Data abnormal value identification method of chronic disease automatic monitoring network

Abstract

The invention relates to the technical field of chronic disease monitoring, and discloses a data abnormal value identification method for an automatic chronic disease monitoring network. The method comprises the following steps: acquiring a physiological index time sequence data set of a target patient at a plurality of monitoring nodes of a chronic disease monitoring network; dividing a time interval according to a data acquisition time sequence and the number of data points to obtain a plurality of sub-time sequence data segments; then associating a first physiological index feature of a first patient node having the same disease feature as the target patient; and finally, outputting abnormal identification data on a preset monitoring interface, wherein the data directly represents the abnormal probability of the physiological indexes of the target patient. According to the method, time sequence data time features and patient group features with the same disease features can be mined, the abnormal value identification effect is optimized, the abnormal probability is visually presented, medical personnel can rapidly master the abnormal conditions of the physiological indexes of the patients, and the method is suitable for automatic monitoring scenes of various chronic diseases.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chronic disease monitoring, in particular to a data outlier identification method of a chronic disease automatic monitoring network. BACKGROUND

[0002] In the current chronic disease management field, with the application of Internet of Things technology and medical monitoring equipment, chronic disease automatic monitoring networks have gradually become an important tool for real-time tracking of patient physiological indicators. Such monitoring networks are usually composed of multiple monitoring nodes distributed in different areas, which can continuously collect physiological indicators such as blood pressure, blood glucose, and heart rate of patients, forming a large amount of physiological indicator time series data. However, in actual application, how to accurately identify outliers from these continuously generated time series data has always been a problem that puzzles medical personnel and technical researchers. Traditional physiological indicator anomaly identification methods mostly rely on isolated analysis of data collected by a single monitoring node, without fully considering the time correlation and continuity characteristics of time series data. For example, some methods only use fixed thresholds to determine whether the data is abnormal. When the physiological indicator value at a certain time exceeds the pre-set threshold range, it is directly determined as abnormal data. However, the physiological indicators of chronic disease patients often have fluctuation rules, and the normal range of indicators for different patients at different time periods may differ. Simply relying on fixed thresholds for judgment may misjudge the indicator data within the normal fluctuation range as abnormal, or may miss some abnormal data that exceeds the threshold slightly but has potential risks, resulting in low accuracy of anomaly identification. Existing technologies usually only analyze the data of a single patient when identifying outliers, ignoring the reference value of physiological indicators of other patients with the same disease characteristics. The onset and development of chronic diseases have certain group characteristics. Patients with the same disease characteristics of the same type and similar course of chronic disease often have common trends in their physiological indicators. For example, patients with type 2 diabetes and a disease course of 5-10 years may have similar rising trends in blood glucose indicators if their diet control is poor. If the physiological indicator characteristics of patients with the same disease characteristics are not referenced during anomaly identification, and only the historical data of a single patient is relied on for judgment, it is difficult to fully grasp the abnormality of the patient's physiological indicators, especially for newly diagnosed patients or patients with less historical monitoring data, as there is not enough data to serve as a basis for comparison, and the reliability of outlier identification is greatly reduced. The abnormal identification result generated by the prior art is usually presented in the form of simple numerical values or words. When viewing, medical personnel needs to analyze comprehensively the monitoring time, historical data and other information of the patient to determine the specific situation and potential risks of the abnormal data. This presentation method not only increases the work burden of medical personnel, but also makes the intuitiveness of the abnormal identification result poor, which is not conducive to the medical personnel to quickly and accurately grasp the health status of the patient, and further affects the timeliness and effectiveness of subsequent diagnosis and treatment decisions. SUMMARY

[0003] The purpose of the present application is to provide a data outlier identification method for a chronic disease automatic monitoring network to solve the problems raised in the background art.

[0004] To achieve the above-mentioned purpose, the present application provides a data outlier identification method for a chronic disease automatic monitoring network, which comprises: At a plurality of monitoring nodes of a chronic disease monitoring network, a set of physiological index time series data corresponding to a target patient is obtained; According to the collection time sequence and the number of data points of the set of physiological index time series data, the time interval of the set of physiological index time series data is divided to obtain a plurality of sub-time series data segments; Correlate the first physiological index characteristics corresponding to the first patient node with the same disease characteristics as the target patient; Output abnormal identification data generated based on the plurality of sub-time series data segments and the first physiological index characteristics on a preset monitoring interface; Wherein, the abnormal identification data represents the abnormal probability of the physiological index of the target patient.

[0005] Preferably, according to the collection time sequence and the number of data points of the set of physiological index time series data, the time interval of the set of physiological index time series data is divided to obtain a plurality of sub-time series data segments, which comprises: Based on the collection time sequence and the number of data points, the cumulative time record of each sub-time series data segment based on the time interval is determined, and the initial time interval when each sub-time series data segment does not exist time gap is determined; When the total number of data points of each sub-time series data segment is less than the expected number of data points of the time interval, the physiological index characteristics of adjacent sub-time series data segments are compared, and the abnormal sub-data segment group with time gap is marked; According to the feature difference degree between the physiological index characteristics, the time gap prediction value of the abnormal sub-data segment group is determined; Insert the time gap prediction value in the initial time interval corresponding to the abnormal sub-data segment group, and recalculate the cumulative time of each sub-time series data segment to obtain the time interval of each sub-time series data segment.

[0006] Preferably, the method further comprises: extracting a physiological index change trend feature of adjacent sub-time sequence data segments; obtaining a peak point and a valley point in each of the physiological index change trend features, and calculating a peak point quantity difference and a valley point quantity difference in each of the physiological index change trend features; when the peak point quantity difference is greater than a first quantity threshold or the valley point quantity difference is greater than a second quantity threshold, determining that the abnormal sub-data segment group has a time gap and marking the abnormal sub-data segment group; when the peak point quantity difference is less than the first quantity threshold and the valley point quantity difference is less than the second quantity threshold, obtaining a matching point set of the closest peak point and valley point in each of the physiological index change trend features; statistically calculating a total distance value of each of the matching point sets, and when the total distance value is greater than a trend fit degree threshold, determining that the abnormal sub-data segment group has a time gap and marking the abnormal sub-data segment group.

