Multi-source data health intervention method

Abstract

The present application relates to the field of medical information technology, and more particularly to a multi-source data health intervention method, which comprises the following steps: first, collecting text, vital signs, image vectors and wearable signals, generating a unified vector through multi-modal encoding and extracting a topological lifetime vector; constructing a pulse neural causal graph based on the unified vector, injecting the lifetime vector into the neural ordinary differential equation as an external field to generate a dynamic pathological manifold; then compressing the manifold into a tensor network state, constructing a quantum optimization model in combination with the causal graph, minimizing the energy expectation and topological risk to obtain an optimal intervention sequence, and injecting the target gradient back into the manifold to adjust it in real time; after the sequence is verified by clinical logic, a commitment value and zero-knowledge proof are generated and written into a Byzantine account book, and a control instruction is issued to an execution terminal. The method realizes millisecond-level decision-making, reduces the false alarm rate and provides on-chain traceability.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical information technology, and in particular to a multi-source data health intervention method. BACKGROUND

[0002] Through continuous prediction and immediate adjustment of the real-time state of the patient, health intervention can actively intervene before the disease worsens, significantly reducing the rate of acute and critical events, reducing the burden on medical staff and improving the quality of chronic disease management. However, the current scheme often uses (1) a single mode cycle network combined with a static threshold rule, which can only capture local time characteristics and lacks cross-modal correlation; (2) simply splicing multi-modal features and selecting actions using greedy or heuristic strategies, without considering the timing topology and global risk, resulting in high false positives and delayed intervention responses; (3) the centralized log recording method lacks tamper-proofing ability and is difficult to meet clinical compliance audits. The above methods lack cross-scale topological stability measurement, have single optimization targets and are not traceable, making it difficult to operate stably in high-noise, low-latency scenarios. SUMMARY

[0003] In view of the many problems existing in the prior art, the present application provides a multi-source data health intervention method, which encodes multi-source health data into a unified vector and extracts a topological lifetime vector to constrain a pulse neural causal graph; generates a dynamic pathological manifold in a neural ordinary differential equation, obtains an optimal intervention sequence through tensor network compression and quantum optimization, and updates the manifold in real time through gradient feedback; the sequence is verified by clinical rules, stored on the chain and drives the terminal, and finally realizes personalized intervention with low false positives, high timeliness and traceability.

[0004] A multi-source data health intervention method, the method comprising:

[0005] Collecting text data, vital sign data, medical image feature vectors and wearable sensor data, performing time synchronization, filtering and standardization processing on all data, and obtaining a unified representation vector through multi-modal encoding processing;

[0006] The unified representation vector sequence is used to extract a topological lifetime vector, a pulse neural causal graph is constructed based on the unified representation vector, and the topological lifetime vector is introduced as a modulation term into a neural ordinary differential equation, and a dynamic pathological manifold is generated by solving the neural ordinary differential equation framework;

[0007] The dynamic pathological manifold is compressed into a tensor network state, a quantum optimization model is established for candidate intervention actions according to the pulse neural causal graph, an optimal intervention sequence is obtained by jointly optimizing the objective function, and the gradient of the joint optimization objective function with respect to the topological lifetime vector is fed back to the neural ordinary differential equation, so that the dynamic pathological manifold is updated in real time with the optimal intervention sequence and forms a public health intervention model;

[0008] The optimal intervention sequence is verified according to a clinical logic rule, and after verification, the optimal intervention sequence is recorded to a distributed ledger, and a control instruction is generated according to the optimal intervention sequence and sent to an execution terminal.

[0009] Preferably, the multi-modal encoding processing adopts a holographic reduction representation, and binding operation and cyclic convolution operation are respectively performed on text data, vital sign data, medical image feature vectors and wearable sensor data to generate a unified representation vector.

[0010] Preferably, when the topological lifetime vector is extracted, a Vietoris-Rips complex is constructed on a sequence of unified representation vectors with a fixed frame length, birth time and death time are calculated, and a lifetime value vector is formed according to the difference between the two to serve as the topological lifetime vector.

[0011] Preferably, when the pulse neural causal graph is constructed, a pulse neural network containing voltage discharge integral firing neurons is established, synaptic weight values are updated using a time difference dependent plasticity rule, and the synaptic weight values are regularized and adjusted using the gradient of the topological lifetime vector.

[0012] Preferably, when the topological lifetime vector is introduced as a modulation term into the neural ordinary differential equation, an external field coefficient is added in the vector field and a self-adaptive step numerical integration method is used for solving to generate a dynamic pathological manifold.

[0013] Preferably, when the dynamic pathological manifold is compressed into a tensor network state, a matrix product representation is obtained through tensor rearrangement and variational singular value decomposition.

[0014] Preferably, when the quantum optimization model is established, the weight values of the pulse neural causal graph are mapped to the Hamiltonian coefficient corresponding to the candidate intervention action, and the energy expectation value and the topological risk value are linearly combined to form a joint optimization objective function.

[0015] Preferably, when the optimal intervention sequence is obtained, the joint optimization objective function is solved using a tensor network virtual time evolution algorithm or a quantum annealing algorithm.

[0016] Preferably, when the optimal intervention sequence is verified, the optimal intervention sequence is converted into a fact set and input to a rule base based on an answer set procedure, and a verified intervention sequence is obtained through reasoning.

[0017] Preferably, when the optimal intervention sequence is recorded to the distributed ledger, a commitment value and a zero-knowledge proof are generated for the optimal intervention sequence, and then written into a chain ledger using a Byzantine fault-tolerant consensus mechanism.

[0018] Compared with the prior art, the advantages and beneficial effects of the present application are that:

[0019] The topological life vector is constructed by persistent homology, the cross-modal time-stable feature extraction is realized, the interpretable continuous health dynamics prediction is realized by coupling the pulse neural causal graph with the neural ordinary differential equation, the millisecond-level global optimal intervention decision is realized by tensor network compression combined with quantum optimization solution, the privacy protection and non-repudiation audit of the intervention process are realized by writing the Byzantine fault-tolerant ledger through zero-knowledge commitment, and the synchronous evolution of the public model and individual differences is realized through the gradient back-echo closed-loop adjustment. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 It is a flowchart of the method of the present application;

[0021] Figure 2 It is a schematic diagram of the quantum optimization model construction of the present application. DETAILED DESCRIPTION

[0022] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure.

[0023] As Figure 1 shown, a multi-source data health intervention method, the method comprising:

[0024] Collecting text data, vital sign data, medical image feature vectors and wearable sensor data, performing time synchronization, filtering and standardization processing on all data, and obtaining a unified representation vector through multi-modal encoding processing;

[0025] The acquisition stage is the only entrance for multi-source health information into the intervention link, and its design directly affects the accuracy and reliability of the subsequent topological-pulse causal modeling. The present application collects four types of original signals in this stage: text data, vital sign data, medical image feature vectors and wearable sensor data. For clarity, the implementation principle and expected effect are described in the following four aspects of "acquisition-alignment-cleaning-encoding", and necessary explanations are given for the first appearing terms.

[0026] The acquisition layer pulls text data in batches through the hospital information system interface, the text data includes medical order records and on-duty doctor-patient conversations; real-time acquisition of vital sign data such as electrocardiogram and blood oxygen saturation is achieved with the help of multi-channel physiological monitor; medical image feature vectors are extracted by convolutional visual transformer on the edge inference node; wearable sensor data comes from the inertial measurement unit and skin temperature sensor integrated in the wristband device. All acquisition nodes use hardware-enhanced network time protocol for time synchronization, ensuring that the cross-device clock error is controlled within microseconds, and ensuring the consistency of multi-modal timestamps from the source.

[0027] The main task of the alignment layer is to ensure that different modal sequences are strictly aligned on the same time axis. Specifically, the absolute time of each frame sample is subtracted from the acquisition window start time to obtain a relative time scale with zero time as the reference point. Then the high sampling rate signal is down-sampled by an integer multiple, and the low sampling rate signal is completed by linear interpolation, until all sequences reach a unified sampling frequency. The text data uses sentence-level trigger timestamps, and the medical image feature vector uses the frame inference completion time as the timestamp. Both are aligned with the high-frequency vital sign sequence through the above relative timestamp strategy. Offline verification shows that this strategy can control the inter-modal synchronization error within milliseconds in a 24-hour continuous acquisition scenario.

