Mental disorder brain network damage and whole body system disease associated dynamic trajectory construction and visual mapping method

Abstract

The invention discloses a dynamic trajectory construction and visual mapping method for association of mental disorder brain network damage and systemic system diseases, and belongs to the field of artificial intelligence medical application. According to the method, high-resolution MRI images, biomarkers and clinical information of major mental disorder patients are collected, and the influence of factors such as age, gender, medication and diagnosis on the braingut axis and the cardio-cerebral axis is evaluated through multi-modal data fusion. By constructing a disease dynamic trajectory model, brain structures and function change modes corresponding to different mental disorders are identified. Large-scale samples are analyzed through machine learning, potential risks and protection factors are extracted, and a visual tool is developed to visually display changes of the brain under different disease systems. A closed-loop feedback mechanism is established through follow-up visit, the disease progress and the intervention effect are dynamically tracked, and key evaluation indexes are identified. According to the invention, theoretical basis and practical guidance are provided for early screening, precise intervention and personalized treatment of mental disorders, and the diagnosis and treatment accuracy and efficiency are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence in medicine, specifically to a method for constructing and visually mapping the dynamic trajectory of the association between brain network damage in mental disorders and systemic diseases. Background Technology

[0002] The diagnosis and treatment of mental disorders currently face multiple challenges. Traditional diagnostic methods mainly rely on clinical symptoms and scale assessments, lacking objective biological markers. Existing brain imaging analysis techniques often focus on single-modality data, failing to fully integrate patients' physiological, biochemical, and clinical information, thus limiting diagnostic accuracy. Research on the association between mental disorders and systemic diseases is still in its early stages, particularly regarding the unclear roles of the brain-gut axis and the heart-brain axis in the pathogenesis of mental disorders.

[0003] Existing technologies suffer from low fusion efficiency and significant information loss when processing multimodal medical data. Traditional statistical analysis methods struggle to capture complex nonlinear relationships and effectively identify multi-factor interactions. Patient disease trajectory tracking often employs cross-sectional study designs, lacking longitudinal dynamic monitoring, making it difficult to accurately assess disease progression patterns and treatment effectiveness. Visualization technologies are limited to two-dimensional static images, failing to intuitively present the dynamic changes in brain networks.

[0004] Furthermore, existing systems lack effective feedback and update mechanisms, and model parameters are fixed, making adaptive optimization based on newly collected data impossible. The identification of risk and protective factors relies primarily on empirical judgment, lacking a systematic data-driven analysis framework. These technical deficiencies severely restrict the accurate diagnosis and treatment of mental disorders and the development of personalized treatment plans. Summary of the Invention This invention aims to address the technical problems in the diagnosis and treatment of mental disorders, such as the lack of objective assessment methods, low efficiency of multimodal data fusion, and imperfect disease trajectory tracking, and provides a method for constructing and visually mapping the dynamic trajectory of the association between brain network damage in mental disorders and systemic diseases.

[0005] The technical solution of the present invention is as follows: The first step was to construct a multimodal data acquisition and preprocessing system. This involved acquiring structural MRI images of the patient at a resolution of 1 mm. 3 Voxel-sized scans were performed using T1-weighted, T2-weighted, and diffusion tensor imaging sequences. Biomarker data were extracted, including serum inflammatory factors, neurotransmitter metabolites, and gut microbiota composition. Detailed clinical information was collected, covering demographic characteristics, medical history, medication records, and symptom scores. Quality control was performed on the collected data, removing images with motion artifacts exceeding 2 mm and standardizing biomarker values ​​to the z-score range. A secure data storage mechanism was established, employing encrypted transmission and distributed storage to ensure patient privacy.

[0006] The second step is to establish a multimodal data fusion model. A deep learning architecture is used to effectively integrate heterogeneous data. The model includes three parallel feature extraction branches. The image feature extraction branch uses a 3D convolutional neural network with 18 layers and a kernel size of 3×3×3. The biomarker feature extraction branch uses a fully connected network with four hidden layers, each containing 256, 128, 64, and 32 neurons respectively. The clinical information feature extraction branch uses embedding layers to process categorical variables, and continuous variables undergo batch normalization.

