A method, device and medium for generating a diagnosis and treatment recommendation for a neuroimmunological disease

Abstract

The application discloses a neural immune disease diagnosis and treatment suggestion generation method and device and a medium, relates to the field of biomedical signal processing, and comprises the following steps: acquiring multi-modal brain-computer interface data of a subject; performing space-time alignment and reparameterization processing on the multi-modal brain-computer interface data to obtain multi-modal space-time alignment data; performing joint optimization on the multi-modal space-time alignment data based on a joint energy function and a deep autoencoder to obtain a unified pathological representation; constraining the unified pathological representation in a pathological manifold space, so that the pathological state of the subject continuously evolves over time on the pathological manifold space to determine manifold pathological representations at each time in a set period; and constructing a diagnosis and treatment decision optimization model based on the manifold pathological representations at each time in the set period, side effects of each intervention measure and a target pathological state, and generating a diagnosis and treatment suggestion according to the diagnosis and treatment decision optimization model. The application improves the accuracy of neural immune disease diagnosis and treatment suggestions.

Description

Technical Field

[0001] This application relates to the field of biomedical signal processing, and in particular to a method, device, and medium for generating diagnostic and treatment recommendations for neuroimmunological diseases. Background Technology

[0002] Neuroimmune diseases are a group of complex diseases caused by abnormal interactions between the nervous and immune systems. Their typical characteristics include significant fluctuations in disease course, marked individual differences, and pathological mechanisms spanning multiple physiological systems. Examples include multiple sclerosis, autoimmune encephalitis, myasthenia gravis, neurological lupus erythematosus, and neurodegenerative diseases accompanied by immune abnormalities.

[0003] The existing diagnostic and treatment technologies have the following main shortcomings: (1) Data dimension fragmentation: Existing diagnosis and assessment are mostly based on single or a small number of physiological indicators, such as brain imaging or serum immune indicators, which are difficult to reflect the dynamic coupling relationship between the neuro-immune system.

[0004] (2) Limitations of brain-computer interface applications: Existing brain-computer interface technologies mainly focus on neural signal acquisition and control applications, lacking systematic modeling of the immune system, physiological system and behavioral system.

[0005] (3) Lack of a unified modeling framework for multimodal data: Even when multimodal data is introduced, existing methods mostly use simple feature splicing or parallel input methods, which do not solve the inconsistency between different modalities in terms of time scale, spatial semantics and pathological significance.

[0006] (4) Lack of closed-loop mechanism for diagnosis and treatment decision making: Most existing intelligent diagnosis and treatment methods are static assessments or one-time decisions, and cannot be adaptively optimized according to the dynamic changes in individual pathological states.

[0007] Therefore, there is an urgent need for a new method that can unify the modeling of neuro-immune multimodal information from the underlying approach and form a closed-loop diagnosis and treatment decision. Summary of the Invention

[0008] The purpose of this application is to provide a method, device, and medium for generating diagnostic and treatment recommendations for neuroimmunological diseases, which can improve the accuracy of such recommendations.

[0009] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for generating diagnostic and treatment recommendations for neuroimmunological diseases, including: Acquire multimodal brain-computer interface data from subjects; The multimodal brain-computer interface data is spatiotemporally aligned and reparameterized to obtain multimodal spatiotemporally aligned data; Based on the joint energy function and deep autoencoder, the multimodal spatiotemporal aligned data is jointly optimized to obtain a unified pathological representation; The unified pathological representation is constrained within the pathological manifold space, so that the pathological state of the subject evolves continuously over time in the pathological manifold space, thereby determining the manifold pathological representation at each moment within a set time period. Based on the manifold pathological representation at each time point within a set time period, the side effects of various interventions, and the target pathological state, a diagnostic and treatment decision optimization model is constructed, and diagnostic and treatment suggestions are generated based on the optimized model.

[0010] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for generating diagnostic and treatment recommendations for neuroimmunological diseases.

[0011] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for generating diagnostic and treatment recommendations for neuroimmunological diseases.

[0012] According to the specific embodiments provided in this application, this application has the following technical effects: First, the use of multimodal brain-computer interface data breaks through the limitations of single-modal information, and can comprehensively capture the physiological and pathological features related to the nervous and immune systems. Then, the multimodal brain-computer interface data is spatiotemporally aligned and reparameterized, which solves the problems of temporal misalignment and scale inconsistency of multi-source data, and improves the consistency and usability of the data. Then, joint optimization is achieved by using joint energy functions and deep autoencoders to transform heterogeneous data into a unified pathological representation, realizing the standardization and abstraction of pathological features, reducing feature redundancy and interference. Furthermore, the pathological representation is constrained to the pathological manifold space to realize continuous dynamic evolution modeling of disease state, which can accurately track the development trend of the disease and overcome the defect of traditional static diagnosis that is difficult to reflect the changes in the course of the disease. Finally, a diagnosis and treatment decision optimization model is constructed based on the manifold pathological representation, intervention side effects and target pathological state, which can automatically generate individualized and optimal diagnosis and treatment suggestions, taking into account both efficacy and safety. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart illustrating a method for generating diagnostic and treatment recommendations for neuroimmunological diseases, provided as an embodiment of this application.

[0015] Figure 2 This is a schematic diagram of the multimodal spatiotemporal alignment process in one embodiment of this application.

[0016] Figure 3 This is a schematic diagram of the neural-immune coupling modeling process in one embodiment of this application.

[0017] Figure 4 This is a schematic diagram of the diagnostic and treatment decision optimization process in one embodiment of this application.

[0018] Figure 5 This is a schematic diagram of the functional modules of a device for generating diagnostic and treatment suggestions for neuroimmunological diseases, provided in an embodiment of this application.

[0019] Figure 6 This is a schematic diagram of the hardware structure of a device for generating diagnostic and treatment recommendations for neuroimmunological diseases in one embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The purpose of this application is to propose a method for generating diagnostic and treatment suggestions for neuroimmunological diseases. By constructing a multimodal brain-computer interface data system for neuroimmunological diseases, a cross-modal and cross-scale spatiotemporal alignment method is proposed, and neuro-immune coupling modeling and diagnostic and treatment decision optimization are achieved in a unified pathological representation space, thereby overcoming the shortcomings of existing technologies.