[0028] The cleaning layer first applies adaptive filtering to physiological signals. The algorithm estimates the noise characteristics using the reference channel, and then implements real-time denoising in the target channel. Then, zero-mean normalization is performed on all signals to obtain dimensionless sequences, avoiding interference from dimensional differences in subsequent cross-modal distance measurement. After the text data is cut by the sub-word marker, the word frequency is calculated, and the low-frequency sub-words are truncated to compress the vector dimension. The sequence length of different modalities is often inconsistent due to differences in sampling frequency. The invention obtains a uniform length in the time domain through fixed window clipping and repeated padding, so that it can be processed in parallel at one time in matrix operations.

[0029] The encoding layer uses a holographic reduction representation method to map the four types of modalities into a single high-dimensional vector. Let the text vector be , the vital sign vector be , the medical image vector be , and the wearable sensor vector be . To ensure reversible retrieval, the invention predefines a set of random phase bases of the same length for each type of modality, denoted as , , , . First, perform the binding operation, that is, multiply the modality vector and the corresponding phase base element by element to obtain the binding result:

[0030] ,

[0031] where the symbol represents the multiplication of the corresponding elements. Then input the four binding results into the cyclic convolution operator. To improve computational efficiency, the convolution process is completed in the frequency domain, and the final unified representation vector is given by:

[0032] ,

[0033] where represents the discrete Fourier transform, denotes its inverse operation. Experimental results show that under the same hardware conditions, the frequency domain convolution can control the single frame multi-modal encoding delay in the order of milliseconds, laying the foundation for real-time health intervention.

[0034] To ensure the reversibility of the encoding result, the application performs a debinding test on the uniformly represented vector after encoding: using the same phase base to perform inverse convolution and recovering the original modal vector through correlation operation, if the cosine similarity reaches the preset threshold, the encoding is valid; if the test fails for 3 times in a row, the system automatically generates a new phase base and re-encodes, avoiding information loss caused by random base collision.

[0035] Theoretically, the holographic reduced representation maintains the approximate orthogonality of each modal in high-dimensional space, thereby reducing the aliasing risk caused by the collapse of different modal distance distributions; at the same time, the binding and circular convolution operations can be micro, providing a continuous gradient path for subsequent topological lifetime vector extraction and pulse neural weight gradient propagation. Simulation results show that when 30% of the input segments are randomly deleted, the encoding method can reduce the subsequent pathological manifold reconstruction error by about 20%.

[0036] Embodiment: In a 48-hour continuous monitoring experiment, the system of the application real-time collects text records, electrocardio and blood oxygen sequence, bedside ultrasound frame features and inertial measurement unit data of 10 subjects. Through the output of the uniformly represented vector by the encoding layer and the input of the topological-pulse causal link, compared with the control scheme (without using the holographic reduced representation), in the early warning task of heart rate abnormalities, the average early warning time of the system of the application is increased to 180 seconds, indicating that the encoding layer improves the real-time and stability of the subsequent decision.

[0037] In summary, the application compresses multi-source health data into a uniformly represented vector with reversibility through five continuous links of unified collection, accurate alignment, noise suppression, scale standardization and holographic reduced representation. The vector has cross-modal orthogonality and gradient continuity, forming a consistent and efficient input basis for topological lifetime vector extraction, pulse neural causal modeling and quantum optimization intervention.

[0038] Preferably, the multi-modal encoding processing adopts the holographic reduced representation, and performs binding operation and circular convolution operation on text data, vital sign data, medical image feature vector and wearable sensor data respectively to generate a uniformly represented vector.

[0039] The multi-modal encoding process is a key hub in the health intervention link of the present application, and its function is to compress four types of heterogeneous original signals into a single high-dimensional vector, both preserving the characteristics of each mode and providing a unified coordinate system for subsequent topology-pulse causal reasoning. This process uses Holographic Reduced Representation (HRR) to complete two-level mapping of "binding operation-circular convolution operation". The principle and effect are as follows.

[0040] First, a phase base with consistent length and randomly generated is assigned to each mode. Let the text vector be , the vital sign vector be , the medical image vector be , and the wearable sensor vector be , and the phase bases be , , , respectively. The principle of binding operation is to use element multiplication to couple the mode content and the phase base to generate a bound vector that cannot be directly decomposed but can be decoded by convolution. The mathematical expression is:

[0041] ,

[0042] where represents element-wise multiplication by index. The obtained after binding still maintains the same length as the original mode, and can be restored separately by the unbinding operation when necessary.

[0043] The circular convolution operation uses discrete Fourier transform to realize efficient vector synthesis. The above four binding results are added in the frequency domain, and then inverse transformed back to the time domain to obtain the unified representation vector :

[0044] ,

[0045] where represents the discrete Fourier transform, represents the inverse transform. Because multiplication in the frequency domain corresponds to convolution in the time domain, and vice versa, the above formula completes the parallel calculation of four-way circular convolution through frequency domain addition, and the operation complexity is .

[0046] Through the present application, the phase base of each mode is unique, and the system can restore the original mode when needed The relevant operation is performed to reconstruct the specified modal vector, and supports subsequent personalized interpretation. The phase distribution of the bound different modalities is approximately independent, maintains high cosine separation in the vector space, and reduces cross-modal feature aliasing. Both binding and convolution are differentiable operations, providing a continuous path for gradient backpropagation to the original modalities. Frequency domain parallel acceleration keeps the single frame encoding delay at the millisecond level, meeting the low latency requirement of clinical intervention.

[0047] Embodiment: 48 hours of continuous data were collected from 8 patients in a real ward environment. Using the coding scheme of the present application, the average cosine reconstruction error was reduced by 0.18 compared with the baseline scheme without coding in the subsequent pathological manifold reconstruction task; in the early warning scene of arrhythmia, the average early warning time was improved to 170 seconds, proving that the coding strategy has a significant contribution to improving the intervention timeliness.

[0048] In summary, the multi-modal coding processing based on holographic reduced representation takes binding and cyclic convolution as the core, not only realizes efficient fusion of multi-source signals, but also lays a unified, highly reversible, differentiable and real-time feature basis for causal graph construction and quantum optimization intervention.

[0049] The unified representation vector sequence is used to extract a topological lifetime vector, a pulse neural causal graph is constructed based on the unified representation vector, the topological lifetime vector is introduced as a modulation term into a neural ordinary differential equation, and a dynamic pathological manifold is generated by solving the neural ordinary differential equation in the framework of the neural ordinary differential equation;

[0050] Mapping the unified representation vector sequence to a dynamic pathological manifold is the most innovative modeling link of the present application, which integrates three types of methods of statistical topology, pulse neural dynamics and differentiable ordinary differential equation, maintaining physical interpretability and supporting end-to-end training.

[0051] The unified representation vector sequence first enters the topological lifetime vector extraction module. This module uses persistent homology theory to measure the connected structure of the sequence in high-dimensional space. Specifically, a Vietoris-Rips complex is constructed on a sliding window of length 64 frames, and then the birth time and death time of 0-dimensional and 1-dimensional are calculated. The persistent lifetime refers to the survival interval of the same topological feature during the scale change, and the longer the lifetime, the more stable the feature. The present application sorts and splices all the lifetime values in the window to obtain a 128-dimensional topological lifetime vector as a compact representation of measuring the internal topological skeleton of the sequence. Medical signals often contain transient artifacts, such as spikes generated by electrode jitter; direct modeling on the original sequence is easy to misjudge, while the lifetime vector preferentially preserves the cross-scale commonality and can naturally ignore transient noise.

[0052] Subsequently, the system constructs a pulse neural causal graph based on the unified representation vector. The pulse neural network uses voltage leak integrate-and-fire neurons to simulate the accumulation and firing process of biological neuron membrane potential. Each neuron corresponds to a slice in the unified vector, and the synaptic weight is initialized as a small random value with a mean of zero. The weight update process includes time-dependent plasticity and topological gradient regularization. Time-dependent plasticity increases or decreases the weight according to the interval between the previous and subsequent pulses, strengthening the time sequence; topological gradient regularization calculates the Euclidean norm gradient of the topological lifetime vector and acts on the weight in the opposite direction, so that the network structure remains sensitive to high-lifetime areas. The double mechanism ensures that the causal graph not only follows the physical timing but also embeds topological stability information.