[0007] The fusion layer employs a multi-head attention mechanism, with 8 attention heads set to the maximum number. The calculation formula is as follows:

[0008] Where Q, K, and V represent the query, key, and value matrices, respectively, and d_k is the dimension of the key vector. The contribution of different modal features is dynamically adjusted through attention weights.

[0009] Introduce a gating mechanism to control the flow of information:

[0010] Where W_g is the gate weight matrix, h_1, h_2, and h_3 are the eigenvectors of the three modes, σ is the sigmoid function, and ⊙ represents element-wise multiplication.

[0011] The third step is to construct a multi-factor influence analysis framework. A hierarchical linear mixed-effects model is established, with the following equation:

[0012] Where Y_ij represents the observation of the i-th patient at the j-th time point, β represents the fixed effect coefficient, X represents the covariate, b represents the random effect, T represents time, and ε represents the residual.

[0013] The model parameters were estimated using restricted maximum likelihood estimation, and the effect size of each factor was calculated. The age factor was modeled using a piecewise linear function, with the piecewise points set at 18, 45, and 65 years of age. Gender was treated as a binary variable, and medication factors were constructed as multilevel categorical variables based on drug type and dosage.

[0014] Introducing time-varying covariates to handle dynamically changing factors:

[0015] Where X_ij(0) is the baseline value, δ_i is the linear rate of change, and γ_i is the nonlinear rate of change.

[0016]

[0017] The fourth step is to generate the patient's dynamic disease trajectory. The brain network connectivity matrix for each time point is extracted, with a dimension of 90×90, corresponding to the AAL brain region template. Network topological attributes are calculated, including clustering coefficients, feature path lengths, and modularity index, using the following formula: Where n is the number of nodes, e_i is the number of edges between the neighbors of node i, k_i is the degree of node i, d_ij is the shortest path between nodes, m is the total number of edges, A_ij is the element of the adjacency matrix, and c_i is the community to which node i belongs.

[0018] Long Short-Term Memory (LSTM) networks were applied to predict brain network states. The network structure consisted of two LSTM layers, each containing 128 hidden units. Mean squared error was used as the loss function, and the Adam optimizer was used for parameter optimization with a learning rate of 0.001.

[0019] Step 5: Assess the function of the gut-brain axis system. Analyze the association between gut microbiota composition and brain network activity; the calculation formula is as follows:

[0020] Where M_gut is the bacterial abundance matrix, B_brain is the brain activity matrix, Cov is the covariance, and σ is the standard deviation.

[0021] Establish a bidirectional regulatory model of the brain-gut axis:

[0022] Where B represents brain state, M represents gut microbiota state, I represents inflammation level, D represents dietary factors, and α, β, and γ are regulatory parameters.

[0023] Step 6: Assess the function of the heart-brain axis system. Calculate the heart-brain coupling index:

[0024] Where PLV is the phase-locked value and MI is the mutual information. The formula for calculating PLV is:

[0025] Where Δφ(t) is the phase difference between heart rate and brain activity.

[0026] Step 7: Develop visualization mapping tools. WebGL technology is used to achieve real-time rendering of 3D brain maps, maintaining a frame rate above 60fps. Vertex and fragment shaders are used to process the brain surface mesh, which contains approximately 32,000 vertices. Multiple color mapping schemes are implemented, supporting linear, logarithmic, and quantile mappings. The D3.js library is integrated to draw statistical charts, including time-series curves, heatmaps, and network diagrams.

[0027] An interactive user interface was developed, supporting rotation, zoom, and panning operations. Brain region selection and information retrieval functions were implemented; clicking on a brain region displays detailed damage indicators and historical trends.

[0028] Step 8: Implement a closed-loop feedback mechanism. Design an incremental learning algorithm to update model parameters, and use an elastic weight consolidation method to prevent catastrophic forgetting. The regularization term is:

[0029] Where L_new is the loss for new data, F_i is the parameter importance, θ is the model parameter, and λ is the regularization coefficient.

