Medical resource management method and system

Abstract

The invention relates to the technical field of medical resource management, in particular to a medical resource management method and system. The method comprises the following steps: a system obtains equipment end operation data, patient monitoring data and equipment operation parameter data, and generates equipment end operation feature data, patient monitoring feature data and equipment operation feature data through feature extraction; multivariable anomaly detection is carried out based on the equipment end operation feature data and the patient monitoring feature data, and equipment end operation anomaly is identified; and performing equipment response consistency detection based on the equipment end operation feature data and the equipment operation feature data, and detecting equipment parameter execution abnormality. And abnormal attribution is enhanced through a knowledge graph, a historical case and an equipment operation mode are combined, potential reasons of anomalies are analyzed, and medical resource abnormal data are generated. And the system sends abnormal information to the medical resource cloud platform, so that intelligent early warning is realized, the equipment utilization rate is improved, and reasonable allocation of medical resources is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of medical resource management, and in particular to a medical resource management method and system. Background Art

[0002] In the modern medical system, as an important life support device in the fields of ICU, emergency department, operating room and home medical care, the operating state of the ventilator directly affects the treatment effect of patients. There are technical bottlenecks in the existing ventilator management methods, resulting in lagged detection of equipment abnormalities, low maintenance efficiency, and unreasonable resource allocation. During the operation of the ventilator, common abnormalities such as sensor failures, pipeline leaks, and humidifier blockages need to be manually checked by medical staff, which takes a long time. The traditional maintenance method mainly relies on regular inspections or manual repairs after a failure occurs, resulting in low equipment repair efficiency and affecting the continuity of medical resource use. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention proposes a medical resource management method and system to solve at least one of the above technical problems.

[0004] The present application provides a medical resource management method, including the following steps:

[0005] Step S1: Obtain device-side operation data, patient monitoring data, and device operation parameter data;

[0006] Step S2: Extract device-side operation characteristics from the device-side operation data to obtain device-side operation characteristic data; extract patient monitoring characteristics from the patient monitoring data to obtain patient monitoring characteristic data; extract device operation parameter characteristics from the device operation parameter data to obtain device operation characteristic data;

[0007] Step S3: Identify device-side operation abnormalities based on the device-side operation characteristic data and the patient monitoring characteristic data to obtain device-side operation abnormality data; identify device-side parameter execution abnormalities based on the device-side operation characteristic data and the device operation characteristic data to obtain device-side parameter execution abnormality data;

[0008] Step S4: Perform knowledge graph enhanced abnormality attribution based on the device-side operation abnormality data and the device-side parameter execution abnormality data to obtain medical resource abnormality data, so as to send it to the medical resource cloud platform for medical resource risk warning operations.

[0009] In the present invention, the comprehensive perception ability of the device operation state is improved through multi-source data collection. The operation characteristics of the device end, the patient monitoring characteristics, and the device operation characteristics are extracted to construct high-dimensional device state data, enabling the system to accurately judge the operation state of the device and improving the refined level of device management. Based on the device operation characteristic data and the patient monitoring characteristic data, the abnormal operation of the device end is identified, and combined with the device operation characteristic data, the abnormal execution of the device end parameters is identified, realizing the intelligent detection of the abnormal state. Through the correlation analysis of the knowledge graph, historical cases can be automatically matched when an abnormality occurs, and the root cause of the abnormality can be inferred based on the operation data of the device, thereby improving the efficiency of device maintenance and reducing the need for manual intervention. The present invention can carry out medical resource risk warning based on the medical resource cloud platform and realize intelligent medical resource scheduling. When the system detects device abnormalities or potential risks, it can automatically send warnings to the cloud medical resource management system.

[0010] Preferably, step S2 is specifically as follows:

[0011] Construct the device operation state space according to the device end operation data to obtain the device end state space data;

[0012] Perform topological simplification on the device end state space data to obtain the device end topology data;

[0013] Extract topological invariants from the device end topology data to obtain the device operation characteristic data;

[0014] Extract periodic characteristics from the patient monitoring data to obtain the patient monitoring characteristic data;

[0015] Perform hypersurface mapping of the device operation parameters according to the device operation parameter data to obtain the device operation parameter hypersurface data;

[0016] Extract the hypersurface curvature characteristics from the device operation parameter hypersurface data to obtain the device operation characteristic data.

[0017] In the present invention, through the construction of the device operation state space, topological simplification, and extraction of topological invariants, the accurate modeling of the device operation state is realized, and the accuracy of abnormal recognition is improved. The periodic characteristic extraction is used to improve the time series analysis ability of the patient monitoring data and enhance the correlation between the device and the patient state. Through the hypersurface mapping and curvature characteristic extraction of the device operation parameters, the modeling of the device operation behavior is optimized, the abnormal operation mode is effectively identified, and the intelligent level of device management is improved. The present invention improves the accuracy of data processing, calculation efficiency, and model generalization ability, and enhances the stability and predictability of medical devices.

[0018] Preferably, the construction of the device operation state space is specifically as follows:

[0019] Extract multi-variable features of the device based on the device-side operation data to obtain multi-variable feature data of the device operation;

[0020] Construct a feature correlation matrix based on the multi-variable feature data of the device operation to obtain feature correlation matrix data;

[0021] Perform device state space modeling on the feature correlation matrix data to obtain device state space data;

[0022] Perform state space mapping based on the device state space data to obtain device-side state space data.

[0023] In the present invention, the data integrity of the device operation state is improved through device multi-variable feature extraction, ensuring the accurate acquisition of key parameters. Based on the construction of the feature correlation matrix, the correlation analysis ability between different device parameters is enhanced, and the accuracy of device state prediction is improved. Through state space modeling, a high-dimensional device operation state distribution is constructed to achieve an accurate characterization of the operation mode. Combined with state space mapping, the data structure is optimized, making the device-side state information more intuitive and efficient for anomaly detection and predictive maintenance, and improving the intelligent level of medical device management.

[0024] Preferably, the topological simplification is specifically as follows:

[0025] Construct a device state space graph based on the device-side state space data to obtain device state space graph data;

[0026] Perform persistent homology analysis on the device state space graph data to obtain persistent homology data;

[0027] Perform persistence diagram calculation based on the persistent homology data to obtain persistence diagram data;

[0028] Perform topological clustering on the persistence diagram data to obtain topological clustering data;

[0029] Perform state classification on the device-side topological graph data according to the topological clustering data to obtain device-side topological graph data.

[0030] In the present invention, through the construction of the device state space graph, the topological structure of the device operation state is accurately characterized, ensuring the comprehensive expression of complex states. Persistent homology analysis is used to extract stable topological features, improving the ability to distinguish long-term stable patterns and abnormal short-term fluctuations. Combined with persistence diagram calculation, the persistence measurement of device state features is optimized, enhancing the interpretability of the device state evolution trend. Similar state patterns are identified through topological clustering, making the device state classification more accurate. State classification optimizes the device-side topological graph, reducing the data dimension and improving the calculation efficiency, providing more accurate topological feature support for device anomaly detection and prediction.

[0031] Preferably, the extraction of topological invariants is specifically as follows:

[0032] Calculate the number of connected components based on the topological graph data of the device end to obtain the number-of-connected-components data;

[0033] Calculate the cyclic pattern based on the topological graph data of the device end to obtain the cyclic pattern data;

[0034] Calculate the hole structure based on the topological graph data of the device end to obtain the hole structure data;

[0035] Calculate the Euler characteristic based on the number-of-connected-components data, the cyclic pattern data, and the hole structure data to obtain the Euler characteristic data;

[0036] Calculate the topological persistent entropy based on the topological graph data of the device end to obtain the topological persistent entropy data;

[0037] Vectorize based on the Euler characteristic data and the topological persistent entropy data to obtain the device-end operation characteristic data.