[0053] After the pulse evolution is completed, the system introduces the topological lifetime vector as an external field modulation term into the neural ordinary differential equation. The neural ordinary differential equation is a framework that extends the neural network to a continuous-time dynamical system, and its core is to describe the state change through a differentiable vector field. The vector field is generated by a gated graph convolution network, the adjacency matrix of the pulse neural causal graph provides information flow paths, and the topological lifetime vector is linearly weighted and added to the vector field as an external field coefficient to adjust the system acceleration. An adaptive step numerical integrator is used to solve it, and a dynamic pathological manifold covering a 5-second prediction window can be obtained. The pathological manifold is a trajectory that describes the patient's health status over time in the embedding space, and its curvature and flow direction directly represent the short-term sign trend.

[0054] In order to make the subsequent decision-making stage fully utilize the global structure of the pathological manifold, the present application compresses the manifold into a tensor network state and selects a matrix product representation. Tensor network is a technology that decomposes high-order tensors into low-rank tensor chains, which can significantly reduce storage and computing overhead. The variational singular value decomposition algorithm is used in the compression stage to control the approximation error within an acceptable range; after compression, any linear analytical operation on the manifold is converted into sequential operation along the chain direction, and the computational complexity is only linearly related to the chain length, which is suitable for real-time scenarios.

[0055] The tensor network state and the pulse neural causal graph are jointly input into a quantum optimization model to evaluate candidate intervention actions. The model constructs a Hamiltonian for each action in Hilbert space, and its coefficients are given by the weighted paths of the causal graph. A low energy expectation means that the action is more matched to the current manifold; the system also calculates the structural disturbance caused by the action to the topological lifetime vector as the topological risk value. The linear combination of energy expectation and topological risk forms a joint optimization objective, and the tensor network virtual time evolution or quantum annealing algorithm is used to solve the optimal intervention sequence. To avoid ignoring the topological contribution in the later training stage, the system reversely injects the gradient of the objective function with respect to the topological lifetime vector into the external field coefficient, realizes the real-time update of the pathological manifold along the optimal sequence, and forms a public health intervention model. The public model maintains consistency at the group level and realizes adaptive convergence through continuous feedback of the topological gradient at the individual level.

[0056] Effect verification shows that the pathological manifold is updated at a frequency of 100 milliseconds, and the intervention response delay is reduced by 30% compared with the offline scheme under the same hardware conditions. In the apnea simulation experiment with 20 subjects, the system predicts the trend of the respiratory waveform according to the curvature change of the manifold, and gives a voice prompt 150 seconds before apnea on average; the control static causal graph scheme is only 60 seconds ahead. Further in the test of 20% random missing of electrocardiogram signals, the pathological manifold remains smooth and continuous, indicating that the topological gradient regularization improves the robustness of the system to data gaps.

[0057] Example implementation: Collect multi-modal data from 10 patients for 48 hours, compare the algorithm of the present application with the control model without topological regularization, the former is 180 seconds ahead on average in predicting ventricular tachycardia events, and the latter is 90 seconds ahead; under the same alarm threshold, the false positive rate of the former decreases by 25%. This verifies the effectiveness of the coupling of the topological life vector and the spiking neural causal graph.

[0058] In summary, the three links of topological life vector extraction, spiking neural causal graph construction and neural ordinary differential equation solving constitute the dynamic pathological manifold generation mechanism of the present application. This mechanism has the advantages of explainability, differentiability and real-time performance, and provides stable and information-rich input for quantum optimization intervention, thereby realizing an active and accurate multi-source data health intervention system.

[0059] Preferably, when extracting the topological life vector, a Vietoris-Rips complex is constructed on the uniform representation vector sequence of fixed frame length, the birth time and death time are calculated, and the difference between the two is used to form a life value vector as the topological life vector.

[0060] Extracting the topological life vector is a key step in structuring feature extraction on time series health data, and its goal is to convert a high-dimensional uniform representation vector sequence into a topological feature that is consistent in length and insensitive to transient noise. This step is realized based on persistent homology theory, and the core process can be summarized as "window division - complex construction - filtering evolution - life calculation - vector mapping".

[0061] Persistent homology is a topological method for measuring the stability of connected structures at different scales. For time series, traditional statistics often focus on the mean or variance, while ignoring the global morphology of point interconnection as the scale changes. Persistent homology records when topological features (connected components and loops) appear and disappear as the scale radius increases, providing a cross-scale, noise-robust description of the sequence.

[0062] In the application of the present application, the uniform representation vector sequence is first sliced according to a fixed frame length. Let the window length be , and the frame vector in the window is denoted as wherein , is the uniform vector dimension. A Vietoris-Rips complex is constructed for the window. The complex is a collection of simplices (points, edges, triangles, etc.) that reflect the geometric skeleton formed by connecting data points according to a distance threshold. Let the distance threshold be , an edge is connected between two points when the metric distance between the two vectors is less than , and a triangle is formed when three points are connected two by two. As continuously increases, new topological features are generated in the complex and are filled or merged at larger scales. The birth time refers to the value at which a feature first appears, and the death time refers to the value at which the feature is filled or merged into other structures. The lifetime is defined as:

[0063] ,

[0064] wherein is the survival length of the th topological feature. The longer the lifetime, the more stable the feature is in the scale space, and the more likely it corresponds to a real structure rather than noise. The present invention sorts all by size and truncates or pads to a fixed dimension to form the topological lifetime vector . The symbol represents the birth time, represents the death time, represents the lifetime, and represents the topological lifetime vector.

[0065] In practical applications, window division, the uniform representation vector sequence is divided into windows in a sliding manner, and the step size can be smaller than the window length to obtain overlap. In the experiment, the window length is set to 64 frames, and the step size is set to 16 frames, taking into account the time resolution and computational burden. Distance measurement, the uniform representation vector has different cross-modal attributes, and the present invention uses cosine distance to measure similarity. A distance matrix is calculated for all vectors in the window and stored in GPU memory to provide data for subsequent parallel complex construction. Complex construction, a parallel algorithm based on edge increasing order is used to generate the Vietoris-Rips complex. As the threshold rrr monotonically increases over the set of pre-set radii, the algorithm dynamically maintains the number of connected branches and loops, and realizes the output of birth and death times in one scan. Filtering evolution, to avoid meaningless short lifetime peaks caused by extreme noise points, the present invention sets a minimum filtering radius for the lifetime threshold. Features with a lifetime below this threshold are not included in the topological lifetime vector. The threshold can be automatically set according to the historical data statistics quantile without human intervention. Vector mapping, the set of lifetimes calculated is arranged in descending order, and if the number exceeds the target dimension then truncate, most of the time take ; if not enough, fill the end with zeros. Finally, the length of the uniform topology life vector is obtained , and is scaled to the [0, 1] interval in a normalized manner, which is convenient for training with other features. The running efficiency benefits from GPU parallelism. The average time consumption of the present application for the persistent homology calculation of a single window is about 2 milliseconds, which meets the 100-millisecond system cycle requirement.

[0066] Through the present application, short-life features are naturally filtered, and transient high-frequency noise is difficult to interfere with subsequent causal modeling. The life vector captures the cross-scale commonality and is not sensitive to slight changes in the sampling rate. The Euclidean norm is calculated , which can calculate the gradient and facilitate the participation of the regular term in the weight update in the spiking neural network. The sequence in the window is still mapped to a fixed length regardless of the dimension, which is convenient for batch processing.

[0067] In an embodiment, the topology life vector is continuously extracted at a 1000-millisecond update cycle for 10 patients in a 24-hour ICU record test, and is compared with the original sequence features without topology processing. In the early detection task of arrhythmia, the model using the topology life vector as input achieves a recall rate of 0.87, while the original sequence feature model achieves a recall rate of 0.74; in the noise injection experiment, when 30% of the frames are randomly deleted, the recall rate only decreases by 0.04, while the control model decreases by 0.15, indicating that the life vector exhibits stronger anti-missing capability than the traditional features.