[0030] Establish a model performance monitoring system to calculate prediction accuracy, sensitivity, and specificity. Trigger a model retraining process when performance metrics drop by more than 10%. Use cross-validation to evaluate the model's generalization ability by dividing the dataset into training, validation, and test sets in an 8:1:1 ratio.

[0031] Step 9: Identify risk and protective factors. Use a random forest algorithm to rank features by importance; the forest contains 500 decision trees with a maximum depth of 10. Construct a Cox proportional hazards model to assess the risk of disease progression; the risk function is:

[0032] Where h_0(t) is the baseline risk function, β is the regression coefficient vector, and x is the covariate vector.

[0033] LASSO regression was used for variable selection, and the regularization parameter was determined through 10-fold cross-validation. Identified risk factors included elevated inflammation levels, decreased brain network efficiency, and poor medication adherence. Protective factors included good social support, high cognitive reserve, and regular exercise.

[0034] Establish a risk stratification system:

[0035] Where w_i represents the factor weight and f_i represents the factor value. Patients are divided into three risk levels: low, medium, and high, based on the risk score.

[0036] Step 10: Assess treatment effectiveness and develop a personalized plan. Establish a multi-dimensional treatment response evaluation system, including dimensions such as symptom improvement, functional recovery, and quality of life improvement. Calculate the treatment response index:

[0037] Where ΔS, ΔF, and ΔQ represent the degree of improvement in symptoms, function, and quality of life, respectively, and w is the weighting coefficient, determined by the analytic hierarchy process.

[0038] Personalized treatment recommendations are generated based on the patient's disease trajectory characteristics and risk factor profile, using reinforcement learning algorithms to optimize treatment options. The algorithm employs a deep Q-network, and the action space includes combinations of drug types, dosages, and adjunctive therapies.

[0039] Establish a predictive model for treatment efficacy:

[0040] Where E(t) represents the therapeutic effect at time t, E_0 represents the maximum therapeutic effect, and k represents the therapeutic effect growth rate.

[0041] Through the above technical solution, this invention achieves multi-dimensional objective assessment of mental disorders, improving the accuracy of disease diagnosis and the personalization of treatment. The system can dynamically track disease progression, identify risk factors in a timely manner, and provide a scientific basis for clinical decision-making.

[0042] This invention has the following significant advantages over the prior art: First, multimodal data fusion technology enabled a comprehensive assessment of mental disorders, integrating brain imaging, biomarkers, and clinical information, improving diagnostic accuracy to over 85%, a 20 percentage point improvement over traditional methods. The multivariate analysis framework quantifies the influence weights of different factors, providing precise guidance for personalized treatment planning.

[0043] Secondly, dynamic trajectory construction technology enables precise tracking of disease progression, achieving a prediction accuracy of 78% and identifying disease deterioration trends 3-6 months in advance. Visual mapping tools provide intuitive 3D displays, allowing doctors to quickly understand complex brain network patterns and improving diagnostic efficiency by over 50%.

[0044] Furthermore, the closed-loop feedback mechanism ensures continuous system optimization, maintains stable model performance, and avoids parameter degradation. The systematic identification of risks and protective factors provides a scientific basis for preventative interventions, increasing the success rate of early interventions by 35%.

[0045] Finally, this invention establishes a standardized data processing and analysis workflow with good scalability and versatility, capable of adapting to the needs of different types of mental disorders and medical institutions. The system has low deployment costs, is easy to maintain, and has broad prospects for clinical application. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the overall system architecture of the present invention; Figure 2 This is a flowchart of the multimodal data fusion module of the present invention; Figure 3 This is a flowchart of the disease trajectory construction and prediction module of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the following embodiments provide a more detailed description of the invention. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention.

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to specific embodiments.

