An information digital management system and method based on critical care medicine department

Abstract

The present invention relates to the field of medical information technology, and provides an information digitalization management system and method based on the critical care medicine department. The system includes a data acquisition module, a data processing module, and a visualization and alarm module. The data acquisition module obtains a patient monitoring data set, and each patient corresponds to a monitoring data sequence. The data processing module performs the following operations on each patient: determines the target monitoring data sequence and four associated monitoring cycles, including the first, second, third monitoring cycles and the fourth monitoring cycle group, and generates a corresponding monitoring subset; generates an early warning signal through a disease deterioration early warning model; if the signal is abnormal, analyzes the cause of the abnormality and the scope of influence, and generates a structured comprehensive disease report. The visualization and alarm module dynamically displays the patient status according to the report, and performs graded alarms on critical areas. The system achieves accurate early warning and monitoring through multi-cycle data fusion analysis and three-dimensional visualization.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and more specifically, to an information digitalization management system and method based on a critical care medicine department. Background Art

[0002] In the modern healthcare system, the intensive care unit (ICU) shoulders the crucial task of centralized management and treatment of critically ill patients. With the continuous advancement of medical technology, patient monitoring equipment in ICUs is becoming increasingly advanced, capable of collecting large amounts of vital sign data in real time, such as heart rate, blood oxygen levels, and blood pressure. This data is crucial for doctors to assess patients' conditions and formulate treatment plans. However, current ICU information management largely relies on traditional electronic medical record systems and manual record-keeping, resulting in low levels of data integration, analysis, and visualization. This requires medical staff to expend considerable time and effort on data collation and retrieval, reducing work efficiency. Furthermore, it is difficult to accurately monitor and provide early warning of changes in patients' conditions in real time, which can easily lead to missed treatment opportunities. Furthermore, existing systems have shortcomings in condition analysis and report generation, lacking structured and multi-dimensional presentations, hindering information sharing and collaboration within the medical team.

[0003] There are at least the following problems or defects in the existing technology: First, traditional information management systems cannot effectively integrate multi-source heterogeneous patient monitoring data, resulting in data dispersion and difficulty in unified management; second, there is a lack of real-time early warning mechanism for changes in patient conditions, and potential risks of disease deterioration cannot be discovered in time; third, the condition analysis and report generation methods are relatively simple and cannot meet the complex and changing clinical needs; finally, the existing visualization display effect is limited and cannot intuitively present the real-time status of patients and critical areas, affecting the efficiency and accuracy of medical decision-making. Summary of the Invention

[0004] The present invention provides an information digital management system and method based on the critical care medicine department.

[0005] In a first aspect of the present invention, there is provided an information digital management system based on a critical care medicine department, comprising:

[0006] A data acquisition module is used to obtain patient monitoring data sets corresponding to each patient in the target critical care department, where each patient has a corresponding patient monitoring data sequence;

[0007] A data processing module is configured to perform, for each of the patients:

[0008] Determine a patient monitoring data sequence corresponding to the patient as a target monitoring data sequence;

[0009] Determine the four monitoring cycle groups corresponding to the patient for the current monitoring cycle;

[0010] Generate a monitoring subset group corresponding to the monitoring period group according to the target monitoring data sequence;

[0011] Based on the monitoring subset group, a pre-trained disease deterioration warning model is used to generate a patient-specific disease deterioration warning signal;

[0012] In response to the early warning signal of worsening condition indicating abnormal condition of the patient, determining the abnormal cause analysis result and impact scope assessment result corresponding to the patient;

[0013] Generate a structured comprehensive report of the disease condition in a predetermined format based on the abnormal cause analysis results, target monitoring data sequence, and impact range assessment results;

[0014] The visualization and alarm module is used to perform dynamic visualization display for each patient based on the comprehensive condition report and to perform graded alarm processing for related critical areas.

[0015] As a further improvement of the present application, the four monitoring cycle groups include a first monitoring cycle, a second monitoring cycle, a third monitoring cycle and a fourth monitoring cycle; the first monitoring cycle includes the second monitoring cycle, the time range corresponding to the first monitoring cycle is greater than the second monitoring cycle, the fourth monitoring cycle includes the third monitoring cycle, and the central monitoring cycle corresponding to the third monitoring cycle is in the same period as the current monitoring cycle.

[0016] As a further improvement of this application, the dynamic visual display includes:

[0017] Determine the impact level of each sub-area in the target critical care department based on the comprehensive reports of each condition;

[0018] Assign visualization strategies to each patient and sub-area based on the disease impact level and disease summary report. The visualization strategies include transparency level mapping and color-coded mapping.

[0019] According to the visualization strategy, abnormal status of patients and sub-areas are rendered in real time in the 3D model of the virtual ward.

[0020] As a further improvement of the present application, the step of determining the monitoring cycle configuration corresponding to the patient includes:

[0021] According to the patient's clinical classification and sub-region, the initial monitoring cycle configuration is matched from the preset association table;

[0022] Obtain historical data on fluctuations in patients' vital signs and calculate their periodic characteristics;

[0023] Based on the periodic characteristics, the initial monitoring period configuration is dynamically adjusted to generate an adapted first adjustment period, a second adjustment period, a third adjustment period, and a fourth adjustment period;

[0024] Combined with the current monitoring cycle, the final first monitoring cycle, second monitoring cycle, third monitoring cycle and fourth monitoring cycle are generated.

[0025] As a further improvement of this application, the steps of generating a comprehensive condition report include:

[0026] Generate multi-dimensional trend graphs corresponding to the target monitoring data sequence, including the temporal changes of heart rate, blood oxygen, and blood pressure;

[0027] Obtain timestamps, interventions, and associated video clips of patients’ abnormal treatment events;

[0028] Generate an encrypted access link to the video clip;

[0029] The abnormal cause analysis results, multi-dimensional trend maps, impact range assessment results, abnormal treatment events and encrypted links are integrated into a structured comprehensive disease report according to the clinical report template.

[0030] As a further improvement of this application, it also includes:

[0031] In response to user-triggered interactive operations on a target patient in the virtual ward model, an information category selection interface including vital signs, treatment records, and video playback pops up;

[0032] Extract corresponding data from the comprehensive disease report according to the category selected by the user;

[0033] The data of the target patient's bed coordinate position is superimposed on the virtual ward model.