[0007] Preferably, the method further comprises: identifying a blank time interval with data missing in the chronic disease monitoring network; selecting a second sub-time sequence data segment corresponding to a second target patient node adjacent to the blank time interval as a reference data segment; calculating a data mean value corresponding to the same physiological index type in each of the reference data segments; determining the data mean value as a filling value of the corresponding physiological index type in the blank time interval, and generating a filling time sequence data of the blank time interval according to the filling value.

[0008] Preferably, the method further comprises: constructing a first feature map corresponding to the target patient, wherein the first feature map comprises an identification feature of the target patient and a physiological index feature corresponding to the plurality of sub-time sequence data segments; obtaining a second feature map corresponding to the first patient node, wherein the second feature map comprises an identification feature of the first patient node and the first physiological index feature; merging the first feature map and the second feature map to generate a comprehensive feature path map; performing feature selection on the physiological index features in the comprehensive feature path map according to a preset time sequence association strategy to obtain a key physiological index feature set.

[0009] Preferably, the feature selection of the physiological indicator features in the comprehensive feature path atlas according to the preset time sequence correlation strategy obtains a key physiological indicator feature set, including: Based on the feature node transition probability defined by the preset time sequence correlation strategy, the connected physiological indicator feature nodes in the comprehensive feature path atlas are traversed; According to the time correlation degree between the physiological indicator feature nodes, target physiological indicator feature nodes satisfying a preset correlation strength threshold are sampled; The physiological indicator features corresponding to the target physiological indicator feature nodes are aggregated to form a key physiological indicator feature set.

[0010] Preferably, the output of the abnormal identification data generated based on the plurality of sub-time sequence data segments and the first physiological indicator feature on the preset monitoring interface includes: The three abnormal probability dimension data corresponding to the plurality of sub-time sequence data segments are respectively converted into operation dimension data; The three abnormal probability dimension data include time offset dimension data, indicator deviation dimension data, and trend deviation dimension data; The operation dimension data includes time dimension representation data, intensity dimension representation data, and duration dimension representation data; The target time dimension representation data is generated based on the conversion of the time offset dimension data; The target intensity dimension representation data is generated based on the conversion of the indicator deviation dimension data; The target duration dimension representation data is generated based on the conversion of the trend deviation dimension data; The abnormal identification data is generated according to the target time dimension representation data, the target intensity dimension representation data, and the target duration dimension representation data.

[0011] Preferably, the greater the time offset value in the time offset dimension data, the lower the trend frequency in the target time dimension representation data corresponding to the conversion; The greater the indicator deviation value in the indicator deviation dimension data, the higher the intensity value in the target intensity dimension representation data corresponding to the conversion; The faster the trend decline rate in the trend deviation dimension data, the shorter the duration in the target duration dimension representation data corresponding to the conversion.

[0012] Preferably, the method further includes: Testing the historical physiological indicator fluctuation range of the target patient to obtain a patient data fluctuation range test result; Adjusting the time interval length of the plurality of sub-time sequence data segments according to the patient data fluctuation range test result to obtain a time interval granularity adjustment result; redetermine time intervals of the physiological index time series data set based on the time interval granularity adjustment result.

[0013] Preferably, the chronic disease monitoring network comprises a plurality of monitoring nodes distributed in different body parts. The generating of the abnormality identification data according to the target time dimension representation data, target intensity dimension representation data and target duration dimension representation data comprises: generating independent abnormality identification data streams for the plurality of monitoring nodes respectively; synchronously outputting the independent abnormality identification data streams corresponding to the plurality of monitoring nodes to form comprehensive abnormality identification data.

[0014] Compared with the prior art, the present application has the following beneficial effects: In the abnormal value identification process, firstly, the target patient physiological index time series data set obtained by the plurality of monitoring nodes of the chronic disease monitoring network is divided into a plurality of sub-time series data segments in combination with the collection time sequence and the number of data points. This division method based on time dimension and data quantity can fully consider the continuity and periodicity of the physiological index time series data, and avoid the problem that the traditional method analyzes the time series data as a whole, and ignores the index change rule in the local time period due to the large data span. By disassembling the whole time series data into a plurality of sub-time series data segments, the fluctuation of the physiological index in each time period can be captured in more detail, so that the subsequent abnormal analysis can focus on the specific time interval, thereby more accurately discovering the potential abnormal data and reducing the abnormal omission or misjudgment caused by the overall data analysis. After associating the first physiological index characteristics of the first patient node with the same disease characteristics of the target patient, more dimensional reference basis is provided for the abnormal value identification of the target patient. Since the physiological index change of chronic disease patients has group commonality, the physiological index characteristics of patients with the same disease characteristics can reflect the typical index change rule of this type of disease in a specific stage. By combining and analyzing the sub-time series data segments of the target patient, the limitations of abnormal judgment relying only on the target patient's own data can be effectively made up. For patients with less historical monitoring data, the physiological index characteristics of patients with the same disease characteristics can be used as an important reference standard to help improve the basis for abnormal judgment; for patients with certain historical data, the index characteristics of patients with the same disease characteristics can further verify the rationality of the abnormality of the own data, reduce the misjudgment caused by individual data fluctuation, and make the abnormal identification result more reliable and comprehensive. The abnormality identification data generated based on the plurality of sub-time sequence data segments and the first physiological index feature is output on a preset monitoring interface, and the abnormality identification data directly represents the abnormal probability of the physiological index of the target patient. This presentation method has strong intuitiveness. Medical personnel do not need to integrate scattered monitoring data and reference information by themselves, but can directly obtain the abnormal probability corresponding to each sub-time sequence data segment through the preset monitoring interface, and can quickly understand the abnormal risk degree of the physiological index of the target patient in different time periods. Meanwhile, the quantitative presentation of the abnormal probability enables medical personnel to clearly compare the abnormal conditions in different time periods, accurately grasp the severity and change trend of the abnormal data, thereby more efficiently judging the health condition of the patient, reducing the time cost of information integration and analysis, and creating favorable conditions for timely taking corresponding medical intervention measures. The method fully utilizes the multi-node data acquisition advantage of the chronic disease monitoring network, integrates physiological index data of a plurality of monitoring nodes, and avoids the influence of errors or data missing of single node data acquisition on the abnormality identification result. The fusion application of multi-node data can construct a more complete and comprehensive physiological index data system of the target patient, provide a more solid data foundation for the division of the sub-time sequence data segment and the association of physiological index features of patients with the same disease feature, and further improve the accuracy of the abnormal value identification. Meanwhile, the operation process of the whole method can be realized by relying on the hardware facilities of the existing chronic disease monitoring network, without the need of additionally building a complex equipment system, and has strong feasibility in actual popularization and application, and can be widely applied to chronic disease monitoring scenes of different scales and different types, and provide more scientific and effective technical support for the health management of chronic disease patients. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A working principle diagram of the data abnormal value identification method of the chronic disease automatic monitoring network described in the present application; Figure 2 A flowchart for dividing time intervals and processing time gaps; Figure 3 A flowchart for filling blank time intervals; Figure 4 A flowchart for associating features of patients with the same disease feature; Figure 5 A flowchart for adjusting the granularity of time intervals. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0017] Referring to Figure 1 The application provides a data outlier identification method of a chronic disease automatic monitoring network, and the method comprises the following steps: At a plurality of monitoring nodes of the chronic disease monitoring network, a wearable device or an implantable sensor is used to acquire a physiological index time series data set of a target patient, including blood pressure, blood glucose, heart rate and other physiological parameters. The physiological index time series data set is stored in chronological order, and each data point contains a timestamp and a measured value. According to the chronological order and the number of data points, a sliding window algorithm is used to divide the physiological index time series data set into a plurality of continuous sub-time series data segments, and each sub-time series data segment corresponds to a time interval of a preset length. A data mining algorithm is used to retrieve a first patient node with the same disease characteristics as the target patient from the chronic disease monitoring network, and a first physiological index feature corresponding to the first patient node is extracted. The plurality of sub-time series data segments of the target patient and the first physiological index feature are input into an anomaly detection model, and the anomaly probability of the physiological index in each time interval is calculated. Finally, anomaly identification data is output in the form of a visual chart on a preset monitoring interface, including an anomaly probability curve on a time axis and an anomaly level mark.