[0049] This study targeted patients with major mental disorders, including schizophrenia, bipolar disorder, and major depressive disorder, enrolling a total of 500 patients and 200 healthy controls. Patients ranged in age from 18 to 65 years, with a mean age of 35.2 ± 12.3 years and a male-to-female ratio of approximately 1:1. Inclusion criteria were: meeting the DSM-5 diagnostic criteria, having a disease duration of more than 2 years, currently in a stable phase, and having no contraindications for MRI examination. Exclusion criteria included: a history of substance dependence, severe physical illness, or neurological disorders. All participants signed informed consent forms, and the study protocol was approved by the hospital's ethics committee.

[0050] Data acquisition was performed at an imaging center equipped with a 3.0T MRI scanner. Structural MRI was performed using a 3D-MPRAGE sequence with the following parameters: TR=2300ms, TE=3ms, flip angle=9°, and voxel size=1×1×1mm. 3 The scan time was approximately 6 minutes. Functional MRI used an EPI sequence with TR=2000ms, TE=30ms, acquiring 240 time points, with a scan time of 8 minutes. Diffusion tensor imaging used 64 orientations with a b-value of 1000s / mm. 2 Voxel size 2×2×2mm 3 Resting-state functional MRI requires participants to close their eyes and relax, without performing any cognitive tasks.

[0051] Biomarker detection was performed in a standard laboratory. 10 ml of fasting venous blood was collected in the morning, centrifuged, and the serum was separated and stored at -80℃. Inflammatory factors such as IL-6 (detection range 0.1-10 pg / ml), TNF-α (0.5-50 pg / ml), and CRP (0.1-10 mg / L) were detected using ELISA, with detection sensitivity reaching the pg / ml level. Neurotransmitter metabolite detection, including serotonin, dopamine, and norepinephrine metabolites, was performed using liquid chromatography-mass spectrometry. Gut microbiota analysis was performed using 16S rRNA sequencing. DNA was extracted from fecal samples, and the V3-V4 region was amplified by PCR. Sequencing was performed using the Illumina MiSeq platform, with a sequencing depth of at least 50,000 sequences per sample.

[0052] Clinical information was collected using standardized scales. The Positive and Negative Syndrome Scale (PANSS) was used to assess schizophrenia symptoms, the Hamilton Depression Rating Scale (HAMD) to assess depressive symptoms, and the Young Mania Rating Scale (YMRS) to assess manic symptoms. Functional assessment used the Personal and Social Functioning Scale (PSP), and quality of life was assessed using the SF-36 scale. Detailed medication history was recorded, including medication type, dosage, duration of use, and adherence. Demographic information, education level, occupational status, family history, and other background data were collected.

[0053] The multimodal data fusion model was trained on a server cluster equipped with four NVIDIA Tesla V100 GPUs. Data preprocessing was performed using the FSL 6.0 and SPM12 software packages. Structural MRI was performed with skull dissection, registration to the MNI152 template, and segmentation into gray matter, white matter, and cerebrospinal fluid. Functional MRI was performed with temporal tomography correction, head motion correction, spatial normalization, and smoothing (FWHM=6mm). The mean time series of 90 AAL brain regions were extracted, and the Pearson correlation construct functional connectivity matrix was calculated.

[0054] The deep learning model is implemented using the PyTorch 1.9 framework, with a batch size of 32 and a training cycle of 100 epochs. A pre-training strategy is employed, first pre-training on a public dataset and then fine-tuning on local data. Model training utilizes a distributed strategy, employing data parallelism and model parallelism. The optimizer used is AdamW with an initial learning rate of 0.001 and a cosine annealing scheduling strategy. An early stopping mechanism is introduced, stopping training if the validation set loss does not decrease for 10 consecutive epochs. After convergence, the model achieves a classification accuracy of 83.5% on the test set with an AUC of 0.89.

[0055] Multivariate analysis was performed using the lme4 package in R language 4.1. A three-level nested model was constructed: time points were nested within individuals, and individuals were nested within diagnostic categories. Model fitting results showed that the standardized effect size of age was 0.42 (95% CI: 0.35–0.49, p<0.001), gender was 0.18 (95% CI: 0.10–0.26, p<0.05), medication type was 0.35 (95% CI: 0.28–0.42, p<0.001), and diagnostic subtype was 0.28 (95% CI: 0.20–0.36, p<0.01). Interaction analysis revealed a significant interaction effect between age and medication type (β=0.15, p<0.05) and between gender and diagnostic subtype (β=0.12, p<0.05).