[0034] As a further improvement of this application, the operating logic of the disease deterioration warning model includes:

[0035] The monitoring subset group includes a first monitoring subset, a second monitoring subset, a third monitoring subset, and a fourth monitoring subset;

[0036] Extracting the data portion in the first monitoring subset that does not overlap with the second monitoring subset as an independent feature analysis set;

[0037] Analyzing the independent feature analysis set through a first feature extraction module to generate a first feature vector reflecting long-term fluctuations;

[0038] fusing the first feature vector and the original data of the first monitoring subset through a second feature extraction module to generate a second feature vector;

[0039] Inputting the second monitoring subset into the second feature extraction module to generate a third feature vector;

[0040] horizontally fusing the first eigenvector and the second eigenvector to generate a first fused feature;

[0041] Vertically fuse the first eigenvector and the third eigenvector to generate a second fused feature;

[0042] The first fusion feature and the second fusion feature are input into the temporal attention module to predict the clinical indicator parameters of the target in the future monitoring period;

[0043] The predicted parameters are compared with the historical data of the same period of the third monitoring subset for similarity. If the difference exceeds the threshold, an early warning is triggered;

[0044] Combined with the cross-cycle consistency verification results of the fourth monitoring subset, a weighted warning signal of disease deterioration is generated, in which the temporal attention module adopts a deep learning algorithm based on the attention mechanism.

[0045] As a further improvement to this application, the processing logic when the similarity comparison does not meet the threshold includes:

[0046] selecting a historical data subset that matches a target future period date from the third monitoring subset;

[0047] Extracting the historical data of the same period of all associated subsets from the fourth monitoring subset;

[0048] Compare the statistical distribution characteristics of the forecast parameters with those of historical data. If the distribution deviation exceeds the preset range, a manual review process is triggered.

[0049] Update the model weights based on the review results and regenerate the early warning signal.

[0050] In a second aspect of the present invention, a method for digital information management based on a critical care medicine department is provided, characterized by comprising:

[0051] Obtaining a patient monitoring data set corresponding to each patient in a target critical care unit, wherein each patient has a corresponding patient monitoring data sequence;

[0052] For each of the individual patients, perform the following generation steps:

[0053] Determine a patient monitoring data sequence corresponding to the patient as a target monitoring data sequence;

[0054] Determine a first monitoring cycle, a second monitoring cycle, a third monitoring cycle, and a fourth monitoring cycle corresponding to the patient for the current monitoring cycle;

[0055] Generate, according to the target monitoring data sequence, a first monitoring subset corresponding to the first monitoring period, a second monitoring subset corresponding to the second monitoring period, a third monitoring subset corresponding to the third monitoring period, and a fourth monitoring subset corresponding to the fourth monitoring period;

[0056] Generate a condition deterioration warning signal for the patient using a pre-trained condition deterioration warning model according to the first monitoring subset, the second monitoring subset, the third monitoring subset, and the fourth monitoring subset;

[0057] In response to the early warning signal of worsening condition indicating abnormal condition of the patient, determining the abnormal cause analysis result and impact scope assessment result corresponding to the patient;

[0058] Generate a structured comprehensive report of the disease condition in a predetermined format based on the abnormal cause analysis results, target monitoring data sequence, and impact range assessment results;

[0059] Based on the comprehensive reports on each condition, dynamic visualization is performed for each patient, and graded alarm processing is performed on the related critical areas.

[0060] As a further improvement of this application, the first monitoring cycle includes the second monitoring cycle, the time range corresponding to the first monitoring cycle is larger than the second monitoring cycle, the fourth monitoring cycle includes the third monitoring cycle, and the central monitoring cycle corresponding to the third monitoring cycle is in the same cycle as the current monitoring cycle.

[0061] The above-described embodiments of the present invention have at least the following beneficial effects: Based on multi-cycle data fusion analysis technology, the present invention can achieve accurate monitoring and early warning of a patient's condition through the synergistic effects of the first, second, third, and fourth monitoring cycles. The temporal attention mechanism and feature fusion algorithm of the disease deterioration warning model can improve the accuracy of disease deterioration prediction. Combined with abnormal cause analysis and impact range assessment, comprehensive and reliable data support can be provided for clinical decision-making.

[0062] This invention can also intuitively present patient condition trends and the risk level of critical areas through dynamic visualization and graded alarm mechanisms. Real-time rendering and interactive data display of virtual ward 3D models can optimize medical staff's monitoring efficiency. Structured reports based on encrypted video clips and multi-dimensional trend maps can also improve the medical team's response speed to critical situations, ultimately enhancing the overall management level and treatment effectiveness of critical care medicine departments. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily apparent by reading the following detailed description with reference to the accompanying drawings, in which several embodiments of the present invention are shown by way of example and not limitation, in which:

[0064] Figure 1 A schematic diagram of the structure of an information digital management system based on the critical care medicine department provided by one embodiment of the present invention;

[0065] Figure 2 A flowchart of a method for digital management of information in a critical care unit according to an embodiment of the present invention;

[0066] Figure 3 The figure is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided solely to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. Rather, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0068] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software.

[0069] It should be noted that any number of elements in the drawings is for illustration only and not for limitation, and any naming is only for distinction and does not have any limiting meaning.

[0070] Reference below Figure 1 , Figure 1 This is a schematic diagram of the structure of an information digital management system based on the critical care medicine department provided by one embodiment of the present invention. Figure 1 As shown, an information digital management system based on the critical care medicine department includes:

[0071] The data acquisition module 101 is used to obtain a patient monitoring data set corresponding to each patient in the target critical care medicine department, wherein each patient has a corresponding patient monitoring data sequence;

[0072] The data processing module 102 is configured to perform the following generation steps for each of the patients:

[0073] Determine a patient monitoring data sequence corresponding to the patient as a target monitoring data sequence;

[0074] Determining a first monitoring cycle, a second monitoring cycle, a third monitoring cycle, and a fourth monitoring cycle corresponding to the patient for the current monitoring cycle, wherein the first monitoring cycle includes the second monitoring cycle, the time range corresponding to the first monitoring cycle is greater than the second monitoring cycle, the fourth monitoring cycle includes the third monitoring cycle, and the central monitoring cycle corresponding to the third monitoring cycle is within the same week as the current monitoring cycle;

[0075] Generate, according to the target monitoring data sequence, a first monitoring subset corresponding to the first monitoring period, a second monitoring subset corresponding to the second monitoring period, a third monitoring subset corresponding to the third monitoring period, and a fourth monitoring subset corresponding to the fourth monitoring period;

[0076] Generate a condition deterioration warning signal for the patient using a pre-trained condition deterioration warning model according to the first monitoring subset, the second monitoring subset, the third monitoring subset, and the fourth monitoring subset;

[0077] In response to the early warning signal of worsening condition indicating abnormal condition of the patient, determining the abnormal cause analysis result and impact scope assessment result corresponding to the patient;

[0078] Generate a structured comprehensive report of the disease condition in a predetermined format based on the abnormal cause analysis results, target monitoring data sequence, and impact range assessment results;

[0079] The visualization and alarm module 103 is used to perform dynamic visualization display for each patient based on the obtained comprehensive reports on each condition, and to perform graded alarm processing on the associated critical areas.

