A 5G-based method and device for scheduling multi-source ICU devices based on clinical event chains.

By constructing patient-centric event nodes and directed graphs of clinical events in the ICU edge gateway, and dynamically adjusting 5G slice scheduling, the problems of data transmission lag and missing data in existing technologies are solved, data dependency order and transmission continuity are realized, and traceability of data integrity verification and scheduling records is provided.

CN123124007APending Publication Date: 2026-11-17CHINA UNIV OF PETROLEUM (EAST CHINA)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611366402.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-04
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing 5G slicing scheduling methods in the ICU fail to dynamically adjust the transmission level of data from the same device based on the clinical event status, resulting in the lag or absence of necessary preliminary data for related events. Furthermore, the lack of a patient-centric event semantic view makes it difficult to maintain data dependency order and processing traceability.

Method used

In the ICU edge gateway, a patient-centered event node and a directed graph of clinical events with prerequisite dependencies are established. The slice mapping is dynamically updated according to the scheduling priority of the event sub-flows. When transmission goes out of bounds or network congestion occurs, resource reallocation and task migration are performed to maintain the dependency order of related data and the verifiability of the running records.

Benefits of technology

It enables dynamic adjustment of the transmission level of device data based on the status of clinical events, maintains the dependency order and transmission continuity of related data, and provides data integrity verification and traceability of the scheduling process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123124007A_ABST
    Figure CN123124007A_ABST
Patent Text Reader

Abstract

This invention discloses a 5G slice scheduling method and apparatus for ICU multi-source devices based on clinical event chains. The ICU edge gateway converts multi-source device data into patient-centric event nodes, constructs a directed graph of clinical events with preceding dependencies based on patient identifiers, instruction identifiers, and time windows; calculates event sub-stream priorities based on normalized remaining time ratio, preceding data integrity, event risk, and control interaction urgency, and dynamically maps them between URLLC, eMBB, and mMTC slices, jointly configuring radio resource blocks, MEC containers, and redundant paths; when latency or reliability exceeds limits, resource reallocation or task migration is implemented; when congestion occurs, preceding dependency closed datasets are extracted and runtime traceability data packets are generated to maintain the dependency order of associated data and scheduling traceability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of medical data communication and wireless network resource scheduling technology, specifically to a 5G slice scheduling method and device for ICU multi-source equipment based on clinical event chains. Background Technology

[0002] Intensive care units (ICUs) simultaneously deploy vital sign monitors, ventilators, infusion pumps, and blood gas analyzers. The multi-source data has varying requirements in sampling frequency, bandwidth, latency, and reliability, and needs to be uploaded to the ICU central station, electronic medical record system, or remote consultation terminal. 5G networks can utilize URLLC, eMBB, and mMTC slices to provide differentiated bearers, laying the foundation for the aforementioned data transmission.

[0003] Existing slice scheduling typically pre-maps and statically allocates resources based on device type or service type such as video, monitoring, and control, without considering how the importance of data from the same device changes with clinical events. For example, after device parameters are adjusted, there are time and data dependencies between device responses, physiological responses, and alarm results. Per-datastream scheduling may result in critical data arriving while necessary prerequisite data is still delayed, causing related data to be out of order or missing; fixed mapping also makes it difficult to improve the transmission level of relevant event data in a timely manner.

[0004] Furthermore, existing resource allocation typically lacks a patient-centric event semantic view. Congestion handling often involves slowing down or discarding data byte-by-byte or in single data streams, failing to identify the necessary data for closing event chains, and making it difficult to retain records that can verify the integrity of source data, device status, and scheduling processes. Therefore, a technical solution is still needed that can dynamically adjust slices and edge resources according to the clinical event status while maintaining the order of related data dependencies and processing traceability. Summary of the Invention

[0005] To address the problems in existing medical scenarios where 5G slicing statically allocates resources based on device type or service type, making it difficult to adjust the transmission level of data from the same device according to the clinical event status, and where independent scheduling per data stream can easily lead to delays or missing necessary prerequisite data for related events, this invention provides a method and apparatus for 5G slice scheduling of multi-source devices in the ICU based on clinical event chains. This method and apparatus establish patient-centric event nodes and a directed graph of clinical events with prerequisite dependencies in the ICU edge gateway. It dynamically updates the slice mapping based on the scheduling priority of event sub-streams and jointly configures radio resource blocks, MEC containers, and redundant paths. Simultaneously, when transmission exceeds limits or network congestion occurs, it performs resource reallocation, task migration, and extraction of prerequisite dependency closed datasets to maintain the dependency order of related data and the verifiability of execution records.