[0056] Brain-gut axis system assessment showed that gut microbiota alpha diversity (Shannon index) was negatively correlated with depression severity (r=-0.42, p<0.001). The Bacteroidetes / Firmwallis ratio was significantly decreased in the patient group (1.2±0.5 vs 1.8±0.6, p<0.01). Bifidobacterium abundance was positively correlated with the strength of functional connectivity in the default mode network (r=0.35, p<0.01). Serum IL-6 levels were positively correlated with the intestinal permeability marker LPS (r=0.48, p<0.001). Brain-gut axis functional connectivity analysis showed that the anterior cingulate cortex was most strongly associated with gut microbiota diversity (β=0.38, p<0.001).

[0057] Assessment of the heart-brain axis revealed that the SDNN index of heart rate variability was decreased in the patient group (42±15 ms vs 58±18 ms, p<0.01). The autonomic balance index (LF / HF ratio) was increased during manic episodes (2.8±1.2 vs 1.5±0.6, p<0.01). Functional connectivity of the brainstem-limbic system-prefrontal cortex network was correlated with heart rate variability (r=0.41, p<0.001). Stress hormone cortisol levels were positively correlated with amygdala activity (r=0.45, p<0.001). The heart-brain coupling index significantly improved after treatment (0.42±0.15 vs 0.58±0.12, p<0.01).

[0058] Disease trajectory tracking lasted for 24 months, with assessments every 3 months. The correlation coefficient between the brain network state predicted by the LSTM network and the actual observed values ​​reached 0.76. Trajectory clustering analysis (using the k-means algorithm, k=3) identified three typical disease progression patterns: rapidly deteriorating (25% of patients), slowly progressive (55% of patients), and relatively stable (20% of patients). The rapidly deteriorating pattern was characterized by a rapid decline in prefrontal-striatal circuit functional connectivity (slope -0.08 / month), the slowly progressive pattern by a gradual decline (slope -0.03 / month), and the stable pattern by smaller fluctuations (slope -0.01 / month).

[0059] The accuracy of individualized trajectory prediction decreased over time: 82% at 3 months, 75% at 6 months, and 68% at 12 months. With the introduction of dynamic updates, the 12-month prediction accuracy improved to 74%. Trajectory transition analysis revealed that 15% of patients transitioned from a rapidly deteriorating pattern to a slowly progressive pattern after 6 months of treatment, highlighting the importance of early intervention.

[0060] The visualization system is deployed on the hospital's intranet server and accessed via a web browser, supporting Chrome, Firefox, and Safari. The system response time is less than 2 seconds, and it supports 10 concurrent users. The 3D brain map rendering uses WebGL 2.0, achieving voxel-level display of damage levels. Color mapping employs a scientific color scheme (Viridis) to avoid difficulties for colorblind users.

[0061] Interactive features include: mouse drag to rotate the view, scroll wheel to zoom, and double-click to focus on a specific brain region. The timeline uses a slider control, allowing for continuous playback or frame-by-frame viewing. Statistical charts are drawn using the ECharts library and support data export to CSV format. System logs record user actions for habit analysis and interface optimization.

[0062] A physician satisfaction survey (n=30) showed that 92% of users believed the system significantly improved diagnostic efficiency, 88% thought the visualization helped with patient communication, and 95% were willing to use the system in their daily clinical work. The average diagnosis time per patient was reduced from 45 minutes to 30 minutes, and diagnostic consistency increased from 72% to 85%.

[0063] The closed-loop feedback model underwent its first update 12 months after implementation. Data from 150 new patients were collected, with 120 used for training and 30 for validation. The update employed a transfer learning strategy, freezing the bottom-level feature extraction layer and fine-tuning only the top-level classification layer. The updated model achieved an 8% performance improvement on the new dataset (accuracy increased from 83.5% to 91.5%).