[0022] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] In one exemplary embodiment, a method for generating diagnostic and treatment recommendations for neuroimmunological diseases is provided. This method is executed by a computer device, specifically a terminal or server, either alone or jointly. Figure 1 As shown, the method for generating diagnostic and treatment recommendations for neuroimmunological diseases includes the following steps 101 to 105.

[0024] Step 101: Obtain multimodal brain-computer interface data from the subjects.

[0025] In a specific application example, multimodal brain-computer interface (BCI) data is acquired through a multimodal data acquisition module and obtained through a computer device. This multimodal BCI data specifically includes data from the following multiple modalities: Central nervous system activity modalities: electroencephalography (EEG), magnetoencephalography (MEG), functional near-infrared spectroscopy (FIR), and neural stimulation-evoked responses, etc.

[0026] Peripheral immune modality: dynamic data such as the proportion of immune cells, inflammatory factors, and immune metabolic indicators.

[0027] Autonomic nervous system and physiological modalities: heart rate variability, skin conductance, respiratory rhythm, etc.

[0028] Behavioral and functional modalities: performance on movement trajectories, postures, and cognitive and language tasks.

[0029] The above data is organized in time series or quasi-continuous form to form multimodal brain-computer interface data.

[0030] Step 102: Perform spatiotemporal alignment and reparameterization processing on the multimodal brain-computer interface data to obtain multimodal spatiotemporal aligned data.

[0031] In a specific application example, such as Figure 2 As shown, step 102 includes steps 21 to 25.

[0032] Step 21: For any two modalities of the multimodal brain-computer interface data, determine the physiological response delay between the two modalities, and perform delay compensation on the brain-computer interface data of the two modalities based on the physiological response delay to obtain global compensation data for the two modalities.

[0033] This application constructs a cross-modal time mapping function to address the differences in sampling frequency and physiological response delay among different modalities: , will the i Modality Time series reparameterization to the first j Modality The time scale enables heterogeneous modal data to be aligned and compared on a unified time axis. Among these, It is a cross-modal time mapping function. For the first i Modality Time series, For the first j Time series of multiple modes.

[0034] First, estimate the first causality using cross-correlation analysis or Granger causality test. i Modality Compared to the first j Modality Physiological response delay : ; ;in, It is a cross-correlation function. This is the time lag. For the first i The signal mean of each mode For the first j The signal mean of each mode For the first i The standard deviation of the signal for each mode For the first j The standard deviation of the signal for each mode for t Time of the first i The values ​​of each modality, for Time of the first j The values ​​that a modality can take.

[0035] Further use of formula For the i Modality Delay compensation will be implemented. Among other things, for t Time of the first i The value after delay compensation for each mode.

[0036] Step 22: Perform dynamic time warp mapping on the global compensation data of the two modes to obtain the optimal time alignment path between the two modes.

[0037] For nonlinear time warps, a dynamic time programming algorithm is used to construct the optimal time alignment path, as follows: ; in, For local distance metrics (such as Euclidean distance or Mahalanobis distance). This is the cumulative distance matrix. For the first i Modality The sampling point index, , For the first j Modality The sampling point index, Sampling points refer to sampling points at different times. P For the first i Modality The number of sampling points For the first j Modality The number of sampling points For the first i Modality The Each sampling point takes a value. For the first j Modality The Each sampling point takes a value.

[0038] Optimal time alignment path By backtracking The backtracking process is as follows: from the endpoint Begin by selecting the current point at each step. Among the three neighbors (top, left, and top left), make The direction with the smallest value, until reached. .in, H This represents the number of alignment points in the optimal time alignment path.

[0039] To construct an adaptive time mapping function, the optimal time alignment path is... The index pairs in the table are converted to physical time pairs. Assume the first... i Modality The time vector is , No. j Modality The time vector is , For the first i Modality The P Each sampling point time, For the first j Modality The Q Each sampling point time. Based on the optimal time alignment path. The discrete-time mapping relationship is defined as follows: .in, Optimal time alignment path The first in The formula represents the alignment point. i Modality exist The time mapping at the index corresponds to the first j Modality exist The time at the index. This mapping establishes the correspondence between discrete sampling points, laying the foundation for subsequent continuous mapping.

[0040] Step 23: Based on the optimal time alignment path between the two modes, construct an adaptive time mapping function through Gaussian process regression, and perform nonlinear time mapping on the global compensation data of the two modes to obtain multimodal time alignment data.

[0041] To handle nonlinear time scaling between sampling points and achieve mapping across the continuous time domain, this application is based on the optimal time alignment path. Building the training set ,in, For the first i The mode in the th ... h The timestamp at each alignment point For the first j The mode in the th ... h The timestamps at each alignment point. An adaptive time mapping is constructed using Gaussian process regression with this training set: ; .

[0042] in, For Gaussian processes, and For input variables (time points); The mean function (about The function is usually set to ); For radial basis kernel functions; The variance of the signal; The length scale parameter is determined by maximizing the length of the training set. Marginal likelihood function, optimization and After training is complete, for any point in time... It can be predicted posteriorly through a Gaussian process. This enables precise alignment in the continuous time domain.

[0043] Step 24: Construct a multi-scale time window set, and perform cross-scale joint modeling on the time-aligned data of each modality based on the multi-scale time window set to obtain the cross-scale joint data of each modality.

[0044] To simultaneously characterize the rapid changes in neural activity and the slow changes in immune status, this application constructs a set of multi-scale time windows: , N For the number of modes, N =4. Different time windows correspond to different response scales of physiological systems: , representing the rapid response scale of neural electrical activity; , indicating the scale of neurovascular coupling; , representing the scale of autonomic nervous system regulation; , which represents the scale of changes in immune status.

[0045] Cross-scale joint modeling can be achieved through weighted or hierarchical methods, as shown in the formula: ;in, For time-aligned data of any modality, For learnable weight coefficients, satisfying , For the first Feature extraction functions at various time scales , for The mean of the signal within the time window. for The standard deviation of the signal within the time window, for The skewness of the signal within the time window. for Kudo of the signal within the time window.

[0046] Step 25: Map the cross-scale joint data of each modality to the standard neural function space and introduce pathological semantic labels to obtain multimodal spatiotemporal aligned data.