[0038] In the present invention, through the calculation of the number of connected components, the independence of the device operation state is accurately identified, and the accuracy of anomaly pattern detection is improved. By using the cyclic pattern calculation, the periodic structure in the device state change is extracted to optimize the operation trend analysis. Combining with the hole structure calculation, the high-dimensional topological features among the device operation parameters are identified to enhance the integrity of the device state characterization. Based on the Euler characteristic calculation, global topological information is provided to improve the reliability of state classification. By calculating the topological persistent entropy, the complexity and stability of the device state are evaluated to optimize the robustness of anomaly detection. Vectorize the device-end operation characteristics to enhance the computational efficiency of the data, making the device anomaly detection and state prediction more accurate and stable.

[0039] Preferably, the hypersurface mapping of the device operation parameters is specifically as follows:

[0040] Perform high-dimensional topological embedding based on the device operation parameter data to obtain the parameter high-dimensional topological data;

[0041] Perform manifold learning dimensionality reduction on the parameter high-dimensional topological data to obtain the high-dimensional manifold structure data;

[0042] Calculate the curvature of the high-dimensional manifold structure data to obtain the parameter curvature data;

[0043] Perform Gaussian-Bonnet curve mapping based on the parameter curvature data to obtain the device operation parameter hypersurface data.

[0044] In the present invention, through high-dimensional topological embedding, a non-linear high-dimensional structure of device operation parameters is constructed to improve the representation ability of complex parameter relationships. Manifold learning is used for dimensionality reduction to reduce the data dimension while preserving key features, improving the calculation efficiency and parameter optimization accuracy. Through curvature calculation, the geometric features of the changes in device operation parameters are extracted to enhance the perception ability of abnormal adjustment patterns. Combining with the Gauss-Bonnet curve mapping, the hypersurface structure of device operation parameters is accurately characterized, making the device state modeling more intuitive and providing efficient and reliable data support for misoperation identification, device optimization, and anomaly detection.

[0045] Preferably, the specific extraction of hypersurface curvature features is as follows:

[0046] Perform the inner product calculation of the tangent vectors on the hypersurface data of device operation parameters to obtain the first fundamental form data;

[0047] Perform the calculation of the rate of change of the normal vector on the hypersurface data of device operation parameters to obtain the second fundamental form data;

[0048] Perform the Gauss curvature calculation based on the first fundamental form data and the second fundamental form data to obtain the Gauss curvature data;

[0049] Perform the mean curvature calculation based on the Gauss curvature data to obtain the mean curvature data;

[0050] Perform the Riemann curvature tensor calculation based on the hypersurface data of device operation parameters to obtain the Riemann curvature tensor data;

[0051] Perform vectorization based on the first fundamental form data, the second fundamental form data, the Gauss curvature data, the mean curvature data, and the Riemann curvature tensor data to obtain the device operation feature data.

[0052] In the present invention, the local geometric characteristics of device operation parameters are extracted through the inner product calculation of tangent vectors to ensure the continuity and differentiability of parameter mapping. The calculation of the rate of change of the normal vector is used to identify the dynamic changes of device operation modes, improving the stability analysis ability of parameter adjustment trends. Combining with the Gauss curvature calculation to quantify the local bending degree of device parameter operations, accurately identifying abnormal adjustment regions. Through the mean curvature calculation to evaluate the overall smoothness of device parameter adjustment, optimizing the sensitivity of misoperation detection. Using the Riemann curvature tensor calculation to extract the global geometric features of device operation states, improving the accuracy of device state modeling. Vectorize the device operation feature data, optimize the data expression, improve the calculation efficiency, and make the device state evaluation, anomaly detection, and parameter optimization more accurate and reliable.

[0053] Preferably, step S3 is specifically as follows:

[0054] Perform multivariate anomaly detection based on the device-end operation feature data and the patient monitoring feature data to obtain the device-end operation anomaly data;

[0055] Perform device response consistency detection based on the device - side operation characteristic data and the device operation characteristic data to obtain device response consistency data;

[0056] Perform misoperation analysis based on the device response consistency data to obtain device - side parameter execution abnormal data.

[0057] In the present invention, through multi - variable anomaly detection, the device - side operation characteristic data and the patient monitoring characteristic data are fused to achieve accurate identification of complex abnormal patterns and improve the accuracy of device anomaly detection. Device response consistency detection is adopted to analyze the matching degree between device parameter adjustment and the actual operation state, ensure that the device response conforms to the set logic, and enhance the ability to identify abnormal responses. Combined with misoperation analysis, non - standard operation modes are identified based on the device response consistency data to detect anomalies caused by mis - setting, parameter drift, and human intervention, and improve the intelligence level of misoperation detection. By means of data - driven methods, device anomaly identification and parameter anomaly attribution are optimized to enhance the security, stability, and intelligence of device management.

[0058] Preferably, step S4 is specifically as follows:

[0059] Associate the device - side operation abnormal data and the device - side parameter execution abnormal data according to the preset device knowledge graph data to obtain device knowledge graph association data;

[0060] Perform knowledge graph reasoning based on the device knowledge graph association data to obtain knowledge graph reasoning data;

[0061] Perform device anomaly impact calculation based on the knowledge graph reasoning data to obtain medical resource abnormal data, and send it to the medical resource cloud platform for medical resource risk warning operations.

[0062] In the present invention, through device knowledge graph association, the device - side operation abnormal data and the parameter execution abnormal data are intelligently matched to construct an association relationship, improving the systematicness of anomaly analysis. Knowledge graph reasoning is adopted to accurately infer the potential causes of anomalies based on historical cases, expert rules, and data patterns, improving the accuracy and interpretability of anomaly attribution. Combined with device anomaly impact calculation, the impact range of device failures on medical resources is quantified to optimize device maintenance and scheduling strategies. Through risk warning on the medical resource cloud platform, intelligent early warning of device anomalies and resource optimization are realized, improving the management efficiency and operation safety of medical devices.

[0063] Preferably, the present application also provides a medical resource management system for executing the medical resource management method as described above. The medical resource management system includes:

[0064] The medical resource edge - side data acquisition module is used to obtain device - end operation data, patient monitoring data, and device operation parameter data;

[0065] The medical resource edge - side data feature extraction module is used to extract device - end operation features from device - end operation data to obtain device - end operation feature data; extract patient monitoring features from patient monitoring data to obtain patient monitoring feature data; extract device operation parameter features from device operation parameter data to obtain device operation feature data;

[0066] The medical resource edge - side data anomaly judgment module is used to identify device - end operation anomalies based on device - end operation feature data and patient monitoring feature data to obtain device - end operation anomaly data; identify device - end parameter execution anomalies based on device - end operation feature data and device operation feature data to obtain device - end parameter execution anomaly data;

[0067] The medical resource edge - side data anomaly attribution module is used to perform knowledge - graph - enhanced anomaly attribution based on device - end operation anomaly data and device - end parameter execution anomaly data to obtain medical resource anomaly data, so as to send it to the medical resource cloud platform for medical resource risk warning operations.

