Intelligent nursing monitoring system after tumor intervention operation

Abstract

The invention relates to the technical field of medical monitoring and early warning, in particular to a tumor intervention postoperative intelligent nursing monitoring system, which comprises an individualized data acquisition module for outputting a standardized data packet and a historical feature index table; the dynamic feature processing module generates three-dimensional feature tensors of coding time, physiological features and pathological mark dimensions; the individual self-adaptive modeling module generates an individual risk prediction model embedded with a genetic and pathological response function; the dynamic threshold optimization module encodes a historical baseline fluctuation range into chromosome gene loci, and iteratively corrects an abnormal judgment boundary in combination with a genetic algorithm; the intelligent early warning decision module calls a clinical knowledge graph to generate a three-level early warning instruction, and constructs a double-closed-loop feedback channel: a first closed loop calibrates and judges boundary parameters through a false alarm feedback signal, and a second closed loop converts disposal effectiveness into weight correction vector optimization model parameters; and full-link closed-loop management of individual dynamic physiological variation from feature fusion and model adaptation to decision optimization is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of medical monitoring and early warning, and in particular to an intelligent nursing monitoring system after tumor intervention. Background Art

[0002] As a minimally invasive cancer treatment, interventional oncology requires close postoperative monitoring to reduce the risk of infection, bleeding, or recurrence. Traditional manual models, relying on intermittent observation and subjective judgment, can easily miss early physiological abnormalities, leading to delayed intervention. In contrast, intelligent nursing incorporates wearable sensors and artificial intelligence systems to collect and analyze vital sign data in real time. Machine learning algorithms are used to identify abnormal trends, predict the probability of complications, and automatically trigger personalized alerts or nursing interventions. Through data-driven proactive management mechanisms, the system optimizes resource allocation, enhances decision-making accuracy, reduces human error, and thus improves patient recovery efficiency and quality of life.

[0003] In the process of monitoring physiological indicators in intelligent nursing after tumor intervention, there is a technical pain point that the artificial intelligence model is not adaptable enough to individual physiological characteristics. The reason for this technical pain point is that machine learning algorithms are usually trained on static group data sets, ignoring dynamic variation factors between individuals such as genetic background or comorbidity status, resulting in decreased prediction accuracy; for example, a postoperative tumor patient had abnormal prothrombin activity fluctuations due to cirrhosis, and the standard algorithm did not integrate patient-specific historical data, which led to the wrong judgment of the bleeding risk level, triggered misleading warnings, and delayed the opportunity for precise intervention. Summary of the Invention

[0004] In response to the shortcomings of existing technologies, the present invention provides an intelligent nursing monitoring system for post-tumor intervention surgery to solve the problem that medical group models are unable to capture individual dynamic physiological variations based on medical information.

[0005] In order to solve the above technical problems, the specific invention of the present invention is as follows: The intelligent nursing monitoring system for tumor interventional surgery provided by the present invention comprises: Individualized data collection module collects patients' real-time vital signs data and structured historical health records after surgery, and outputs standardized data packages and historical feature index tables; a dynamic feature processing module, connected to the individualized data acquisition module, receiving the standardized data packet and the historical feature index table, performing motion artifact removal processing, extracting key pathological markers, and generating a fused feature tensor with a time tag; An individual adaptive modeling module is connected to the dynamic feature processing module, obtains the fused feature tensor, reconstructs the pre-trained population model parameters using an incremental transfer learning mechanism, generates an individualized risk prediction model including the patient's genetic background and pathophysiological characteristics, and outputs the model to the dynamic threshold optimization module; A dynamic threshold optimization module connects the individual adaptive modeling module and the dynamic feature processing module, loads the individualized risk prediction model, intercepts the feature change rate through a sliding time window, combines the genetic algorithm to iteratively correct the abnormality judgment boundary, generates feature data with annotated deviation levels, and transmits it to the intelligent early warning decision module; The intelligent early warning decision module is connected to the dynamic threshold optimization module, responds to the received feature data marked with deviation levels, calls the clinical knowledge graph to match the patient-specific contraindication rules, generates three-level early warning instructions and treatment suggestions, and simultaneously evaluates the clinical effectiveness of the treatment suggestions to obtain clinical effectiveness evaluation results. When there is a deviation between the treatment suggestions and the real-time vital signs data of the patient collected by the individualized data acquisition module, a false alarm feedback signal is generated, and the false alarm feedback signal is transmitted to the dynamic threshold optimization module to update the judgment boundary parameters. At the same time, the clinical effectiveness evaluation results are converted into weight correction vectors and fed back to the incremental transfer learning mechanism of the individual adaptive modeling module for correcting the pre-trained group model parameters.

[0006] Furthermore, in the intelligent nursing monitoring system for post-interventional tumor surgery of the present invention, the individualized data acquisition module includes: Wearable biosensor interface that receives blood oxygen saturation, ECG waveform, and blood pressure fluctuation data at a preset frequency to generate a real-time vital sign data stream; The medical information extraction port calls genetic testing conclusions, comorbidity diagnosis results, and test index time series data from the hospital information system through a standard interface and outputs a structured historical data set; The data integration component receives the real-time vital sign data stream and the structured historical data set, performs data standardization processing to generate a standardized data packet, and extracts key fields to construct the historical feature index table.

[0007] Furthermore, in the intelligent nursing monitoring system for post-interventional tumor surgery of the present invention, the dynamic feature processing module includes: a signal filtering component, receiving the standardized data packet, eliminating motion artifacts and retaining effective physiological bands, and outputting filtered physiological data; A pathology mark parsing component receives the historical feature index table, calculates the individual baseline fluctuation range of key pathology indicators, and outputs a pathology mark feature set; The time alignment component obtains the filtered physiological data and the pathological marker feature set, fuses the real-time monitoring data with the historical features based on the timestamp alignment mechanism, and generates a three-dimensional feature tensor with a time tag.

[0008] Furthermore, the intelligent nursing monitoring system for post-interventional tumor surgery of the present invention further includes: The individual adaptive modeling module obtains the three-dimensional feature tensor with the time label, inputs the three-dimensional feature tensor into the residual connection network layer, and learns the patient-specific parameter offset; The pre-trained convolutional layer weights are dynamically adjusted based on the learning results, the weights of the pre-trained convolutional layer are reconstructed to generate an individualized risk prediction model, and the model is output to the dynamic threshold optimization module.