[0047] This application, based on time alignment, maps central nervous system modalities to a standard neural functional space, and projects immune modalities to their corresponding functional spaces using a neuro-immune functional association model; simultaneously, it introduces pathological semantic labels to achieve unification of different modalities at the pathological meaning level. The specific processing steps are as follows: (1) Spatial mapping of central nervous system modalities.

[0048] Central nervous system modal data (EEG, MRI, functional near-infrared spectroscopy) are spatially standardized and mapped to a predefined standard neural functional space. For neural stimulation-evoked responses, evoked potential components (P100, N200, etc.) within a specific time window (e.g., 0ms~500ms) after stimulation are first extracted, and then mapped to the corresponding brain regions (e.g., motor cortex, prefrontal cortex) in the standard brain space through source tracing analysis, serving as the activation intensity of that brain region. It participates in subsequent modeling. The standard neural functional space is constructed based on the MNI (Montreal Neurological Institute) coordinate system or standard brain atlases (such as the AAL atlas and the Desikan-Killiany atlas), and includes predefined anatomical regions and functional networks (such as the default pattern network, sensorimotor network, salience network, etc.).

[0049] For EEG / MEG signals, the scalp sensor spatial signals are inverted to the cortical source space using source localization algorithms (such as minimum norm estimation and beamforming), and then mapped to the standard brain space using a registration algorithm. For functional near-infrared signals, the propagation path of photons in brain tissue is calculated using Monte Carlo simulation or the finite element method, the channel spatial signals are inverted to the changes in cortical hemoglobin concentration, and then registered to the standard brain space.

[0050] The mapped neural feature vector is: ;in, express A 3D real space, A The number of functional regions in a standard brain atlas. For the first The activation intensity of individual brain regions or functional networks .

[0051] (2) Spatial mapping of peripheral immune modalities.

[0052] The peripheral immune modality data is projected onto the neural functional space using a neuro-immune functional association model. The neuro-immune functional association model is constructed as follows: First, a neuro-immune association matrix is ​​constructed based on prior medical knowledge. ,in, Number of immune markers. Matrix elements. Indicates the first The brain region and the first The association strength of the immune indicators is based on experimental evidence reported in the literature or expert knowledge.

[0053] Then, the projection of the immune modality into the neural functional space is calculated as follows: ;in, For immune feature vectors, This is the original immune feature vector, which includes all peripheral immune indicators (proportion of immune cells, concentration of inflammatory factors, and immune metabolic indicators). Indicates the first m The concentration or proportion of each immune indicator, m =1~ M .

[0054] (3) Introduce pathological semantic tags.

[0055] Pathological semantic tags Binding with multimodal data, where, This represents the number of pathological categories. Label definitions include: disease diagnosis labels (such as "relapse phase of multiple sclerosis", "acute phase of autoimmune encephalitis", etc.), pathological state labels (such as "active inflammation", "suppressed neurological function", "immune tolerance", etc.), and treatment response labels (such as "hormone sensitive", "effective immunosuppressant", etc.).

[0056] For historical data, pathological semantic labels are annotated by clinical experts through retrospective medical record analysis. For real-time data, pathological semantic labels are automatically predicted using a pre-trained classifier and dynamically updated as the diagnosis and treatment process progresses.

[0057] (4) Integration of autonomic nervous system with physiological and behavioral functional modalities.

[0058] The autonomic nervous system and physiological modal data are integrated into a unified representation in the following way: first, the time-domain features (mean, standard deviation) and frequency-domain features (power spectral density) of each modality are extracted; then, the dimensionality is reduced to a low-dimensional feature space through principal component analysis or autoencoder; finally, the data are mapped to the corresponding dimension of the standard neural function space through a fully connected layer.

[0059] Behavioral functional modal data (motion trajectory, posture, task performance) are integrated into a unified representation in the following way: motion trajectory and posture data are used to extract behavioral features through a spatiotemporal graph convolutional network; task performance scores are directly used as scalar features; and then fused with neural functional spatial features through a cross-modal attention mechanism.

[0060] In another specific application example, the cross-modal time mapping function can be replaced by a hidden Markov model, Gaussian process regression, or a Transformer-based temporal alignment network. Multi-scale time windows can be replaced by wavelet transform multi-resolution analysis or a multi-rate sampling system. Spatial alignment can be replaced by anatomical landmark-based registration or functional connectivity-based mapping.

[0061] Step 103: Based on the joint energy function and deep autoencoder, the multimodal spatiotemporal aligned data is jointly optimized to obtain a unified pathological representation.

[0062] In a specific application example, the multimodal spatiotemporal aligned data includes neural feature vectors and immune feature vectors. For example... Figure 3 As shown, step 103 includes steps 31 to 33.

[0063] Step 31: Based on the neural feature vector and the immune feature vector, construct a joint energy function and optimize the joint energy function to obtain neural-immune coupling features.

[0064] To avoid simple feature splicing, this application introduces a joint energy function to uniformly constrain multimodal information, which takes the form of: ;in, This represents the total energy of the neuro-immune system, indicating the degree to which the neuro-immune system deviates from homeostasis. The lower the energy, the closer the system is to a healthy homeostasis. This is the energy term for neural activity states, representing the functional state of the nervous system; This is an energy term representing changes in immune status, indicating the degree of activation of the immune system; This is a coupling constraint term between the nervous system and the immune system, representing the strength of the interaction between the two systems. All of these are adaptively adjusted weight parameters. , Weight parameters representing the state of neural activity. Weighted parameters representing immune status, The weighting parameters represent the neural-immune coupling.

[0065] Energy Items of Neural Activity State This characterizes the degree of deviation between the current neural state and a healthy state; the greater the deviation, the higher the energy. The formula is: ;in, The neural feature vector; This represents the mean vector of neural characteristics in healthy individuals. Let be the covariance matrix of the neural feature vectors; Represents Mahalanobis distance, ; For regularization terms of neural feature vectors, such as L1 norm sparsity constraints or graph Laplacian smoothing constraints; This is the first regularization coefficient.

[0066] Energy Term of Changes in Immune Status It represents the degree of deviation between the current immune status and the healthy status. The higher the inflammation level, the higher the energy. The formula is: ;in, This is an immune feature vector, including the proportion of immune cells, the concentration of inflammatory factors, and immune metabolic indicators; This represents the mean vector of immune characteristics in healthy individuals. Let be the covariance matrix of the immune feature vectors; This is the regularization term for the immune feature vector. This is the second regularization coefficient.