[0068] The present invention realizes precise device management and intelligent early warning. By adopting device - end operation feature extraction, patient monitoring feature extraction, and device operation feature extraction, it improves the comprehensiveness and accuracy of medical device status assessment. Combining multivariate anomaly detection and device response consistency analysis, it optimizes device operation monitoring, accurately identifies abnormal patterns, and reduces the false alarm rate. Through knowledge - graph - enhanced anomaly attribution, based on historical cases and device operation patterns, it realizes efficient anomaly diagnosis and improves the intelligent level of medical device management. Through risk warning on the medical resource cloud platform, it optimizes device maintenance scheduling, improves resource utilization rate, reduces the impact of medical device failures on hospital operations, and enhances the stability and reliability of the medical system. Brief Description of the Drawings

[0069] By reading the detailed description of the non - restrictive embodiments with reference to the following drawings, other features, purposes, and advantages of the present application will become more obvious:

[0070] Figure 1 Shows the step - flow chart of a medical resource management method in an embodiment;

[0071] Figure 2 Shows the step - flow chart of a medical resource edge - side data feature extraction method in an embodiment;

[0072] Figure 3 Shows the step - flow chart of a medical resource edge - side data anomaly judgment method in an embodiment;

[0073] Figure 4 The flowchart shows the steps of a method for attributing data anomalies at the edge of medical resources in one embodiment. Detailed implementation

[0074] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative work fall within the scope of protection of the present invention.

[0075] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. Functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0076] It should be understood that although terms such as "first", "second", etc. may be used here to describe each unit, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit can be called the second unit, and similarly, the second unit can be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed associated items.

[0077] There are 20 ventilators in the ICU of a certain hospital, 15 of which are in use and 5 are in reserve. Five ventilators in use triggered high PEEP alarms (>12 cmH2O) within the past 24 hours. The oxygen saturation (SpO2) of 3 patients decreased by more than 5%. Patient A: PEEP = 14 cmH2O, SpO2 decreased from 98% to 92% (abnormal). Patient B: PEEP = 13 cmH2O, SpO2 decreased from 96% to 89% (high risk). Patient C: PEEP = 12.5 cmH2O, SpO2 decreased by 3% (possibly abnormal). It was found that the medical staff adjusted the PEEP of 3 ventilators 5 times within 30 minutes, and the parameter changes were abnormal. During the adjustment of PEEP, the tidal volume (Vt) of the equipment fluctuated by more than 20%, indicating unstable parameter settings. Through knowledge graph reasoning, it was found that in the past 6 months, 80% of the PEEP abnormal cases of similar models of ventilators were related to humidifier failures. In the current ICU, 80% of the abnormal ventilator humidifiers had a usage time > 1000 hours (close to the maintenance threshold). 80% of the possible causes → humidifier blockage; 15% of the possible causes → incorrect parameter settings by medical staff; 5% of the possible causes → equipment sensor drift. Three ventilators in the current ICU are in a high-risk state, affecting 3 critically ill patients. Among the 5 reserve ventilators, 2 have just been maintained and are available first. The dispatching suggestion is to allocate 2 reserve ventilators to Patients A and B to reduce the impact of abnormal PEEP settings on the patients. It is recommended that the medical staff check the condition of the humidifier and replace it if it is confirmed to be blocked. Send a remote maintenance request to notify the equipment manufacturer to check the equipment sensor drift. The system sends an ICU risk alarm to the medical resource cloud platform: the current abnormal rate of ventilators in the ICU = 15% (3 / 20), exceeding the warning line (10%). It is recommended that the hospital arrange the maintenance of 2 ventilators in advance to prevent equipment shortages in the ICU within the next 48 hours.

[0078] Please refer to Figures 1 to 4 , this application provides a medical resource management method, including the following steps:

[0079] Step S1: Obtain device-side operation data, patient monitoring data, and device operation parameter data;

[0080] In one embodiment, data such as air flow rate, pressure, oxygen concentration, temperature and humidity, and vibration are collected through sensors installed on devices such as ventilators and monitors, and uploaded to the data processing system through wireless communication (such as Wi-Fi, 5G). Data is obtained from the patient's physiological monitoring equipment, such as heart rate, oxygen saturation (SpO2), respiratory rate (RR), carbon dioxide (EtCO2), etc., to evaluate the patient's ventilation status. Record the operation instructions of medical staff, such as the adjustment of parameters such as tidal volume (VT), respiratory mode (VCV, PCV), inspiratory oxygen concentration (FiO2), respiratory rate (RR), etc.

[0081] Step S2: Extract device - end operation characteristics from the device - end operation data to obtain device - end operation characteristic data; extract patient monitoring characteristics from the patient monitoring data to obtain patient monitoring characteristic data; extract device operation parameter characteristics from the device operation parameter data to obtain device operation characteristic data;

[0082] In one embodiment, signal analysis is performed on the collected device - end operation data. For example, the Fourier transform (FFT) is used to extract the frequency characteristics of the airflow and pressure waveforms. Time - series models (LSTM, HMM) are used to identify changes in the operation mode of the device. Statistical characteristics (mean, variance, slope, etc.) are calculated to characterize the device state. Machine learning methods (such as PCA dimensionality reduction, Autoencoder, etc.) are used to extract the change trends of key physiological parameters. The behaviors of medical staff in adjusting the ventilator parameters are recorded, such as whether the frequency and amplitude of device parameter adjustment are abnormal.

[0083] Step S3: Identify device - end operation anomalies based on the device - end operation characteristic data and the patient monitoring characteristic data to obtain device - end operation anomaly data; identify device - end parameter execution anomalies based on the device - end operation characteristic data and the device operation characteristic data to obtain device - end parameter execution anomaly data;

[0084] In one embodiment, an anomaly detection algorithm (such as DBSCAN, isolation tree algorithm) is used to identify the abnormal state of the device. For example, leak detection: the volume ratio of airflow input and output is abnormal. Blockage detection: the airway pressure suddenly increases but the airflow decreases. Motor failure: the vibration signal of the device changes abnormally. Device - end parameter execution anomaly identification detects abnormal behaviors in device parameter adjustment through a time - series prediction model (such as LSTM, GRU). After the medical staff adjusts the tidal volume, the patient's SpO2 still continues to decline, indicating improper parameter settings. The mode of device adaptive adjustment deviates from the historical normal operation range, indicating a software failure.

[0085] Step S4: Perform knowledge - graph enhanced anomaly attribution based on the device - end operation anomaly data and the device - end parameter execution anomaly data to obtain medical resource anomaly data, so as to send it to the medical resource cloud platform for medical resource risk warning operations.

[0086] In one embodiment, a knowledge graph associating the operation state of the ventilator, patient physiological data, and device operation records is established, including device abnormal modes (such as leakage, pipeline blockage, motor failure). Patient abnormal manifestations (such as hypoxemia, respiratory failure). Operation behaviors (such as improper parameter adjustment). A graph neural network (GNN) is used to analyze the potential causes of anomalies. where is the feature representation of node v at the l + 1 layer, σ is the activation function, W(l) is the training weight matrix, u is the neighbor node, is the feature of neighbor node u at the l-th layer, b (l) is the bias term. For example, if airway obstruction is detected, the knowledge graph can provide reasons (such as humidifier water accumulation, patient sputum blockage). If a decrease in SpO2 is detected, the system prompts whether it is caused by improper ventilator parameter settings and provides adjustment suggestions. Combining historical data with the current abnormal situation, intelligent alarms are pushed. Low risk (recommended for monitoring): slight parameter deviation. Medium risk (recommended for adjustment): patient hypoxemia, but the device is operating normally. High risk (urgent intervention): severe device failure or acute respiratory failure of the patient.