[0068] In summary, the present application realizes the structural stability measurement of the uniformly represented vector sequence by constructing the Vietoris-Rips complex on the fixed frame length window, calculating the birth time and death time, and forming the life value vector. The topology life vector not only provides a noise-robust global feature, but also injects a topological prior into the spiking neural network and neural ordinary differential equation in a differentiable manner, laying a foundation for subsequent dynamic pathological manifold generation and intervention decision-making.

[0069] Preferably, when constructing the spiking neural causal graph, a spiking neural network containing voltage release integral firing neurons is established, the synaptic weight is updated using the time difference dependent plasticity rule, and the gradient of the topology life vector is used to implement regularization adjustment of the synaptic weight.

[0070] The construction link of the spiking neural causal graph maps the uniformly represented vector to an interpretable time-dependent structure, providing a causal inference basis for subsequent intervention decision-making. This link includes four core steps: neuron definition, weight initialization, double-channel weight update, and regularization feedback.

[0071] The present application selects voltage leaky integrate-and-fire neuron as basic computing unit. Such neuron takes resistor-capacitor circuit as physical analogy, membrane potential accumulates in exponential curve under external input current driving, and resets after instantaneously firing pulse when exceeding threshold. Unlike common rectifying linear unit, integrate-and-fire neuron naturally possesses pulse timing information, and can directly correspond to vital signs with obvious instantaneous peak in medical signals.

[0072] The unified representation vector dimension is set as . The system maps each dimension to an input channel of a neuron, totaling neurons and sequentially numbered. The synaptic weight matrix is denoted as . Initially, is randomly assigned according to normal distribution with zero mean to avoid network falling into symmetric state. The weight update adopts a double-channel mechanism, jointly acted by time difference-dependent plasticity and topological gradient regularization term. Time difference-dependent plasticity describes the influence of pulse time difference of two neurons on their synaptic weight. Let the pre-neuron fire at time and the post-neuron fire at time . The weight increment can be represented as

[0073] ,

[0074] , where and are positive and negative learning rates respectively, and are time constants. The formula makes synapse strengthen when , and vice versa, thereby recording time sequence at network level and forming causal directed edge. Simply relying on time difference-dependent plasticity is prone to cause network to be disturbed by instantaneous noise. The present application introduces topological gradient regularization term to suppress unstable connection. The topological lifetime vector output by the previous step reflects stable structure of data in scale space. The present application defines regularization loss as

[0075] ,

[0076] and calculates gradient for :

[0077] ,

[0078] The gradient direction points to the direction of weight increment with fastest increasing lifetime value, so that small step subtraction of gradient during network training can keep high-lifetime topological features from being damaged. Since has been sorted according to scale stability, regularization will not introduce additional noise. In order to be compatible with the discreteness of pulse update, the system further uniformly updates Add the regularization correction.

[0079] After the evolution of the network ends, The non-zero elements in the matrix are the edge weights of the pulse neural causal graph. The edge direction points from the synapse starting point to the ending point, and the absolute value of the edge weight represents the causal strength. The positive and negative values of the weight correspond to the excitatory or inhibitory relationship. The causal graph itself is a sparse structure, which can be directly converted into an adjacency matrix to participate in gated graph convolution and neural differential equation solving.

[0080] In terms of implementation, in order to balance real-time performance and stability, the present application uses batch integration to update membrane potential on GPU in parallel, uses event-driven queue to store pulse timestamps, and applies the time difference-dependent plasticity formula immediately after each event processing. Topological gradient regularization is performed in a batch update form, so that the two types of updates do not interfere with each other. The overall evolution iteration step number increases linearly with the data window length, and the operation amount of each step increases linearly with the number of sparse edges, meeting the real-time demand of milliseconds.

[0081] In terms of effect, the introduction of the topological gradient regularization term significantly improves the stability of the causal graph. Experimental results show that, under the condition of randomly occluding 20% of the electrocardiogram frames, the average overlap degree of the causal edges of the network without regularization is 40%, and after regularization, it is increased to 70%. In the dynamic pathological manifold prediction task, the fluctuation of the curvature change rate of the manifold generated by the regularized network is reduced by 1 / 3, which is conducive to the convergence of the subsequent quantum optimization algorithm.

[0082] Embodiment: For 48-hour multi-modal recordings of 10 patients, the average node degree of the pulse neural causal graph constructed by the present application is 12, and the edge weight update time interval is not more than five milliseconds. Compared with the control group using only time difference-dependent plasticity, the average advance time is increased by 90 seconds and the false positive rate is reduced by 25% in the apnea advance warning task after adding topological gradient regularization.

[0083] In summary, by fusing time difference-dependent plasticity and topological gradient regularization in the voltage discharge integral firing neural network, the present application realizes the joint coding of time sequence and topology for the unified representation vector sequence. The generated pulse neural causal graph not only preserves the temporal relationship, but also highlights the stable structure, providing high-quality adjacency information for neural differential equations, and significantly improving the prediction accuracy and intervention timeliness in downstream quantum optimization intervention.

[0084] Preferably, the topological lifetime vector is introduced as a modulation term into the neural differential equation, an external field coefficient is added in the vector field, and an adaptive step numerical integration method is used to solve it, to generate a dynamic pathological manifold.

[0085] Introducing topological lifetime vector as a modulation term to construct neural ordinary differential equation is the core step of the present application to realize real-time health state prediction and interpretable evolution. This step directly embeds the scale stable information obtained by persistent homology into continuous-time dynamical system, so that the dynamic pathological manifold not only reflects the time sequence dependence, but also explicitly obeys high-dimensional topological constraints.

[0086] At the principle level, the neural ordinary differential equation can be regarded as converting the discrete transformation under the infinite layer limit of the deep network into the integral of the continuous vector field. Let represent the hidden state at time , and the vector field is denoted as , where is a learnable parameter, then the system evolution satisfies:

[0087] ,

[0088] where is the adjacency matrix, which comes from the pulse neural causal graph constructed in the previous link. The traditional method takes as fixed structure information, but does not consider the stability of data in the scale space. The present application introduces the topological lifetime vector , maps it to the external field coefficient , and injects into the vector field with weight:

[0089] ,

[0090] In the formula, , is a trainable scaling parameter, represents zero-mean normalization of the lifetime vector; is the product of corresponding elements; is a single hidden layer fully connected network, which is used to project the state to a channel with the same dimension as . Through this additive external field, the topological features with long lifetime produce more significant acceleration or deceleration effects in dynamics, guiding the evolution of hidden state towards stable structure.

[0091] The implementation details contain three steps. The first step is to normalize the adjacency matrix to improve numerical stability. The degree matrix is used to normalize the row of , and is obtained, which is then taken as the input of the gated graph convolution. The second step adopts the gating mechanism in the vector field :

[0092] ,

[0093] where and are learnable weights, Sigmoid activation. Gating structure can suppress gradient explosion. Third step uses adaptive step solver. Experiments compare fixed step fourth order Runge-Kutta and Dormand-Prince method. Adaptive solver reduces 40% steps on average with same error tolerance, ensuring 5s prediction window finishes integration within 50ms.

[0094] Effectively, external field modulation significantly improves manifold explainability. Compared with baseline without external field, curvature continuity of pathological manifold increases by 0.12 and torsion standard deviation decreases by 0.09 after adding external field, indicating that the trajectory is smoother and more robust to noise. In addition, through the backtracking of abnormal events, it is found that the manifold changes direction obviously tens of seconds before the event, and the change amplitude is positively correlated with the longest life value in the life vector, which can be used as a basis for clinical interpretation.

[0095] Embodiment: Deploy the application in a chest pain center and continuously monitor 20 unstable angina patients for 24 hours. The system generates a topological life vector every 100ms, and outputs a 5s prediction window of pathological manifold after introducing neural ordinary differential equation. The correlation coefficient between the subjective score of chest tightness marked by doctors and the peak value of manifold curvature reaches 0.78, which is significantly higher than 0.55 of traditional cycle network. Further in the dose adjustment test, when the system predicts the trend of ECG voltage drop 180s in advance according to the manifold, it automatically recommends a dose adjustment scheme, and the average blood pressure stabilization time is shortened to 7min, while the manual adjustment scheme needs 12min.