[0080] It should be noted that the information digital management system based on the critical care department proposed in the present invention is intended to achieve accurate monitoring and management of critical care patients through functional modules such as data acquisition, processing, visualization and alarm. Among them, the data acquisition module refers to a component used to obtain the vital signs and related monitoring data of each patient in the critical care department. These data include but are not limited to continuous monitoring signals such as heart rate, blood oxygen, blood pressure, and the patient's treatment records, imaging data, etc. These data are organized into patient monitoring data sets, and each patient corresponds to a data sequence for subsequent analysis and processing. The data processing module is the core part for analyzing and processing the collected data. By processing the patient monitoring data sequence, it generates subsets for different monitoring cycles, and uses the pre-trained disease deterioration warning model to predict changes in the patient's condition, thereby achieving real-time monitoring and early warning of the patient's condition.

[0081] Specifically, the first monitoring cycle, the second monitoring cycle, the third monitoring cycle and the fourth monitoring cycle are monitoring time periods divided according to different time ranges and monitoring needs. The first monitoring cycle is usually a longer time range, such as a week or a month, used to observe the overall trend of changes in the patient's condition; the second monitoring cycle is a shorter time period included in the first monitoring cycle, such as a day or a few days, used to capture short-term fluctuations in the patient's condition; the third monitoring cycle is a specific time period centered on the current monitoring cycle, usually a specific time point within a week, used to analyze the patient's condition changes around that time point; the fourth monitoring cycle includes multiple third monitoring cycles for cross-cycle comparative analysis. The division of these monitoring cycles can be dynamically adjusted according to the patient's clinical characteristics and the severity of the disease to better meet the needs of different patients. For example, for patients with relatively stable conditions, the monitoring cycle can be appropriately extended; for patients with rapidly changing conditions, the monitoring cycle needs to be shortened to detect changes in the condition more promptly.

[0082] For example, in response to the cross-cycle monitoring needs of critically ill patients, the system uses a master-slave time window design to construct a fourth monitoring cycle. The main cycle is set to 72 hours, which is internally divided into 9 overlapping 8-hour sub-cycles, and a 2-hour overlap area is set between adjacent sub-cycles. Data collection adopts a sliding window mechanism, with a sampling interval of one minute, and a circular buffer to achieve seamless splicing of time series data. For example, in the monitoring scenario of sepsis patients, when the central venous pressure fluctuates by more than ±5 mmHg, the intensive sampling mode of the fourth monitoring cycle is automatically activated, the sub-cycle is shortened to 4 hours, and the video monitoring synchronization mark is started.

[0083] Preferably, in the process of generating a warning signal for worsening of the condition, a model construction method based on deep learning can be adopted. Specifically, features are first extracted from a large amount of historical patient monitoring data, including time series variation features of vital signs, timestamp features of treatment events, etc., as input parameters of the model. Then, the model is trained using labeled samples of worsening of the condition so that the model can learn the characteristic patterns of worsening of the patient's condition. In actual applications, the patient monitoring data collected in real time is input into the trained model, and the model will output the probability of worsening of the patient's condition or a warning signal based on the characteristics of the input data. In addition, during the data processing process, the collected data can also be pre-processed, such as filtering, denoising, normalization and other operations, to improve the quality of the data and the accuracy of the model. For example, for heart rate data, high-frequency noise can be removed by a filtering algorithm to make it smoother and facilitate subsequent analysis and processing.

[0084] In some embodiments, the dynamic visual presentation includes:

[0085] Determine the impact level of each sub-area in the target critical care department based on the comprehensive reports of each condition;

[0086] Assign visualization strategies to each patient and sub-area based on the disease impact level and disease summary report. The visualization strategies include transparency level mapping and color-coded mapping.

[0087] According to the visualization strategy, abnormal status of patients and sub-areas are rendered in real time in the 3D model of the virtual ward.

[0088] It should be noted that the dynamic visualization display function mentioned in the present invention is an important component of the system. It visualizes the key information in the comprehensive medical report so that medical staff can more intuitively understand the distribution and changes of patients' conditions in the critical care department. Among them, the disease impact level refers to the level divided according to the severity of the patient's condition, the trend of disease change, and the potential impact on the surrounding environment (such as other patients in the ward), usually divided into mild, moderate, severe and other levels. The visualization strategy includes transparency level mapping and color coding mapping. These strategies are used to display the abnormal status of patients and sub-areas with different visual effects in the three-dimensional model of the virtual ward, so that medical staff can quickly identify the problem.

[0089] Specifically, the three-dimensional model of the virtual ward is a digital virtual environment that simulates the actual layout and structure of the intensive care unit ward, including beds, medical equipment, ward areas, etc. In this model, the location and status of each patient can be displayed through visualization strategies. Transparency level mapping refers to adjusting the transparency of patients or sub-areas in the virtual model according to the level of disease impact. The more serious the disease, the lower the transparency, thereby highlighting the areas with more serious diseases. Color-coded mapping uses different colors to represent different disease states or impact levels, such as red for severe abnormalities, yellow for moderate abnormalities, and green for normal conditions. The settings of these parameters can be adjusted according to actual needs to ensure that the visualization effect can clearly convey key information.

[0090] The severity of illness can be defined using the SOFA scoring system, combined with a comprehensive assessment of vital sign parameters. For example, the SOFA scoring system assesses organ dysfunction in six areas: consciousness, circulatory system, respiratory system, liver function, coagulation function, and renal function. Each area is scored out of 4, for a total of 24 points. Higher scores indicate more severe illness. Furthermore, we consider the degree of abnormality in vital sign parameters (such as heart rate, blood pressure, and blood oxygen saturation) and integrate them with the SOFA score. For example, if a patient's heart rate is consistently above 120 beats / minute or below 40 beats / minute, an additional 2 points are added; if blood pressure is consistently below 90 / 60 mmHg, an additional 3 points are added. This comprehensive scoring system categorizes the severity of illness into four levels: mild (0-4 points), moderate (5-9 points), severe (10-14 points), and extremely severe (15 points and above), providing a basis for subsequent dynamic visualization and graded alerting.

[0091] Preferably, in the process of realizing dynamic visualization display, the system can further optimize the processing and display methods of data. For example, when generating a comprehensive report on the condition, the system can automatically extract key data, such as abnormal values ​​of the patient's vital signs, time points of treatment events, etc., and associate them with the corresponding positions in the three-dimensional model of the virtual ward. When medical staff view a patient or sub-area in the virtual model, the system can render and display the relevant abnormal status in real time, including changes in transparency and color. In addition, the system can also provide interactive functions, allowing medical staff to choose to view detailed information of specific patients or sub-areas through mouse clicks or gestures, such as real-time trend charts of vital signs, treatment records, etc., thereby further improving the readability and practicality of the information.