[0087] Preferably, step S2 is specifically as follows:

[0088] Step S21: Construct the device operation state space based on the device-side operation data to obtain the device-side state space data;

[0089] In one embodiment, through the phase space reconstruction and Markov chain discretization method, the device operation data is transformed into a state space for further topological analysis and device operation mode modeling. Key state variables are selected (such as air flow rate, airway pressure, oxygen concentration, humidity, etc.). Using phase space reconstruction, the time series signal is embedded into a high-dimensional state space. For example, let the airway pressure P(t) be the state variable, and the state vector is constructed through the time delay coordinate: X(t) = [P(t), P(t - τ), P(t - 2τ), …, P(t - nτ)], where X(t) is the state vector representing the system state at a certain moment, P(t) is the airway pressure reflecting the patency of the airway and the patient's breathing state, P(t - τ) is the airway pressure value at time t - τ, P(t - 2τ) is the airway pressure value at time t - 2τ, P(t - nτ) is the airway pressure value at time t - nτ, where τ is the time delay and n is the embedding dimension. The continuous state is discretized through the Markov chain, so that the state space has a finite number of state nodes for topological analysis. For example, different airway pressure intervals are set: low pressure (0 - 5 cmH2O), normal (5 - 15 cmH2O), high pressure (> 15 cmH2O). The transition probabilities of different states are statistically calculated to form a state transition matrix, and the discrete state space is constructed.

[0090] Step S22: Simplify the topology according to the device-side state space data to obtain the device-side topology graph data;

[0091] In one embodiment, principal component analysis (PCA), t-SNE, or UMAP is used for topological structure dimensionality reduction to extract the main influencing factors. For example, the original state dimension of device operation is (10000, 20) (i.e., 10,000 pieces of data, each with 20-dimensional features). After dimensionality reduction by PCA to (10000, 3), only the main features are retained. Regarding the state space as a complex network, a community detection algorithm is used for partitioning. For example, a similarity graph between "state points" is established. The minimum spanning tree (MST) is calculated to reduce redundant connections and extract the core topological structure.

[0092] Each point in the state space represents the state of the system at a certain time point and contains multiple variables, such as X = [P, V, O2, T,...], where X is the point in the state space, P is the airway pressure, V is the tidal volume, O2 is the oxygen concentration, and T is the time. Each state point is regarded as a node in the graph, and the similarity (or distance) between two state points is used as the weight of the edge. The similarity between each other is calculated by cosine similarity. If the distance between state points is small, they belong to the same state cluster. The graph is constructed using the k-nearest neighbor (k-NN) or fully connected method. For each state point X i , the k nearest neighbors are found to establish edges. The weight is a similarity measure (such as the negative exponential transformation of the Euclidean distance).

[0093] Each node is regarded as an independent community. Traverse each node and try to move it into an adjacent community, calculating the change in modularity. Modularity is used to measure the quality of community partitioning in the network. Q is the modularity, measuring the quality of community partitioning, m is the total number of edges in the graph, i is the row index of the graph, j is the column index of the graph, A ij is the adjacency matrix, k i is the row degree of the node, k j is the column degree of the node, δ(c i , c j ) is 1 if i and j belong to the same community, otherwise 0. c i is the community label of node i, and c j is the community label of node j. Nodes in the same community are merged to form a new simplified graph, and the modularity change calculation is performed again. Repeat the steps until the modularity no longer improves.

[0094] The core topology is extracted through the minimum spanning tree and sorted in ascending order of edge weights. The edge with the smallest weight is selected in turn. If no loop is formed after adding it, it is retained. Until all nodes are connected. Calculate the minimum spanning tree within the community to remove redundant connections and make the topological structure more compact.

[0095] Step S23: Extract topological invariants based on the device-side topology graph data to obtain device-side operation characteristic data;

[0096] In one embodiment, the topological invariant is extracted as follows: The Betti numbers are extracted to represent the connectivity of the topological structure: β0: the number of connected components (the number of independent states). β1: the number of loops (the complexity of state changes). For example, low complexity: β0 = 1, β1 = 0 (indicating a stable state space). High complexity: β0 = 3, β1 = 5 (indicating that the system has multiple operating modes).

[0097] In one embodiment, the state complexity is measured by calculating the topological entropy to achieve the extraction of topological invariants, and the device-side operation characteristic data is obtained. H = -∑p i log p i , where H is the topological entropy, representing the system state complexity, p i is the occurrence probability of the i-th state, and log is the logarithmic function in the information entropy calculation formula. The topological entropy of the normal operating state is relatively low (indicating regular state changes). When the device fails, the topological entropy suddenly increases (indicating disordered state changes).

[0098] Step S24: Periodically extract features from the patient monitoring data to obtain patient monitoring feature data;

[0099] In one embodiment, wavelet transform is performed on the respiratory signal to extract periodic components at different scales. High-frequency signals reflect short-term acute events (such as the periodic representation of the characteristics of apnea). Low-frequency signals reflect long-term trends (such as the periodic representation of the characteristics of chronic obstructive pulmonary disease). X(a, b) is the transformation result, x(t) is the original signal, ψ * is the complex conjugate of the function ψ, t is the integral component, a is the scale factor, and b is the translation factor.

[0100] In one embodiment, X(f) is the spectrum function, x(t) is the original signal, e is the base of the natural logarithm, j is the imaginary unit, π is the pi, f is the frequency, and t is the time variable. Calculate the respiratory frequency spectrum of the patient and extract the dominant frequency. Normal breathing: The main frequency is in the range of 12 - 20 Hz. Abnormal breathing (apnea, cardiogenic respiration): The main frequency decreases or disappears.

[0101] In one embodiment, empirical mode decomposition is used to decompose the respiratory waveform into different intrinsic mode functions (IMFs). Calculate the instantaneous frequency of the IMFs and analyze the periodic changes of the respiratory pattern.

[0102] Step S25: Perform a hypersurface mapping of the device operation parameters based on the device operation parameter data to obtain device operation parameter hypersurface data;

[0103] In one embodiment, multiple device operation parameters (such as tidal volume, respiratory rate, oxygen concentration) are selected as dimensions to construct a high-dimensional hypersurface. Let the operation parameter vector be X = [VT ,RR,F i O2, PEEP], V T where V is the tidal volume, RR is the respiratory rate, F i O2 is the inspired oxygen concentration, PEEP is the positive end-expiratory pressure, and the mapping relationship of these parameters to the patient's physiological response is fitted by multiple regression: Y = f(X) + ε, where Y represents the patient's SpO2 or respiratory mechanics parameters, and f(X) is a high-dimensional mapping function.

[0104] In one embodiment, a deep neural network (DNN) is used to learn the hypersurface mapping of the parameter space. The input is set as: device operation parameters (tidal volume, respiratory mode, pressure). The output is: patient physiological responses (blood oxygen, respiratory rate). The training objective is to minimize the prediction error. Through convolutional layers, pooling layers, fully connected layers, and weight calculations for iterative fitting, a deep neural network model is obtained to achieve the hypersurface mapping of the parameter space.

[0105] Step S26: Extract hypersurface curvature features from the hypersurface data of the device operation parameters to obtain device operation feature data.