[0006] To address the aforementioned technical problems, one aspect of the present invention provides a 5G slice scheduling method for ICU multi-source devices based on clinical event chains, the scheduling method comprising steps S1 to S6.

[0007] Step S1: Collect and standardize monitoring and treatment data from multiple devices in the ICU. Aggregate the standardized data into patient-centric event nodes and record device status, sampling window, data source, and command response relationship.

[0008] Step S2: Using the patient-centered event node as the vertex and the data dependencies between parameter adjustment events, device response events, physiological response events, and alarm result events as directed edges, construct a directed graph of clinical events with preceding dependencies, and update the determineability and uploadability status of subsequent related events based on the completeness of preceding event data.

[0009] Step S3: For each event sub-flow in the directed graph of clinical events, calculate the scheduling priority based on the normalized remaining time ratio, the completeness of the preceding data, the event risk quantity, and the urgency of the control interaction.

[0010] Step S4: According to the scheduling priority and the first priority threshold, dynamically update the mapping relationship of event sub-streams among URLLC slices, eMBB slices and mMTC slices, generate a joint configuration request for radio resource blocks, MEC containers and redundant paths, and send the joint configuration request to the slice management unit and MEC orchestration unit of the 5G network.

[0011] Step S5: Monitor the transmission status of the allocated event sub-streams. When the latency or reliability exceeds the limit, release the radio resource blocks occupied by the low-priority event sub-streams and reassign them to the high-priority event sub-streams that have exceeded the limit, or migrate the computation and forwarding tasks of the event sub-streams that have exceeded the limit and update the redundant paths.

[0012] Step S6: When the network is congested, backtrack the predecessor dependencies from the target event according to the directed graph of clinical events, remove duplicate nodes and retain the necessary predecessor nodes of the target event to form a closed dataset of predecessor dependencies, and generate a runtime traceback data packet.

[0013] Specifically, in step S1, the multi-source devices in the ICU include vital signs monitors, ventilators, infusion pumps, blood gas analyzers, and bedside control terminals. The data generated by each device differs in protocol format, sampling period, and dimensionality. Device data access and standardization processing unifies data from different protocols and sampling periods into standardized data with a unified timestamp. Using the patient identifier as an index, the multi-source device data for the same patient is aligned by sampling time and written to the corresponding patient-centric event node. The patient-centric event node records device status, sampling window, data source, and command-response relationship. Device status includes online and abnormal sampling status; the sampling window includes sampling start and end times and sampling frequency; the data source includes device identifier and data channel identifier; and the command-response relationship records the control commands that trigger device responses and their corresponding relationships. Thus, data scattered across different devices and channels is organized into patient-centric event nodes with traceable sources and response relationships, providing a unified data foundation for subsequent scheduling based on data dependency chains.

[0014] Specifically, in step S2, the directed graph of clinical events starts with the parameter adjustment event as the initial node, sequentially connecting the device response event, physiological response event, and alarm result event, and uses directed edges to represent the dependency of subsequent related events on the preceding event data. The data dependency relationship is determined based on the patient identifier, instruction identifier, and time window, and is used for data organization and sequence control in network scheduling, but not for outputting medical diagnoses or treatment decisions. The edge gateway updates the determineability and uploadability status of subsequent related events based on the completeness of the preceding event data; this status update only constrains the upload of related events generated by the edge gateway and does not change the independent operation of the medical device's own alarm and control channels. The directed graph of clinical events is continuously updated as new event nodes are written, and its preceding dependencies are used for scheduling priority calculation and extraction of the preceding dependency closed dataset during network congestion.

