<0.001).

[0093] b) The network strength of the healthy control group (HCs) is between the <1 month group and the 6-24 month group (H 4,141 =46.203, P<0.001).

[0094] The method can identify the MUD abstinence neural network: the correlation of the model to the abstinence duration is r=0.51, and it can be reproduced on an independent data set. Compared with existing research, the cortical area composition of the neural recovery during the abstinence period is identified for the first time, providing a treatment target for clinical intervention treatment. The degree of neural recovery during the abstinence period of MUD is dynamically monitored: the nonlinear growth law of the network strength during the abstinence period is revealed, and a stage quantitative index is provided for rehabilitation evaluation.

[0095] Figure 3 The present application is an embodiment of the present application, and further provides a neural recovery brain network prediction system for the abstinence period of a methamphetamine disorder, comprising:

[0096] A data preprocessing module acquires multi-modal brain neural image data and performs data standardization preprocessing;

[0097] A whole brain functional connection network and feature extraction module extracts functional connection edges in a brain region functional matrix that are significantly related to the withdrawal duration, and establishes a positive network for predicting longer withdrawal time through connection enhancement and a negative network for predicting longer withdrawal through connection weakening;

[0098] A connection group prediction model and verification training module assigns weights to the positive network and the negative network respectively, and takes the logarithm of the withdrawal duration as the target variable to construct a multi-source regression model to calculate the individual network strength;

[0099] A dynamic network strength analysis and clinical correlation mapping module calculates the individual positive network strength and negative network strength, generates a dynamic recovery curve, and verifies the independence of the network prediction by including age, medication history and other covariates in the model;

[0100] A multi-modal data fusion and model optimization module uses high-precision head motion data correction to verify the stability of the core network, integrates functional connection data and clinical indicators to construct a multi-dimensional prediction model, and improves the long-term withdrawal prediction accuracy.

[0101] Figure 4 is a structural schematic diagram of an electronic device according to an embodiment of the present application. As shown in Figure 4 The electronic device according to an embodiment of the present application includes one or more input devices 1000, one or more output devices 1000, one or more processors 3000 and a memory 4000.

[0102] In an embodiment of the present application, the processor 1000, the input device 2000, the output device 3000 and the memory 4000 can be connected through a bus or other means. The input device 2000 and the output device 3000 can be standard wired or wireless communication interfaces.

[0103] The processor 1000 can be a central processing module (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0104] The memory 4000 can be a high speed RAM memory, or a non-volatile memory such as a disk memory. The memory 4000 is used to store a set of computer programs that the input device 2000, the output device 3000 and the processor 1000 can invoke the program codes stored in the memory 4000.

[0105] The computer programs stored in the memory 4000 include program instructions that, when executed by the processor, cause the processor to perform the steps of the prediction method as described in the above embodiments.

[0106] One embodiment of the present application also provides a computer readable storage medium. The computer readable storage medium can be a high speed RAM memory, or a non-volatile memory such as a disk memory. The computer readable storage medium can be connected through an external computing device or a network to read a set of computer programs stored in the computer readable storage medium. The computer programs stored in the computer readable storage medium include program instructions that, when executed by the processor, cause the processor to perform the steps of the prediction method as described in the above embodiments.

[0107] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and are not to be construed as limiting the present application, and those ordinarily skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting neural recovery during the withdrawal period of methamphetamine disorder using brain networks, characterized in that, Includes the following steps: Step S100: Acquire multimodal brain neural imaging data and perform data standardization preprocessing; Step S200: Establish a whole-brain functional connectivity network and extract features. Extract functional connectivity edges that are significantly related to abstinence duration from the brain region functional matrix. Establish a positive network that enhances connectivity to predict longer abstinence time and a negative network that weakens connectivity to predict longer abstinence time. Step S300: Train and validate the connection group prediction model. Assign weights to the positive and negative networks respectively, and construct a multivariate regression model to calculate the individual network strength using the logarithm of abstinence duration as the target variable. Step S400: Dynamic network strength analysis and clinical association mapping, calculate individual positive and negative network strength, generate dynamic recovery curves, and include covariates such as age and medication history in the model to verify the independence of network predictions; Step S500 involves multimodal data fusion and model optimization. High-precision head movement data is used for correction to verify the stability of the core network. Functional connection data and clinical indicators are integrated to construct a multidimensional prediction model and improve the accuracy of long-term withdrawal prediction.