[0106] In one embodiment, the Gaussian curvature at different points of the device parameter hypersurface is calculated. K is the Gaussian curvature, which describes the degree of curvature of the hypersurface at a certain point. f is the high-dimensional mapping function of the device operation parameter hypersurface, representing the influence of device parameters on the patient's physiological response. x is the device operation parameter, such as tidal volume, and y is another device operation parameter, such as inspired oxygen concentration. is the derivative symbol. If the curvature is small, it indicates that the parameter adjustment has a stable impact on the patient. If the curvature is large, it indicates that the parameter adjustment causes significant physiological fluctuations (unstable state). Calculate the principal curvature of the hypersurface. |k1| >> |k2| indicates that the system is in a strongly nonlinear region (sensitive to parameter adjustment). |k1| ≈ |k2| indicates that the system is in a stable region (the effect of parameter adjustment is predictable).

[0107] Preferably, the construction of the device operating state space is specifically as follows:

[0108] Extract multi-variable features of the device based on the device-side operation data to obtain device operation multi-variable feature data;

[0109] In one embodiment, the device operating state variables include pressure: airway pressure, supply pressure, flow rate: tidal volume of the ventilator, inhalation / exhalation flow rate, oxygen concentration: FiO2, temperature: sensor temperature, ambient temperature, vibration: motor operating state, current: device power consumption, and multi-variable feature vector: X = [P, F, O2, T, V, I], where P is pressure, F is flow rate, O2 is oxygen concentration, T is temperature, V is vibration, and I is current. Features are extracted using a sliding window. uX is the average value feature, w is the sliding window length, i is the data point sequence term, and X i is the i-th data point, σ X is the degree of fluctuation of the data within the window, max(X) is the maximum value of the data within the window, and max(X) - min(X) is the difference between the maximum and minimum values of the data within the window.

[0110] Construct a feature correlation matrix based on the multi-variable characteristic data of the device operation to obtain the feature correlation matrix data;

[0111] In one embodiment, calculate the correlation between each feature and establish a correlation matrix, where: R ij is the correlation coefficient between feature X i and feature X j , with a value range of (0,1](0,1](0,1]. The larger the value, the stronger the correlation. d(X i ,X j ) is the Euclidean distance between feature X i and X j (used to measure the similarity of variables). X ik is the value of the k-th sample in the variable X i dimension, and X jk is the value of the k-th sample in the variable X j dimension. k is an adjustment term used to adjust the influence degree of different variables.

[0112] Perform device state space modeling on the feature correlation matrix data to obtain the device state space data;

[0113] In one embodiment, establish a graph structure of the device state space through the feature correlation matrix to represent the topological structure of the device operation state. The device state is regarded as a node, and the feature variables with higher correlation are connected as edges. If the correlation between two variables is high then establish a connection in the topological graph to obtain the device state space data.

[0114] Perform state space mapping based on the device state space data to obtain the device-side state space data.

[0115] In one embodiment, the state space data is mapped to a low-dimensional representation. Calculate the Laplacian matrix of the graph: L = D - A, where L is the Laplacian matrix of the graph, D is the degree matrix (the diagonal elements are the degrees of the nodes), and A is the adjacency matrix. Construct a low-dimensional embedding through the eigenvectors of the Laplacian matrix. Select the first two eigenvectors, then each state point can be mapped to a 2D plane, and the device state is mapped from the original high-dimensional features (such as 6D) to a 2D / 3D low-dimensional representation.

[0116] Preferably, the topological simplification is specifically as follows:

[0117] Construct a device state space graph according to the device-side state space data to obtain device state space graph data;

[0118] In one embodiment, a device state space graph is generated from the device state space data to represent the relationship between different states of the device in a graph structure. The state points are regarded as the nodes of the graph. The similarity between the state points is used as the weight of the edge. Adopt the K-nearest neighbor (k-NN) or complete graph construction method. If the similarity between the state points is greater than a certain threshold, then connect them. Where S(X i , X j ) is the similarity between state points i and j, which can be calculated using the Euclidean distance or cosine similarity, and Θ is the set similarity threshold.

[0119] Perform persistent homology analysis on the device state space graph data to obtain persistent homology data;

[0120] In one embodiment, through the topological data analysis (TDA) method, extract topological features from the device state space graph and analyze the topological relationship between different states. Persistent homology analyzes the topological structures in different dimensions: 0D: connected components; 1D: loops; 2D: cavities. Calculate persistent homology using the VR complex. Set the radius r. If d(X i , X j ) < r, then establish simplices between the state points. Calculate the Betti numbers: β0 = the number of connected components, β1 = the number of loops, β2 = the number of cavities.

[0121] Perform persistent diagram calculation according to the persistent homology data to obtain persistent diagram data;

[0122] In one embodiment, a persistent bar chart and a persistence diagram are calculated to visualize the survival times of different topologies. The persistent bar chart shows the durations of different topologies varying with scale. The persistence diagram represents the birth time and death time of topologies as a scatter plot, facilitating the observation of important topological features. Persistent bar chart: X-axis: radius parameter, Y-axis: duration of topological features. Persistence diagram: X-axis: emergence time of topologies. Y-axis: disappearance time of topologies. Construct birth-death coordinates to plot the persistence diagram.

[0123] Perform topological clustering on the persistence diagram data to obtain topological clustering data;

[0124] In one embodiment, topological clustering is performed based on persistent data, grouping state points with similar topological features into one category. On the persistence diagram, points close to the diagonal represent noise. Points far from the diagonal represent important topological features (i.e., stable device states). Cluster points with similar coordinate structures to discover categories of device operation modes and identify abnormal states. Use HDBSCAN for topological clustering: HDBSCAN is suitable for high-dimensional data and performs clustering based on topological persistence features. Set the minimum cluster size and automatically adjust the shape of the clusters.

[0125] Classify the device-side topology graph data according to the topological clustering data to obtain the device-side topology graph data.

[0126] In one embodiment, according to the topological clustering result, classify the device-side state points to generate the device-side topology graph data. Map the topological clustering result to the device state graph: Color the device state points according to the clustering labels to form the device-side topology graph.

[0127] Preferably, the extraction of topological invariants is specifically as follows:

[0128] Calculate the number of connected components based on the device-side topology graph data to obtain the number-of-connected-components data;

[0129] In one embodiment, calculate the number of connected components in the device-side topology graph to measure the degree of separation of the network. Let the topology graph be G, then the number of connected components β0 (in the graph G, the number of the largest connected subgraphs that are mutually reachable) is: β0 = number of connected subgraphs in G. If β0 = 1, the entire graph is connected; if β0 > 1, it means that the device state is split into multiple independent regions. Use breadth-first search (BFS) or depth-first search (DFS) to traverse the unvisited nodes and count the number of independent connected subgraphs to obtain the number-of-connected-components data.

[0130] Calculate the cyclic pattern based on the device-side topology graph data to obtain the cyclic pattern data;

[0131] In one embodiment, the cyclic pattern (Betti-1 number) in the computational topology graph is calculated to measure the cyclic behavior of the device state. The first Betti number β1 (the number of independent cycles in the graph G, used to describe the cyclic behavior of the network), that is, the number of cyclic patterns: β1 = number of independent cycles in G. For example, if β1 = 0, it means that there is no closed cycle in the device state. If β1 > 0, it means that the device has a periodic operation mode or an abnormal state loop. β1 = |E(G)| - |V(G)| + β0, where β1 is the number of independent cycles in the graph G, E(G) is the number of edges in the graph G, V(G) is the number of nodes in the graph G, and β0 is the number of connected components.