[0009] Furthermore, in the intelligent nursing monitoring system for post-interventional tumor surgery of the present invention, the dynamic threshold optimization module includes: The change rate calculation component divides the three-dimensional feature tensor with time tags according to a fixed time window and outputs the instantaneous change rate of physiological indicators; the genetic optimization controller retrieves the historical baseline fluctuation range extracted from the pathological marker feature set from the dynamic feature processing module, encodes the historical baseline fluctuation range as a chromosome gene site, calculates the false alarm rate and the number of missed events based on the deviation value between the three-level warning instructions generated by the intelligent warning decision module and the actual clinical status, evaluates the adaptability of the chromosome configuration to abnormality judgment, and generates optimized judgment boundary parameters.

[0010] Furthermore, the intelligent nursing monitoring system for post-interventional tumor surgery of the present invention further includes: The genetic optimization controller in the dynamic threshold optimization module receives the false alarm feedback signal transmitted by the intelligent early warning decision module, uses the false alarm feedback signal as a correction factor for fitness evaluation, performs a chromosome crossover mutation operation, generates an updated decision boundary parameter and replaces the optimized decision boundary parameter.

[0011] Furthermore, in the intelligent nursing monitoring system for post-interventional tumor surgery of the present invention, the intelligent early warning decision module includes: A clinical rule reasoning component receives feature data annotated with deviation levels, associates key pathological markers extracted by the dynamic feature processing module with drug interaction data from the patient contraindication database, and outputs a preliminary warning strategy; An ambiguity resolution component obtains the preliminary warning strategy. If the abnormal attributes in the feature data marked with the deviation level trigger a rule conflict in the preliminary warning strategy, the treatment strategy is screened from the preliminary warning strategy based on the similarity of the current medical history characteristics collected by the individualized data acquisition module, and integrated into the three-level warning instructions. The treatment strategy is screened from the preliminary warning strategy based on the similarity of the current medical history characteristics of the structured historical health records collected by the individualized data acquisition module, and integrated into the three-level warning instructions.

[0012] Furthermore, the intelligent nursing monitoring system for post-interventional tumor surgery of the present invention further includes: The intelligent early warning decision module obtains effectiveness data of the treatment recommendation based on the clinical effectiveness evaluation results, encodes the mapping relationship between the effectiveness data and the execution results of the three-level early warning instructions generated by the intelligent early warning decision module into a vector matrix, generates a weight correction vector and transmits it to the individual adaptive modeling module; The individual adaptive modeling module receives the weight correction vector and dynamically adjusts the back propagation learning rate parameter of the residual connection network layer to update the individualized risk prediction model.

[0013] Furthermore, the intelligent nursing monitoring system for post-interventional tumor surgery of the present invention further includes: The individual adaptive modeling module, the dynamic threshold optimization module and the intelligent early warning decision module constitute a dual closed-loop feedback channel, which includes a first closed-loop channel and a second closed-loop channel: First closed-loop channel: the intelligent early warning decision module outputs a false alarm feedback signal to the genetic optimization controller of the dynamic threshold optimization module, triggering iterative update of the decision boundary parameter; Second closed-loop channel: the intelligent early warning decision module outputs the weight correction vector to the individual adaptive modeling module, and continuously optimizes the individualized risk prediction model by adjusting the residual connection network layer parameters.

[0014] Furthermore, the intelligent nursing monitoring system for post-interventional tumor surgery of the present invention further includes: After the treatment suggestion is executed, the intelligent early warning decision module monitors the characteristic data of the deviation level output by the dynamic threshold optimization module in real time, extracts the change trend of physiological indicators in the characteristic data of the deviation level, and sends an optimization instruction to the dynamic threshold optimization module when no abnormal characteristic rebound of the key pathological marker is detected within a preset time; The genetic optimization controller responds to the optimization instruction, increases the fitness score weight of the successful matching anomaly judgment in the current chromosome configuration, generates an enhanced version of the judgment boundary parameter and replaces the optimized judgment boundary parameter.

[0015] Beneficial effects of the present invention: The present invention effectively solves the technical defect that the medical group model has difficulty in capturing individual physiological dynamic variations through an individualized dynamic feature fusion mechanism and a dual closed-loop adaptive optimization system. The individualized data acquisition module synchronously integrates real-time vital signs and structured historical records to construct a feature tensor to eliminate the statistical bias caused by the group model ignoring individual specificity; the individual adaptive modeling module uses an incremental transfer learning mechanism to reconstruct the pre-trained model parameters, learns patient-specific offsets through the residual network layer, and generates a risk prediction model embedded in genetic and pathological response functions to improve the perception sensitivity of dynamic physiological variations; the dynamic threshold optimization module encodes historical fluctuation tolerance into chromosomal gene sites based on a genetic algorithm, and iteratively corrects the judgment boundary in combination with the real-time false alarm rate to avoid misjudgment of individual variation scenarios such as abnormal coagulation function in patients with cirrhosis; the intelligent early warning decision module drives the dual closed-loop feedback channel. The first closed loop calibrates the abnormal judgment boundary through false alarm signals, and the second closed loop converts the treatment effectiveness into a weight correction vector to optimize the model parameters to achieve the continuous evolution of individual feature perception accuracy and clinical decision adaptability, and ultimately achieves full-link closed-loop management of individual dynamic physiological variations from feature fusion model adaptation to decision optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings without paying any creative labor.

[0017] Figure 1 This is a system architecture diagram of the intelligent nursing monitoring system for post-tumor interventional surgery provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and corresponding drawings. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The present invention provided by each embodiment of the present invention is described in detail below in conjunction with the drawings. In order to better understand the purpose of the present invention, the present invention is further described below.

[0019] See also Figure 1 The intelligent nursing monitoring system for tumor interventional surgery provided by the present invention includes: The intelligent nursing monitoring system for tumor interventional surgery provided by the present invention comprises: Individualized data collection module collects patients' real-time vital signs data and structured historical health records after surgery, and outputs standardized data packages and historical feature index tables; a dynamic feature processing module, connected to the individualized data acquisition module, receiving the standardized data packet and the historical feature index table, performing motion artifact removal processing, extracting key pathological markers, and generating a fused feature tensor with a time tag; An individual adaptive modeling module is connected to the dynamic feature processing module, obtains the fused feature tensor, reconstructs the pre-trained population model parameters using an incremental transfer learning mechanism, generates an individualized risk prediction model including the patient's genetic background and pathophysiological characteristics, and outputs the model to the dynamic threshold optimization module; A dynamic threshold optimization module connects the individual adaptive modeling module and the dynamic feature processing module, loads the individualized risk prediction model, intercepts the feature change rate through a sliding time window, combines the genetic algorithm to iteratively correct the abnormality judgment boundary, generates feature data with annotated deviation levels, and transmits it to the intelligent early warning decision module; The intelligent early warning decision module is connected to the dynamic threshold optimization module, responds to the received feature data marked with deviation levels, calls the clinical knowledge graph to match the patient-specific contraindication rules, generates three-level early warning instructions and treatment suggestions, and simultaneously evaluates the clinical effectiveness of the treatment suggestions to obtain clinical effectiveness evaluation results. When there is a deviation between the treatment suggestions and the real-time vital signs data of the patient collected by the individualized data acquisition module, a false alarm feedback signal is generated, and the false alarm feedback signal is transmitted to the dynamic threshold optimization module to update the judgment boundary parameters. At the same time, the clinical effectiveness evaluation results are converted into weight correction vectors and fed back to the incremental transfer learning mechanism of the individual adaptive modeling module for correcting the pre-trained group model parameters.