[0096] In summary, topological life vector external field modulation seamlessly combines statistical topological information with neural ordinary differential equation, providing global stability prior while maintaining vector field differentiable structure. Adaptive step integrator ensures real-time calculation requirements within the prediction window, while external field feedback injects interpretable dynamic characteristics into subsequent intervention decisions, thus constituting a key link in the active health intervention closed loop of the application.

[0097] Compress the dynamic pathological manifold into a tensor network state, establish a quantum optimization model for candidate intervention actions according to the impulse neural causal graph, obtain the optimal intervention sequence by jointly optimizing the objective function, and feed the gradient of the joint optimization objective function with respect to the topological life vector back to the neural ordinary differential equation, so that the dynamic pathological manifold is updated in real time with the optimal intervention sequence and forms a public health intervention model.

[0098] Preferably, when compressing the dynamic pathological manifold into a tensor network state, a matrix product representation is obtained through tensor rearrangement and variational singular value decomposition.

[0099] Dynamic pathological manifolds describe the multimodal health state trajectory of patients over a continuous time axis, providing a high-dimensional, continuous, and interpretable predictive basis for intervention strategies. However, the original manifold is spanned by a large number of time steps and high-dimensional states, which consumes storage resources and complicates subsequent optimization calculations. To improve real-time performance and operability, this invention first compresses the manifold into tensor network states, then combines it with a spiking neural causal graph to construct a quantum optimization model, and finally achieves closed-loop updates through a feedback mechanism.

[0100] The core of the compression phase is the tensor network state. Tensor networks are a class of methods that factorize higher-order tensors into sets of lower-order tensors using a graph structure; matrix product representation is one of the most commonly used forms. Let the dynamic pathological manifold at discrete time points... The hidden state is By concatenating these states in time, we can obtain an order of... tensor If directly in The above operation not only requires storage space The magnitude is enormous, and subsequent optimization requires traversing a massive number of parameters. The matrix multiplication integral solution... Rewrite as a chain structure:

[0101] ,

[0102] in The rank does not exceed The third-order tensor, This is called the virtual key dimension. It is adaptively determined through variational singular value decomposition. This allows the approximate error to be controlled within the order of 1% while reducing the parameter size to... The present invention selects Adaptive to manifold curvature: segments with high curvature are assigned higher dimensions to preserve local details, while segments with low curvature are compressed more aggressively, thus balancing accuracy and computational speed.

[0103] After obtaining the tensor network state, it needs to be used in conjunction with the spiking neural causal graph for interventional action assessment. The spiking neural causal graph stores the excitation or inhibition relationships between neurons in the form of an adjacency matrix, with matrix elements denoted as... The absolute value reflects the strength of causality. For each candidate intervention action... The present invention constructs the Hamiltonian:

[0104] ,

[0105] in For the first Vipalil Operator, It is generated by weighting the original adjacency matrix according to action type. The expected energy of the Hamiltonian:

[0106] ,

[0107] In tensor network ground state The computation reflects how well the action matches the current manifold. At the same time, the system computes the topology risk vector and the action risk vector Element-wise multiplication of the two vectors gives the topology risk:

[0108] ,

[0109] Topologies with longer lifespans are more vulnerable to improper intervention. Linear combination of energy and risk defines the joint objective:

[0110] ,

[0111] The smaller the objective, the safer and more effective the action.

[0112] The optimization process has two branches: quantum and tensor. Without quantum hardware, the tensor network is evolved in virtual time: by inserting variational gates in the chain of matrix multiplications and fine-tuning along the negative energy gradient, the objective minimum is quickly approached. With quantum hardware, the is mapped to a binary disordered quadratic optimization problem submitted to a quantum annealer for sampling, followed by local optimization with the tensor network for refinement. The hybrid strategy takes into account both hardware availability and real-time performance.

[0113] To avoid neglecting topology contributions in the later stages of training, the invention computes the gradient of the objective with respect to the topology lifespan vector after each optimization round:

[0114] ,

[0115] and feeds the gradient back to the external field coefficients of the neural ordinary differential equation in small steps. The new values of the external field coefficients:

[0116] ,

[0117] cause the vector field to pay more attention to the repair of damaged topology holes in the next cycle, thus prompting the pathological manifold to deform in real time with the optimal sequence. Updates shared by multiple patients constitute a public health intervention model, which is synchronized to all edge nodes through a broadcast mechanism, achieving both group knowledge accumulation and individualization through patient private gradients.

[0118] Effect evaluation shows that the system can limit the optimization time to 50 milliseconds within 100 intervention cycles, and the cosine similarity between the topology lifetime vector of each iteration and the external field coefficient increases by 0.08. Compared with the control group without feedback, the lead time of abnormal heart rate prediction is improved by 40% after introducing feedback, and the speed of blood pressure stabilization after intervention is improved by 30%. Further in the stress test with artificial noise inserted, the system greatly weakens the curvature oscillation caused by noise through external field adaptation, so that the pathological manifold always maintains a predictable direction.

[0119] Embodiment: In a 48-hour monitoring experiment, the tensor network state of 12 patients is updated every 200 milliseconds, and four intervention actions are evaluated within a 1-second window. The quantum annealing hardware provides 100 samples per batch, and the tensor fine-tuning takes 25 milliseconds, with the overall cycle completing in 45 milliseconds. After 24 hours of training, the optimal action sequence has an agreement rate of 87% with the recommendations of human physicians, while the agreement rate of the topology feedback-free model is 71%.

[0120] In summary, by compressing the dynamic pathological manifold into the tensor network state, constructing a quantum optimization model using the pulse neural causal graph, and injecting the target function gradient back into the external field of the ordinary differential equation, the present application realizes the unification of explainable, trainable, and real-time iterative health intervention closed loop. This closed loop retains individual adaptive characteristics while being shared by multiple patients, improving intervention timeliness and stability, and is the core innovation of the multi-source data health intervention system.

[0121] As shown in Figure 2 , preferably, when establishing the quantum optimization model, the weight values of the pulse neural causal graph are mapped to the Hamiltonian coefficient corresponding to the candidate intervention action, and the energy expectation value and the topology risk value are linearly combined to form a joint optimization objective function.

[0122] The multi-source data health intervention system needs to select the intervention action with the lowest risk and the highest benefit from a large number of alternative measures in a very short time. The present application uses a quantum optimization model to complete this task, and the core idea is to map the weight values of the pulse neural causal graph to the quantum Hamiltonian coefficient, to measure the degree of fit between the action and the pathological manifold with the energy expectation value, and to form a joint optimization objective by combining the topology risk. This step not only shortens the search time for the optimal action, but also maintains real-time performance in the absence of quantum hardware by taking advantage of the decomposable nature of the tensor network.

[0123] In terms of principles, the directed edge weight of the pulse neural causal graph records the triggering relationship between neurons, and the absolute value of the edge weight represents the causal strength, while the sign indicates excitation or inhibition. Let the total number of nodes be , and the adjacency matrix element be . For a candidate intervention action , the system first selects the relevant node set according to the action range, and then adjusts the weight values in the set by multiplying the edge weight by the action sensitivity coefficient Then the Hamiltonian is constructed:

[0124] ,

[0125] in For the first Pauli qubits Operator. This operator takes values ​​in a binary state. Therefore, it can be used to describe the energy contribution when the states of two nodes are consistent or opposite. The effective state obtained by compressing the dynamic pathological manifold through matrix product representation is denoted as... Then the expected value of the action energy is:

[0126] ,

[0127] The lower the energy, the more the action conforms to the current physiological evolution trend.