[0018] Embodiment 1: Referring to Figure 2 In the chronic disease automatic monitoring network system, the integrity of time series data directly affects the accuracy of anomaly detection. This embodiment proposes a time gap detection and repair method based on feature comparison to solve the problem of possible collection discontinuity in the physiological index time series data set. The system first receives the original physiological index data stream from a plurality of monitoring nodes, which is stored in a ring buffer in chronological order to form a continuous time series data set. Each data point contains a time marker accurate to milliseconds and a calibrated physiological index value.

[0019] In the initial time interval division stage, the system uses a sliding partition algorithm with adaptive window size. The algorithm calculates the standard number of data points for each time interval according to the preset basic time unit (e.g. 5 minutes) and data acquisition frequency (e.g. 1 time per minute). For example, in a 5-minute interval, it should contain 5 data points under the standard acquisition frequency. The system traverses the entire time series data set in chronological order and calculates the cumulative time for each continuous data segment. When it detects that the actual number of data points in a certain time interval reaches more than 90% of the standard value, it marks the interval as a complete data segment and records its start and end timestamps. When the system detects that the number of data points in a certain sub-time series data segment is significantly insufficient (less than 80% of the standard value), it starts the time gap detection process. The system extracts the statistical features of the data segment and its adjacent two data segments, including moving average, standard deviation, and autocorrelation coefficient, and other key indicators. These features can reflect the basic change rule and short-term fluctuation pattern of physiological indicators. By comparing the feature differences of adjacent data segments, the system can determine whether there is a data acquisition interruption.

[0020] The system calculates the relative difference between adjacent data segments in the mean level. If the difference exceeds 2 times the standard deviation of the historical fluctuation range, further analysis is triggered. Then check the variance change rate between data segments. When the variance mutation exceeds the preset threshold, it indicates that there may be data missing. The comparison of autocorrelation coefficients is used to evaluate the consistency of data patterns. A significantly reduced autocorrelation coefficient suggests that there may be a time gap. For abnormal data segment groups marked as possibly having a time gap, the system starts the predictive repair mechanism. This mechanism analyzes the overall trend of data before and after the gap, taking into account factors such as circadian rhythm and patient individual characteristics. By establishing a regression model of the data segments before and after the gap, it predicts the possible data change path during the missing period. The system will insert virtual data points at the corresponding position on the time axis according to the prediction results. These data points have special markers to distinguish them from actual measurement values.

[0021] The prediction of the length of the time gap uses a feature-based estimation algorithm to analyze the slope change and extreme point distribution of the data before and after the gap, and estimates the most likely data missing duration based on the historical data patterns of such patients. For example, when the mean level of the data before and after the gap jumps, but the trend remains continuous, the system will judge that there is a short-term interruption in data acquisition; while the data before and after the gap presents completely different fluctuation patterns, it may correspond to a longer period of device offline.

[0022] After the gap prediction is completed, the system recalculates the time interval boundaries of each sub-sequence. The adjustment process maintains the original time sequence of the data, only marking and compensating at the detected gap location. The new time interval division ensures that each sub-sequence has complete time coverage, while preserving the true acquisition time information of the original data. This processing method not only guarantees the time sequence continuity of subsequent analysis, but also avoids distortion caused by data filling. The system also sets up a multi-level verification mechanism to ensure the accuracy of time gap detection. The primary verification confirms the specific time of the interruption by checking the device log; the intermediate verification compares the data integrity of multiple related physiological indicators; the advanced verification refers to the data pattern of patients with the same disease characteristics at the same period. When there is a contradiction between the results of each level of verification, the system will automatically select the most conservative estimation scheme to avoid error accumulation caused by excessive filling.

[0023] In terms of implementation details, the system adopts a hierarchical processing architecture, with the bottom layer data processing module responsible for receiving and buffering the original data; the intermediate analysis module performing time gap detection and feature comparison; and the upper decision module completing the final time interval adjustment. Asynchronous communication is carried out between modules through message queues to ensure the response performance of the system when processing large-scale data. The marking information of the time gap is stored in a special metadata area, associated with but stored independently from the original data, facilitating subsequent tracing and correction. For different types of physiological indicators, the system sets differentiated processing parameters, using shorter time windows and looser gap determination thresholds for indicators with greater fluctuations, such as instantaneous heart rate, and longer time windows and stricter determination conditions for relatively stable indicators, such as body temperature. This differentiated processing can adapt to the unique variation patterns of different physiological indicators, improving the accuracy of gap detection.