[0064] Model drift detection uses KL divergence to monitor changes in the input data distribution, with a threshold set at 0.1. When data drift is detected, the system automatically marks abnormal samples for manual review. During incremental learning, an experience replay buffer is used to store historical samples, with a buffer size of 1000 samples, employing a priority sampling strategy.

[0065] Risk factor analysis identified five major risk factors: elevated inflammation levels (IL-6 > 10 pg / ml, OR = 2.3), decreased brain network efficiency (global efficiency < 0.3, OR = 1.8), poor medication adherence (< 80%, OR = 2.1), insufficient social support (SSRS < 30 points, OR = 1.9), and poor sleep quality (PSQI > 10 points, OR = 1.7).

[0066] Protective factors included: good social support (SSRS score > 40, HR = 0.52, 95% CI: 0.35-0.78), high cognitive reserve (years of education > 12 years, HR = 0.61, 95% CI: 0.42-0.89), and regular exercise (> 150 minutes per week, HR = 0.68, 95% CI: 0.48-0.96). Multivariate Cox regression analysis showed that patients with all three protective factors had a 65% reduced risk of disease progression.

[0067] Treatment efficacy evaluation showed that the symptom improvement rate in the group receiving personalized treatment (n=100) was 72%, which was higher than the 54% in the conventional treatment group (n=100) (χ²). 2 =6.92, p<0.01). Decrease in PANSS total score: Personalized group 35.2±12.1 points, Standard group 24.6±10.8 points (t=6.54, p<0.001). Functional improvement (PSP score increase >10 points): Personalized group 68%, Standard group 45% (p<0.01).

[0068] Improved quality of life: The SF-36 total score increased, with the personalized treatment group scoring 18.5±7.2 points and the conventional group scoring 12.3±6.5 points (p<0.01). After 12 months of follow-up, the relapse rate in the personalized treatment group was 18% (18 / 100), significantly lower than the 31% (31 / 100) in the control group (RR=0.58, 95%CI: 0.35-0.97, p<0.05). Readmission rate: 12% in the personalized treatment group and 23% in the conventional group (p<0.05).

[0069] Cost-benefit analysis showed that although the initial investment in personalized treatment was higher (approximately RMB 3,000 per case for additional examinations), the average annual medical cost actually decreased by 15% (RMB 42,000 vs. RMB 49,500) due to the reduced recurrence and readmission rates. Patient work capacity recovery rate: 45% in the personalized group, 28% in the conventional group (p<0.01).

[0070] The system has been implemented in three psychiatric hospitals, serving over 2,000 patients. A regional data sharing platform has been established to enable multi-center data aggregation and analysis. A mobile application has been developed to support patient self-monitoring and remote doctor-patient communication. Standard operating procedures (SOPs) have been established, including data collection, quality control, and result interpretation.

[0071] This embodiment fully verifies the feasibility and effectiveness of the method in the diagnosis and treatment of major mental disorders, provides a complete technical solution for precision medicine of mental disorders, and has important clinical application value and social benefits.

Claims

1. A method for constructing and visualizing the dynamic trajectory of the association between brain network damage in mental disorders and systemic diseases, characterized in that, Includes the following steps: This study systematically collects high-resolution MRI imaging data, biomarker data, and detailed clinical information from patients with major mental disorders; assesses brain imaging changes corresponding to different mental disorders using multimodal data fusion technology; constructs comprehensive dynamic disease trajectories for patients, identifying potential risk and protective factors; explores the influence of multiple factors such as age, gender, medication, and diagnosis on the brain-gut axis and heart-brain axis systems; develops visualization mapping tools to intuitively display structural and functional changes in the brain under different disease systems; examines the correlation between these changes and clinical diagnoses; explores the long-term effects of influencing factors on brain changes through regular follow-ups, forming a closed-loop feedback mechanism; and identifies key indicators of disease progression and intervention effectiveness.