[0132] Calculate the void structure based on the device-side topology graph data to obtain the void structure data;

[0133] In one embodiment, the void structure (Betti-2 number) in the computational topology graph is calculated to measure the high-dimensional topological structure of the device state. The second Betti number β2 (the number of closed cavities in the network): β2 = number of voids in the network. For example, in three-dimensional space, β2 represents a closed cavity. If β2 > 0, it means that the device state forms a topological structure. β2 = |C(G)| - |F(G)| + |E(G)| - |V(G)| + β0, where C(G) is the number of three-dimensional closed bodies (such as closed cavities), F(G) is the number of two-dimensional faces (such as closed toroidal surfaces), E(G) is the number of edges in the graph G, V(G) is the number of nodes in the graph G, and β0 is the number of connected components.

[0134] Perform Euler characteristic calculation based on the connected component number data, cyclic pattern data, and void structure data to obtain the Euler characteristic data;

[0135] In one embodiment, the Euler characteristic is calculated to describe the overall topological structure of the device state. The Euler characteristic χ(G) is calculated from the Betti numbers: χ(G) = β0 - β1 + β2, where β0 is the number of connected components, β1 is the number of cyclic patterns, and β2 is the number of void structures. The Euler characteristic is a topological invariant that can reflect the basic topological characteristics of the network, including connectivity, cyclic structure, and the number of closed cavities.

[0136] Calculate the topological persistence entropy based on the device-side topology graph data to obtain the topological persistence entropy data;

[0137] In one embodiment, the topological persistence entropy is calculated to measure the stability of the topological structure. Let the proportion of the persistent homology feature p i in the persistence diagram: H topo = -∑p i log p i , Htopo For topological persistence entropy data, log is the logarithmic function (the natural logarithm log or the base-2 logarithm can be used), and p i is the relative persistence degree of topological feature i in the persistence diagram. l i is the persistence of topological feature i, representing the difference between the birth time and the death time at different scales. ∑l i is the sum of the persistences of all topological features, which is used for normalization. Calculate the topological persistence entropy to measure the stability of the topological structure. The topological persistence entropy comes from persistent homology and is used to characterize the persistence of topological features (such as connected components, cyclic patterns, and hole structures) at different scales, and reflects the topological complexity and stability of the system through entropy measurement.

[0138] Vectorize according to the Euler characteristic data and the topological persistence entropy data to obtain the device-side operation feature data.

[0139] In one embodiment, vectorize the Euler characteristic and the topological persistence entropy, and integrate the Euler characteristic and the topological persistence entropy into the device-side topological feature vector.

[0140] Preferably, the device operation parameter hypersurface mapping is specifically:

[0141] Perform high-dimensional topological embedding based on the device operation parameter data to obtain the parameter high-dimensional topological data;

[0142] In one embodiment, through the high-dimensional topological embedding method, place the device operation parameter data in the high-dimensional topological space to better represent its internal structure. Construct a K-nearest neighbor graph to represent the local topological structure between the parameters. Calculate the manifold distance between the parameters instead of the Euclidean distance.

[0143] Perform manifold learning dimensionality reduction based on the parameter high-dimensional topological data to obtain the high-dimensional manifold structure data;

[0144] In one embodiment, the manifold learning method is used to reduce the dimension of high-dimensional topological data, extract the key manifold structure, and avoid losing non-linear relationships. Such as t-SNE (t-Distributed Stochastic Neighbor Embedding), LLE (Locally Linear Embedding), Isomap (Isometric Mapping), which are used to preserve the manifold distance, and UMAP (Uniform Manifold Approximation), which is used for non-linear dimensionality reduction. A K-nearest neighbor graph is constructed, that is, for each parameter point, its K closest neighbors are found and connected by an edge to form a graph structure. The construction steps of the K-nearest neighbor graph are as follows: calculate the distance between all parameters. Here, the Euclidean distance is not directly used, but the geodesic distance or Mahalanobis distance suitable for manifold learning is selected to more accurately reflect the internal distribution of the data. Select the value of K (such as K = 5 or K = 10), determine the neighbors of each parameter point, and establish connections. In order to more accurately reflect the potential relationship of the parameters, the manifold distance is introduced instead of the simple Euclidean distance. The calculation method of the manifold distance is like locally linear embedding, which calculates the weighted linear combination based on the K neighbors of each data point, so as to map to the low-dimensional space. Or, t-SNE, which is used for dimensionality reduction of high-dimensional data and keeps the local neighbors in the low-dimensional space. Or, ISOMAP, which calculates the manifold distance through the shortest path algorithm and constructs a globally consistent low-dimensional representation. After completing the construction of the K-nearest neighbor graph and the calculation of the manifold distance, the data is mapped to the high-dimensional topological space to obtain the high-dimensional topological data representation.

[0145] Calculate the curvature of the high-dimensional manifold structure data to obtain the parameter curvature data;

[0146] In one embodiment, a suitable number of neighbors K (such as 5 or 10) is selected, and the K-nearest neighbor relationship between data points is calculated. A local manifold structure is constructed to establish the topological connection between points. Fit a local surface around each point to obtain the principal curvature. Calculate the Gaussian curvature and mean curvature corresponding to the local surface. Construct a differential Laplacian matrix for the data points and calculate the LBO value of each point, that is, calculate the local curvature of the high-dimensional manifold to obtain the parameter curvature data.

[0147] Perform a Gaussian-Bonnet curve mapping according to the parameter curvature data to obtain the device operation parameter hypersurface data.

[0148] In one embodiment, surface mapping is performed through the Gauss-Bonnet theorem to obtain the device operation parameter hypersurface data. Gauss-Bonnet theorem: f M KdA = 2πX(M), f M KdA is the total curvature on the surface M. K is the Gaussian curvature, which describes the local bending degree of the surface, π is the constant term of the circumference ratio, χ(M) is the Euler characteristic number, which describes the topological properties of the surface. The surface equation: S(x, y) = ∫Kdxdy. If K > 0, it means the surface is convex; if K < 0, it means the surface is saddle-shaped; if H > 0, it means the surface has a local maximum; if H < 0, it means the surface has a local minimum.

[0149] Preferably, the extraction of the hypersurface curvature feature is specifically as follows:

[0150] Calculate the inner product of the tangent vectors of the hypersurface data of the device operation parameters to obtain the first fundamental form data;

[0151] In one embodiment, I is the first fundamental form data, which is used to describe the local metric properties of the surface. It can quantify geometric properties such as length and angle on the surface. E is the first fundamental coefficient, representing the self-inner product of the tangent vector along the u direction, measuring the stretching degree of the surface in this direction. F is the mixed fundamental coefficient, representing the inner product of the tangent vector in the u direction and the tangent vector in the v direction measuring the change in the angle between these two directions. G is the second fundamental coefficient, representing the self-inner product of the tangent vector along the v direction measuring the stretching degree of the surface in this direction, S is the hypersurface data of the device operation parameters. u is the main parameter, representing the main change direction, that is, the main control variable along a specific path on the surface. S is the hypersurface data of the device operation parameters. v is the auxiliary parameter, representing the auxiliary change direction, that is, the secondary control variable on the surface, used to supplement the definition of u to make the surface a two-dimensional parametric surface.