[0067] Coupling constraints between the nervous system and the immune system The mutual predictive consistency between neural and immune states is ensured through bidirectional constraints, and the formula is as follows: ;in, This is a neural-immune coupling matrix, with elements... Indicates the first The brain region and the first The coupling strength of several immune indicators; for The transpose of .

[0068] Neural-immune coupling matrix Learn through the following methods: ;in, Denotes the Frobenius norm. The L1 norm indicates that sparsity is promoted. The denominator represents the sparsity penalty coefficient. This optimization problem can be solved using either the alternating direction multiplier method or the proximal gradient method.

[0069] By optimizing the joint energy function, the neuro-immune coupling characteristics were obtained. : That is, in the neural feature vector Immune feature vectors and neuro-immune coupling matrix Under the common constraints and influences, by searching for ways to increase the total energy of the neuro-immune system... The parameter combinations that reach their minimum values ​​are ultimately determined to identify the core features that characterize the coupling between the nervous and immune systems, serving as neuroimmune coupling features. .

[0070] Under the aforementioned energy constraints, multimodal features are mapped to the same pathological representation space, enabling changes in neural signals and immune status to have a consistent explanation at the pathological level.

[0071] Step 32: Concatenate the neural feature vector, the immune feature vector, and the neuroimmune coupling feature into a joint feature vector: .in, Joint eigenvectors; This indicates matrix vectorization operations.

[0072] Step 33: Map the joint feature vector to a low-dimensional pathological representation space using a deep autoencoder to obtain a unified pathological representation, as shown in the formula: ; .in, To standardize the representation of pathology, For encoders (such as multilayer perceptrons or variational autoencoders). For encoder parameters, For decoder, These are the parameters for the decoder. To standardize the representation of pathology, For pathological representation dimensions (usually) ).

[0073] The encoder and decoder parameters are optimized by minimizing the reconstruction error. What I learned: To make pathological representations discriminative, a supervised contrastive learning loss is introduced: ;in, For cosine similarity, For temperature parameters, For anchor point samples, As a positive sample, This is a negative sample.

[0074] Unified Pathological Representation Obtained through joint optimization: ;in, and This is the balance coefficient.

[0075] In another specific application example, the joint energy function can be replaced by a joint probability distribution, an information bottleneck, or an adversarial generative network framework. The unified pathological representation can be replaced by a graph neural network representation or a knowledge graph embedding. Furthermore, to enhance the interpretability of the mechanism, causal inference models can be introduced instead of correlation analysis.

[0076] Step 104: Constrain the unified pathological representation within the pathological manifold space, so that the pathological state of the subject evolves continuously over time in the pathological manifold space, in order to determine the manifold pathological representation at each moment within a set time period.

[0077] In a specific application example, the process of determining the pathological manifold space includes: (1) Use manifold learning algorithms (such as t-SNE, UMAP, or diffusion mapping) to learn low-dimensional manifold structures from high-dimensional pathological representation data: ;in, The pathological representation of the training samples, For manifold dimension (usually) or For visualization, (Used for modeling). The core assumption of manifold learning is that high-dimensional data is distributed on a low-dimensional Riemannian manifold, and the geodesic distance between data points reflects the similarity of pathological states.

[0078] (2) Learn the local coordinate graph and transition function of the manifold: for each point on the manifold There exists a neighborhood and mapping , making This is a homeomorphism. Globally consistent low-dimensional embeddings are constructed using locally linear embeddings or isometric mappings: ;in, This is a manifold embedding function.

[0079] This application uses the formula The unified pathological representation is constrained within the pathological manifold space; wherein... For a unified pathological representation within the pathological manifold space, To standardize the representation of pathology, For pathological manifold space, This represents the pathological state within the pathological manifold space. The projection process can be solved approximately through iterative optimization or a neural network.

[0080] Furthermore, a manifold regularization term is introduced during the pathological representation learning process: ;in, For similarity weights, based on geodesic distance on the manifold or Nearest neighbor graph calculation: , The geodesic distance on the manifold, , For sample indexing. During the offline training phase of the model: and All samples are from the training dataset. This regularization term is minimized to constrain the unified pathological representation. Maintaining in the learned pathological manifold space Above. In the actual diagnostic and treatment application stage (online inference): this formula is used to constrain the projection of the pathological state of new patients. At this time, This represents a unified pathological representation of the current patient, collected and mapped in real time. Representative manifold reference database and The most recent historical samples. In application, by calculating and minimizing this term (or using it as a constraint), the current patient's pathological state is "corrected" to a manifold surface that conforms to the pathological patterns of the population, thereby eliminating noise interference and ensuring that diagnostic and treatment decisions are generated based on reliable pathological evolution paths.

[0081] The temporal evolution of pathological states on the manifold is modeled using stochastic differential equations: ;in, Let be the drift function, describing the deterministic evolution trend; Input for therapeutic intervention; Let be the diffusion function, describing random fluctuations; This is a Wiener process (Brownian motion). This manifold is used to describe disease progression, remission, and fluctuations, avoiding the discretization of disease states into static categories.

[0082] In another specific application example, the low-dimensional pathological manifold space can be replaced by a dynamic system model, stochastic differential equations, or Markov decision processes. Furthermore, it can be combined with prior medical knowledge to construct structured manifolds, rather than purely data-driven methods.

[0083] Step 105: Based on the manifold pathological representation at each time point within the set time period, the side effects of various intervention measures, and the target pathological state, construct a diagnosis and treatment decision optimization model, and generate diagnosis and treatment suggestions based on the diagnosis and treatment decision optimization model.

[0084] In a specific application example, the diagnostic and treatment decision optimization model minimizes the objective function. Generate treatment recommendations; among them, The objective function value, For disease risk assessment items, For safety constraints of intervention measures, This refers to the degree of deviation of the pathological state from the target pathological state. The weighting parameters for disease risk assessment items. For the weighting parameters of the safety constraints of the intervention measures, This is a weighted parameter for the degree of deviation of the pathological state from the target pathological state.

[0085] Disease risk assessment items The formula for assessing the risk of disease progression or activity corresponding to the current pathological state is as follows: ;in, For risk prediction model parameters; For feature transformation functions (such as polynomial basis or neural network); for t Time-bound manifold pathological representation; The length of the historical time window; For moments within a historical time window, ; These are time-dependent weights.