[0020] The personalized data acquisition module obtains the patient's real-time vital signs data stream after surgery at a preset frequency through the wearable biosensor interface, including dynamic physiological parameters such as blood oxygen saturation, electrocardiogram waveform and blood pressure fluctuation. The medical information extraction port is synchronously connected to the hospital information system, calling the genetic test conclusions, comorbidity diagnosis results and test index time series data to generate a structured historical data set. The data integration component standardizes the real-time vital signs data stream and the structured historical data set, unifies the data format and dimension, and generates a standardized data packet; at the same time, it extracts key fields to construct a historical feature index table. The index table includes genetic background labels and pathological marker time series pointers, providing a rapid retrieval basis for subsequent feature fusion. The core of this step is to integrate multi-source heterogeneous medical data and establish a spatiotemporal data structure.

[0021] After the dynamic feature processing module receives the standardized data packet, the signal filtering component uses an adaptive filtering algorithm to eliminate motion artifacts, retain valid physiological bands, and output filtered physiological data. The pathology marker parsing component parses the historical feature index table, calculates the individual baseline fluctuation range of key pathological indicators, and outputs a pathology marker feature set, including abnormal coagulation function thresholds and inflammatory response thresholds. The time alignment component, based on a timestamp alignment mechanism, performs spatiotemporal matching of the real-time monitoring values ​​of the filtered physiological data with the historical baseline of the pathology marker feature set, generating a three-dimensional feature tensor with a time label. This three-dimensional tensor structure encodes the time dimension, the physiological feature dimension, and the pathology marker dimension, respectively, enabling spatiotemporal modeling of individualized physiological dynamic variation.

[0022] After receiving the three-dimensional feature tensor, the individual adaptive modeling module learns patient-specific parameter offsets through a residual connection network layer. This process utilizes an incremental transfer learning mechanism: based on the pretrained population model, the convolutional layer weights are dynamically adjusted to reconstruct the model parameters. This reconstruction process integrates the patient's genetic background with the pathophysiological response function to generate a personalized risk prediction model. This model output includes a dynamic risk probability value and pathological marker sensitivity parameters, providing a patient-specific quantitative basis for abnormality determination. This step addresses the problem of prediction bias caused by population models ignoring individual dynamic variation.

[0023] After the dynamic threshold optimization module loads the individualized risk prediction model, the rate-of-change calculation component segments the three-dimensional feature tensor according to a fixed time window and calculates the instantaneous rate of change of physiological indicators. The genetic optimization controller encodes the historical baseline fluctuation range as chromosomal gene loci. Based on the deviation between the three-level warning instructions and the actual clinical status, it calculates the false alarm rate and the number of missed events, and evaluates the fitness score of the chromosome configuration for abnormality determination. Based on the fitness assessment results, the abnormality determination boundary parameters are iteratively modified, and feature data with annotated deviation levels is output. This process achieves adaptive optimization of the determination boundary through a genetic algorithm, avoiding the risk of misjudgment of specific pathological conditions such as cirrhosis.

[0024] The intelligent early warning decision-making module calls the clinical knowledge graph to match patient-specific contraindication rules, and the clinical rule reasoning component generates a preliminary early warning strategy. When the ambiguity resolution component detects that an abnormal attribute triggers a rule conflict, it screens the treatment strategy based on the similarity of the current medical history characteristics and integrates it into three-level early warning instructions and treatment suggestions. The module synchronously monitors the trend of changes in physiological indicators in the deviation level feature data after the execution of the treatment suggestion, and generates a clinical effectiveness evaluation result. When there is a deviation between the treatment suggestion and the real-time vital sign data, a double closed-loop feedback is triggered: the first closed-loop drives the dynamic threshold optimization module to update the judgment boundary parameters through the false alarm feedback signal; the second closed-loop converts the clinical effectiveness evaluation results into a weight correction vector, which is fed back to the individual adaptive modeling module to correct the pre-trained group model parameters.

[0025] Specifically, the intelligent nursing monitoring system for tumor interventional surgery of the present invention includes the following individualized data acquisition module: Wearable biosensor interface that receives blood oxygen saturation, ECG waveform, and blood pressure fluctuation data at a preset frequency to generate a real-time vital sign data stream; The medical information extraction port calls genetic testing conclusions, comorbidity diagnosis results, and test index time series data from the hospital information system through a standard interface and outputs a structured historical data set; The data integration component receives the real-time vital sign data stream and the structured historical data set, performs data standardization processing to generate a standardized data packet, and extracts key fields to construct the historical feature index table.

[0026] The wearable biosensor interface integrates a medical-grade photoelectric sensor and accelerometer to collect raw signals from patients' postoperative blood oxygen saturation, electrocardiogram (ECG) waveforms, and blood pressure fluctuations. This interface uses an adaptive sampling mechanism to dynamically adjust the data acquisition frequency based on the patient's activity level. After analog-to-digital conversion, the raw signals are filtered through a Butterworth filter to remove high-frequency noise, generating a real-time vital sign data stream compliant with the IEEE 11073 protocol. This data stream includes a timestamp, device identifier, and signal quality label, providing a standardized input source for subsequent processing.

[0027] The medical information extraction portal connects to the hospital information system via the HL7 / FHIR standard interface, accessing pathogenic variants from genetic testing results, ICD-11 codes for comorbidity diagnoses, and time-series data on test indicators. The portal performs data desensitization, removing patient identifying information while retaining medical ontology terminology. This access process utilizes the OAuth 2.0 authorization protocol to generate a structured historical dataset that includes genotype-phenotype associations, complication severity grading, and biomarker trends. This dataset is embedded with a time series index, enabling the spatiotemporal tracing of historical health status.

[0028] The data integration component receives the real-time vital sign data stream and the structured historical data set and performs multi-level data processing: Perform time window segmentation and motion artifact compensation on real-time data streams to repair signal interruptions; Perform ontology term mapping on historical datasets to link genetic testing conclusions to the OMIM disease database; Use Z-score to standardize and unify the dimensions to eliminate measurement deviations between devices; The two types of data are spliced ​​together through a timing alignment algorithm to generate a standardized data packet; Based on the medical ontology tree, we extract key fields to construct a historical feature index table. The index fields include genetic variant sites, pathology marker baseline values, and time decay coefficients. The index table uses a key-value pair storage structure to support fast retrieval.