[0024] The system continuously updates the time gap detection model during operation, automatically adjusting the weight parameters of feature comparison by recording the deviation between each gap prediction and the actual device log. This self-learning mechanism enables the system to gradually adapt to the individual characteristics of specific patients and the characteristics of different monitoring devices, continuously improving the accuracy of time gap identification. All adjustment records are recorded in detail in the system log for quality control and algorithm optimization.

[0025] At the user interface level, the system provides time gap visualization tools. Medical personnel can view a complete timeline display, with actual measurement data and predicted filling data identified by different colors. The interface supports drilling functions, allowing users to view the detailed basis and adjustment process of each gap detection. This transparent display method helps medical personnel understand the decision-making process of the system and intervene manually when necessary.

[0026] Example 2: see Figure 3In the actual operation of the chronic disease monitoring network, the waveform feature analysis of physiological index data is a key link for identifying data abnormalities and time gaps. The system uses a multi-level waveform feature comparison method to achieve accurate detection and intelligent repair of data integrity. When the monitoring node collects the patient's blood pressure data, the original signal first passes through the pre-processing module for filtering and noise reduction to eliminate noise components caused by motion artifacts or device interference.

[0027] The system scans the entire time series data in a sliding window manner, identifying potential peak and trough points within each window. The window size is dynamically adjusted according to the signal sampling rate, for example, for ECG data sampled at 60 times per minute, a short window of 1 second is used; while for blood glucose data sampled at 1 time per hour, a longer window of 15 minutes is used. Each extreme point must meet the amplitude condition and duration condition to avoid misjudging short-term noise as valid extreme values. The identified peak and trough points are recorded in the feature point list and labeled with their time position and amplitude value.

[0028] The waveform comparison of adjacent data segments is carried out from three dimensions: extreme point number, extreme point distribution, and waveform shape. The system first counts the number of peaks and troughs in the two data segments and calculates the absolute difference. Considering the normal fluctuations under different physiological states, the system does not simply compare the number difference, but standardizes it with the typical fluctuation range of similar patients. For example, in the resting state, the allowed peak number difference of heart rate data is small; while in the exercise recovery period, the number difference is allowed. This dynamic threshold mechanism avoids false positives caused by excessive sensitivity.

[0029] When the difference in the number of extreme points does not exceed the threshold, the system enters a more detailed waveform matching analysis. The closest peak and trough points in time are selected from the two data segments to form a matching point pair. The composite distance of each pair of extreme points in time and amplitude is calculated, which considers the combined effects of time shift and amplitude change. The system performs a weighted summation of the distance values of all matching point pairs, with the weight coefficients dynamically adjusted according to the significance of the extreme points, with the weight of the main peak higher than that of the secondary fluctuation. The summation result reflects the overall fit degree of the two waveforms, and a lower threshold indicates continuous data, while a higher threshold indicates a possible time gap.

[0030] The system analyzes the trend of the waveform before and after the gap, including periodic characteristics, amplitude change rate, and phase information. For indicators with obvious periodicity such as respiratory rate, the system uses periodic extension to predict the missing waveform; for trend indicators such as blood glucose changes, the system performs linear prediction based on the historical change slope. The prediction process takes into account the influence of circadian rhythm, for example, physiological parameter fluctuations during the night are generally flatter than during the day. The generated predicted waveform is smoothly connected with the actual data before and after, maintaining the natural transition of the waveform.

[0031] For the completely blank time interval that appears in the monitoring network, the system starts the data reconstruction mechanism. The reconstruction process is not a simple mean filling, but a comprehensive calculation based on multi-source information. The system first searches for the same patient's data at the same time in other monitoring nodes to find relevant indicators as a reference. For example, when blood pressure data is missing, reference the heart rate and activity data at the same time. Second, query the monitoring data of other patients with the same disease characteristics at the same time to establish a group reference model. Finally, combined with the patient's personal historical data pattern, the most likely physiological parameter change curve is generated.

[0032] The data reconstruction adopts a hierarchical generation strategy. The basic layer establishes the overall trend of change, reflecting the macro trend of physiological indicators; the middle layer adds periodic fluctuations to simulate normal physiological rhythms; and the detail layer introduces reasonable random variations to maintain the natural characteristics of the data. The reconstruction results are verified multiple times: time series verification checks the reasonableness of the time distribution of the data points, range verification ensures that the values are within the physiological possible interval, and trend verification confirms that the change rate conforms to medical common sense. Filling data that fails the verification will be regenerated until all the constraints are met. When an extreme abnormal value is detected, it will not be immediately excluded, but its context environment will be analyzed. Some real clinical events such as arrhythmia can indeed produce abnormal waveforms, and these medically meaningful abnormalities need to be preserved. The system distinguishes between real pathological events and acquisition abnormalities by comparing the changes in multiple related indicators. For example, a sudden drop in heart rate alone may be a device malfunction, but if it is accompanied by a drop in blood pressure and blood oxygen, it is more likely to be a real clinical event.

[0033] The system continuously monitors the reliability indicators of the analysis results. When a large number of suspected gaps are found to be false positives by manual review, the decision threshold is automatically adjusted higher; otherwise, the threshold is lowered to improve sensitivity. This adaptive mechanism enables the system to adapt to the characteristics of different monitoring environments and patient populations. All parameter adjustments are recorded in the configuration log, forming a complete traceability chain. Medical staff can view the waveform superimposition of adjacent data segments side by side to intuitively understand the basis for the system's gap determination. The interface supports labeling and annotation functions for automatically filled data, allowing medical personnel to add clinical context information. These manual feedback information will be collected by the system to optimize subsequent analysis algorithms.

[0034] The data storage adopts a separate architecture of raw data and derived data. Raw measurement data is saved in read-only mode, and all filled and repaired data is stored in a separate derived area with detailed records of processing reasons and methods. This design ensures the integrity of the original data and provides appropriate data versions for different application scenarios. The data access interface provides corresponding data views according to the user's role. Researchers can obtain complete data containing all filling marks, while the clinical alarm system uses verified continuous data streams.