[0086] Safety constraints of intervention measures The safety of treatment decisions, ensuring that side effects are within acceptable limits, is governed by the following formula: ;in, The number of interventions (e.g., types of drugs, neuromodulation parameters); For the first The probability of a intervention causing side effects; For the first The dosage (or intensity) of the intervention; For the first The severity of side effects of an intervention is a function of dose, typically a monotonically increasing function. For the first The fixed cost (or risk baseline) of an intervention.

[0087] The degree of deviation of the pathological state from the target pathological state. The formula for driving the pathological state to converge to the target state is: ;in, For the target pathological state (such as a healthy state or a treatment target state), predefine it in the manifold space or set it through expert knowledge; This is a weight matrix, including deviation weights for different pathological dimensions; For weighted Euclidean distance, .

[0088] Weight parameters in the objective function , , It needs to be dynamically adjusted based on the individual characteristics of the subjects. The adjustment method is as follows: (1) Static adjustment based on patient risk stratification. Patients are divided into different risk levels according to their baseline characteristics (age, disease duration, comorbidities, etc.), with each level corresponding to a preset weight configuration: High-risk patients (acute phase, rapid progression): (Prioritize risk control); Intermediate-risk patients (stable phase): (Balancing risk, safety, and goals); Low-risk patients (remission phase, maintenance therapy): (Prioritize security and target maintenance).

[0089] (2) Dynamic adjustment based on treatment response. Weights are dynamically adjusted during treatment based on the actual response: ;in, For learning rate, Through automatic differential calculation .

[0090] Specifically, if disease activity continues to rise ( Increase), increase Enhance risk control; if significant side effects occur ( Increase), increase Strengthen safety constraints; if approaching the treatment goal ( Decrease), Increase Strengthen goal orientation.

[0091] (3) Personalized adjustment based on preference learning. Personalized weights are inferred by learning the historical decision-making preferences of patients or doctors: given a historical decision-making dataset ,in, For the patient's condition, For expert-selected decisions, the weights are learned through maximum likelihood estimation: This optimization problem is solved using inverse reinforcement learning, a method within reinforcement learning.

[0092] In another specific application example, the objective function can be replaced with a multi-objective optimization framework, a reinforcement learning reward function, or a Bayesian optimization objective.

[0093] In another exemplary embodiment, the method for generating diagnostic and treatment recommendations for neuroimmunological diseases further includes the step 106.

[0094] Step 106: After intervening with the subject according to the treatment recommendations, collect the multimodal brain-computer interface data of the subject after the intervention, and dynamically update the weight parameters in the manifold pathological representation, the joint energy function, and the treatment decision optimization model based on the multimodal brain-computer interface data after the intervention.

[0095] Specifically, multimodal data after intervention is continuously collected through a brain-computer interface to represent the pathological state. Weighting parameters of the joint energy function Weight parameters of the diagnosis and treatment decision optimization model Dynamic updates are performed to form a closed-loop optimization process of perception, modeling, decision-making, and feedback.

[0096] The update rules are as follows: ; ; .in, For the update function, For the multimodal data after intervention, For time intervals, For weight parameters learning rate, joint energy function For weight parameters gradient, To optimize the model weight parameters for diagnosis and treatment decisions learning rate, Indicates projection onto the feasible region , .

[0097] like Figure 4 As shown, the overall diagnostic and treatment decision optimization process includes: multimodal data acquisition, pathological status assessment, treatment recommendation generation, intervention execution, feedback data acquisition, and parameter updating. The pathological status assessment process includes neural status, immune status, coupling analysis, disease severity, and risk stratification. The treatment recommendation generation process includes diagnostic classification, treatment selection, dosage optimization, intervention timing, and prognostic prediction, while incorporating a decision-making knowledge system (internally storing a knowledge base, clinical guidelines, expert system, risk assessment process, and prognostic prediction process). The intervention execution process includes neuromodulation, drug delivery, cognitive training, and immune regulation. Feedback data acquisition includes neural response, immune response, clinical outcomes, and adverse events. The parameter updating process includes online learning, adaptive adjustment, knowledge updates, and personalized model adjustments.

[0098] Furthermore, the closed-loop optimization mechanism can also employ a federated learning architecture to achieve multi-center data sharing and model optimization while protecting privacy. Digital twin technology can also be introduced to construct patient-specific virtual models for treatment plan simulation and optimization.

[0099] Taking multiple sclerosis patients as an example, the processing procedure of this application is described as follows: ① Collect EEG, heart rate variability, inflammatory factors and motor behavior data; ② Construct a unified pathological representation through multimodal spatiotemporal alignment; ③ Identify the state of inflammation-driven neural function inhibition in the pathological manifold space; ④ Automatically generate joint decision-making for immune intervention and neural regulation; ⑤ Optimize subsequent decisions through brain-computer interface feedback.

[0100] Compared with existing technologies, this application has the following advantages: it realizes the underlying unified modeling of neuro-immune multimodal data; it solves the problem of inconsistent time scales of different physiological systems; it upgrades from "disease classification" to "continuous modeling of pathological states"; it realizes closed-loop, adaptive optimization of diagnosis and treatment decisions; and it has high versatility and scalability.

[0101] This application does not limit the specific signal type, algorithm model or hardware implementation method. Any solution based on the same technical idea and implemented using equivalent or equivalent technical means should fall within the protection scope of this invention.

[0102] Based on the same inventive concept, this application also provides a neuroimmune disease diagnosis and treatment suggestion generation device for implementing the above-mentioned method for generating diagnosis and treatment suggestions for neuroimmune diseases. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the neuroimmune disease diagnosis and treatment suggestion generation device provided below can be found in the limitations of the neuroimmune disease diagnosis and treatment suggestion generation method described above, and will not be repeated here.

[0103] In one exemplary embodiment, such as Figure 5 As shown, a device for generating diagnostic and treatment suggestions for neuroimmunological diseases is provided, comprising: a multimodal data acquisition module 501, a spatiotemporal alignment module 502, a joint optimization module 503, a manifold evolution module 504, a suggestion generation module 505, and a feedback optimization module 506.

[0104] The multimodal data acquisition module 501 is used to acquire multimodal brain-computer interface data of the subjects.