[0029] Specifically, in the intelligent nursing monitoring system for tumor interventional surgery of the present invention, the dynamic feature processing module includes: a signal filtering component, receiving the standardized data packet, eliminating motion artifacts and retaining effective physiological bands, and outputting filtered physiological data; A pathology mark parsing component receives the historical feature index table, calculates the individual baseline fluctuation range of key pathology indicators, and outputs a pathology mark feature set; The time alignment component obtains the filtered physiological data and the pathological marker feature set, fuses the real-time monitoring data with the historical features based on the timestamp alignment mechanism, and generates a three-dimensional feature tensor with a time tag.

[0030] The signal filtering component receives standardized data packets and performs multi-stage signal processing. Accelerometer data is used as a reference source for motion noise, and an adaptive noise cancellation algorithm is employed to isolate motion artifacts. Wavelet packet decomposition and reconstruction techniques are applied to suppress myoelectric interference and environmental noise while preserving the effective frequency bands of the ECG ST segment and blood oxygen perfusion index. The filtering process is embedded in a real-time signal quality assessment module, which performs Kalman filtering compensation on low signal-to-noise ratio data segments. The output of filtered physiological data with quality labels includes time-synchronized series of physiological parameters such as heart rate variability and blood oxygen trend.

[0031] The pathology marker parsing component parses key fields from the historical feature index table. Using a medical ontology mapping engine, the rsID identifiers in the genetic test results are linked to the PharmGKB pharmacogenomic database and the coefficient of variation of metabolic enzyme activity is calculated. ICD-11 codes for comorbidity diagnoses are parsed and mapped to the Charlson Comorbidity Index. Dynamic time warping is performed on the time series data of test indicators to extract individualized baseline fluctuation ranges for prothrombin time and inflammatory factor concentrations. This outputs a structured pathology marker feature set, including genetic pharmacodynamic parameters, complication risk levels, and dynamic thresholds for pathology indicators.

[0032] The time alignment component establishes a unified time reference system. First, a millisecond timestamp sequence is embedded into the filtered physiological data. Next, a time decay model is constructed based on the pathological marker feature set. A dynamic time warping algorithm is used to align the time axes of the two data types, addressing the issue of sampling frequency discrepancies. Using tensor slice reconstruction technology, real-time physiological parameters and historical pathological markers are integrated along the time dimension. The three dimensions of the generated three-dimensional feature tensor are: the time series axis encodes minute-level monitoring points; the physiological feature axis maps parameters such as blood oxygen and blood pressure; and the pathological marker axis carries quantitative indicators such as genetic risk values.

[0033] Specifically, the intelligent nursing monitoring system for post-operative tumor intervention according to the present invention further includes: The individual adaptive modeling module obtains the three-dimensional feature tensor with the time label, inputs the three-dimensional feature tensor into the residual connection network layer, and learns the patient-specific parameter offset; The pre-trained convolutional layer weights are dynamically adjusted based on the learning results, the weights of the pre-trained convolutional layer are reconstructed to generate an individualized risk prediction model, and the model is output to the dynamic threshold optimization module.

[0034] The individual adaptive modeling module receives the time-tagged 3D feature tensor and then performs feature learning via a residual connection network layer. This network layer includes a cross-layer connection structure that retains the basic features of the pretrained model while learning patient-specific parameter offsets. The parameter offsets are calculated using a backpropagation algorithm to quantify the difference between the current patient's physiological characteristics and the population model, focusing on capturing individual dynamic variability factors such as genetic background and drug metabolism.

[0035] Based on the learned parameter offsets, an incremental transfer learning mechanism is used to dynamically adjust the pretrained convolutional layer weights. This adjustment follows the gradient masking principle, selectively fine-tuning the underlying convolutional kernels of the population model. This approach enhances the modeling of individual pathological response functions while retaining general feature extraction capabilities. The reconstructed convolutional layer weights are then integrated with the patient-specific parameter offsets to generate a personalized risk prediction model that includes genetic pharmacodynamic responses and complication evolution pathways.

[0036] The generated personalized risk prediction model outputs multidimensional risk probability values, including bleeding risk index, infection probability, and organ dysfunction score. When the model is output to the dynamic threshold optimization module, the pathological marker sensitivity parameters are simultaneously transmitted, providing a quantitative basis for optimizing the abnormality judgment boundary. This model continuously integrates dynamic patient data to establish an adaptive prediction mechanism that evolves over time.

[0037] Specifically, in the intelligent nursing monitoring system for tumor interventional surgery of the present invention, the dynamic threshold optimization module includes: a change rate calculation component that segments the three-dimensional feature tensor with time tags according to a fixed time window and outputs the instantaneous change rate of the physiological indicator; The genetic optimization controller retrieves the historical baseline fluctuation range extracted from the pathological marker feature set from the dynamic feature processing module, encodes the historical baseline fluctuation range as a chromosome gene site, calculates the false alarm rate and the number of missed events based on the deviation value between the three-level warning instructions generated by the intelligent warning decision module and the actual clinical status, evaluates the adaptability of the chromosome configuration to abnormality judgment, and generates optimized judgment boundary parameters.

[0038] The rate-of-change calculation component receives a time-tagged 3D feature tensor and slices it into fixed time windows. Using a sliding window mechanism, it calculates the instantaneous rate of change of physiological indicators such as blood oxygen saturation and blood pressure within each time window. This rate-of-change calculation integrates first-order derivatives and dynamic time warping algorithms to capture the sudden changes in indicator fluctuation trends and outputs a timestamped sequence of physiological indicator rate-of-change.

[0039] The genetic optimization controller executes multi-level optimization logic: it retrieves the pathological marker feature set from the dynamic feature processing module, extracts historical fluctuation range data such as coagulation function threshold and inflammatory response baseline, quantifies the historical fluctuation range into discrete interval values, and encodes it into a chromosome gene site sequence. Based on the deviation between the three-level warning instructions generated by the intelligent warning decision module and the actual vital signs data collected by the individualized data acquisition module, it calculates the number of false alarm events and the number of missed events. The abnormality judgment efficiency of the chromosome configuration is evaluated through the fitness function. The fitness value is proportional to the number of missed events and inversely proportional to the false alarm rate. The elite retention strategy is used for chromosome crossover mutation, and the optimized judgment boundary parameter set is iteratively generated.