[0035] The system continuously monitors the performance of the waveform feature analysis module during operation, recording operating indicators such as processing time, resource consumption, and result consistency for each gap detection. When performance degradation or abnormal results are detected, a diagnostic process is automatically triggered to check hardware resources, data quality, and algorithm status. The diagnostic results generate maintenance recommendations, prompting possible manual intervention measures.

[0036] Diabetes monitoring focuses on trend analysis of blood glucose changes, using longer time windows and trend-sensitive feature extraction parameters; cardiovascular disease monitoring focuses on instantaneous heart rate variability, using short time windows and high time resolution analysis methods. These specialized configuration templates can be combined and applied according to the specific diagnosis of patients, providing personalized monitoring solutions. The entire waveform feature analysis process emphasizes the balance between medical rationality and engineering practicality. The system does not pursue mathematical perfect fitting, but ensures that the generated filling data conforms to clinical reality. All automatic processing retains the way for manual review and correction, forming a human-machine collaborative workflow. This design concept makes the system have the efficiency advantage of intelligent processing, while maintaining the necessary medical professional supervision.

[0037] Example 3: see Figure 4 In the chronic disease monitoring network, the construction and association of feature maps is the core technical path to achieve accurate anomaly detection. In the system initialization stage, a multi-dimensional feature map is created for the target patient, which is represented by an attribute graph structure. The nodes in the graph are divided into entity nodes and feature nodes: entity nodes contain patient identification information (medical record number, age layer, gender identification, diagnosed disease code, etc.), and feature nodes correspond to physiological indicator time series features (systolic pressure trend vector, blood glucose change matrix, etc.). Nodes are semantically associated through directed edges, such as "patient - has - diabetes" or "blood glucose feature - affects - systolic pressure feature". Each edge is attached with a time decay weight factor, reflecting the time effectiveness of the association relationship.

[0038] When the system retrieves the same disease feature patient node, it uses a semantic matching algorithm based on medical ontology. The algorithm first analyzes the disease feature description of the target patient and maps it to the node position in the standard medical terminology tree. When traversing in the monitoring network database, the semantic distance between the candidate patient and the ontology is calculated, and the patients within the distance threshold range are selected as similar nodes. For each selected same disease feature patient, the system extracts their physiological indicator time series data within the past three months and converts it into a standardized feature vector set through feature engineering.

[0039] The atlas merging process adopts a hierarchical fusion strategy. The first layer performs node alignment, matching nodes representing the same semantics in different atlases (e.g., the "morning fasting blood glucose" node). The second layer handles unmatched nodes, deciding whether to create new nodes or merge them into existing nodes based on node attribute similarity. The third layer performs relationship fusion, weighting and superimposing multiple edges between the same nodes. The final integrated feature path atlas contains a cross-patient feature association network, with edge weight calculation formula as follows: Wherein: represents the fusion weight of node i to j; is the number of associated paths; is the confidence coefficient of the kth original path (0.8~1.2); is a time decay function (take a high value for recent association); is the minimum time factor (default 30 days); is the association time span (days). This formula ensures that recent high-frequency associations get higher weights.

[0040] A transition probability matrix is established, and the matrix element represents the transition probability from feature node u to v, which is determined by three factors: time adjacency (the inverse of the time interval between the two features), numerical association (Pearson correlation coefficient), and semantic similarity (path distance in medical ontology). Second, perform random walk with restart, starting from the current feature node of the target patient, with a probability of 0.15 to jump back to the starting point to avoid walking away from the core area. Record the node access frequency during the walking process to form the importance ranking of the feature nodes.

[0041] The primary screening retains nodes with a transition probability greater than 0.6, and the secondary screening requires a time association degree (normalized value of dynamic time warping distance) less than 0.3. For nodes that pass the screening, the system calculates their mutual information value with the core features, and removes weakly related nodes with mutual information less than 0.05. The final target node set needs to meet the coverage index: it can explain more than 80% of the physiological index variation of the target patient.

[0042] The feature aggregation stage adopts attention mechanism weighted graph convolution operation. For each target node, the system calculates the weighted sum of its neighbor node feature vectors, and the weight is determined by the attention score. The attention score is calculated by a three-layer perception machine, and the input includes node feature difference, time interval and relationship type encoding. The aggregated feature vector is transformed by a nonlinear activation function to form a 128-dimensional feature representation. The feature representation of all target nodes forms the key physiological index feature set, which retains the spatiotemporal association characteristics of the original features.

[0043] The system implements a dynamic graph update mechanism during operation, automatically performing incremental graph updates every morning: newly added monitoring data triggers feature node updates, and outdated associations decay over time. When detecting changes in a patient's clinical status (e.g., new complications), the system initiates a graph reconstruction process, recalculating the patient set for the same disease characteristics and feature association paths. The graph version management uses snapshot technology, preserving the last seven days of historical versions and supporting traceability analysis of abnormal detection results.

[0044] In terms of computational optimization, the system uses a partitioned parallel processing strategy, dividing the comprehensive feature path graph into multiple subgraphs based on disease types and distributing them across different computing nodes. Sparse processing is implemented for large-scale graphs, pruning edges with weights below 0.1 and isolated nodes with degrees less than 3. Frequently accessed subgraphs are cached in memory databases, with the LRU algorithm used to manage cache replacement.

[0045] The system has a built-in medical rule engine that regularly scans for conflicting associations in the graph (e.g., "diabetes patients" and "normal insulin secretion" conflict nodes). Detected conflicting associations trigger an artificial review process, and clinicians confirm or correct the association relationships through the review interface. All correction operations are recorded in the audit log for subsequent optimization of the automatic construction algorithm.

[0046] Node positions are automatically arranged based on association strength, with different types of nodes using differentiated color coding (disease nodes are red, and physiological indicator nodes are blue). The interface supports time axis filtering, allowing viewing of feature association states within a specific time period. Healthcare professionals explore the implicit paths between feature nodes through drag-and-drop methods, and the system displays the path's confidence score and clinical interpretation in real-time.