[0092] When it comes to selecting a graphics engine for a 3D model rendering strategy, the Unity3D engine can be chosen. Using Unity3D's ShaderLab shader programming language, custom shader programs can be written to achieve real-time rendering of the 3D model of the virtual ward. For example, in order to render abnormal conditions of patients and their sub-areas, a transparency level mapping and color coding mapping strategy can be adopted. Different transparency values ​​and color values ​​are assigned to each patient and sub-area based on the severity of the disease. In the shader program, dynamic rendering of the model is achieved by controlling the transparency and color properties of pixels. At the same time, the Unity3D engine also supports scene management, lighting calculation, and texture mapping, which can ensure rendering performance and effects in complex scenarios (such as large critical care medicine departments), providing medical staff with a smooth visualization experience.

[0093] In some embodiments, the step of determining the monitoring cycle configuration corresponding to the patient includes:

[0094] According to the patient's clinical classification and sub-region, the initial monitoring cycle configuration is matched from the preset association table;

[0095] Obtain historical data on fluctuations in patients' vital signs and calculate their periodic characteristics;

[0096] Based on the periodic characteristics, the initial monitoring period configuration is dynamically adjusted to generate an adapted first adjustment period, a second adjustment period, a third adjustment period, and a fourth adjustment period;

[0097] Combined with the current monitoring cycle, the final first monitoring cycle, second monitoring cycle, third monitoring cycle and fourth monitoring cycle are generated.

[0098] It should be noted that the step of determining the corresponding monitoring cycle configuration for the patient mentioned in the present invention is to better adapt to the individual differences and changes in the condition of different patients, so as to achieve accurate monitoring. Among them, clinical typing refers to the classification based on the patient's disease type, disease severity, etc., which is an important basis for determining the initial monitoring cycle configuration. The vital signs fluctuation historical data refers to the changes in the patient's heart rate, blood pressure, blood oxygen and other vital signs data recorded during the previous monitoring process. These data reflect the patient's physiological characteristics and the periodic laws of disease changes. By analyzing the periodic characteristics of these data, the initial monitoring cycle configuration can be dynamically adjusted to generate a monitoring cycle group that is more in line with the patient's current condition.

[0099] Specifically, the initial monitoring cycle configuration is matched from a preset association table, which pre-sets different monitoring cycle templates based on the patient's clinical classification and sub-region. For example, for patients with mild conditions, the initial monitoring cycle may be longer; for patients with critical conditions, the initial monitoring cycle is shorter. Periodic characteristics refer to the regular changes in the patient's vital sign data over time, such as the circadian rhythm of heart rate and the fluctuation period of blood pressure. By calculating these characteristics, the initial monitoring cycle can be adjusted. For example, if the patient's vital sign fluctuations have a clear circadian rhythm, the monitoring cycle can be set to dynamically adjust in 24-hour units. The resulting first, second, third, and fourth adjustment cycles will be combined with the current monitoring cycle to form the final monitoring cycle group for subsequent disease monitoring and analysis.

[0100] Preferably, in the process of determining the monitoring cycle configuration, the following steps can be used for optimization. First, the initial monitoring cycle configuration is matched from the preset association table according to the patient's clinical classification and the sub-region in which it is located. For example, for patients with severe respiratory failure, the initial monitoring cycle may be set to monitor blood oxygen saturation once an hour. Next, the patient's vital sign fluctuation historical data over the past period of time is obtained, and its periodic characteristics are calculated through a time series analysis algorithm. For example, it is found that the patient's blood oxygen saturation fluctuates greatly at night, with a period of about 8 hours. Then, the initial monitoring cycle is dynamically adjusted according to these periodic characteristics. For example, the night monitoring cycle is shortened to once every half hour to better capture changes in the condition. Finally, combined with the current monitoring cycle, the final monitoring cycle group is generated to ensure that the monitoring cycle can accurately reflect the patient's condition changes and provide more accurate data support for subsequent disease deterioration warnings and comprehensive report generation.

[0101] When calculating the periodic characteristics of historical data on vital sign fluctuations, time-frequency analysis methods such as Fourier transform and wavelet analysis can be used. Fourier transform converts time-domain signals into frequency-domain signals, thereby identifying periodic frequency components within the signal. For example, for heart rate data, Fourier transform can produce a spectrum that reveals the periodic frequencies of heart rate fluctuations, such as circadian rhythms (approximately once per 24 hours) and respiratory-related frequencies (approximately once per several seconds). Wavelet analysis is suitable for analyzing non-stationary signals. By decomposing the signal at different scales, it can capture periodic variations at different time scales. For example, for blood pressure data, wavelet analysis can reveal the periodic fluctuation characteristics of blood pressure over both short-term (minutes to hours) and long-term (days to weeks) periods. Using these time-frequency analysis methods, we can accurately calculate the periodic characteristics of vital sign data, providing a basis for dynamic adjustment of subsequent monitoring cycles.

[0102] In some embodiments, the step of generating a comprehensive medical report includes:

[0103] Generate multi-dimensional trend graphs corresponding to the target monitoring data sequence, including the temporal changes of heart rate, blood oxygen, and blood pressure;

[0104] Obtain timestamps, interventions, and associated video clips of patients’ abnormal treatment events;

[0105] Generate an encrypted access link to the video clip;

[0106] The abnormal cause analysis results, multi-dimensional trend maps, impact range assessment results, abnormal treatment events and encrypted links are integrated into a structured comprehensive disease report according to the clinical report template.

[0107] It should be noted that the step of generating a comprehensive medical report mentioned in the present invention is to integrate the patient's monitoring data, abnormal cause analysis results, and impact range assessment results to form a structured report so that medical staff can quickly and comprehensively understand the patient's condition. Among them, the multi-dimensional trend map refers to the visualization of the patient's vital signs data (such as heart rate, blood oxygen, blood pressure, etc.) in the form of time series changes, so as to intuitively present the changing trend of the patient's condition. The timestamps, intervention measures and associated video clips of abnormal treatment events refer to special situations and their treatment measures that occurred during the patient's treatment process. This information is of great reference value for analyzing changes in the patient's condition. The encrypted access link is to protect the patient's privacy and data security, ensuring that only authorized personnel can access the relevant video clips.

[0108] Specifically, the multi-dimensional trend map is formed by analyzing the target monitoring data sequence, extracting the time series data of vital signs such as heart rate, blood oxygen, and blood pressure, and plotting them in the same coordinate system. For example, the horizontal axis represents time, and the vertical axis represents the values ​​of heart rate, blood oxygen, and blood pressure respectively. By distinguishing different colors or lines, the changing trends of various vital signs can be clearly displayed. Abnormal treatment events refer to special situations that occur during the patient's treatment process, such as sudden arrhythmia, respiratory arrest, etc. The timestamp records the specific time point of the event, and the intervention measures are the treatment methods taken by medical staff for the event, such as drug injection, electric defibrillation, etc. The associated video clip is a real-time video recording of the event, which is used to assist analysis and review. Integrating this information into a structured comprehensive report on the condition can provide medical staff with a comprehensive reference for the condition.