[0040] Specifically, the intelligent nursing monitoring system for post-operative tumor intervention according to the present invention further includes: The genetic optimization controller in the dynamic threshold optimization module receives the false alarm feedback signal transmitted by the intelligent early warning decision module, uses the false alarm feedback signal as a correction factor for fitness evaluation, performs a chromosome crossover mutation operation, generates an updated decision boundary parameter and replaces the optimized decision boundary parameter.

[0041] The genetic optimization controller receives false alarm feedback signals from the intelligent early warning decision module and converts them into correction factors for fitness assessment. The correction factor quantification rule is as follows: when a false alarm feedback signal is activated, the fitness score of the corresponding chromosome locus decreases by a preset attenuation coefficient. When performing chromosome crossover and mutation operations, the controller uses an adaptive probability adjustment strategy—increasing the mutation probability when the false alarm rate increases and increasing the crossover probability when the false alarm rate decreases. This operation retains elite chromosome individuals to maintain optimization stability.

[0042] After the chromosome crossover mutation operation generates a new population, the chromosome configuration is re-evaluated using a fitness function. This evaluation integrates historical baseline fluctuations with real-time false positive feedback to calculate the optimization potential of the decision boundary parameters. The chromosome configuration with the highest potential is selected and decoded to generate an updated decision boundary parameter set. Parameter replacement utilizes a version control mechanism: when the fitness score of the new parameter set exceeds the currently optimized decision boundary parameters, an atomic replacement operation is triggered to overwrite the original parameters.

[0043] Specifically, the intelligent nursing monitoring system for post-operative tumor intervention according to the present invention includes the following intelligent early warning decision module: A clinical rule reasoning component receives feature data annotated with deviation levels, associates key pathological markers extracted by the dynamic feature processing module with drug interaction data from the patient contraindication database, and outputs a preliminary warning strategy; An ambiguity resolution component obtains the preliminary warning strategy. If the abnormal attributes in the feature data marked with the deviation level trigger a rule conflict in the preliminary warning strategy, the treatment strategy is screened from the preliminary warning strategy based on the similarity of the current medical history characteristics collected by the individualized data acquisition module, and integrated into the three-level warning instructions. The treatment strategy is screened from the preliminary warning strategy based on the similarity of the current medical history characteristics of the structured historical health records collected by the individualized data acquisition module, and integrated into the three-level warning instructions.

[0044] The clinical rule reasoning component receives the annotated deviation level feature data output by the dynamic threshold optimization module and associates key pathology markers with a library of drug contraindications using a medical ontology mapping engine. This association process performs multi-dimensional matching: abnormal coagulation markers are associated with anticoagulant contraindication rules, and inflammatory response markers are associated with immunosuppressant usage guidelines. The inference logic is constructed using the SWRL semantic rule language. When pathology markers trigger drug interaction conditions, a preliminary warning strategy is generated, including risk levels and recommended actions. The output strategy is embedded in the HL7 CDA clinical document schema, including timestamps and decision path identifiers.

[0045] After obtaining the initial warning strategy, the ambiguity resolution component monitors anomaly attributes in the data that deviates from the hierarchical feature in real time. When anomaly attributes trigger multiple conflicting rules simultaneously, the resolution mechanism is activated: the structured historical health records of the individualized data acquisition module are called upon to extract current medical history features and construct a similarity matrix. The similarity calculation uses a dynamic time warping algorithm to quantify the degree of match between the current abnormal state and the temporal evolution pattern of historical cases. Disposition strategies are screened based on the similarity ranking results, prioritizing historically proven effective solutions. Finally, the conflict resolution results are integrated to generate a three-level warning instruction, which includes the priority of the disposal measures, contraindication avoidance solutions, and timeliness parameters.

[0046] Specifically, the intelligent nursing monitoring system for post-operative tumor intervention according to the present invention further includes: The intelligent early warning decision module obtains effectiveness data of the treatment recommendation based on the clinical effectiveness evaluation results, encodes the mapping relationship between the effectiveness data and the execution results of the three-level early warning instructions generated by the intelligent early warning decision module into a vector matrix, generates a weight correction vector and transmits it to the individual adaptive modeling module; The individual adaptive modeling module receives the weight correction vector and dynamically adjusts the back propagation learning rate parameter of the residual connection network layer to update the individualized risk prediction model.

[0047] The intelligent early warning decision-making module quantifies the effectiveness of treatment recommendations based on clinical effectiveness assessment results. This effectiveness data includes metrics such as the improvement rate of physiological indicators and the efficiency of complication avoidance after treatment. This module constructs a mapping matrix between the execution results of three-level early warning instructions and effectiveness data. The row vectors of the matrix correspond to different warning levels, and the column vectors map clinical benefit indicators such as vital sign stability and pathological marker recovery. The singular value decomposition algorithm compresses the matrix dimensions, generating a low-dimensional dense vector as the weight correction vector, which is transmitted via an encrypted channel to the individual adaptive modeling module.

[0048] After receiving the weight correction vector, the individual adaptive modeling module analyzes the vector's dimensions to determine the direction of model parameter adjustment. For the residual connection network layer, the learning rate parameter is dynamically adjusted during backpropagation: Dimensions with high sensitivity to the weight correction vector are increased to accelerate convergence, while dimensions with low sensitivity are reduced to maintain stability. This learning rate adjustment utilizes an adaptive momentum algorithm, incorporating historical gradient information to smooth parameter updates. The optimized network layer updates the output layer weight parameters of the individualized risk prediction model through forward propagation, enhancing the model's responsiveness to treatment effectiveness patterns.

[0049] Specifically, the intelligent nursing monitoring system for post-operative tumor intervention according to the present invention further includes: The individual adaptive modeling module, the dynamic threshold optimization module and the intelligent early warning decision module constitute a dual closed-loop feedback channel, which includes a first closed-loop channel and a second closed-loop channel: First closed-loop channel: the intelligent early warning decision module outputs a false alarm feedback signal to the genetic optimization controller of the dynamic threshold optimization module, triggering iterative update of the decision boundary parameter; Second closed-loop channel: the intelligent early warning decision module outputs the weight correction vector to the individual adaptive modeling module, and continuously optimizes the individualized risk prediction model by adjusting the residual connection network layer parameters.

[0050] The first closed-loop channel establishes a real-time calibration mechanism: The intelligent early warning decision module encodes false alarm feedback signals into standardized event messages and transmits them to the genetic optimization controller of the dynamic threshold optimization module. After analyzing the signals, the controller activates a chromosome mutation operation, mapping false alarm events into penalty terms in the fitness function. This drives the iterative update of the decision boundary parameters to reduce the false alarm rate. The updated boundary parameters overwrite the original parameters through a version control protocol, forming a dynamic optimization loop for the anomaly detection threshold.