[0015] Specifically, in step S3, the scheduling priority P is determined according to... The calculation is performed, where Δ is the normalized remaining time ratio between 0 and 1, C is the pre-data integrity between 0 and 1, R is the event risk between 0 and 1, U is the control interaction urgency between 0 and 1, α, β, γ, and δ are dimensionless non-negative weights, and ε is a positive dimensionless smoothing constant. The first term reflects the degree to which the event sub-stream is close to the deadline, the second term is the pre-data incompleteness gain term, used to incorporate the degree of pre-data dependency incompleteness into the ranking, the third term reflects the event risk, and the fourth term reflects the control interaction urgency. The higher the scheduling priority, the earlier the event sub-stream is in the slice mapping and resource allocation.

[0016] Furthermore, the weighting coefficients α, β, γ, and δ are all not less than 0 and satisfy the following conditions: The normalized remaining time ratio Δ is obtained by dividing the difference between the event's deadline and the current time by the total event time and is limited to between 0 and 1. Δ is set to 0 when the deadline is exceeded. The smoothing constant ε is a positive dimensionless constant used to limit the maximum value of the time urgency term. The scheduling priority is used for the relative ordering of event sub-streams, and its absolute value does not correspond to the medical risk level.

[0017] Specifically, in step S4, a first priority threshold is preset. Event sub-streams with scheduling priorities not lower than the first priority threshold are mapped to URLLC slices, while event sub-streams with scheduling priorities lower than the first priority threshold are mapped to eMBB slices or mMTC slices according to their data carrying characteristics of high bandwidth or a large number of low-frequency connections. When the scheduling priority of an event sub-stream crosses the first priority threshold, the edge gateway updates its slice mapping relationship. The edge gateway forms a radio resource block configuration order based on the scheduling priority and generates MEC container and redundant path configuration requests for event sub-streams mapped to URLLC slices, which are sent to the 5G network's slice management unit and MEC orchestration unit through the communication interface.

[0018] Specifically, in step S5, the transmission delay and reliability of the allocated event sub-streams are monitored. When the transmission delay of an event sub-stream exceeds the delay limit corresponding to its clinical event or the transmission reliability is lower than the reliability limit, an out-of-bounds condition is determined. When an out-of-bounds condition occurs, some radio resource blocks occupied by low-priority event sub-streams are released within the URLLC slice in ascending order of scheduling priority, and the released radio resource blocks are allocated to the out-of-bounds high-priority event sub-streams; or the computation and forwarding tasks of the out-of-bounds event sub-streams are migrated to a backup MEC container or adjacent edge nodes, and redundant paths are updated synchronously to restore transmission quality without interrupting the critical event chain.

[0019] Specifically, in step S6, starting from the target event to be uploaded, backtracking is performed along the direction of the prerequisite dependencies in the directed graph of clinical events. Duplicate event nodes are removed, and necessary prerequisite event nodes reachable from the target event are retained, forming a closed dataset of prerequisite dependencies. A source data summary is calculated for the data in this dataset, and the source data summary, the device status recorded by the event nodes, and the intermediate scheduling quantities and their calculation results used for scheduling priority calculation are encapsulated into a runtime traceability data package for data integrity verification and scheduling process review.

[0020] Another aspect of the present invention provides a 5G slice scheduling device for ICU multi-source devices based on clinical event chains. The scheduling device includes an event node construction module, an event directed graph maintenance module, a priority calculation module, a slice mapping and joint allocation module, an out-of-bounds emergency module, and a congestion emergency module. The modules are deployed on the ICU edge gateway. The event node construction module standardizes multi-source device data and aggregates it into patient-centric event nodes; the event directed graph maintenance module establishes data dependencies and maintains the clinical event directed graph based on patient identifiers, instruction identifiers, and time windows; the priority calculation module calculates scheduling priorities based on normalized remaining time ratio, pre-existing data integrity, event risk, and control interaction urgency; the slice mapping and joint allocation module dynamically updates the mapping relationship between event sub-streams in URLLC slices, eMBB slices, and mMTC slices based on the first priority threshold, and generates joint configuration requests for radio resource blocks, MEC containers, and redundant paths; the boundary outage emergency module releases radio resource blocks occupied by low-priority event sub-streams and allocates the released resources to high-priority event sub-streams that have crossed the boundary, or migrates out-of-bounds tasks; and the congestion emergency module forms a pre-existing dependency closed dataset and a runtime traceability data packet.