[0109] Preferably, in the process of generating a comprehensive report on the condition, the way of integrating and displaying information can be further optimized. For example, when generating a multi-dimensional trend map, a data smoothing algorithm can be used to process the original monitoring data to reduce noise interference and make the trend map clearer and more accurate. At the same time, for the records of abnormal treatment events, the type, severity and processing results of the event can be marked in detail, and the key frames of the associated video clips can be extracted so that medical staff can quickly locate the critical moment when the event occurred. When integrating the report, according to the requirements of the clinical report template, the abnormal cause analysis results, multi-dimensional trend maps, impact range assessment results and abnormal treatment events and other information can be reasonably arranged to form a comprehensive report on the condition with rich content and clear structure. In addition, indexing and keyword retrieval functions can be added to the report to facilitate medical staff to quickly find the required information.

[0110] Encrypted access links can use a temporary access key generation scheme based on JWT tokens. JWT (JSON Web Token) is an open standard (RFC 7519) that defines a compact and self-contained method for transmitting secure information between parties as a JSON object. Each encrypted access link contains a JWT token, which contains the following information: a unique identifier for the accessed video clip, the identity of the accessor, the validity period of the access (for example, expiring after 1 hour), and a signing key (used to verify the authenticity of the token). The signing key is generated using the HMAC SHA256 algorithm to ensure the security of the token. For example, when medical staff need to access the video clip associated with an abnormal treatment event, the system will generate an encrypted access link with a JWT token. After clicking the link, the medical staff can view the video clip within the validity period. After the validity period expires, the link will expire, and the video content will be inaccessible, thereby protecting the patient's privacy and data security.

[0111] In some embodiments, it further includes:

[0112] In response to user-triggered interactive operations on a target patient in the virtual ward model, an information category selection interface including vital signs, treatment records, and video playback pops up;

[0113] Extract corresponding data from the comprehensive disease report according to the category selected by the user;

[0114] The data of the target patient's bed coordinate position is superimposed on the virtual ward model.

[0115] It should be noted that the interactive operation function mentioned in the present invention is to enhance the interactivity between medical staff and the system, so that medical staff can obtain key information of patients more conveniently. When the user interacts with the target patient in the virtual ward model, the system will pop up an information category selection interface, which allows the user to choose to view different categories of information such as vital signs, treatment records, video playback, etc. The implementation of this function is based on the system's structured storage and rapid retrieval capabilities for comprehensive medical reports, ensuring that users can quickly obtain relevant information as needed. By superimposing the display data on the bed coordinate position of the target patient in the virtual ward model, medical staff can quickly understand the patient's detailed condition in an intuitive virtual environment, thereby improving work efficiency and decision-making accuracy.

[0116] Specifically, interactive operations refer to actions in which users select or query target patients in the virtual ward model through mouse clicks, touch screen operations, or other input devices. When the user triggers an interactive operation, an information category selection interface will pop up, which provides multiple options for the user to choose from. For example, the vital signs option can display the patient's heart rate, blood pressure, blood oxygen and other real-time data; the treatment record option can display the patient's medication records, surgical records, etc.; the video playback option can play monitoring video clips related to the patient. These information categories are classified according to the data content stored in the comprehensive medical report to ensure that users can quickly locate the required information. The target patient's bed coordinate position refers to the specific location information of the patient's bed in the virtual ward model. The system realizes the visualization of information by superimposing the data on this location, allowing medical staff to intuitively view patient information in a virtual environment.

[0117] Preferably, when implementing the interactive operation function, the user experience and the response speed of the system can be further optimized. For example, the system can pre-index and classify the data in the comprehensive medical report so that the required information can be quickly retrieved when the user triggers the interactive operation. For the display of vital signs data, the system can obtain the latest data from the monitoring equipment in real time and display it in the information category selection interface in the form of a chart. At the same time, it provides a comparison function for historical data to help medical staff quickly understand the changing trends of the patient's vital signs. For treatment record data, the system can sort it by chronological order or event type, so that users can quickly find specific treatment events. For the video playback function, the system can extract and annotate key frames of video clips, and users can quickly locate important moments in the video by clicking on key frames. In addition, the system can also provide personalized interface layout and data display methods based on the user's operating habits and preferences, further improving the ease of use and practicality of the system.

[0118] In some embodiments, the operating logic of the disease progression warning model includes:

[0119] Extracting the data portion in the first monitoring subset that does not overlap with the second monitoring subset as an independent feature analysis set;

[0120] Analyzing the independent feature analysis set through a first feature extraction module to generate a first feature vector reflecting long-term fluctuations;

[0121] fusing the first feature vector and the original data of the first monitoring subset through a second feature extraction module to generate a second feature vector;

[0122] Inputting the second monitoring subset into the second feature extraction module to generate a third feature vector;

[0123] horizontally fusing the first eigenvector and the second eigenvector to generate a first fused feature;

[0124] Vertically fuse the first eigenvector and the third eigenvector to generate a second fused feature;

[0125] The first fusion feature and the second fusion feature are input into the temporal attention module to predict the clinical indicator parameters of the target in the future monitoring period;

[0126] The predicted parameters are compared with the historical data of the same period of the third monitoring subset for similarity. If the difference exceeds the threshold, an early warning is triggered;

[0127] Combined with the cross-cycle consistency verification results of the fourth monitoring subset, a weighted warning signal of disease deterioration is generated, in which the temporal attention module adopts a deep learning algorithm based on the attention mechanism.

[0128] It should be noted that the operating logic of the disease deterioration warning model mentioned in the present invention is to predict the possible deterioration of the patient's condition in advance through multi-dimensional analysis of the patient's monitoring data, so that medical staff can take timely intervention measures. Among them, the independent feature analysis set refers to the long-term data part extracted from the monitoring data that does not overlap with the short-term monitoring period. These data are used to analyze the long-term change trend of the patient's condition. The first feature extraction module and the second feature extraction module are algorithm modules for extracting key features from the monitoring data. These features can reflect the changing patterns of the patient's condition. The temporal attention module is an algorithm based on deep learning that can analyze time series data and predict clinical indicator parameters for future monitoring periods. By comparing the predicted parameters with historical data, the system can determine whether there are abnormal changes in the patient's condition and generate a warning signal.

[0129] Specifically, the first monitoring subset refers to the patient monitoring data collected during the first monitoring cycle, which usually contains information on changes in the disease over a longer period; the second monitoring subset refers to the data collected during the second monitoring cycle, which reflects changes in the disease over a shorter period. The first eigenvector and the second eigenvector are features extracted from these monitoring subsets by the feature extraction module for subsequent fusion analysis. The third monitoring subset and the fourth monitoring subset are used to generate the third eigenvector and verify consistency across cycles, respectively. In the feature fusion process, horizontal fusion refers to merging different feature vectors within the same monitoring cycle to obtain a more comprehensive feature representation; vertical fusion refers to merging feature vectors from different monitoring cycles to analyze the long-term trend of changes in the disease. Finally, the fused features are analyzed through the temporal attention module to predict the clinical indicator parameters of the future monitoring cycle, and are compared with historical data for similarity to determine whether an early warning is triggered.