[0128] Simply minimizing energy may lead to the selection of strategies that compromise topological stability. Therefore, this invention introduces a topological risk index: Topological lifetime vector. Each component in Indicates the first The survival length of a topological feature in scale space; the longer the lifetime, the more the structure should be protected. (Regarding action...) Pre-calculate the historical perturbation probability of each topological feature to form a risk vector. Topological risk is defined as:

[0129] ,

[0130] in Represents the product of corresponding elements. Let the sum be the absolute values. The final joint optimization objective function can be written as:

[0131] ,

[0132] To balance the parameters, cross-validation was used to determine them. The objective function considers both physiological fit and suppression of potential disruption to key topologies.

[0133] In the implementation process, the imaginary-time evolution of tensor networks is first used to coarsely search for the optimal action sequence on classical hardware, compressing the search space to a scale that can be processed by a quantum processor; if quantum hardware is available, then... The result is rewritten as a binary unordered quadratic optimization form and fed into a quantum annealing machine for rapid sampling, then refined using the tensor network local gradient method. To ensure the result remains synchronized with the topology, the system immediately calculates the gradient of the objective function with respect to the topological lifetime vector after completing one action optimization.

[0134] ,

[0135] and multiply the gradient by a small step size to feed back to the neural differential equation external field coefficient, so that the next round of pathological manifold evolution adjusts to the direction of risk reduction. Since the topological lifetime vector is derived from window-level persistent homology, this feedback mechanism can self-heal noise disturbances within a few iterations and gradually form a public health intervention model covering all patients.

[0136] Effect evaluation shows that in a real-time system with a sampling interval of 200 milliseconds, the quantum optimization model outputs a 3-step action sequence in an average of 50 milliseconds, which is 70% shorter than the traditional greedy ray search; the risk weight brought by the joint target reduces the false alarm rate by 25%. In a 24-hour ICU record comparison experiment, the system improves the lead time to 180 seconds in the arrhythmia warning task, and maintains a false alarm rate of at most 5%. After using the topological gradient feedback, even if 20% of the random electrocardiogram frames are deleted, the manifold curvature remains smooth, indicating that the feedback mechanism is naturally robust to data loss.

[0137] Example implementation: 12 heart failure patients were continuously monitored for 36 hours, and 4 types of intervention actions were evaluated every 1 second. Using only virtual time evolution in the non-quantum hardware scenario, the average cycle takes 45 milliseconds to complete; after connecting to quantum annealing hardware, the overall time is shortened to 30 milliseconds, and the average early warning of heart failure exacerbation events is 200 seconds. The above experiments verify the synergistic value of the quantum optimization model and the topological feedback of the present invention.

[0138] Preferably, when obtaining the optimal intervention sequence, the tensor network virtual time evolution algorithm or the quantum annealing algorithm is used to solve the joint optimization objective function.

[0139] In the multi-source data health intervention process of the present invention, the solution of the optimal intervention sequence is the core operation of the decision layer. The system needs to select a combination that meets the safety and efficacy requirements from the candidate action library within tens of milliseconds. Therefore, the present invention designs a quantum-tensor hybrid optimization framework: first, use the tensor network virtual time evolution algorithm to complete the coarse-grained potential energy valley search on classical hardware, and if a quantum annealing processor is deployed, further call the quantum annealing algorithm to find the approximate global minimum value of the joint optimization objective function. The two paths can be run independently or in series; the former ensures that the results can still be output in real time without quantum hardware, and the latter significantly improves the optimality when hardware is available.

[0140] The joint optimization objective function is denoted as:

[0141] ,

[0142] wherein The action is an energy expectation value on the tensor network ground state, quantifying the match between the action and the current pathological manifold; The topological risk value reflects the potential disturbance of the action to the topological lifetime vector; The trade-off parameter. The energy term is calculated by the Hamiltonian generated by the pulse neural causal graph weight mapping, and the risk term is obtained by taking the absolute value of the element-wise product of the lifetime vector and the action risk vector. The minimization problem of the objective function is equivalent to finding the state with the lowest energy and risk constraint in the high-dimensional Hilbert space.

[0143] The idea of virtual time evolution algorithm comes from the infinitesimal time evolution operator in quantum mechanics. Let The initial tensor network state is obtained by acting on the operator And normalized at each time slice, the ground state of the Hamiltonian Can be gradually approached when the virtual time Increases. The present application selects As the effective Hamiltonian with risk penalty term:

[0144] ,

[0145] Where Is the matrix obtained by embedding the risk vector into the diagonal operator. Using the separable property of the matrix product decomposition, the exponential operator can be decomposed into a multi-segment gate sequence, each segment only affecting the local tensor on the chain. By variational singular value decomposition, the virtual bond dimension is truncated at each step, ensuring that the computational complexity is linearly related to the chain length. Iteration to the preset error threshold can obtain the locally optimal action sequence.

[0146] Quantum annealing algorithm belongs to quantum approximate optimization strategy, which encodes the combinatorial optimization problem as Ising model, and then uses quantum tunneling effect to cross the classical barrier. Re-express As a binary unordered quadratic optimization form, the elements of the coefficient matrix are determined by the Hamiltonian and the risk term. After giving the annealing path and time, the quantum annealing machine returns a batch of low-energy samples, and the system selects the sample with the lowest energy as the candidate action sequence, and uses the local gradient descent of the tensor network to refine, ensuring that it is consistent with the micro-local constraints of the pathological manifold.

[0147] In practical applications, each candidate action is mapped to a binary vector of length , and the element 1 indicates that the action is selected and executed at the current time. The size of the action library is reduced to no more than 32 through domain knowledge in advance, so as to be efficiently coded on the tensor network and quantum annealing machine. The dynamic pathological manifold is obtained by matrix product decomposition to obtain the initial effective state . Chain length corresponds to the predicted window frame number, and the virtual key dimension is adaptively set. The greater the curvature, the higher the section dimension. The first 3 steps use a large step to quickly descend, and the subsequent steps decrease until the energy decreases by less than 1e-4 to terminate; on average, 6 steps are needed to converge. After converting the optimization function into an Ising model, 100 low-energy solutions are sampled in batches on a quantum annealing machine, and the top 20% are selected as the starting point for refinement. After completing the action sequence determination, the gradient of the topological lifetime vector is calculated , and the neural differential equation external field coefficient is updated with a learning rate of 0.05, so that the pathological manifold adjusts in the direction of risk reduction in the next round of evolution.

[0148] In a 100-millisecond update cycle, the pure tensor network path takes an average of 45 milliseconds to output a 3-step optimal sequence; when a quantum annealing machine is available, the total time is reduced to 30 milliseconds. After introducing the risk term and gradient feedback, the system's false positive rate decreases by 25%, and the blood pressure recovery time after intervention shortens by 30%. In the test of simulating the random missing of 20% of the vital signs frames, the action selection accuracy remains 92%, proving the robustness of the algorithm to data gaps.

[0149] Embodiment, 15 severe patients are continuously monitored for 48 hours. The system evaluates 4 types of intervention actions every 1 second: voice prompt, drip speed adjustment, vibration reminder, and emergency alarm. The experiment is divided into three groups:

[0150] Group A only uses virtual time evolution; Group B uses virtual time evolution and quantum annealing; Group C turns off the topological risk term based on Group B. The results show that Group B can provide early warning of room speed events by an average of 180 seconds, Group A by 140 seconds, and Group C by 160 seconds but with a false positive rate of 18%. This shows that quantum annealing combined with topological risk can improve both the advance amount and accuracy.

[0151] The present application solves the joint optimization target by tensor network virtual time evolution and quantum annealing cooperation, realizes millisecond-level multi-step action planning, introduces topological risk to suppress destructive decisions, and gradient feedback to reshape the pathological manifold in time, ultimately forming a safe and efficient public health intervention model, providing reliable technical support for real-time clinical decision-making assistance.

[0152] According to the clinical logic rule, the optimal intervention sequence is verified, and after verification, the optimal intervention sequence is recorded to the distributed ledger, and control instructions are generated according to the optimal intervention sequence and sent to the execution terminal.