[0051] The second closed-loop channel enables long-term model evolution: The intelligent early warning decision-making module extracts clinical effectiveness assessment data and constructs a mapping matrix between the execution results of the three-level early warning instructions and the treatment effects. Principal component analysis and dimensionality reduction are used to generate weight correction vectors, which are transmitted to the individual adaptive modeling module via an encrypted channel. This module analyzes the dimensional characteristics of the vectors and dynamically adjusts the backpropagation learning rate of the residual connection network layer. It increases the learning rate for high-response dimensions to accelerate parameter convergence, while maintaining a low learning rate for basic dimensions to ensure stability. After the network layer weights are updated, an enhanced individualized risk prediction model is output to improve the response accuracy to clinical treatment patterns.

[0052] Dual closed-loop systems form a hierarchical optimization architecture: The first closed-loop, leveraging a rapid feedback mechanism for false alarms, optimizes boundary parameters on a minute-by-minute timescale, addressing threshold drift in real-time monitoring scenarios. The second closed-loop, based on in-depth analysis of clinical effectiveness data, evolves risk prediction models on a daily timescale, enhancing long-term decision adaptability. The dual channels share the output status of the intelligent early warning decision module via a data bus to avoid parameter conflicts. The system coordinator ensures synergy between boundary parameter updates and model optimization, aligning threshold adjustment with risk prediction capabilities.

[0053] Specifically, the intelligent nursing monitoring system for post-operative tumor intervention according to the present invention further includes: After the treatment suggestion is executed, the intelligent early warning decision module monitors the characteristic data of the deviation level output by the dynamic threshold optimization module in real time, extracts the change trend of physiological indicators in the characteristic data of the deviation level, and sends an optimization instruction to the dynamic threshold optimization module when no abnormal characteristic rebound of the key pathological marker is detected within a preset time; The genetic optimization controller responds to the optimization instruction, increases the fitness score weight of the successful matching anomaly judgment in the current chromosome configuration, generates an enhanced version of the judgment boundary parameter and replaces the optimized judgment boundary parameter.

[0054] After the recommended action is executed, the intelligent early warning decision module acquires the annotated deviation level feature data output by the dynamic threshold optimization module in real time. This module uses time series analysis to extract the changing trend characteristics of physiological indicators within a sliding time window, including dimensions such as blood oxygen recovery slope and blood pressure fluctuation stability. The monitoring process is correlated with the expected evolution path of key pathological markers. If pathological rebound characteristics such as abnormal coagulation function and sudden increase in inflammatory factors do not appear within the preset monitoring period, the optimization instruction generation logic is triggered.

[0055] After receiving the optimization instruction, the genetic optimization controller parses the embedded monitoring cycle parameters and pathological stability labels. It then performs an adaptive enhancement operation on the current chromosome population: Gene loci that successfully match abnormality judgments are marked as elite chromosomes, and their fitness scores are increased. Weight adjustment uses an exponential weighting mechanism, with recent successful judgments receiving higher weight coefficients. Based on the enhanced fitness distribution, a directed chromosome crossover mutation is performed to generate an enhanced set of judgment boundary parameters. After the new parameter set passes version verification, it atomically replaces the original optimized judgment boundary parameters.

[0056] Physiological indicator trend monitoring establishes the foundation for treatment effectiveness evaluation, with the absence of pathological rebound features serving as a positive feedback signal. Optimization instructions transform clinical treatment effectiveness into a reinforced training signal for the genetic algorithm, driving chromosome configuration toward high-precision judgment. The fitness weight adjustment mechanism focuses on historically successful judgment patterns, enhancing the stability of parameter optimization by retaining elite chromosomes. Parameter replacement utilizes a transactional update protocol to ensure decision continuity during the optimization process, forming a positive reinforcement loop from treatment effectiveness to anomaly detection capabilities.

[0057] This invention addresses the limitations of medical population models in capturing individual dynamic physiological variation through a multi-layered technical architecture. The personalized data acquisition module integrates real-time vital signs with structured historical records to construct a historical feature index table containing genetic background labels and time series pointers to pathological markers, eliminating statistical bias caused by population models that ignore patient specificity. The dynamic feature processing module generates a three-dimensional feature tensor encoding time, physiological feature, and pathological marker dimensions. Using a spatiotemporal alignment mechanism, it fuses real-time monitoring values ​​with historical baselines to establish a quantitative vector for individual dynamic physiological variation.

[0058] The individual adaptive modeling module uses an incremental transfer learning mechanism to reconstruct a pretrained population model. Residual-connected network layers learn patient-specific parameter offsets, and convolutional layer weights are dynamically adjusted to integrate the genetic pharmacodynamic response function with the complication evolution path. This generates a personalized risk prediction model that outputs quantitative indicators such as the bleeding risk index. This model continuously integrates dynamic data to establish an adaptive mechanism that evolves over time.

[0059] The dynamic threshold optimization module encodes the historical baseline fluctuation range as chromosomal gene loci and iteratively modifies the judgment boundary based on clinical decision feedback. The intelligent early warning decision module drives a dual closed-loop feedback channel: the first closed-loop drives the genetic algorithm to optimize the abnormal judgment boundary through false alarm feedback signals to solve the threshold drift caused by short-term physiological fluctuations; the second closed-loop maps the effectiveness of the treatment to a weight correction vector and adjusts the learning rate of the residual network layer to optimize the long-term adaptability of the model. The dual channels work together to achieve the continuous evolution of individual feature perception accuracy and clinical decision adaptability, ultimately achieving full-link closed-loop management of individual dynamic physiological variations from feature fusion, model adaptation to decision optimization.

[0060] The intelligent post-operative nursing and monitoring system for tumor interventions presented in this invention addresses the pain points described in the background art and is specifically implemented in the postoperative setting for cancer patients. The personalized data acquisition module collects real-time blood oxygen saturation and blood pressure data at a preset frequency via a wearable biosensor interface. Simultaneously, the medical information extraction port accesses the hospital information system to obtain the patient's liver cirrhosis history and prothrombin time series. The data integration component performs standardization processing to generate standardized data packets, extracts gene mutation sites and abnormal coagulation thresholds, and constructs a historical feature index table, addressing the data gap caused by population models that ignore individual genetic background and comorbidities.

[0061] The signal filtering component in the dynamic feature processing module removes motion artifacts from standardized data packets, preserving valid ECG waveforms. The pathology marker parsing component parses historical feature index tables to calculate individual coagulation function fluctuations. The time alignment component aligns real-time blood pressure data with historical coagulation thresholds using millisecond-level timestamps, generating a three-dimensional feature tensor encoding the time dimension, physiological indicator dimension, and pathology marker dimension, enabling spatiotemporal dynamic modeling of coagulation abnormalities in patients with cirrhosis.