[0130] The time series attention module can be based on the Transformer architecture. The Transformer architecture uses a self-attention mechanism to capture dependencies between different time points in time series data. The input feature vector sequence is processed through a multi-head self-attention mechanism, calculating attention weights in multiple subspaces to capture richer feature information. Furthermore, to better handle the long-term dependencies of time series data, the Transformer architecture can be combined with an LSTM (Long Short-Term Memory) network to form a hybrid LSTM+Attention model. The LSTM performs preliminary processing of time series data, extracting both short-term and long-term features. These features are then fed into the Attention module, which further emphasizes key features and ignores less important ones. For example, when processing electrocardiogram (ECG) data, the LSTM can capture the waveform characteristics of the ECG signal, while the Attention mechanism can identify characteristic segments associated with arrhythmias, thereby improving the accuracy of early warnings of worsening conditions.

[0131] Preferably, the following steps can be employed when constructing a disease progression warning model. First, the data portion that does not overlap with the second monitoring subset is extracted from the first monitoring subset to form an independent feature analysis set. The independent feature analysis set is then analyzed using the first feature extraction module to extract a first eigenvector reflecting long-term fluctuations. The first eigenvector and the raw data from the first monitoring subset are then input into the second feature extraction module to generate a second eigenvector. Simultaneously, the second monitoring subset is input into the second feature extraction module to generate a third eigenvector. The first and second eigenvectors are then horizontally fused to generate a first fused feature, and the first and third eigenvectors are vertically fused to generate a second fused feature. Finally, the first and second fused features are input into a temporal attention module, which utilizes a deep learning algorithm based on an attention mechanism to automatically learn key features from time series data and predict clinical indicator parameters for the target future monitoring cycle. The predicted parameters are then compared for similarity with historical data from the third monitoring subset for the same period. If the difference exceeds a preset threshold, a warning signal is triggered. In addition, the early warning signal can be weighted in combination with the cross-cycle consistency verification result of the fourth monitoring subset to improve the accuracy and reliability of the early warning.

[0132] Cross-period consistency verification uses the Kolmogorov-Smirnov test (KS test) to assess the distributional consistency of the predicted parameters with the historical data from the same period. The KS test compares the cumulative distribution functions (CDFs) of two samples, calculates the maximum vertical distance (D value) between them, and determines whether to accept the null hypothesis (i.e., the two samples come from the same distribution) based on the sample size and significance level (usually 0.05). If the p-value of the KS test result for the predicted parameters and the historical data is greater than 0.05, the two distributions are considered consistent and the verification passes; otherwise, the verification fails. For example, when verifying the blood pressure prediction parameters for a patient's third monitoring period, the predicted blood pressure values ​​are compared with the historical blood pressure data for the same period through a KS test. If the p-value is 0.03 (less than 0.05), the predicted parameters are considered inconsistent with the historical data distribution, triggering a manual review process. After manual review, the model weights are adjusted based on the review results, and prediction and verification are repeated until verification passes.

[0133] In some embodiments, the processing logic when the similarity comparison does not meet the threshold includes:

[0134] selecting a historical data subset that matches a target future period date from the third monitoring subset;

[0135] Extracting the historical data of the same period of all associated subsets from the fourth monitoring subset;

[0136] Compare the statistical distribution characteristics of the forecast parameters with those of historical data. If the distribution deviation exceeds the preset range, a manual review process is triggered.

[0137] Update the model weights based on the review results and regenerate the early warning signal.

[0138] It should be noted that the processing logic mentioned in the present invention when the similarity comparison does not meet the threshold is to further verify the accuracy of the model's prediction results when the disease deterioration warning model fails to trigger an early warning, to ensure that the patient's potential risk of disease deterioration is not missed due to model misjudgment. Among them, similarity comparison refers to comparing and analyzing the clinical indicator parameters of the future monitoring period predicted by the model with the historical data of the same period to determine whether the difference between the two is within the allowable range. If the difference exceeds the preset threshold, an early warning is triggered; and when the similarity comparison does not meet the threshold, that is, the prediction result is relatively close to the historical data, but there may still be potential risks, and further verification and processing are required at this time.

[0139] Specifically, the third monitoring subset refers to the patient monitoring data collected during the third monitoring cycle, which are used to perform similarity comparison with the clinical indicator parameters of the future monitoring cycle predicted by the model. The fourth monitoring subset refers to the patient monitoring data collected during the fourth monitoring cycle, which are used for cross-cycle consistency verification to ensure the stability and reliability of the model prediction results. The historical data subset is the data that matches the target future cycle date screened out from the third monitoring subset and is used for further comparative analysis. Statistical distribution characteristics refer to statistical characteristics such as the mean, variance, and distribution morphology of the data. By comparing the statistical distribution characteristics of the predicted parameters with those of the historical data, it can be determined whether the two have similar distribution patterns. If the distribution deviation exceeds the preset range, it may mean that there is a potential change in the patient's condition, and a manual review process needs to be triggered.

[0140] Preferably, in the processing process when the similarity comparison does not meet the threshold, the following steps can be further refined. First, a subset of historical data that matches the target future cycle date is screened out from the third monitoring subset to ensure that the compared data is comparable. Then, the historical data of the same period of all associated subsets are extracted from the fourth monitoring subset for cross-cycle consistency verification. Next, the clinical indicator parameters of the future monitoring cycle predicted by the model are compared with the screened historical data subsets for statistical distribution characteristics, such as calculating statistical indicators such as the mean difference and variance difference between the two. If these differences exceed the preset range, it may mean that there is a deviation between the model prediction results and the actual condition, and the manual review process is triggered. During the manual review process, medical staff can evaluate and adjust the model's prediction results based on the patient's specific situation and clinical experience. Finally, based on the results of the manual review, the model weights are updated and the early warning signal is regenerated to improve the accuracy and adaptability of the model.

[0141] With specific examples, the following is a complete implementation case of sepsis early warning in diabetic patients:

[0142] Patient Li, a 55-year-old male with a 10-year history of type 2 diabetes, was admitted to the hospital with symptoms including fever, cough, and dyspnea. He was diagnosed with diabetes and sepsis. Upon admission, he was enrolled in the critical care unit's monitoring and management system.

[0143] Monitoring cycle configuration: Set 72 hours as the primary monitoring cycle (first monitoring cycle), including a 48-hour secondary monitoring cycle (second monitoring cycle), the third monitoring cycle is 24 hours, and the fourth monitoring cycle includes 3 third monitoring cycles, each with an interval of 8 hours.

[0144] During the main monitoring cycle, pay attention to the patient's blood sugar, lactate, heart rate, blood pressure, blood oxygen saturation and other indicators.

[0145] Cross-cycle feature fusion: In the fourth monitoring cycle, cross-cycle feature fusion is performed on blood glucose and lactate indicators.