[0047] The system monitors real-time changes in key features and triggers a multi-feature collaborative warning when three or more key features simultaneously deviate from the normal range. The warning level is dynamically calculated based on feature deviation and clinical importance, with priority given to features directly related to core diseases. All warning events are displayed in conjunction with the association paths in the feature graph, helping healthcare professionals quickly locate the root cause of the anomaly.

[0048] The construction of the feature graph not only follows the disease relationships defined by medical ontology but also incorporates statistical associations from actual monitoring data. Key feature selection balances temporal correlation and clinical interpretability, ensuring that the generated anomaly identifiers are both consistent with data rules and medically meaningful. The system continuously learns from medical staff feedback, gradually optimizing the feature association model, and forms a closed-loop knowledge evolution mechanism.

[0049] Embodiment 4: In the abnormality identification generation link of the chronic disease monitoring system, multidimensional data conversion and visual presentation constitute the core technical module. The system receives three abnormal probability dimension data from the preprocessing process: time offset dimension data reflects the periodic abnormality of physiological indicator fluctuation, indicator deviation dimension data represents the degree of measured value deviation from the benchmark, and trend deviation dimension data describes the irregularity of change rate. These original dimension data generate operational dimension data through specific conversion rules, and finally form an abnormality identification that can be directly interpreted.

[0050] The system calculates the cross-correlation function of the target patient's physiological indicator sequence and the reference sequence of the same type of patients, and determines the time lag value corresponding to the maximum correlation coefficient as the original time offset. This offset is input into a nonlinear converter, and the conversion rule is set as follows: when the lag value is in the 0-5 minute interval, the high-frequency trend (1 fluctuation per minute) is output; in the 5-15 minute interval, the medium-frequency trend (1 every 3 minutes) is converted; and more than 15 minutes is mapped to the low-frequency trend (1 every 10 minutes). The converted target time dimension representation data directly reflects the frequency characteristics of abnormal fluctuations, and high-frequency abnormalities prompt urgent attention.

[0051] The processing of the indicator deviation dimension data goes through two stages of standardization and normalization. The original deviation value is calculated by Mahalanobis distance, representing the degree of current measured value deviation from the patient's personal baseline distribution. The system normalizes this distance value by dividing it by the historical maximum deviation value to obtain a relative deviation coefficient in the 0-1 interval. The converter uses a piecewise linear interpolation strategy: coefficients 0-0.3 are mapped to intensity values 0-30; 0.3-0.6 correspond to 30-70; and 0.6 and above are mapped to 70-100. The generated target intensity dimension representation data forms a unified scale that quantifies the severity of abnormalities.

[0052] The system performs linear fitting on the physiological indicator sequence in the current time window, and takes the absolute value of the slope as the original trend change rate. The conversion rule establishes an inverse relationship between rate and duration: rate 0-0.1 units / minute corresponds to 120 minutes of duration; 0.1-0.2 corresponds to 60 minutes; 0.2-0.5 corresponds to 30 minutes; and more than 0.5 is mapped to 15 minutes. This nonlinear mapping ensures that rapidly deteriorating trends receive more urgent short-term warnings, see Table 1.

[0053] Table 1: Abnormal dimension data conversion mapping table The system automatically adjusts the weight coefficients of each dimension according to the patient's current clinical state: stable patients focus on the time dimension (weight 0.5), acute episode patients focus on the intensity dimension (weight 0.6), and recovery patients pay attention to the duration dimension (weight 0.4). The weighted calculation result generates a comprehensive abnormality score of 0-100, and a score exceeding 70 triggers a yellow warning, and a score exceeding 85 upgrades to a red alarm.

[0054] A spherical coordinate system is established in the operation interface: the radial axis represents the target intensity dimension data, the angular axis corresponds to the target time dimension data, and the height axis maps the target duration dimension data. Each abnormal event is represented as a colored sphere in the coordinate system, and the sphere diameter expands as the abnormal score increases. The system has an automatic perspective cruise function, which automatically focuses on the corresponding sphere when a high-risk abnormality is detected and intensifies the warning with a pulsating light effect.

[0055] When the user selects a specific abnormal sphere, the system automatically expands the associated panel: the left side displays the original physiological indicator curve, marking the abnormal occurrence position; the middle part presents the reverse mapping path of three-dimensional coordinates to the original dimension; the right side lists the historical treatment schemes of similar patients. The bottom of the panel is equipped with a time axis zoom control, supporting the viewing of changes in related indicators before and after the abnormal event.

[0056] Abnormalities with a comprehensive score of 70-85 trigger internal system records and are displayed as amber-colored markers in the monitoring interface; abnormalities with a score of 85-95 are pushed to the responsible nurse via instant message; abnormalities with a score of 95 or above activate the voice alarm and connect the on-duty doctor terminal simultaneously. All warning events generate structured logs, including detailed numerical values before and after three-dimensional data conversion and decision path traceability codes.

[0057] The system projects the patient's abnormal events within three months into the spherical coordinate system, forming a spatiotemporal distribution cloud map. The cloud map density coloring function can identify high-frequency abnormal regions, and the heat map layer displays the abnormal aggregation period. Medical staff can batch review analysis reports of similar abnormalities by framing the cloud map area.

[0058] The medical team can adjust the mapping parameters according to clinical experience, such as adjusting the blood glucose trend deviation threshold for diabetic patients from 0.2 to 0.15 units / minute. All custom rules need to be reviewed by a medical safety verification engine to avoid misjudgment caused by conflicting settings. Modified rules automatically generate version branches for differentiated application in different wards. The system is adapted to multiple devices from mobile terminals to command center large screens. The touch interface supports gesture operations: two-finger rotation adjusts the observation angle, pinch zoom switches the display granularity, and long-pressing the sphere retrieves the detail card. The command center large screen mode additionally provides a holographic projection option, projecting the three-dimensional model of the abnormal event into the physical space for multi-angle consultation.

[0059] The original three-dimensional data is saved in a time series database, the conversion rules are stored in a configuration library, and the visual state information is cached in an in-memory database. This architecture ensures that the core data processing performance is not affected by frequent interface interactions. The system automatically generates interface snapshots every five minutes to store as audit tracks, supporting post-mortem analysis.