[0062] The individual adaptive modeling module feeds the three-dimensional feature tensor into a residual connection network layer to learn patient-specific coagulation parameter offsets. Using an incremental transfer learning mechanism, the pretrained convolutional layer weights are dynamically adjusted to generate an individualized risk prediction model that includes a cirrhosis pathological response function. This model outputs a bleeding risk probability value and coagulation sensitivity parameter, providing a quantitative basis for dynamic threshold optimization.

[0063] The rate-of-change calculation component of the dynamic threshold optimization module segments the feature tensor into minute-level time windows to calculate the instantaneous rate of change of blood pressure. A genetic optimization controller encodes historical coagulation fluctuations as chromosomal loci and calculates the false alarm rate based on the deviation between the three-level warning instructions and real-time vital signs. If the system initially misjudges the bleeding risk in a cirrhotic patient, the false alarm feedback signal triggers chromosomal mutation, iteratively generating decision boundary parameters tailored to the individual's coagulation abnormality.

[0064] The clinical rule reasoning component of the intelligent early warning decision module associates abnormal coagulation markers with a library of anticoagulant contraindications to generate water restriction recommendations. When the ambiguity resolution component detects blood pressure fluctuations that trigger conflicting rules, it screens treatment strategies based on historical similar cases. Dual closed-loop feedback channels operate in real time: the first closed-loop optimizes the decision boundary through false positive signals; the second closed-loop generates a weight correction vector based on clinical effectiveness data showing no rebound bleeding after treatment, which is fed back to the residual network layer via an encrypted channel to adjust the learning rate parameters, continuously improving the model's perception accuracy of coagulation abnormalities. Ultimately, full-link optimization of individual dynamic variations in cirrhosis patients, from data collection and risk modeling to early warning decision-making, is achieved.

[0065] The personalized data acquisition module collects raw signals of blood oxygen saturation, electrocardiogram (ECG) waveforms, and blood pressure fluctuations at a preset frequency through a wearable biosensor interface. These signals undergo analog-to-digital conversion and Butterworth filtering to generate a standardized real-time vital sign data stream. The medical information extraction port connects to the hospital information system, accessing genetic testing results, comorbidity diagnosis results, and time-series data on test indicators. This data is desensitized and transmitted using the OAuth 2.0 authorization protocol and HL7 / FHIR interfaces. The data integration component performs multi-level processing: time window segmentation to repair signal interruptions, SNOMED CT terminology mapping to unify semantics, and Z-score normalization to eliminate dimensional differences. The final output is a historical feature index table containing genetic variant sites and pathology marker indicators. This process establishes a spatiotemporal correlation structure for multi-source heterogeneous medical data, addressing the data foundational flaw of population models that ignore individual specificity.

[0066] The signal filtering component in the dynamic feature processing module uses an adaptive wavelet transform to separate motion artifacts while preserving valid physiological bands such as the ST segment of the ECG signal. The pathology marker parsing component uses a medical ontology engine to associate genetic test rsID identifiers with the PharmGKB database, calculates the coefficient of variation of metabolic enzyme activity, parses ICD-11 codes, maps them to the Charlson Comorbidity Index, and outputs a set of personalized pathology marker features, such as coagulation function thresholds. The time alignment component uses a dynamic time warping algorithm to align real-time physiological data with historical pathology baselines, constructing a three-dimensional feature tensor: the first dimension encodes minute-by-minute time series, the second dimension maps physiological parameters such as blood pressure, and the third dimension carries genetic risk values. This architecture enables spatiotemporal quantitative modeling of individual dynamic physiological variation.

[0067] The individual adaptive modeling module feeds the three-dimensional feature tensor into a residually connected network layer. This layer retains the common features of the pretrained model through cross-layer connections. The backpropagation algorithm learns patient-specific parameter offsets, focusing on capturing individual variability factors such as genetic pharmacodynamic responses. An incremental transfer learning mechanism dynamically adjusts the convolutional layer weights: gradient masking fine-tunes the underlying convolution kernels. The reconstructed model incorporates the complication evolution path and outputs quantitative metrics such as the bleeding risk index. This process transforms the population model into an adaptive prediction system embedded in the individual pathological response function.

[0068] The rate-of-change calculation component of the dynamic threshold optimization module segments the feature tensor into fixed time windows and combines first-order derivatives with a dynamic time warping algorithm to calculate the instantaneous rate of change of physiological indicators. A genetic optimization controller encodes the historical baseline fluctuation range as chromosomal loci. A fitness function correlates the deviation between the three-level warning instructions and actual physical signs: increasing the probability of mutation as the false alarm rate rises, and increasing the probability of crossover as missed alarms increase. An elite retention strategy iteratively generates boundary parameters, and a version control mechanism enables atomic replacement. This design addresses the issue of threshold drift in pathological conditions such as cirrhosis.

[0069] The clinical rule reasoning component of the intelligent early warning decision-making module uses SWRL semantic rules to associate abnormal coagulation markers with a library of anticoagulant contraindications to generate a preliminary strategy. The ambiguity resolution component calculates the similarity of current medical history using a dynamic time warping algorithm and selects historically verified effective treatment plans in conflicting rule scenarios. A dual closed-loop feedback channel operates hierarchically: the first closed-loop converts false positive signals into chromosome fitness penalty terms, optimizing the decision boundary at the minute level; the second closed-loop maps the effectiveness of the treatment into a weight correction vector, adjusts the residual network layer learning rate using the Nesterov momentum algorithm, and evolves model parameters at the daily level. This hierarchical optimization mechanism balances real-time response and long-term adaptability, achieving full-link management of individual dynamic variation.