[0146] First, statistical features such as the mean, standard deviation, maximum, and minimum values ​​of blood glucose and lactate are calculated for each third monitoring cycle. These features are then normalized to fall within the range [0, 1]. Next, a weighted fusion strategy is used to assign weights based on feature importance (e.g., a weight of 0.4 for the mean blood glucose value and a weight of 0.3 for the maximum lactate value), and the fused feature values ​​are calculated.

[0147] Finally, the fused feature values ​​are input into the disease deterioration warning model as one of the bases for judging disease deterioration.

[0148] An example of orange warning rendering in 3D visualization: In a 3D model of a virtual ward, based on the patient's condition impact level score, if the patient's condition impact level reaches moderate (5-9 points), orange warning rendering is applied to the patient's bed and the sub-area where it is located. Through shader programming in the Unity3D engine, the bed model's diffuse color is set to RGB (255,165,0) (orange), and its transparency is adjusted to 0.7 to highlight abnormal areas. At the same time, color coding mapping is applied to the area surrounding the patient, setting the ground color of the sub-area to orange. The color saturation is gradually reduced as the degree of condition impact decreases, forming an orange warning area that spreads outward from the patient's bed, allowing medical staff to intuitively identify potential risk areas.

[0149] The above-mentioned various embodiments of the present invention have the following beneficial effects: the critical care medicine information digital management system can realize accurate monitoring and early warning of the patient's condition based on the multi-cycle data collaborative analysis mechanism, through the dynamic configuration and data fusion of the first monitoring cycle, the second monitoring cycle, the third monitoring cycle and the fourth monitoring cycle. The system can significantly improve the accuracy of disease prediction by means of the temporal attention mechanism and feature fusion algorithm in the disease deterioration warning model, combined with the cross-validation of long-cycle fluctuation characteristics and real-time monitoring data. Through the automatic analysis of abnormal causes and the intelligent assessment of the scope of influence, data support can be provided for clinical decision-making. At the same time, the structured comprehensive disease report can integrate multi-dimensional trend maps, treatment event records and encrypted video clips to form a complete basis for diagnosis and treatment.

[0150] The system can rely on 3D visualization technology to dynamically render sub-areas and patients based on the level of disease impact, and intuitively present risk distribution through transparency hierarchical mapping and color coding strategies. The interactive operation function can support medical staff to quickly retrieve key information such as patient vital signs, treatment records and video playback, thereby optimizing clinical work efficiency. The graded alarm mechanism can combine cross-cycle consistency verification results to achieve intelligent grading and early warning of risks in critical areas, and the model self-learning function can automatically trigger the manual review process when the warning error exceeds the standard, continuously optimize the warning accuracy, and ultimately improve the overall management efficiency and treatment quality of the critical care department.

[0151] like Figure 2 As shown, some embodiments of a method for digital information management based on a critical care medicine department include:

[0152] Obtaining a patient monitoring data set corresponding to each patient in a target critical care unit, wherein each patient has a corresponding patient monitoring data sequence;

[0153] For each of the individual patients, perform the following generation steps:

[0154] Determine a patient monitoring data sequence corresponding to the patient as a target monitoring data sequence;

[0155] Determining a first monitoring cycle, a second monitoring cycle, a third monitoring cycle, and a fourth monitoring cycle corresponding to the patient for the current monitoring cycle, wherein the first monitoring cycle includes the second monitoring cycle, the time range corresponding to the first monitoring cycle is greater than the second monitoring cycle, the fourth monitoring cycle includes the third monitoring cycle, and the central monitoring cycle corresponding to the third monitoring cycle is within the same week as the current monitoring cycle;

[0156] Generate, according to the target monitoring data sequence, a first monitoring subset corresponding to the first monitoring period, a second monitoring subset corresponding to the second monitoring period, a third monitoring subset corresponding to the third monitoring period, and a fourth monitoring subset corresponding to the fourth monitoring period;

[0157] Generate a condition deterioration warning signal for the patient using a pre-trained condition deterioration warning model according to the first monitoring subset, the second monitoring subset, the third monitoring subset, and the fourth monitoring subset;

[0158] In response to the early warning signal of worsening condition indicating abnormal condition of the patient, determining the abnormal cause analysis result and impact scope assessment result corresponding to the patient;

[0159] Generate a structured comprehensive report of the disease condition in a predetermined format based on the abnormal cause analysis results, target monitoring data sequence, and impact range assessment results;

[0160] Based on the comprehensive reports on each condition, dynamic visualization is performed for each patient, and graded alarm processing is performed on the related critical areas.

[0161] It is understandable that the steps and references in the information digital management method based on the critical care medicine department are Figure 1 Therefore, the modules, features and beneficial effects described above for the information digital management system based on the critical care department are also applicable to the information digital management method based on the critical care department and the operations contained therein, and will not be repeated here.

[0162] like Figure 3 As shown, electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) 302 or programs loaded from a storage device 308 into a random access memory (RAM) 303. RAM 303 also stores various programs and data required for the operation of electronic device 300. Processing device 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to bus 304.

[0163] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as needed.

[0164] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or a terminal device such as a computer, server, mobile phone, or tablet.

[0165] The above descriptions merely illustrate some preferred embodiments of the present invention and the underlying technical principles. Those skilled in the art should understand that the scope of the invention encompassed by the embodiments of the present invention is not limited to technical solutions formed by specific combinations of the aforementioned technical features. It also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents, without departing from the aforementioned inventive concept. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. An information digital management system based on the Department of Critical Care Medicine, characterized in that: include: The data acquisition module obtains the patient monitoring data set corresponding to each patient in the target critical care department. Each patient has a corresponding patient monitoring data sequence; The data processing module performs the following operations on each patient: Determine a patient monitoring data sequence corresponding to the patient as a target monitoring data sequence; Determine the four monitoring cycle groups corresponding to the patient for the current monitoring cycle; According to the target monitoring data sequence, the monitoring period groups are generated to correspond to the monitoring subset groups respectively; Based on the monitoring subset group, a pre-trained disease deterioration warning model is used to generate a patient-specific disease deterioration warning signal; In response to the early warning signal of worsening condition indicating abnormal condition of the patient, determining the abnormal cause analysis result and impact scope assessment result corresponding to the patient; Generate a structured comprehensive report on the disease condition based on the abnormal cause analysis results, target monitoring data sequence and impact range assessment results; The visualization and alarm module is used to perform dynamic visualization of each patient based on the comprehensive condition report and perform graded alarm processing on the associated critical areas; The operating logic of the disease deterioration warning model includes: The monitoring subset group includes a first monitoring subset, a second monitoring subset, a third monitoring subset, and a fourth monitoring subset; Extracting the data portion in the first monitoring subset that does not overlap with the second monitoring subset as an independent feature analysis set; Analyzing the independent feature analysis set through a first feature extraction module to generate a first feature vector reflecting long-term fluctuations; fusing the first feature vector and the original data of the first monitoring subset through a second feature extraction module to generate a second feature vector; Inputting the second monitoring subset into the second feature extraction module to generate a third feature vector; horizontally fusing the first eigenvector and the second eigenvector to generate a first fused feature; Vertically fuse the first eigenvector and the third eigenvector to generate a second fused feature; The first fusion feature and the second fusion feature are input into the temporal attention module to predict the clinical indicator parameters of the target in the future monitoring period; The predicted parameters are compared with the historical data of the same period of the third monitoring subset for similarity. If the difference exceeds the threshold, an early warning is triggered; Combined with the cross-cycle consistency verification results of the fourth monitoring subset, a weighted warning signal of disease deterioration is generated. The temporal attention module adopts a deep learning algorithm based on the attention mechanism; The processing logic when the similarity comparison does not meet the threshold includes: selecting a historical data subset that matches a target future period date from the third monitoring subset; Extracting the historical data of the same period of all associated subsets from the fourth monitoring subset; Compare the statistical distribution characteristics of the forecast parameters with those of historical data. If the distribution deviation exceeds the preset range, a manual review process is triggered. Update the model weights based on the review results and regenerate the early warning signal.