[0060] Example 5: see Figure 5The dynamic optimization mechanism of the chronic disease monitoring network improves the monitoring accuracy by continuously adapting to the individual characteristics of the patients. In the system initialization phase, the basic monitoring frequency is established, and a uniform time window length is used for data segmentation for all physiological indicators. During operation, a background analysis thread is started to extract the fluctuation characteristics of the historical physiological indicator data of the target patient. This analysis uses a sliding window statistical method, with a window length configurable from 24 hours to 30 days, to calculate the coefficient of variation and the range ratio of each physiological indicator within the corresponding time period. The coefficient of variation reflects the degree of data dispersion, and the range ratio represents the fluctuation amplitude, which together form the data stability profile of the individual patient.

[0061] The system sets fluctuation level thresholds: when the coefficient of variation is less than 0.05 and the range ratio is less than 0.1, it is determined to be in a low fluctuation state, and the time interval of the sub-time sequence data segment is automatically extended to 150% of the basic value; when the coefficient of variation is between 0.05 and 0.15 or the range ratio is between 0.1 and 0.3, the basic time window is maintained; when any indicator exceeds the upper threshold, the time window is shortened to 80% of the basic value. The adjustment process uses a gradual change strategy, with each change not exceeding 20% of the current window, to avoid sudden changes in monitoring rhythm affecting trend analysis.

[0062] The system maintains a global time window configuration table, which records the optimal window parameters for each physiological indicator of each patient. When the configuration is updated, the relevant monitoring nodes are notified through a publish-subscribe mechanism. After receiving the new configuration, the nodes apply the new window settings at the start of the next complete monitoring cycle. During the transition period, dual-window parallel processing is used: the old window completes the current data analysis, and the new window starts collecting data for the next cycle, ensuring data continuity is not affected by configuration switching.

[0063] For monitoring nodes distributed in different body parts, the system implements a distributed anomaly detection architecture, with a lightweight analysis engine deployed on each monitoring node to independently process physiological indicator data collected at that site. The analysis engine consists of three processing units: a signal preprocessing unit for real-time filtering and noise reduction, a feature extraction unit for calculating statistical indicators and waveform features, and an anomaly scoring unit for generating independent anomaly probability values based on site characteristics. Each node outputs an independent anomaly identification data stream every second, with a data packet containing a timestamp, node number, anomaly score, and confidence indicator.

[0064] The system deploys a time synchronization server in the monitoring network, which coordinates the clocks of each node through the Precision Time Protocol. Each anomaly data packet is attached with a high-precision time stamp, with an error controlled within milliseconds. The central processing unit establishes a data reception buffer to align and match the data packets of multiple nodes within the same time window. The time alignment algorithm allows a time tolerance of ±100 milliseconds, and data packets exceeding the tolerance are placed in an asynchronous processing queue.

[0065] The central processor performs spatio-temporal correlation analysis on the independent abnormal data streams reported by each node: first, it detects the time correlation of abnormal events in different parts, and marks multiple parts that occur within 5 seconds as coordinated events; second, it analyzes the spatial distribution pattern of abnormal intensity to identify abnormal hotspots; finally, it calculates the comprehensive index of cross-part abnormalities, which is composed of the weighted sum of the scores of each part, and the weight coefficient is set according to the clinical importance of the part.

[0066] The main view displays a schematic diagram of the whole body, and each monitoring node position is marked with a color-coded dot, with the color depth reflecting the real-time abnormal score. Clicking on the node enters the local view, which displays the detailed physiological parameter curve and abnormal probability time series graph of that part. When a coordinated event triggers, a panoramic analysis panel automatically pops up, displaying the abnormal data streams of related parts and marking the spatio-temporal correlation path with a connection line. The bottom of the interface is set with a time axis zoom control, supporting observation granularity adjustment from seconds to days.

[0067] The system determines the alarm level according to the comprehensive abnormal index and duration: an index of 70-85 for 5 minutes triggers a primary reminder, and the interface displays an amber warning bar; an index of 85-95 for 3 minutes activates a medium-level alarm and pushes a message to the mobile terminal; an index of 95 or more or high-score abnormalities in two or more parts occur simultaneously, immediately starting an emergency alarm, triggering sound and light alarms and establishing a remote consultation channel. All alarm events generate a comprehensive report containing multi-node data, automatically associating with the patient's electronic medical record.

[0068] Real-time tracking of each node's calculation delay, data quality, and resource occupancy, automatically switching to a backup analysis strategy during abnormality. A network health report is generated every week, with key indicators such as time window adjustment frequency and multi-node synchronization success rate. The configuration management interface supports medical personnel to manually adjust the window parameters of specific patients, and all manual operations are recorded in the audit log and marked with the operator's identity.

[0069] Raw sensor data is stored locally on the edge node for 7 days, feature data is uploaded to the regional server for 6 months, and abnormal event data is permanently saved in the central database. The archiving system establishes a cross-layer index, supporting complete path queries from abnormal identification to original waveform. The data access interface sets role-based permission control, and researchers can access anonymized data sets for algorithm optimization.

[0070] Dynamic adjustment of the time window respects individual differences of patients and maintains the consistency of monitoring standards. Multi-node coordinated analysis considers both local abnormal sensitivity and overall systemic state, and converts complex data relationships into intuitive clinical decision support through a visual interface. The system continuously records the effects of configuration adjustments and alarm response results, forming a closed-loop optimization mechanism, enabling the monitoring network to continuously adapt to the evolving needs of chronic disease management.

[0071] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0072] While the embodiments of the application have been shown and described herein, it is to be understood that the scope of the application, jointly pointed out in the appended claims, is not to be limited to the above-described embodiments but can be otherwise variously changed, modified, replaced, and altered within the principles and spirit of the present application.

Claims

1. A method for identifying abnormal values ​​in data of a chronic disease automatic monitoring network, characterized in that: include: Obtaining a set of physiological indicator time series data corresponding to target patients at multiple monitoring nodes in the chronic disease monitoring network; Dividing the time interval of the physiological indicator time series data set into a plurality of sub-time series data segments according to the collection time sequence and the number of data points of the physiological indicator time series data set; Associating a first physiological indicator feature corresponding to a first patient node having the same disease feature as the target patient; Outputting abnormal identification data generated based on the multiple sub-time series data segments and the first physiological indicator feature on a preset monitoring interface; The abnormal identification data indicates the abnormal probability of the target patient's physiological indicators.