[0021] Specifically, the event node construction module associates patient identifiers and writes multi-source device data of the same patient into the corresponding event node after aligning them according to sampling time. Device status includes online status and sampling abnormal status. Sampling window includes sampling start and end times and sampling frequency. Data source includes device identifier and data channel identifier. Command response relationship is used to record the control commands that trigger device response and their corresponding relationships. The event directed graph maintenance module takes the parameter adjustment event as the starting node and connects the device response event, physiological response event and alarm result event in sequence. Directed edges represent the dependence of subsequent events on the preceding event data. After the preceding event data is complete, the edge gateway marks the subsequent related events as decidable and uploading. This state only restricts the uploading of related events generated by the edge gateway and does not affect the alarm channel and control channel of the medical device itself.

[0022] Specifically, the priority calculation module follows The scheduling priority P is calculated, where Δ is the normalized remaining time ratio, C is the pre-existing data integrity, R is the event risk, U is the control interaction urgency, α, β, γ, and δ are dimensionless non-negative weights, and ε is a positive dimensionless smoothing constant. The slice mapping and joint allocation module maps event sub-streams with scheduling priorities not lower than the first priority threshold to URLLC slices, and maps event sub-streams with priorities lower than the first priority threshold to eMBB or mMTC slices according to their data carrying characteristics. The mapping relationship is updated when the scheduling priority crosses the first priority threshold. During joint allocation, the radio resource block configuration order is formed based on the scheduling priority, and MEC containers and redundant path configuration requests are generated for event sub-streams mapped to URLLC slices.

[0023] Specifically, the boundary crossing emergency module monitors the transmission latency and reliability of the allocated event sub-streams. When an event sub-stream crosses the boundary, it releases some of the radio resource blocks occupied by low-priority event sub-streams in the URLLC slice according to the scheduling priority from low to high. The released radio resource blocks are then allocated to the high-priority event sub-streams that have crossed the boundary. Alternatively, the computation and forwarding tasks of the event sub-streams that have crossed the boundary are migrated to a backup MEC container or an adjacent edge node, and redundant paths are updated synchronously. The congestion emergency module extracts the pre-dependency closure dataset, which contains parameter adjustment event data, device response event data, physiological response event data, and alarm result event data necessary to close the target event chain. During extraction, the pre-dependencies are traced back from the target event along the directed graph of clinical events. The traceback data packet includes the source data summary, device status, and scheduling intermediate quantities and their calculation results.

[0024] The above modules are deployed on the ICU edge gateway. The event node construction module, the event directed graph maintenance module, the priority calculation module, the slice mapping and joint allocation module, the boundary crossing emergency module, and the congestion emergency module can be implemented by different processing units within the edge gateway, or they can be implemented by the same processing unit as software functional modules.

[0025] This invention converts ICU multi-source device data into patient-centric event nodes and establishes a directed graph of clinical events with pre-dependencies based on patient identifiers, instruction identifiers, and time windows. Scheduling priorities are calculated based on the normalized remaining time ratio, pre-data integrity, event risk, and control interaction urgency. The mapping relationship between event sub-streams in URLLC slices, eMBB slices, and mMTC slices is dynamically updated based on a first priority threshold, and joint configuration requests for radio resource blocks, MEC containers, and redundant paths are generated. When latency or reliability exceeds limits, resources occupied by low-priority event sub-streams are released or out-of-limit tasks are migrated. When network congestion occurs, a closed dataset of pre-dependencies and a runtime traceability data packet are formed, thus creating a scheduling process that integrates event modeling, priority calculation, dynamic mapping, joint configuration, and anomaly handling.

[0026] Compared to the method of fixedly allocating slice resources according to device type, this invention updates the slice mapping based on the scheduling priority of event sub-streams and its relationship with the first priority threshold. This allows data generated by the same device to receive different transmission and edge computing configurations as the event state changes. The directed graph of clinical events incorporates the data dependencies of related events into the scheduling process, facilitating the maintenance of the upload order of related data and enabling each priority calculation to be verified by normalized remaining time ratio, prior data integrity, event risk, and control interaction urgency. When transmission out of bounds or network congestion occurs, resource reallocation, task migration, and extraction of prior dependency closed datasets can maintain the transmission continuity of key related data and provide corresponding information for data integrity verification and operation record traceability. Attached Figure Description

[0027] Figure 1 This is an overall flowchart of the 5G slice scheduling method for ICU multi-source devices based on clinical event chains, as described in this invention.