[0153] The three links of clinical logic verification, distributed ledger record, and control instruction issuance together constitute the "execution security layer" of the intervention link. Its core goal is to ensure that the optimal intervention sequence derived by the algorithm is not only optimal in energy and topological sense, but also meets the established medical specifications, has traceability, and can trigger terminal device action in a low-latency manner.​

[0154] The present application verifies the clinical compliance of intervention sequences using an answer set procedure reasoning engine. Answer set procedure is a formalized knowledge representation method based on satisfiability solving, which is particularly suitable for fast reasoning on "if-then" type rules. The system is preloaded with a rule base containing items such as vital sign thresholds, drug contraindications, and rescue priority. For example, when systolic blood pressure is continuously below a threshold and the patient has a do-not-resuscitate order, an automatic call for emergency help should be prohibited. To avoid conflicts between rules, the rule base is checked for consistent answer sets before deployment. When the optimal intervention sequence is generated, the system splits the sequence into structured facts such as

[0155] action(step_index, action_type, delay) and submits them to the reasoning engine together with the translated facts from the latest vital sign data. If the engine returns a stable answer set with no conflicting predicates, the sequence is deemed compliant; otherwise, it enters the manual confirmation or rollback process. This verification process takes an average of 5 milliseconds, meeting the real-time requirement of hundreds of milliseconds.

[0156] Compliant sequences must have non-repudiation. The present application stores intervention records in a distributed ledger based on Byzantine fault-tolerant consensus. The ledger nodes are deployed on the nurse station server, the doctor's order system backup machine, and the hospital-level master room, forming a 3-write-2-read redundancy structure. To reduce the risk of data leakage on the chain, the system first performs a commitment operation on the intervention sequence: using a random number and the sequence digest to generate a Pedersen commitment:

[0157] ,

[0158] where and are different generators of the cyclic group of the same order, is a large prime number. This commitment has both hiding and binding properties, i.e., an external observer cannot infer from , and the operator cannot adjust after the fact to match the same . To prove fairly that it is within the legal domain (e.g., the sequence length is no more than 8 steps, and the delay of each step is non-negative), the system uses the Bulletproof zero-knowledge proof protocol to generate a proof of length about 1.3 kilobytes. The ledger transaction contains three fields: timestamp, commitment value, and proof, which are confirmed within 2 seconds through Byzantine fault-tolerant consensus algorithm. Only the irreversible digest is stored on the chain, and the plaintext sequence is stored encrypted in the hospital's private object storage, with key shards kept in 2 nodes to prevent single-point leakage.

[0159] After verification, the system generates control instructions according to the intervention sequence. The instruction format is

[0160] {action_type, parameter, target_id, delay_ms}, where action_type includes voice broadcast, drip rate setting, vibration reminder, and alarm call; parameter contains drug pump rate or prompt text; target_id is the target device address; delay_ms is the relative delay. The instruction is sent to the execution terminal using gRPC bidirectional stream, and requires an ACK within 2 seconds. If no ACK is received, retry once; if still fails, record the error and trigger suboptimal action rollback.

[0161] The control instruction generator is also responsible for feeding back the optimal sequence to the public health intervention model. The feedback includes sequence action type distribution, execution success rate, and delay statistics. These data are sent to the model aggregation service in an asynchronous manner, and are updated together with the topology gradient when training the next cycle. This design ensures continuous accumulation of group knowledge, while preserving local differences to support personalized intervention.

[0162] Empirical evaluation shows that the answer set program reasoning has a 100% verification accuracy rate for 500 randomly generated intervention sequences, with an average time consumption of 4.8 milliseconds; the distributed ledger takes 1.9 seconds to confirm a single transaction, but does not block the control instruction delivery process. The zero-knowledge proof passes at a rate of 100%, with no false positive cases. The device ACK success rate is 99%, and the main failure scenario is network jitter. The retry mechanism can improve the final success rate to 99.8%. In a 48-hour ICU monitoring experiment, the system recorded 320 interventions, none of which violated the clinical rules; the on-chain transaction and object storage operation correspond completely, proving the traceability.

[0163] For example: A patient has sustained hypotension and atrial fibrillation at 02:15:30. The optimal sequence contains 3 actions: voice prompt, drip rate increase, and vibration reminder, with delays of 0 milliseconds, 2000 milliseconds, and 5000 milliseconds respectively. The rule base retrieves the patient's existing vibration stimulation restriction entry, and the reasoning engine immediately excludes the vibration reminder and returns a compliant sequence of 2 steps. The system records the commitment and chains it at 02:15:31, issues the voice prompt instruction at 02:15:32, and issues the drip rate adjustment instruction at 02:15:34. Subsequently, the blood pressure rises to a safe range at 02:15:50, 120 seconds earlier than the manual solution, completing the intervention.

[0164] In summary, the combination of clinical logic rule verification and distributed ledger provides a double security barrier for the intervention sequence generated by the algorithm; the low-latency delivery of control instructions ensures the immediacy of the intervention effect. This complete link not only meets the requirements of traceability and auditability of medical regulations, but also maintains the efficiency of real-time decision-making, which is an important execution guarantee for the multi-source data health intervention system of the present application.

[0165] Preferably, when verifying the optimal intervention sequence, the optimal intervention sequence is converted into a fact set and input into the rule base based on the answer set procedure, and the verified intervention sequence is obtained through reasoning.

[0166] In the execution layer of the multi-source data health intervention closed loop, any algorithm-generated intervention action must be verified by clinical rules to avoid violating treatment conventions, patient contraindications, or medical ethics. The present application uses a "fact set + answer set procedure" verification framework to formalize the optimal intervention sequence into machine-readable facts, which are then input into the reasoning engine together with the rule base to obtain the verified intervention sequence; if the sequence meets all the rules, it enters the on-chain storage and device execution process, otherwise it is rolled back or requests manual confirmation.

[0167] The answer set procedure is a kind of knowledge representation method based on satisfiability solution. Its core is to encode the domain rules into Horn clause form, and obtain the "stable model" that satisfies all the rules through a logic programming solver. In the context of the present application, the rule set covers clinical logic such as drug interactions, vital sign thresholds, rescue priorities, patient instructions (such as prohibition of resuscitation), and resource occupation order. Any intervention sequence that causes rule conflicts will not be able to generate a stable model, and the reasoning engine will return an empty solution or a model containing conflict identifiers. In this way, the legality of the algorithm recommendation can be determined within milliseconds.

[0168] In practical applications, the facts are converted, and the optimal intervention sequence is first disassembled into a fact set. Each step of action is represented by a predicate:

[0169] ,

[0170] wherein is the serial number, type is an enumerated string (such as drip_rate, voiceprompt, call_ems, vibrate_band), is the relative delay in milliseconds. The latest vital sign facts such as hr (80) and sbp (68) are injected synchronously. Text medical orders and patient instructions are also converted into Boolean facts, such as do_not_resuscitate.

[0171] The rule base is constructed using unambiguous predicates to avoid confusion caused by abbreviations. An example entry is:

[0172] low_sbp :- sbp(V), V < 70.

[0173] trigger(call_ems) :- low_sbp, not inhibit(call_ems).

[0174] inhibit(call_ems) :- do_not_resuscitate.

[0175] conflict :- action(_, call_ems, _), inhibit(call_ems).

[0176] Where conflict is used to mark illegal situations. Rule base updates are managed through version control, ensuring consistent reasoning baseline.

[0177] Reasoning flow, the system invokes Clingo solver in command line form: clingofacts.lprules.lp --time-limit=100 If the output stable model and does not contain conflict, It is considered as a verified intervention sequence If it contains conflict, automatically search for the minimum modification: delete the riskiest action in turn and retry until it passes or back to the medical staff.

[0178] Time complexity, since the length of the intervention sequence No more than 8 steps, the number of facts is thousands of levels; Clingo solves the complexity close to linear under hundreds of rules. Actual measurement completes reasoning in an average of 4.5 milliseconds, meeting the 100-millisecond cycle requirement.

[0179] On-chain storage and execution, once verified, the system immediately generates Pedersen commitment and zero-knowledge proof , and calls the Byzantine fault-tolerant consensus interface to submit transactions . At the same time, generate gRPC control instructions according to each step action / 3 fact, and issue them to the pump control, voice broadcast, or wearable terminal. ACK timeout automatically retries and records logs.

[0180] Benchmark 500 random sequences, reasoning accuracy 100%, average time 4.8 milliseconds; In the real ICU scene of 48 hours running, the system intercepted 17 sequences containing contraindicated drug combinations, avoiding potential adverse events. On-chain transactions and object storage entries correspond one by one, traceable in full; Terminal execution ACK success rate 99.8%.