2. The method according to claim 1, characterized in that, The systematic data collection process includes: acquiring high-resolution structural MRI images of the patient, including T1-weighted, T2-weighted, FLAIR sequences, and diffusion tensor imaging, with a spatial resolution of 1 mm; collecting blood biomarkers, including inflammatory factors, neurotransmitter metabolites, and oxidative stress markers; collecting gut microbiota samples for 16S rRNA sequencing analysis; recording detailed clinical information, including disease course, symptom scores, medication history, and family history; and using a deep learning network for multimodal data fusion, comprising convolutional neural networks for image data, fully connected networks for numerical data, and embedding layers for categorical data, achieving feature-level fusion through an attention mechanism.

3. The method according to claim 1, characterized in that, The multifactor impact analysis framework adopts a hierarchical linear model, with individual patients as the first level and time points as the second level, and calculates the fixed and random effects of each factor using the maximum likelihood estimation method. Establish a covariate matrix including variables such as age, gender, type of medication, dosage of medication, and diagnosis type; use interaction analysis to identify synergistic or antagonistic effects between different factors and generate a factor influence weight matrix.

4. The method according to claim 1, characterized in that, The process of constructing the patient's dynamic disease trajectory includes: extracting the brain network connectivity matrix at each time point, and using graph theory analysis to calculate the network topology attributes; applying the sliding window technique to dynamically analyze the time series data, with the window size set to three months and the step size to one month; using the long short-term memory network to predict the brain network state at future time points; and identifying typical disease progression patterns through trajectory clustering algorithms to generate personalized disease trajectory curves.

5. The method according to claim 1, characterized in that, The assessment method for the brain-gut axis system includes: collecting 16S rRNA sequencing data of gut microbiota and calculating the microbiota diversity index; extracting serum inflammatory factor levels, including interleukins, tumor necrosis factor, and C-reactive protein; analyzing the correlation between the brain default mode network and the composition of gut microbiota; and constructing a brain-gut axis functional connectivity model to quantify the functional status of bidirectional regulatory pathways.

6. The method according to claim 1, characterized in that, The assessment method for the heart-brain axis system includes: collecting heart rate variability data and calculating time-domain and frequency-domain indices; measuring blood pressure, blood lipids, and echocardiographic parameters; analyzing the correlation between autonomic nervous system function and brain network activity; establishing a heart-brain axis coupling index to assess the synchronicity between the cardiovascular system and the central nervous system.

7. The method according to claim 1, characterized in that, The visualization mapping tool uses WebGL technology to render 3D brain maps and supports real-time interactive operation; it uses color coding to show the degree of damage in different brain regions, with red indicating severe damage, yellow indicating moderate damage, and green indicating normal; it provides a timeline sliding function to dynamically display changes in the brain network during disease progression; it integrates a statistical analysis result display module to display the values ​​and trend graphs of key assessment indicators in real time; and it supports multi-view switching and brain region localization navigation functions.

8. The method according to claim 1, characterized in that, The implementation steps of the closed-loop feedback mechanism include: setting the follow-up interval to once every three months, with a continuous follow-up period of no less than two years; repeatedly collecting multimodal data at each follow-up time point; using incremental learning algorithms to update the parameters of the disease trajectory model; comparing the difference between the actual observed values ​​and the model predicted values, and calculating the prediction accuracy; and triggering the model retraining process when the accuracy is lower than a set threshold.

9. The method according to claim 1, characterized in that, It also includes steps for identifying risk and protective factors: using machine learning classification algorithms to analyze large-scale sample data; extracting characteristic variables related to the rate of disease progression; constructing a risk scoring model to generate an individualized risk score for each patient; identifying protective factors, including indicators such as good drug response, adequate social support, and high cognitive reserve; and establishing a dynamic adjustment mechanism for factor weights to adapt to the characteristics of different disease stages.

10. The method according to claim 1, characterized in that, The method also includes a treatment effect evaluation function: establishing a baseline database to record various indicators before treatment; tracking the changing trends of brain network indicators during treatment; calculating the treatment response index to quantify the effects of different treatment plans; generating personalized treatment suggestion reports to provide data support for clinical decision-making; and constructing a treatment plan optimization algorithm to recommend the best intervention strategy based on historical data.

Images