[0028] Figure 2 This is a structural block diagram of the ICU multi-source equipment 5G slice scheduling device based on clinical event chain of the present invention.

[0029] Figure 3 This is a schematic diagram illustrating the process of constructing patient-centered event nodes and generating a directed graph of clinical events according to the present invention.

[0030] Figure 4 This is a schematic diagram of the data dependency chain between parameter adjustment events, device response events, physiological response events, and alarm result events in the directed clinical event graph of this invention.

[0031] Figure 5 This is a diagram showing the dynamic mapping and joint allocation relationship of event sub-streams in the URLLC slice, eMBB slice, and mMTC slice of the present invention.

[0032] Figure 6 This is a flowchart illustrating the process of resource redistribution, edge migration, and congestion emergency handling for this invention.

[0033] Figure 7 This is a schematic diagram illustrating the application scenario of the present invention in multi-source equipment monitoring and remote diagnosis and treatment in ICU wards. Detailed Implementation

[0034] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The drawings are used to illustrate the connection relationships, control relationships, and execution relationships between the components and method steps, and do not represent the actual dimensions, installation ratio, or unique arrangement of the equipment. The following specific embodiments are used to illustrate the technical solutions and possible implementations of the present invention and do not constitute a limitation on the scope of protection; where there is no technical contradiction, the technical features described in the various specific embodiments can be combined with each other.

[0035] Example 1 This embodiment illustrates the complete execution process of a 5G slice scheduling method for ICU multi-source devices based on clinical event chains. The overall flow of the scheduling method in this embodiment is as follows: Figure 1 As shown, the process of converting multi-source device data into a directed graph of patient-centric event nodes and clinical events is as follows: Figure 3 As shown, the dynamic mapping and joint allocation relationship of event substreams among URLLC slices, eMBB slices, and mMTC slices is as follows: Figure 5 As shown, the process for redistributing out-of-bounds resources, edge migration, and congestion emergency handling is as follows: Figure 6 As shown.

[0036] Figure 1 Steps S1 to S6 are represented sequentially according to the scheduling process. Step S1 standardizes multi-source device data and constructs patient-centric event nodes. Step S2 constructs a directed graph of clinical events with prerequisite dependencies. Step S3 calculates the scheduling priority of event sub-streams. Step S4 dynamically updates slice mappings based on the first priority threshold and forms a joint configuration request. Step S5 continuously monitors transmission latency and reliability. When an out-of-bounds event occurs, the radio resource blocks occupied by low-priority event sub-streams are released and reallocated, or the out-of-bounds task is migrated and redundant paths are updated. If no out-of-bounds event occurs, the network congestion is assessed. If the network is congested, step S6 is executed to form a closed dataset of prerequisite dependencies and run trace data packets. If there is no congestion, normal transmission is performed according to the mapping and allocation results. After processing each branch, the next scheduling cycle begins.

[0037] Figure 3 It consists of event nodes constructed from multi-source device layers and edge gateways, as well as event-directed... Figure 3 The multi-source device layer consists of vital signs monitors, ventilators, infusion pumps, blood gas analyzers, and bedside control terminals. Data generated by each device is standardized and aggregated into patient-centric event nodes. These patient-centric event nodes record four types of information: device status, sampling window, data source, and command response relationship. The directed graph of clinical events uses directed edges to form event chains with preceding dependencies, based on the data dependencies between parameter adjustments, device responses, physiological responses, and alarm results. Figure 3Solid arrows represent data aggregation between the multi-source device layer and the event node construction, and solid arrows represent event writing between the event node construction and the event directed graph.

[0038] Figure 5 This section describes the scheduling priority calculation, threshold determination, slice mapping, and joint configuration relationships for event sub-streams. The edge gateway calculates the scheduling priority P based on Δ, C, R, and U. Event sub-streams with a scheduling priority not lower than the first priority threshold enter the URLLC slice, while event sub-streams that do not reach the threshold enter the eMBB slice or mMTC slice based on their data carrying characteristics. URLLC slices, eMBB slices, and mMTC slices are all connected to a joint allocation object, which includes radio resource blocks, MEC containers, and redundant paths. The scheduling priority is recalculated and the mapping relationships are updated in the next scheduling cycle.