2. The method for predicting neural recovery during the withdrawal period of methamphetamine disorder using brain networks as described in claim 1, characterized in that, In step 100, the multimodal neuroimaging data includes high-resolution T1-weighted images covering the entire brain, and the preprocessing includes: head movement and distortion correction, spatial normalization, noise regression analysis, and time series cleaning.

3. The method for predicting neural recovery during the withdrawal period of methamphetamine disorder using brain networks as described in claim 2, characterized in that, In step S200, brain region segmentation and time series extraction are first performed, including: S201, based on the Shen-268 brain map, performs brain region segmentation, dividing the whole brain into 268 functional nodes; S202, extract the standardized BOLD time series of each node, calculate the Pearson correlation coefficient between nodes, and generate a 268×268 functional connectivity matrix.

4. The method for predicting neural recovery during the withdrawal period of methamphetamine disorder using brain networks as described in claim 3, characterized in that, In step S200, dynamic network modeling is performed, including: S203 uses a double threshold method and permutation test to correct and screen functional connections that are significant with abstinence duration; S204 divides significant edges into positive networks that enhance connectivity and predict longer withdrawals, and negative networks that weaken connectivity and predict longer withdrawals.

5. The method for predicting neural recovery during the withdrawal period of methamphetamine disorder using brain networks as described in claim 4, characterized in that, In step S300, the model training includes: S301, Feature Weighting: Assign weights to the positive and negative networks respectively, and obtain the individual network strength by calculating the sum of the weights of significant edges; S302, Regression Modeling: Using the logarithm of abstinence duration as the target variable, a multiple linear regression model is constructed, and leave-one-out cross-validation is used to optimize the parameters.

6. The method for predicting neural recovery during the withdrawal period of methamphetamine disorder using brain networks as described in claim 5, characterized in that, In step S300, the validation and generalization of the model includes: S303, Permutation test: The significance of the model is verified by generating a zero distribution through 5,000 random permutations; S304, Independent Sample Validation: Test the model's predictive ability on a new sample set and calculate the Spearman correlation between the predicted and actual values.

7. The method for predicting neural recovery during the withdrawal period of methamphetamine disorder as described in claim 6, characterized in that, Step S400 further includes: S401, Network Strength Quantization: Calculates the positive and negative network strength of an individual network and generates a dynamic recovery curve; S402, clinical variable control, incorporates covariates such as age and medication history into the model to verify the independence of network predictions.

8. The method for predicting neural recovery during the withdrawal period of methamphetamine disorder as described in claim 7, characterized in that, Step S500 includes: S501, motion artifact correction, adopts high-precision head motion correction, with high-quality data of FD<0.3mm, to verify the stability of the core network; The S502 integrates functional data and clinical indicators to build a multidimensional predictive model and improve the accuracy of long-term withdrawal prediction.

9. A brain network prediction system for neural recovery during the withdrawal period of methamphetamine disorder, characterized in that, include: The data preprocessing module acquires multimodal brain neural imaging data and performs data standardization preprocessing. The whole-brain functional connectivity network and feature extraction module extracts functional connectivity edges that are significantly related to abstinence duration from the brain region functional matrix, and establishes a positive network that enhances connectivity to predict longer abstinence time and a negative network that weakens connectivity to predict longer abstinence time. The connection group prediction model and validation training module assign weights to the positive and negative networks respectively, and construct a multi-source regression model to calculate the individual network strength using the logarithm of abstinence duration as the target variable. The dynamic network strength analysis and clinical association mapping module calculates the positive and negative network strength of individuals, generates dynamic recovery curves, and incorporates covariates such as age and medication history into the model to verify the independence of network predictions. The multimodal data fusion and model optimization module uses high-precision head movement data correction to verify the stability of the core network, integrates functional connection data and clinical indicators, constructs a multidimensional prediction model, and improves the accuracy of long-term withdrawal prediction.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the steps of the method as described in any one of claims 1 to 9.

Images