[0152] Calculate the change rate of the normal vector of the hypersurface data of the device operation parameters to obtain the second fundamental form data;

[0153] In one embodiment, the second fundamental form is used to describe the bending degree of the surface. It measures the change rate of the normal vector on the surface, thereby characterizing the local curvature characteristics of the surface, L is the first curvature coefficient, representing the inner product of the second-order derivative along the u direction and the normal vector N, measuring the curvature change of the surface along this direction. M is the mixed curvature coefficient, representing the inner product of the second-order derivatives in the u direction and the v direction and the normal vector N, measuring the coupled bending degree of the surface in two directions, is the second curvature coefficient, representing the inner product of the second-order derivative along the v direction and the normal vector N, measuring the curvature change of the surface along this direction, N is the unit normal vector, representing the normal direction of the surface at this point.

[0154] Perform Gaussian curvature calculation based on the first fundamental form data and the second fundamental form data to obtain Gaussian curvature data;

[0155] In one embodiment, the Gaussian curvature is calculated from the first fundamental form data and the second fundamental form data, and Gaussian curvature data is obtained. det(II) is the second fundamental form matrix, and det(I) is the first fundamental form matrix. The parameter meanings are as described in the previous steps.

[0156] Based on the Gaussian curvature data, the mean curvature is calculated to obtain mean curvature data.

[0157] In one embodiment, the mean curvature is calculated from the Gaussian curvature data, and mean curvature data is obtained. H is the mean curvature, E is the first fundamental coefficient, N is the second curvature coefficient, G is the second fundamental coefficient, L is the first curvature coefficient, F is the mixed fundamental coefficient, and M is the mixed curvature coefficient.

[0158] Based on the hypersurface data of the device operation parameters, the Riemann curvature tensor is calculated to obtain Riemann curvature tensor data.

[0159] In one embodiment, the Riemann curvature tensor is calculated to describe the curvature change of the surface in different directions. R ijkl is a component of the Riemann curvature tensor, used to measure the curvature change in different directions on the surface. is the Christoffel symbol, x j is the partial derivative coordinate component. is the second Christoffel symbol, x k is the local coordinate component. is the connection transformation term. is the coordinate bending coefficient. is the local curvature interaction term. is the geodesic transformation term; the specific calculation method of the Christoffel symbol Γ: g im is the inverse metric tensor, g mj is the local metric tensor, x k is the partial derivative coordinate variable, g mk is the second local metric tensor, x j is the additional partial derivative coordinate variable, g jk is the mixed metric tensor, x m is the differential coordinate of the metric tensor, and the metric tensor is the first fundamental form data.

[0160] Based on the first fundamental form data, the second fundamental form data, the Gaussian curvature data, the mean curvature data, and the Riemann curvature tensor data, vectorization is performed to obtain device operation characteristic data.

[0161] In one embodiment, the first basic form data, the second basic form data, the Gaussian curvature data, the mean curvature data, and the Riemann curvature tensor data are integrated into a data group.

[0162] Preferably, step S3 is specifically as follows:

[0163] Step S31: Perform multivariate anomaly detection based on the device - end operation characteristic data and the patient monitoring characteristic data to obtain device - end operation anomaly data;

[0164] In one embodiment, multivariate anomaly detection is performed through the device - end operation characteristic data and the patient monitoring characteristic data to identify abnormal states (such as device failures or abnormal patient breathing). The Mahalanobis distance is used to calculate the anomaly of data points: D M (X)=(X - μ) T S -1 (X - μ), D M (X) is the Mahalanobis distance of the data point, X is the feature vector at the current moment, μ is the historical data mean vector, T is the transpose symbol, and S ―1 is the inverse matrix of the covariance matrix; a threshold is set. When D M (X) is greater than the threshold, the data point is determined to be abnormal. Alternatively, isolation forest is used for anomaly detection, and based on randomly splitting the data set, a few abnormal points are identified.

[0165] Step S32: Perform device response consistency detection based on the device - end operation characteristic data and the device operation characteristic data to obtain device response consistency data;

[0166] In one embodiment, after calculating the adjustment of the device parameters, the change in the device state is calculated, and a time - series window (such as 10s) is set. Calculate the mean change before and after the adjustment of the device operation parameters, that is, whether the mean change of the device - end operation characteristic data and the device operation characteristic data matches. If there is no obvious change after adjustment, it means that the device fails to respond to the operation.

[0167] Step S33: Perform misoperation analysis based on the device response consistency data to obtain device - end parameter execution anomaly data.

[0168] In one embodiment, based on the device response consistency data, misoperation situations are identified, including medical staff's misoperation (unreasonable device parameter adjustment), and device system anomalies (correct operation input but the device does not respond). Misoperation classifications include type 1: ineffective parameter adjustment, where the device operation parameters change, but the device state does not respond. Type 2: abnormal overshoot, where the device adjusts the parameters too much, resulting in a drastic change in the state. Type 3: adjustment lag, where after the device adjusts the parameters, the state response is delayed too long.

[0169] Preferably, step S4 is specifically as follows:

[0170] Step S41: Correlate the device - side operation anomaly data and the device - side parameter execution anomaly data according to the preset device knowledge graph data to obtain device knowledge graph correlation data;

[0171] In one embodiment, the device operation anomaly data and the device parameter execution anomaly data are correlated through the device knowledge graph to establish a causal relationship, improving the intelligence of anomaly analysis. The device knowledge graph G constructs the following triples: (device status, device operation, device response). Device status (nodes): normal, overloaded, failed. Device operation (edges): parameter adjustment, mode switching. Device response (nodes): normal, abnormal, lag. Define the inference rule as: That is, when the device performs a certain operation in a certain state, it leads to a new state and generates corresponding fault reasons. Calculate the similarity between the device anomaly data and the knowledge graph: A is the anomaly data vector (device - side operation anomaly data and device - side parameter execution anomaly data), B is the knowledge graph anomaly pattern vector, representing the typical anomaly patterns stored in the knowledge graph, ||A|| is the norm of vector A, ||B|| is the norm of vector B. If S(A, B)>θ, then it is determined that the anomaly is correlated with the knowledge graph.

[0172] Step S42: Perform knowledge graph reasoning based on the device knowledge graph correlation data to obtain knowledge graph reasoning data;

[0173] In one embodiment, reasoning is performed based on the knowledge graph relationship to find the potential causes of anomalies, improving the intelligence level of diagnosis. Set inference rules, such as Rule 1 (airway obstruction → ventilator alarm → cause: pipeline blockage), Rule 2 (excessive pressure → device overheating → cause: heat dissipation failure).

[0174] Step S43: Perform device anomaly impact calculation based on the knowledge graph reasoning data to obtain medical resource anomaly data, and send it to the medical resource cloud platform for medical resource risk warning operations.

[0175] In one embodiment, calculate the impact of device anomalies on medical resources and perform medical resource risk warnings. Calculate the impact degree based on the device anomaly weight. Let I be the medical resource anomaly impact: where W i is the device anomaly impact factor, and S i is the anomaly severity. Set the medical resource risk levels: low risk (L): I < 0.3, medium risk (M): 0.3 ≤ I < 0.7, high risk (H): I ≥ 0.7. Send data to the cloud platform through the API or MQTT protocol.