[0039] Figure 6 This describes the emergency handling process for boundary violations and congestion. The edge gateway continuously monitors the transmission latency and reliability of event substreams. When a boundary violation occurs, it performs radio resource block release and reallocation, or migrates the boundary violation task to a backup MEC container or adjacent edge node and updates redundant paths. After recovery, it returns to continuous monitoring. If no boundary violation occurs, it determines whether the network is congested. If there is no congestion, transmission proceeds normally according to the mapping and allocation results. If congestion occurs, it traces back the preceding dependencies from the target event, removes duplicate nodes and retains necessary preceding nodes, forming a closed dataset of preceding dependencies and running traceback data packets. After processing, it enters the next monitoring cycle.

[0040] This embodiment uses an ICU ward as an application scenario. The ward has 8 beds, each equipped with a vital signs monitor, ventilator, infusion pump, and blood gas analyzer, as well as a bedside control terminal. The edge gateway communicates with the ICU central station, remote consultation terminal, and electronic medical record system via a 5G network. The 5G network is configured with URLLC slices, eMBB slices, and mMTC slices. URLLC slices are for low-latency, high-reliability services, eMBB slices are for high-bandwidth services, and mMTC slices are for massive connection services.

[0041] In step S1, the edge gateway collects and standardizes data from each device. The vital signs monitor uploads data such as heart rate, blood oxygen saturation, and blood pressure with a 1-second sampling cycle; the ventilator uploads ventilation parameters with a 0.5-second sampling cycle; the infusion pump uploads the infusion rate with a 2-second sampling cycle; and the blood gas analyzer uploads test results every 30 minutes. The edge gateway unifies the data from these different protocols and sampling cycles into standardized data with a unified timestamp, and writes the data for the same patient, aligned by sampling time, into the patient center event node using the patient identifier as an index. The event node records the device's online status and sampling anomaly status, sampling start and end times and sampling frequency, device identifier and data channel identifier, as well as the control commands that trigger device responses and their corresponding relationships.

[0042] In step S2, the edge gateway constructs a directed graph of clinical events with patient-centric event nodes as vertices. Taking a reduction in ventilator support level as an example, the parameter adjustment event is when clinical staff issue a command to reduce the support level through the bedside control terminal; the device response event is the parameter change confirmation frame returned by the ventilator; the physiological response event is the change in the patient's blood oxygen saturation as the support level is reduced; and the alarm result event is the hypoxia alarm generated when the blood oxygen saturation is below the threshold. These four types of events are sequentially connected in the directed graph of clinical events, with directed edges representing the dependency of subsequent events on the preceding event data. After the preceding event data is complete, the edge gateway marks the subsequent related events as decisionable and uploadable. This state only restricts the upload of related events generated by the edge gateway and does not affect the alarm and control channels of the medical device itself.

[0043] In step S3, the edge gateway calculates the scheduling priority of each event sub-stream. In this embodiment, the weighting coefficients α, β, γ, and δ are 0.30, 0.20, 0.30, and 0.20, respectively, and the smoothing constant ε is 0.05. Taking the alarm result event sub-stream as an example, its normalized remaining time ratio Δ is 0.5, the prior data integrity C is 0.7, the event risk R is 0.9, and the control interaction urgency U is 0.8, according to... Calculation. Substituting the above parameters, the calculated scheduling priority P is 1.035. This alarm result event substream has a higher scheduling priority than the event substream that only carries regular monitoring data during the same period, and is therefore scheduled first.

[0044] In step S4, the edge gateway dynamically updates the slice mapping based on the comparison between the scheduling priority and the first priority threshold. Event sub-streams with a scheduling priority not lower than the first priority threshold are mapped to URLLC slices, while event sub-streams with a scheduling priority lower than the first priority threshold are mapped to eMBB slices or mMTC slices according to their data carrying characteristics of high bandwidth or a large number of low-frequency connections. Based on the mapping results, the edge gateway generates a joint configuration request for radio resource blocks, MEC containers, and redundant paths, and sends it to the slice management unit and MEC orchestration unit through the communication interface.