[0105] In a specific application example, such as Figure 6As shown, the multimodal data acquisition module 501 includes a central nervous system activity acquisition unit, a peripheral immune data acquisition unit, an autonomic nervous system and physiological data acquisition unit, and a behavioral and functional data acquisition unit. The neuroimmunological disease diagnosis and treatment suggestion generation device also includes a data storage and management module and a display and output module, which are connected. The central nervous system activity acquisition unit, peripheral immune data acquisition unit, autonomic nervous system and physiological data acquisition unit, behavioral and functional data acquisition unit, and data storage and management module are all connected to the central processing unit via a data bus and synchronization module. The spatiotemporal alignment module 502, joint optimization module 503, manifold evolution module 504, suggestion generation module 505, and feedback optimization module 506 are all deployed in the central processing unit.

[0106] (1) The central nervous system activity acquisition unit includes an electroencephalogram (EEG) sensor, a magnetoencephalography (MEG) sensor, a functional near-infrared spectroscopy (fNIRS) sensor, and a neural stimulation-evoked response acquisition component, used to capture dynamic activity signals of the central nervous system. The dynamic activity signals of the central nervous system include: EEG signals (reflecting the electrical activity of neuronal clusters), magnetoencephalography (reflecting the magnetic field generated by neuronal synaptic currents), functional near-infrared signals (reflecting changes in cerebral blood oxygen metabolism), and neural stimulation-evoked response signals (reflecting the excitability of neural pathways).

[0107] The specific hardware structure is as follows: The EEG sensor uses a dry or wet electrode array, which contacts the scalp via conductive gel or capacitive coupling. The electrode material is Ag / AgCl, with an electrode diameter of 8mm~10mm. The electrode spacing follows the international 10-20 or 10-10 system extended layout. After the signal is amplified by a preamplifier (input impedance ≥100MΩ, common-mode rejection ratio ≥80dB), it is converted into a digital signal by a 24-bit analog-to-digital converter. The sampling rate is set to 1000Hz~2000Hz, the bandwidth to 0.5Hz~100Hz, and the number of channels to 32~128.

[0108] The magnetoencephalogram (MEG) sensor employs a superconducting quantum interference device (SQUID) magnetometer or an optically pumped atomic magnetometer (OPM). The sensor array covers the entire brain region, with a sensor spacing of approximately 20 nm to 30 mm. MEG signals are acquired in a magnetically shielded chamber (shielding coefficient ≥ 60 dB), with a magnetic field resolution of 0.1 fT / √Hz to 1 fT / √Hz and a sampling rate ≥ 1000 Hz.

[0109] The functional near-infrared sensor uses a dual-wavelength (760nm and 850nm) or triple-wavelength (730nm, 805nm, 850nm) LED light source, with a light source-detector spacing of 3cm to 4cm, covering target brain regions such as the prefrontal cortex and motor cortex. The photodetector is a silicon photodiode or avalanche photodiode (APD), with a sampling rate of 1Hz to 10Hz and a light intensity detection accuracy of ±1%.

[0110] The neural stimulation-evoked response acquisition component includes a transcranial magnetic stimulation (TMS) coil or transcranial direct current stimulation (tDCS) electrode, and a synchronous trigger acquisition module. TMS stimulation parameters: stimulation intensity is 80%–120% of the resting motor threshold (RMT), and stimulation frequency is adjustable from 0.1 Hz to 10 Hz. tDCS stimulation parameters: current intensity is 1 mA–2 mA, and stimulation time is 10–30 minutes. Evoked responses are recorded synchronously via EEG or EMG.

[0111] (2) Peripheral immune data acquisition unit: integrates microfluidic chip and immune detection module, supports dynamic monitoring of immune cell ratio (detection accuracy ±1%), inflammatory factors (detection limit 1pg / mL), and immune metabolic indicators (detection error ±3%), with a configurable sampling interval of 0.5 hours to 24 hours.

[0112] The specific hardware structure and connection relationships are as follows: The microfluidic chip is fabricated using polydimethylsiloxane (PDMS) or polymethyl methacrylate (PMMA) materials, and integrates sample pretreatment, reaction, and detection zones. The chip channel width ranges from 100 μm to 500 μm, the depth from 50 μm to 200 μm, and the required sample volume is 10 μL to 100 μL. The microfluidic chip drives sample flow via a micropump (flow rate accuracy ±1%), enabling automated sample dilution, mixing, and reaction.

[0113] The immunoassay module comprises an optical detection unit and an electrochemical detection unit. The optical detection unit employs fluorescence immunoassay or chemiluminescence immunoassay, with an LED or laser diode (wavelength 450nm~650nm) as the excitation source and a photomultiplier tube (PMT) or CCD camera as the detector. It detects fluorescent labels or chemiluminescent signals to achieve quantitative detection of inflammatory factors (such as IL-6, TNF-α, IL-10, etc.). The electrochemical detection unit employs electrochemical impedance spectroscopy (EIS) or amperometrics. The working electrode is a gold or carbon electrode, the reference electrode is an Ag / AgCl electrode, and the counter electrode is a platinum electrode. It is used to detect the proportion of immune cells (such as the CD4+ / CD8+ T cell ratio) and immunometabolic indicators (such as lactate, glucose, ATP, etc.).

[0114] The sample inlet of the microfluidic chip is connected to a blood collection device or body fluid collection device via microtubes (0.5mm~1mm inner diameter). The reaction zone of the microfluidic chip is pre-immobilized with capture antibodies or probe molecules, and the detection zone is aligned with the optical detection unit or electrochemical detection unit. The microfluidic chip maintains the reaction temperature through a temperature control module (temperature control accuracy ±0.5℃), and the detection results are transmitted to the central processing unit via a data acquisition card.

[0115] The proportion of immune cells, inflammatory factors, and immune metabolic indicators are all collected through the immune detection module: the proportion of immune cells is calculated by the impedance change of the electrochemical detection unit or the fluorescence intensity of the optical detection unit; inflammatory factors are quantified by the fluorescence intensity or chemiluminescence intensity of the optical detection unit; and immune metabolic indicators are obtained by the current response or impedance spectroscopy analysis of the electrochemical detection unit.