2. The method for identifying abnormal values ​​of data in a chronic disease automatic monitoring network according to claim 1, characterized in that: The step of dividing the time interval of the physiological indicator time series data set to obtain a plurality of sub-time series data segments according to the collection time sequence and the number of data points of the physiological indicator time series data set includes: Based on the acquisition time sequence and the number of data points, cumulative time records are performed on each of the sub-time series data segments based on the time interval, and an initial time interval when each of the sub-time series data segments does not have a time gap is determined; When the total number of data points in each of the sub-time series data segments is less than the expected number of data points in the time interval, comparing the physiological indicator characteristics of adjacent sub-time series data segments and marking abnormal sub-data segment groups with time gaps; Determining a time gap prediction value of the abnormal sub-data segment group according to the characteristic difference between the physiological indicator characteristics; The time gap prediction value is inserted into the initial time interval corresponding to the abnormal sub-data segment group, and the cumulative time of each sub-time series data segment is recalculated to obtain the time interval of each sub-time series data segment.

3. The method for identifying abnormal values ​​of data in a chronic disease automatic monitoring network according to claim 2, characterized in that: The step of comparing the physiological indicator characteristics of adjacent sub-time series data segments and marking abnormal sub-data segment groups with time gaps includes: Extracting the physiological index change trend characteristics of adjacent sub-time series data segments; Obtaining peak points and trough points in the trend characteristics of each physiological indicator change, and calculating the difference in the number of peak points and the difference in the number of trough points in the trend characteristics of each physiological indicator change; When the difference in the number of peak points is greater than a first threshold value or the difference in the number of trough points is greater than a second threshold value, it is determined that the abnormal sub-data segment group has a time gap and is marked; When the difference in the number of peak points is less than a first threshold value and the difference in the number of trough points is less than a second threshold value, obtaining a matching point set of the nearest peak points and trough points in the change trend characteristics of each physiological indicator; The total distance value of each matching point set is counted. When the total distance value is greater than a trend fit threshold, it is determined that the abnormal sub-data segment group has a time gap and is marked.

4. The method for identifying abnormal values ​​of data in a chronic disease automatic monitoring network according to claim 2, characterized in that: The method further comprises: Identifying gap time periods where data are missing in the chronic disease surveillance network; Selecting a second sub-time series data segment corresponding to a second target patient node adjacent to the blank time interval as a reference data segment; Calculating the mean value of the data corresponding to the same physiological indicator type in each reference data segment; The data mean is determined as a filling value of a corresponding physiological indicator type in the blank time interval, and filling time series data of the blank time interval is generated according to the filling value.

5. The method for identifying abnormal values ​​of data in a chronic disease automatic monitoring network according to claim 1, characterized in that: The associating the first physiological indicator feature corresponding to the first patient node having the same disease feature as the target patient includes: Constructing a first feature map corresponding to the target patient, wherein the first feature map includes an identification feature of the target patient and physiological indicator features corresponding to the multiple sub-time series data segments; Acquire a second feature map corresponding to the first patient node, wherein the second feature map includes an identification feature of the first patient node and the first physiological indicator feature; Combining the first feature map and the second feature map to generate a comprehensive feature path map; The physiological indicator features are selected in the comprehensive feature path map according to a preset time series association strategy to obtain a key physiological indicator feature set.

6. The method for identifying abnormal values ​​of data in a chronic disease automatic monitoring network according to claim 5, characterized in that: The feature selection of physiological indicator features in the comprehensive feature path map according to the preset time series association strategy is performed to obtain a key physiological indicator feature set, including: Based on the characteristic node transition probability defined by the preset time series association strategy, traversing the connected physiological indicator characteristic nodes in the comprehensive characteristic path graph; According to the temporal correlation between the physiological indicator feature nodes, sampling target physiological indicator feature nodes that meet a preset correlation strength threshold; The physiological indicator features corresponding to the target physiological indicator feature nodes are aggregated to form a key physiological indicator feature set.

7. The method for identifying abnormal values ​​of data in a chronic disease automatic monitoring network according to claim 1, characterized in that: Outputting abnormal identification data generated based on the multiple sub-time series data segments and the first physiological indicator feature on a preset monitoring interface includes: Convert the three abnormal probability dimension data corresponding to the multiple sub-time series data segments into operation dimension data respectively; The three abnormal probability dimension data include time offset dimension data, indicator deviation dimension data and trend deviation dimension data; The operation dimension data includes time dimension representation data, intensity dimension representation data and duration dimension representation data; Generate target time dimension representation data based on the time offset dimension data conversion; Generate target intensity dimension representation data based on the indicator deviation dimension data conversion; Generate target continuous dimension representation data based on the trend deviation dimension data conversion; The abnormal identification data is generated according to the target time dimension representation data, the target intensity dimension representation data and the target duration dimension representation data.

8. The method according to claim 7, wherein: The larger the time offset value in the time offset dimension data is, the lower the trend frequency in the corresponding converted target time dimension representation data is; The greater the indicator deviation value in the indicator deviation dimension data, the higher the intensity value in the corresponding converted target intensity dimension representation data; The faster the trend decline rate in the trend deviation dimension data is, the shorter the duration in the corresponding converted target duration dimension representation data is.

9. The method for identifying abnormal values ​​of data in a chronic disease automatic monitoring network according to claim 7, characterized in that: The method further comprises: Testing the fluctuation range of historical physiological indicators of the target patient to obtain a test result of the fluctuation range of patient data; Adjusting the time interval lengths of the plurality of sub-time series data segments according to the patient data fluctuation range test result to obtain a time interval granularity adjustment result; The time intervals of the physiological indicator time series data set are re-divided based on the time interval granularity adjustment result.

10. The method for identifying abnormal values ​​of data in a chronic disease automatic monitoring network according to claim 7, characterized in that: The chronic disease monitoring network includes a plurality of monitoring nodes distributed in different parts of the body; Generating the abnormality identification data according to the target time dimension representation data, the target intensity dimension representation data, and the target duration dimension representation data includes: generating independent abnormality identification data streams for the plurality of monitoring nodes respectively; The independent abnormality identification data streams corresponding to the multiple monitoring nodes are output synchronously to form comprehensive abnormality identification data.

Images