[0176] Preferably, the present application further provides a medical resource management system for implementing the medical resource management method as described above. The medical resource management system includes:

[0177] A medical resource edge - end data acquisition module for obtaining device - end operation data, patient monitoring data, and device operation parameter data;

[0178] A medical resource edge - end data feature extraction module for extracting device - end operation features from the device - end operation data to obtain device - end operation feature data; extracting patient monitoring features from the patient monitoring data to obtain patient monitoring feature data; extracting device operation parameter features from the device operation parameter data to obtain device operation feature data;

[0179] A medical resource edge - end data anomaly judgment module for identifying device - end operation anomalies based on the device - end operation feature data and the patient monitoring feature data to obtain device - end operation anomaly data; identifying device - end parameter execution anomalies based on the device - end operation feature data and the device operation feature data to obtain device - end parameter execution anomaly data;

[0180] A medical resource edge - end data anomaly attribution module for performing knowledge - graph - enhanced anomaly attribution based on the device - end operation anomaly data and the device - end parameter execution anomaly data to obtain medical resource anomaly data, so as to send it to the medical resource cloud platform for medical resource risk warning operations.

[0181] Therefore, from any perspective, the embodiments should be regarded as exemplary and non - restrictive. The scope of the present invention is defined by the appended application documents rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be encompassed within the present invention.

[0182] The above - described are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather will conform to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A medical resource management method, characterized in that, It includes the following steps: Step S1: Obtain device - end operation data, patient monitoring data, and device operation parameter data; Step S2: Extract device - end operation feature based on the device - end operation data to obtain device - end operation feature data; Extract patient monitoring feature based on the patient monitoring data to obtain patient monitoring feature data; Extract device operation parameter feature based on the device operation parameter data to obtain device operation feature data; Step S3: Identify device - end operation anomalies based on the device - end operation feature data and the patient monitoring feature data to obtain device - end operation anomaly data; Identify device - end parameter execution anomalies based on the device - end operation feature data and the device operation feature data to obtain device - end parameter execution anomaly data; Step S4: Perform knowledge - graph enhanced anomaly attribution based on the device - end operation anomaly data and the device - end parameter execution anomaly data to obtain medical resource anomaly data, and send it to the medical resource cloud platform for medical resource risk warning operations.

2. The method according to claim 1, wherein Specifically, step S2 is as follows: Construct the device operation state space based on the device - end operation data to obtain device - end state space data; Perform topological simplification based on the device - end state space data to obtain device - end topological graph data; Extract topological invariants based on the device - end topological graph data to obtain device - end operation feature data; Extract periodic features from the patient monitoring data to obtain patient monitoring feature data; Perform device operation parameter hypersurface mapping based on the device operation parameter data to obtain device operation parameter hypersurface data; Extract hypersurface curvature features from the device operation parameter hypersurface data to obtain device operation feature data.

3. The method according to claim 2, wherein Specifically, the construction of the device operation state space is as follows: Extract device multivariable features based on the device - end operation data to obtain device operation multivariable feature data; Construct a feature correlation matrix based on the device operation multivariable feature data to obtain feature correlation matrix data; Perform device state space modeling on the feature correlation matrix data to obtain device state space data; Perform state space mapping based on the device state space data to obtain device - end state space data.

4. The method according to claim 2, wherein Specifically, the topological simplification is as follows: Construct a device state space graph based on the device - end state space data to obtain device state space graph data; Perform persistent homology analysis on the device state space graph data to obtain persistent homology data; Perform persistence diagram calculation based on the persistent homology data to obtain persistence diagram data; Perform topological clustering on the persistence diagram data to obtain topological clustering data; Perform state classification on the device - end topological graph data according to the topological clustering data to obtain device - end topological graph data.

5. The method according to claim 2, wherein Specifically, the extraction of topological invariants is as follows: Calculate the number of connected components based on the device - end topological graph data to obtain connected component number data; Calculate the cyclic pattern based on the device - end topological graph data to obtain cyclic pattern data; Calculate the void structure based on the device - end topological graph data to obtain void structure data; Calculate the Euler characteristic based on the connected component number data, the cyclic pattern data, and the void structure data to obtain Euler characteristic data; Perform topological persistent entropy calculation based on the device - side topology graph data to obtain topological persistent entropy data; Perform vectorization based on the Euler characteristic data and the topological persistent entropy data to obtain device - side operation characteristic data.

6. The method according to claim 2, characterized in that, Among them, the device operation parameter hypersurface mapping is specifically as follows: Perform high - dimensional topological embedding based on the device operation parameter data to obtain parameter high - dimensional topological data; Perform manifold learning dimensionality reduction on the parameter high - dimensional topological data to obtain high - dimensional manifold structure data; Perform curvature calculation on the high - dimensional manifold structure data to obtain parameter curvature data; Perform Gaussian - Bonnet curve mapping based on the parameter curvature data to obtain device operation parameter hypersurface data.

7. The method according to claim 2, wherein Among them, the hypersurface curvature feature extraction is specifically as follows: Perform tangent vector inner product calculation on the device operation parameter hypersurface data to obtain the first fundamental form data; Perform normal vector change rate calculation on the device operation parameter hypersurface data to obtain the second fundamental form data; Perform Gaussian curvature calculation based on the first fundamental form data and the second fundamental form data to obtain Gaussian curvature data; Perform mean curvature calculation based on the Gaussian curvature data to obtain mean curvature data; Perform Riemann curvature tensor calculation on the device operation parameter hypersurface data to obtain Riemann curvature tensor data; Perform vectorization based on the first fundamental form data, the second fundamental form data, the Gaussian curvature data, the mean curvature data, and the Riemann curvature tensor data to obtain device operation characteristic data.

8. The method according to claim 1, wherein Step S3 is specifically as follows: Perform multivariate anomaly detection based on the device - side operation characteristic data and the patient monitoring characteristic data to obtain device - side operation anomaly data; Perform device response consistency detection based on the device - side operation characteristic data and the device operation characteristic data to obtain device response consistency data; Perform misoperation analysis based on the device response consistency data to obtain device - side parameter execution anomaly data.

9. The method according to claim 1, characterized in that Step S4 is specifically as follows: Associate the device - side operation anomaly data and the device - side parameter execution anomaly data according to the preset device knowledge graph data to obtain device knowledge graph association data; Perform knowledge graph reasoning based on the device knowledge graph association data to obtain knowledge graph reasoning data; Perform device anomaly impact calculation based on the knowledge graph reasoning data to obtain medical resource anomaly data, so as to send it to the medical resource cloud platform for medical resource risk warning operations.

10. A medical resource management system, characterized in that, For executing the medical resource management method as described in claim 1, the medical resource management system includes: A medical resource edge - side data acquisition module, configured to obtain device - side operation data, patient monitoring data, and device operation parameter data; A medical resource edge - side data feature extraction module, configured to perform device - side operation feature extraction based on the device - side operation data to obtain device - side operation characteristic data; perform patient monitoring feature extraction based on the patient monitoring data to obtain patient monitoring characteristic data; perform device operation parameter feature extraction based on the device operation parameter data to obtain device operation characteristic data; The medical resource edge - end data anomaly judgment module is used to identify device - end operation anomalies based on device - end operation characteristic data and patient monitoring characteristic data, and obtain device - end operation anomaly data; identify device - end parameter execution anomalies based on device - end operation characteristic data and device operation characteristic data, and obtain device - end parameter execution anomaly data. The medical resource edge - end data anomaly attribution module is used to perform knowledge - graph enhanced anomaly attribution based on device - end operation anomaly data and device - end parameter execution anomaly data, obtain medical resource anomaly data, and send it to the medical resource cloud platform for medical resource risk warning operations.

Images