2. The system according to claim 1, wherein: The four monitoring cycle groups include a first monitoring cycle, a second monitoring cycle, a third monitoring cycle, and a fourth monitoring cycle; The first monitoring cycle includes the second monitoring cycle, the time range corresponding to the first monitoring cycle is greater than the second monitoring cycle, the fourth monitoring cycle includes the third monitoring cycle, and the central monitoring cycle corresponding to the third monitoring cycle is in the same cycle as the current monitoring cycle.

3. The system according to claim 2, characterized in that Dynamic visualization includes: Determine the impact level of each sub-area in the target critical care department based on the comprehensive reports of each condition; Assign visualization strategies to each patient and sub-area based on the disease impact level and disease summary report. The visualization strategies include transparency level mapping and color-coded mapping. According to the visualization strategy, abnormal status of patients and sub-areas are rendered in real time in the 3D model of the virtual ward.

4. The system according to claim 2, characterized in that The steps for determining the monitoring cycle configuration corresponding to a patient include: According to the patient's clinical classification and sub-region, the initial monitoring cycle configuration is matched from the preset association table; Obtain historical data on fluctuations in patients' vital signs and calculate their periodic characteristics; Based on the periodic characteristics, the initial monitoring period configuration is dynamically adjusted to generate an adapted first adjustment period, a second adjustment period, a third adjustment period, and a fourth adjustment period; Combined with the current monitoring cycle, the final first monitoring cycle, second monitoring cycle, third monitoring cycle and fourth monitoring cycle are generated.

5. The system according to claim 3, wherein: The steps to generate a comprehensive medical report include: Generate multi-dimensional trend graphs corresponding to the target monitoring data sequence, including the temporal changes of heart rate, blood oxygen, and blood pressure; Obtain timestamps, interventions, and associated video clips of patients’ abnormal treatment events; Generate an encrypted access link to the video clip; The abnormal cause analysis results, multi-dimensional trend maps, impact range assessment results, abnormal treatment events and encrypted links are integrated into a structured comprehensive disease report according to the clinical report template.

6. The system according to claim 4, characterized in that Also includes: In response to user-triggered interactive operations on a target patient in the virtual ward model, an information category selection interface including vital signs, treatment records, and video playback pops up; Extract corresponding data from the comprehensive disease report according to the category selected by the user; The data of the target patient's bed coordinate position is superimposed on the virtual ward model.

7. A method for digital management of information based on critical care medicine, characterized in that: include: Obtaining a patient monitoring data set corresponding to each patient in a target critical care department, wherein each patient has a corresponding patient monitoring data sequence; For each of the individual patients, perform the following generation steps: Determine a patient monitoring data sequence corresponding to the patient as a target monitoring data sequence; Determine a first monitoring cycle, a second monitoring cycle, a third monitoring cycle, and a fourth monitoring cycle corresponding to the patient for the current monitoring cycle; Generate, according to the target monitoring data sequence, a first monitoring subset corresponding to the first monitoring period, a second monitoring subset corresponding to the second monitoring period, a third monitoring subset corresponding to the third monitoring period, and a fourth monitoring subset corresponding to the fourth monitoring period; Generate a condition deterioration warning signal for the patient using a pre-trained condition deterioration warning model according to the first monitoring subset, the second monitoring subset, the third monitoring subset, and the fourth monitoring subset; In response to the early warning signal of worsening condition indicating abnormal condition of the patient, determining the abnormal cause analysis result and impact scope assessment result corresponding to the patient; Generate a structured comprehensive report of the disease condition in a predetermined format based on the abnormal cause analysis results, target monitoring data sequence, and impact range assessment results; Based on the comprehensive reports of each condition, dynamic visualization of each patient is performed, and graded alarm processing is performed on the related critical areas; The operating logic of the disease deterioration warning model includes: The monitoring subset group includes a first monitoring subset, a second monitoring subset, a third monitoring subset, and a fourth monitoring subset; Extracting the data portion in the first monitoring subset that does not overlap with the second monitoring subset as an independent feature analysis set; Analyzing the independent feature analysis set through a first feature extraction module to generate a first feature vector reflecting long-term fluctuations; fusing the first feature vector and the original data of the first monitoring subset through a second feature extraction module to generate a second feature vector; Inputting the second monitoring subset into the second feature extraction module to generate a third feature vector; horizontally fusing the first eigenvector and the second eigenvector to generate a first fused feature; Vertically fuse the first eigenvector and the third eigenvector to generate a second fused feature; The first fusion feature and the second fusion feature are input into the temporal attention module to predict the clinical indicator parameters of the target in the future monitoring period; The predicted parameters are compared with the historical data of the same period of the third monitoring subset for similarity. If the difference exceeds the threshold, an early warning is triggered; Combined with the cross-cycle consistency verification results of the fourth monitoring subset, a weighted warning signal of disease deterioration is generated. The temporal attention module adopts a deep learning algorithm based on the attention mechanism; The processing logic when the similarity comparison does not meet the threshold includes: selecting a historical data subset that matches a target future period date from the third monitoring subset; Extracting the historical data of the same period of all associated subsets from the fourth monitoring subset; Compare the statistical distribution characteristics of the forecast parameters with those of historical data. If the distribution deviation exceeds the preset range, a manual review process is triggered. Update the model weights based on the review results and regenerate the early warning signal.

8. The method according to claim 7, characterized in that The first monitoring cycle includes the second monitoring cycle, the time range corresponding to the first monitoring cycle is greater than the second monitoring cycle, the fourth monitoring cycle includes the third monitoring cycle, and the central monitoring cycle corresponding to the third monitoring cycle is in the same cycle as the current monitoring cycle.

Images