[0116] (3) The autonomic nervous and physiological data acquisition unit includes a heart rate variability (HRV) sensor, a Galvanic skin response (GSR) sensor, and a respiratory rhythm sensor to collect autonomic nervous related physiological signals in real time.

[0117] The specific hardware structure is as follows: The heart rate variability sensor uses a photoplethysmography (PPG) sensor or an electrocardiogram (ECG) electrode. The PPG sensor includes a green LED (wavelength 525nm or 565nm) and a photodiode, worn on the finger or wrist, with a sampling rate of 50Hz to 200Hz and a pulse wave detection accuracy of ±1 bpm. The ECG electrode uses Ag / AgCl electrodes, configured with three or five leads, with a sampling rate of 250Hz to 1000Hz and a bandwidth of 0.05Hz to 100Hz.

[0118] The skin conductance sensor uses a dual-electrode configuration with Ag / AgCl electrode material and an electrode area of ​​1 cm². 2 ~2cm 2 The electrodes are spaced 2cm to 3cm apart and worn on the palm or fingers. The sensor is excited by a constant voltage source (0.5V) to measure changes in skin conductivity, with a range of 1μS to 50μS, a resolution of 0.01μS, and a sampling rate of 10Hz to 50Hz.

[0119] The respiratory rhythm sensor employs a chest strap-type respiratory sensing band (piezoelectric or strain gauge type) or a nasal airflow thermistor. The chest strap-type sensor is worn around the chest to detect the amplitude of chest expansion, with a sampling rate of 10Hz~50Hz and an amplitude resolution of ±1mm. The nasal airflow sensor is placed below the nostrils to detect changes in the temperature of the respiratory airflow, with a sampling rate of 10Hz~50Hz.

[0120] (4) The behavior and function data acquisition unit acquires motion trajectory, posture, and task performance data through motion capture components (spatial resolution ±1mm, sampling rate 30fps~100fps) and the cognitive language task interaction module. Among them, the motion trajectory is obtained through the acceleration integral of the IMU system or the marked spherical coordinate sequence of the optical motion capture system, and the output is three-dimensional spatial coordinates with a time resolution of 10ms~33ms. The posture is obtained through the posture calculation of the IMU system (quaternion or Euler angle representation) or the rigid body registration of the optical motion capture system, and the output is the angle of each joint or the direction vector of the body segment. The task performance score is automatically calculated by the software system of the cognitive language task interaction module. The cognitive task score is calculated based on reaction time and accuracy (such as d-prime value, reaction time coefficient of variation), and the language task score is calculated based on speech recognition results and semantic analysis (such as the number of correct names, semantic fluency score).

[0121] The specific hardware structure, setup location, and data acquisition method are as follows: The motion capture component employs an Inertial Measurement Unit (IMU) system or an optical motion capture system. The IMU system comprises 6–17 wireless IMU nodes, each integrating a triaxial accelerometer (range ±16g, 16-bit resolution), a triaxial gyroscope (range ±2000° / s, 16-bit resolution), and a triaxial magnetometer (range ±4800μT, 16-bit resolution). The IMU nodes are secured to various parts of the subject's body using elastic straps: head (forehead or top of head), trunk (sternum or lumbar vertebrae), upper limbs (upper arm, forearm, back of hand), and lower limbs (thigh, calf, dorsum of foot). The sampling rate is 60Hz–100Hz, and data transmission is via Bluetooth or a 2.4GHz wireless protocol with a latency of <20ms. Sensor fusion algorithms (such as Kalman filtering or complementary filtering) are used to calculate joint angles and motion trajectories. The optical motion capture system comprises 6-12 infrared cameras (resolution ≥2MP, frame rate ≥100fps), arranged around the perimeter of the acquisition space to form a acquisition area with a coverage of ≥3m×3m×2m. Reflective marker balls (10mm-15mm in diameter) are affixed to the subject's body surface. The positions of the marker balls are calculated using multi-camera triangulation, with a spatial accuracy of ±1mm. The marker ball placement locations include: head (4-6), trunk (3-5), upper limbs (6-8 per side), and lower limbs (6-8 per side).

[0122] The cognitive-language task interaction module includes a touchscreen display (≥15 inches, ≥1920×1080 resolution), a microphone array (16kHz sampling rate, 16-bit bit depth), and response buttons. Participants complete cognitive tasks (such as the Stroop task, N-back task, and Trail Making Test) via the touchscreen and language tasks (such as naming tests and fluency tests) via the microphone. Task performance data includes indicators such as reaction time (accuracy ±1ms), accuracy, and completion time.

[0123] In another specific application example, central nervous system activity modalities can also utilize intracranial electroencephalography (EEG), functional magnetic resonance imaging (fMRI), or positron emission tomography (PET). Peripheral immune modalities can further employ single-cell sequencing data, T-cell receptor library analysis, or metabolomics data; gut microbiome data can also be added as a supplementary modal to reflect the interactions of the gut-brain-immune axis.

[0124] The spatiotemporal alignment module 502 is used to perform spatiotemporal alignment and reparameterization processing on the multimodal brain-computer interface data to obtain multimodal spatiotemporal aligned data.

[0125] The joint optimization module 503 is used to perform joint optimization on the multimodal spatiotemporal aligned data based on the joint energy function and the deep autoencoder to obtain a unified pathological representation.

[0126] The manifold evolution module 504 is used to constrain the unified pathological representation within the pathological manifold space, so that the pathological state of the subject evolves continuously over time in the pathological manifold space, in order to determine the manifold pathological representation at each moment within a set time period.

[0127] The suggestion generation module 505 is used to construct a diagnosis and treatment decision optimization model based on the manifold pathological representation at each time point within a set time period, the side effects of various intervention measures, and the target pathological state, and to generate diagnosis and treatment suggestions based on the diagnosis and treatment decision optimization model.

[0128] Specifically, it is recommended that the generation module 505 be equipped with a diagnosis and treatment decision engine, which supports objective function optimization calculation and closed-loop adaptive update, with a decision output response time of ≤500ms, and feedback forms including visual reports and parameter command output (such as neuromodulation parameters), and supports linkage with clinical diagnosis and treatment systems and home monitoring devices.

[0129] The feedback optimization module 506 is used to collect multimodal brain-computer interface data of the subject after intervention according to the treatment recommendations, and dynamically update the weight parameters in the manifold pathology representation, the joint energy function, and the treatment decision optimization model based on the multimodal brain-computer interface data after intervention.