[0181] In the embodiment, patient A has systolic pressure of 65 mmHg at 02:15:25, and the algorithm outputs the sequence: voice prompt (0 ms); drip speed increase (2000 ms); vibration reminder (5000 ms).

[0182] The rule base contains the entry "ban vibration stimulus". The reasoning engine detects that action(_,vibrate_band,_) conflicts with ban_vibrate, and marks conflict. The system deletes the vibration reminder and re-solves, generates commitment after verification, and records on-chain. The voice instruction is broadcast at 02:15:26, the drip speed adjustment is completed at 02:15:28, and the systolic pressure rises to the safe range at 02:15:48, which is 110 seconds earlier than the manual process.

[0183] By formalizing the intervention sequence into a set of facts and inputting the answer set program, the system completes the clinical rule verification in milliseconds; on-chain storage provides non-repudiation and traceability; control instructions and feedback mechanisms ensure a closed-loop execution. This execution security layer ensures that the algorithm output not only meets medical standards but also can be quickly implemented, improving the clinical usability and regulatory compliance of multi-source data health intervention systems.

[0184] Preferably, when recording the optimal intervention sequence to the distributed ledger, a commitment value and a zero-knowledge proof are generated for the optimal intervention sequence, and then written to a chain ledger using a Byzantine fault-tolerant consensus mechanism.

[0185] The design goal of the distributed ledger storage module is to create a non-repudiable, auditable, and quickly confirmed on-chain record for each intervention decision without exposing patient privacy details. To this end, the present invention uses a "commitment-zero-knowledge proof-Byzantine fault-tolerant consensus" three-step method to lock the integrity and legality of the intervention sequence simultaneously in the chain ledger.

[0186] The commitment scheme uses Pedersen commitment. Let the intervention sequence digest be denoted as , the random number be denoted as , the cyclic group of order be denoted as , and the different generators be denoted as and . The commitment value is defined as:

[0187] ,

[0188] For the commitment value, for the sequence digest, for the random number, for the group order. This formula has both hiding property (external cannot deduce ) and binding property (cannot replace on the same chain after submission) ).

[0189] To prove For sequences falling in the legal domain (e.g. sequence length no more than 8, delay non-negative), the invention introduces Bulletproof zero-knowledge proof. Bulletproof generates short proof without trusted setup through inner product commitment protocol. The proof only exposes the fact that the sequence satisfies the constraints, without leaking any action details.

[0190] At the chain level, nodes are deployed on 3 servers in the hospital, using Byzantine fault-tolerant consensus. Each transaction contains 3 fields of timestamp, commitment value, and zero-knowledge proof. After the majority of nodes sign, it is considered successful to be chained, ensuring that even if some nodes fail or are malicious, it does not affect the consistency of the entire network.

[0191] The implementation process includes: abstract generation, taking SHA256 hash of the byte string encoding of the optimal intervention sequence to get the abstract Commitment calculation, call secure random number generator to generate , calculate Proof generation,

[0192] According to the sequence constraints, construct the constraint vector, call the Bulletproof library to output the proof Transaction assembly, assemble and broadcast to the consensus network. Consensus confirmation, node verification , after signing, enter two-phase commit, and complete block generation in about 2 seconds. Asynchronous plaintext storage, encrypted sequence plaintext is saved in object storage with double copies, and the index is bound with the on-chain hash.

[0193] Through the invention, only the commitment and proof are saved on the chain Anyone needs to hold to open the commitment, which meets the requirement of homomorphic invisibility. The binding ensures that it cannot be tampered with afterwards, and the append-only write mode provides global order. The commitment and proof generation takes about 3 milliseconds, and the verification takes 2 milliseconds; transaction confirmation takes 1.8 seconds, but does not block the intervention execution path. Byzantine fault-tolerant consensus allows 1 node to be malicious or offline, and still can generate blocks, and the deployment of three nodes in the hospital meets high availability. The supervisor can require the operator to open the box when needed to verify the commitment and confirm that the intervention details and timestamp are consistent after comparing with the plaintext object storage.

[0194] In the 24-hour continuous monitoring experiment, the system generated a total of 320 intervention sequences. On average, each sequence corresponds to a commitment size of 32 bytes, a proof size of 1300 bytes, and a total data volume of about 430 kilobytes. Random sampling of 10 transactions for offline unpacking found no hash inconsistencies or proof failures. In a simulated node failure test, the remaining two nodes still confirmed the new block within 2.3 seconds after one node went offline, demonstrating the robustness of Byzantine fault tolerance.

[0195] In summary, by introducing Pedersen commitment, Bulletproof zero-knowledge proof and Byzantine fault tolerance consensus in the recording phase, the present application not only protects the privacy of patient intervention strategies, but also provides auditable and non-repudiable on-chain credentials, providing a solid technical guarantee for the compliance and landing of multi-source data health intervention systems in clinical environments.

[0196] The above is only an embodiment of the present application and is not intended to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the scope of the claims of the present application.

Claims

1. A multi-source data-driven health intervention method, characterized in that, The method includes: Collect text data, vital sign data, medical image feature vectors, and wearable sensor data. Perform time synchronization, filtering, and standardization on all data, and obtain a unified representation vector through multimodal coding. A unified representation vector sequence is used to extract the topological lifetime vector. Based on the unified representation vector, a spiking neural causal graph is constructed. The topological lifetime vector is introduced as a modulation term into the neural ordinary differential equation. The dynamic pathological manifold is generated by solving the neural ordinary differential equation within its framework. When extracting the topological lifetime vector, a Vitoris-Lipps complex is constructed on a unified representation vector sequence with a fixed frame length. The birth time and death time are calculated, and the lifetime value vector is formed based on the difference between the two as the topological lifetime vector. When constructing the spiking neural causal graph, a spiking neural network containing voltage discharge integral firing neurons is established. The synaptic weights are updated using time difference-dependent plasticity rules, and the synaptic weights are regularly adjusted using the gradient of the topological lifetime vector. The dynamic pathological manifold is compressed into a tensor network state. A quantum optimization model is established based on the spiking neural causal graph as candidate intervention actions. The optimal intervention sequence is obtained by jointly optimizing the objective function. The gradient of the joint optimization objective function with respect to the topological lifetime vector is fed back to the neural ordinary differential equation so that the dynamic pathological manifold is updated in real time with the optimal intervention sequence and forms a public health intervention model. When the compressed dynamic pathological manifold is a tensor network state, the matrix product representation is obtained through tensor rearrangement and variational singular value decomposition. When establishing the quantum optimization model, the weights of the spiking neural causal graph are mapped to the Hamiltonian coefficients corresponding to the candidate intervention actions, and the energy expectation value and the topological risk value are linearly combined to form a joint optimization objective function; The optimal intervention sequence is verified according to clinical logic rules. After successful verification, the optimal intervention sequence is recorded in the distributed ledger, and control instructions are generated based on the optimal intervention sequence and sent to the execution terminal. When validating the optimal intervention sequence, the optimal intervention sequence is converted into a set of facts and input into a rule base based on the answer set program, and the validated intervention sequence is obtained through reasoning. When recording the optimal intervention sequence to the distributed ledger, a commitment value and a zero-knowledge proof are generated for the optimal intervention sequence, and then written into the chain ledger that adopts the Byzantine fault-tolerant consensus mechanism.

2. The method according to claim 1, characterized in that, Multimodal coding processing employs holographic reduction representation, performing binding and circular convolution operations on text data, vital sign data, medical image feature vectors, and wearable sensor data respectively to generate a unified representation vector.

3. The method according to claim 1, characterized in that, When the topological lifetime vector is introduced as a modulation term into the neural ordinary differential equation, external field coefficients are added to the vector field and the adaptive step-size numerical integration method is used to solve the problem, so as to generate a dynamic pathological manifold.

4. The method according to claim 1, characterized in that, When the optimal intervention sequence is obtained, the joint optimization objective function is solved using the tensor network imaginary time evolution algorithm or the quantum annealing algorithm.

Images