Claims

1. Intelligent nursing monitoring system for post-operative tumor intervention, characterized by: include: Individualized data collection module collects patients' real-time vital signs data and structured historical health records after surgery, and outputs standardized data packages and historical feature index tables; a dynamic feature processing module, connected to the individualized data acquisition module, receiving the standardized data packet and the historical feature index table, performing motion artifact removal processing, extracting key pathological markers, and generating a fused feature tensor with a time tag; An individual adaptive modeling module is connected to the dynamic feature processing module, obtains the fused feature tensor, reconstructs the pre-trained population model parameters using an incremental transfer learning mechanism, generates an individualized risk prediction model including the patient's genetic background and pathophysiological characteristics, and outputs the model to the dynamic threshold optimization module; A dynamic threshold optimization module connects the individual adaptive modeling module and the dynamic feature processing module, loads the individualized risk prediction model, intercepts the feature change rate through a sliding time window, combines the genetic algorithm to iteratively correct the abnormality judgment boundary, generates feature data with annotated deviation levels, and transmits it to the intelligent early warning decision module; The intelligent early warning decision module is connected to the dynamic threshold optimization module, responds to the received feature data marked with deviation levels, calls the clinical knowledge graph to match the patient-specific contraindication rules, generates three-level early warning instructions and treatment suggestions, and simultaneously evaluates the clinical effectiveness of the treatment suggestions to obtain clinical effectiveness evaluation results. When there is a deviation between the treatment suggestions and the real-time vital signs data of the patient collected by the individualized data acquisition module, a false alarm feedback signal is generated, and the false alarm feedback signal is transmitted to the dynamic threshold optimization module to update the judgment boundary parameters. At the same time, the clinical effectiveness evaluation results are converted into weight correction vectors and fed back to the incremental transfer learning mechanism of the individual adaptive modeling module for correcting the pre-trained group model parameters.

2. The intelligent nursing monitoring system for post-operative tumor intervention according to claim 1 is characterized in that: The individualized data acquisition module includes: Wearable biosensor interface that receives blood oxygen saturation, ECG waveform, and blood pressure fluctuation data at a preset frequency to generate a real-time vital sign data stream; The medical information extraction port calls genetic testing conclusions, comorbidity diagnosis results, and test index time series data from the hospital information system through a standard interface and outputs a structured historical data set; The data integration component receives the real-time vital sign data stream and the structured historical data set, performs data standardization processing to generate a standardized data packet, and extracts key fields to construct the historical feature index table.

3. The intelligent nursing monitoring system for post-operative tumor intervention according to claim 2 is characterized in that: The dynamic feature processing module includes: a signal filtering component, receiving the standardized data packet, eliminating motion artifacts and retaining effective physiological bands, and outputting filtered physiological data; A pathology mark parsing component receives the historical feature index table, calculates the individual baseline fluctuation range of key pathology indicators, and outputs a pathology mark feature set; The time alignment component obtains the filtered physiological data and the pathological marker feature set, fuses the real-time monitoring data with the historical features based on the timestamp alignment mechanism, and generates a three-dimensional feature tensor with a time tag.

4. The intelligent nursing monitoring system for post-operative tumor intervention according to claim 3 is characterized in that: Also includes: The individual adaptive modeling module obtains the three-dimensional feature tensor with the time label, inputs the three-dimensional feature tensor into the residual connection network layer, and learns the patient-specific parameter offset; The pre-trained convolutional layer weights are dynamically adjusted based on the learning results, the weights of the pre-trained convolutional layer are reconstructed to generate an individualized risk prediction model, and the model is output to the dynamic threshold optimization module.

5. The intelligent nursing monitoring system for post-operative tumor intervention according to claim 4 is characterized in that: The dynamic threshold optimization module includes: The change rate calculation component divides the three-dimensional feature tensor with time tags according to a fixed time window and outputs the instantaneous change rate of physiological indicators; the genetic optimization controller retrieves the historical baseline fluctuation range extracted from the pathological marker feature set from the dynamic feature processing module, encodes the historical baseline fluctuation range as a chromosome gene site, calculates the false alarm rate and the number of missed events based on the deviation value between the three-level warning instructions generated by the intelligent warning decision module and the actual clinical status, evaluates the adaptability of the chromosome configuration to abnormality judgment, and generates optimized judgment boundary parameters.

6. The intelligent nursing monitoring system for post-operative tumor intervention according to claim 5, characterized in that: Also includes: The genetic optimization controller in the dynamic threshold optimization module receives the false alarm feedback signal transmitted by the intelligent early warning decision module, uses the false alarm feedback signal as a correction factor for fitness evaluation, performs a chromosome crossover mutation operation, generates an updated decision boundary parameter and replaces the optimized decision boundary parameter.

7. The intelligent nursing monitoring system for post-interventional tumor surgery according to claim 6, characterized in that: The intelligent early warning decision module includes: A clinical rule reasoning component receives feature data annotated with deviation levels, associates key pathological markers extracted by the dynamic feature processing module with drug interaction data from the patient contraindication database, and outputs a preliminary warning strategy; An ambiguity resolution component obtains the preliminary warning strategy. If the abnormal attributes in the feature data marked with the deviation level trigger a rule conflict in the preliminary warning strategy, the treatment strategy is screened from the preliminary warning strategy based on the similarity of the current medical history characteristics collected by the individualized data acquisition module, and integrated into the three-level warning instructions. The treatment strategy is screened from the preliminary warning strategy based on the similarity of the current medical history characteristics of the structured historical health records collected by the individualized data acquisition module, and integrated into the three-level warning instructions.

8. The intelligent nursing monitoring system for post-operative tumor intervention according to claim 7 is characterized in that: Also includes: The intelligent early warning decision module obtains effectiveness data of the treatment recommendation based on the clinical effectiveness evaluation results, encodes the mapping relationship between the effectiveness data and the execution results of the three-level early warning instructions generated by the intelligent early warning decision module into a vector matrix, generates a weight correction vector and transmits it to the individual adaptive modeling module; The individual adaptive modeling module receives the weight correction vector and dynamically adjusts the back propagation learning rate parameter of the residual connection network layer to update the individualized risk prediction model.

9. The intelligent nursing monitoring system for post-interventional tumor surgery according to claim 8, characterized in that: Also includes: The individual adaptive modeling module, the dynamic threshold optimization module and the intelligent early warning decision module constitute a dual closed-loop feedback channel, which includes a first closed-loop channel and a second closed-loop channel: First closed-loop channel: the intelligent early warning decision module outputs a false alarm feedback signal to the genetic optimization controller of the dynamic threshold optimization module, triggering iterative update of the decision boundary parameter; Second closed-loop channel: the intelligent early warning decision module outputs the weight correction vector to the individual adaptive modeling module, and continuously optimizes the individualized risk prediction model by adjusting the residual connection network layer parameters.

10. The intelligent nursing monitoring system for post-operative tumor intervention according to claim 9, characterized in that: Also includes: After the treatment suggestion is executed, the intelligent early warning decision module monitors the characteristic data of the deviation level output by the dynamic threshold optimization module in real time, extracts the change trend of physiological indicators in the characteristic data of the deviation level, and sends an optimization instruction to the dynamic threshold optimization module when no abnormal characteristic rebound of the key pathological marker is detected within a preset time; The genetic optimization controller responds to the optimization instruction, increases the fitness score weight of the successful matching anomaly judgment in the current chromosome configuration, generates an enhanced version of the judgment boundary parameter and replaces the optimized judgment boundary parameter.

Images