[0130] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0131] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0132] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0133] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0134] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0135] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, etc., and are not limited to these.

[0136] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0137] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for generating diagnostic and treatment recommendations for neuroimmunological diseases, characterized in that, The method for generating diagnostic and treatment recommendations for neuroimmunological diseases includes: Acquire multimodal brain-computer interface data from subjects; The multimodal brain-computer interface data is spatiotemporally aligned and reparameterized to obtain multimodal spatiotemporally aligned data; Based on the joint energy function and deep autoencoder, the multimodal spatiotemporal aligned data is jointly optimized to obtain a unified pathological representation; The unified pathological representation is constrained within the pathological manifold space, so that the pathological state of the subject evolves continuously over time in the pathological manifold space, thereby determining the manifold pathological representation at each moment within a set time period. Based on the manifold pathological representation at each time point within a set time period, the side effects of various interventions, and the target pathological state, a diagnostic and treatment decision optimization model is constructed, and diagnostic and treatment suggestions are generated based on the optimized model.

2. The method for generating diagnostic and treatment recommendations for neuroimmunological diseases according to claim 1, characterized in that, The multimodal brain-computer interface data is spatiotemporally aligned and reparameterized to obtain multimodal spatiotemporally aligned data, including: For any two modalities of the multimodal brain-computer interface data, the physiological response delay between the two modalities is determined, and the brain-computer interface data of the two modalities is delayed based on the physiological response delay to obtain global compensation data for the two modalities. Dynamic time warp mapping is performed on the global compensation data of the two modes to obtain the optimal time alignment path between the two modes; Based on the optimal time alignment path between the two modes, an adaptive time mapping function is constructed through Gaussian process regression to perform nonlinear time mapping on the global compensation data of the two modes in order to obtain multimodal time alignment data; Construct a multi-scale time window set, and based on the multi-scale time window set, perform cross-scale joint modeling on the time-aligned data of each modality to obtain the cross-scale joint data of each modality; The cross-scale joint data of each modality were mapped to the standard neural function space and pathological semantic labels were introduced to obtain multimodal spatiotemporally aligned data.

3. The method for generating diagnostic and treatment recommendations for neuroimmunological diseases according to claim 1, characterized in that, The multimodal spatiotemporal aligned data includes neural feature vectors and immune feature vectors; based on a joint energy function and a deep autoencoder, the multimodal spatiotemporal aligned data is jointly optimized to obtain a unified pathological representation, including: Based on the neural feature vector and the immune feature vector, a joint energy function is constructed and optimized to obtain neural-immune coupling features; The neural feature vector, the immune feature vector, and the neuroimmune coupling feature are concatenated into a joint feature vector; The joint feature vector is mapped to a low-dimensional pathological representation space by a deep autoencoder to obtain a unified pathological representation.

4. The method for generating diagnostic and treatment recommendations for neuroimmunological diseases according to claim 3, characterized in that, The joint energy function is: ； ； ； ； in, This represents the total energy of the neuro-immune system, indicating the degree to which the neuro-immune system deviates from homeostasis. This is the energy term representing the state of neural activity, indicating the functional state of the nervous system. For neural feature vectors, Let be the mean vector of neural features of healthy individuals. Let be the covariance matrix of the neural feature vectors. Represents Mahalanobis distance, For the regularization term of the neural feature vector, The first regularization coefficient is... This is an energy term representing changes in immune status, indicating the degree of activation of the immune system. immune feature vector This represents the mean vector of immune characteristics in healthy individuals. Let be the covariance matrix of the immune feature vectors. For the regularization term of the immune feature vector, This is the second regularization coefficient. This represents the coupling constraint term between the nervous and immune systems, indicating the strength of their interaction. This is a neural-immune coupling matrix. All of these are adaptively adjusted weight parameters. , Weight parameters representing the state of neural activity. Weighted parameters representing immune status, The weighting parameters represent the neural-immune coupling.

5. The method for generating diagnostic and treatment recommendations for neuroimmunological diseases according to claim 4, characterized in that, The neuroimmune coupling characteristics are obtained by optimizing the joint energy function using the following formula: ； in, This is a characteristic of neuroimmune coupling.

6. The method for generating diagnostic and treatment recommendations for neuroimmunological diseases according to claim 1, characterized in that, The unified pathological representation is constrained within the pathological manifold space using the following formula: ； in, For a unified pathological representation within the pathological manifold space, To standardize the representation of pathology, For pathological manifold space, This refers to the pathological state within the pathological manifold space.

7. The method for generating diagnostic and treatment recommendations for neuroimmunological diseases according to claim 1, characterized in that, The diagnostic and treatment decision optimization model generates diagnostic and treatment recommendations by minimizing the following objective function: ； ； ； ； in, The objective function value, For disease risk assessment items, For risk prediction model parameters, For the characteristic transformation function, for t Time-dependent manifold pathology representation, The length of the historical time window, For moments within a historical time window, , For time-dependent weights, For safety constraints of intervention measures, For the number of interventions, For the first The probability of side effects from an intervention measure. For the first The dosage of the intervention, For the first The severity of side effects of an intervention measure is a function. For the first The fixed costs of an intervention measure This refers to the degree of deviation of the pathological state from the target pathological state. For the target pathological state, This is a weight matrix, including deviation weights for different pathological dimensions. For weighted Euclidean distance, The weighting parameters for disease risk assessment items. For the weighting parameters of the safety constraints of the intervention measures, This is a weighted parameter for the degree of deviation of the pathological state from the target pathological state. , and The adjustments are made dynamically based on the individual characteristics of the subjects.

8. The method for generating diagnostic and treatment recommendations for neuroimmunological diseases according to claim 1, characterized in that, The method for generating diagnostic and treatment recommendations for neuroimmunological diseases also includes: After intervening in the subjects according to the treatment recommendations, multimodal brain-computer interface data of the subjects after the intervention is collected, and the weight parameters in the manifold pathology representation, the joint energy function, and the treatment decision optimization model are dynamically updated based on the multimodal brain-computer interface data after the intervention.

9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method for generating diagnostic and treatment recommendations for neuroimmunological diseases according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for generating diagnostic and treatment recommendations for neuroimmunological diseases as described in any one of claims 1-8.

Images