[0045] In step S5, the edge gateway monitors the transmission latency and transmission reliability of the allocated event sub-streams. When the transmission latency of an alarm result event sub-stream exceeds the latency limit corresponding to its clinical event or the transmission reliability is lower than the reliability limit, it is determined to be out of bounds. Within the URLLC slice, some radio resource blocks occupied by low-priority event sub-streams are released in ascending order of scheduling priority and allocated to the out-of-bounds alarm result event sub-stream, or the calculation and forwarding tasks of the alarm result event sub-stream are migrated to a backup MEC container or an adjacent edge node.

[0046] In step S6, when the network is congested, the edge gateway backtracks along the direction of the pre-dependencies of the directed graph of clinical events, starting from the target event, removes duplicate event nodes and retains the necessary pre-dependency nodes of the target event, forming a closed dataset of pre-dependencies; the edge gateway generates a source data summary for the data, and generates a runtime traceability data packet together with the device status, scheduling intermediate quantities and their calculation results.

[0047] The parameters described above are used to illustrate the calculation and control process of this embodiment and do not constitute a limitation on the actual equipment configuration, sampling period, and slice resource parameters.

[0048] Example 2 This embodiment provides a detailed explanation of the first key technical means, namely the construction of patient-centered event nodes and the maintenance of directed graphs of clinical events. The process of converting multi-source device data into patient-centered event nodes and generating directed graphs of clinical events is as follows: Figure 3 As shown in the diagram, the structure of the data dependency chain between parameter adjustment, device response, physiological response, and alarm result in the directed clinical event graph is as follows: Figure 4 As shown.

[0049] like Figure 3 As shown, the event node construction of the multi-source device layer and edge gateway, as well as the event-directed... Figure 3 This constitutes the main data processing layer of this embodiment. The multi-source device layer includes vital signs monitors, ventilators, infusion pumps, blood gas analyzers, and bedside control terminals. The data generated by each device differs in protocol format, sampling period, and dimensions. Vital signs monitors upload parameters such as heart rate, blood oxygen saturation, and blood pressure in the first sampling period; ventilators upload ventilation parameters in the second sampling period; infusion pumps upload infusion rates in the third sampling period; blood gas analyzers upload test results at longer intervals; and the bedside control terminal generates control commands when clinicians adjust parameters. The device data access and standardization processing of the edge gateway unifies data from different protocols and sampling periods into standardized data with a unified timestamp. Using the patient identifier as an index, the multi-source device data for the same patient is aligned according to sampling time and written to the corresponding patient center event node.

[0050] In terms of the specific implementation of device data access and standardization processing, the edge gateway is configured with corresponding protocol parsers for different devices. This unifies the communication protocols used by vital sign monitors, ventilators, infusion pumps, blood gas analyzers, and bedside control terminals into structured fields, and standardizes heart rate to beats per minute, blood oxygen saturation to percentage, blood pressure to millimeters of mercury, ventilation parameters to centimeters of water column or milliliters, and infusion rate to milliliters per hour, among other standardized units. The local clocks of each device are aligned with the edge gateway clock through a time synchronization mechanism, and the edge gateway assigns a unified timestamp to each structured data record. Device data access and standardization processing unifies data from different protocols and sampling periods into standardized data with a unified timestamp, eliminating differences in protocols, units, and time bases between devices.

[0051] In terms of the specific implementation of alignment by sampling time, the edge gateway uses the patient identifier as an index to align multi-source device data of the same patient according to a unified timestamp. Alignment can be achieved using an alignment window method, which groups data from devices within the same preset alignment window into the same time slice, and if multiple data points for the same indicator exist within that time slice, the data with the most recent timestamp is selected or the average is applied. Alignment can also be achieved using the nearest timestamp matching method, which uses the timestamp of a certain device's data as a benchmark and matches the data from other devices with the one with the smallest time difference. The aligned data is written to the corresponding patient center event node, enabling data from the same patient generated on different devices with different sampling periods to establish a correspondence under the same time benchmark.

[0052] The patient center event nodes use structured data to record four types of information. The first type is device status, including online and sampling anomaly status, used to identify whether the device generating the event node data is currently online and whether sampling is abnormal. The online status is determined as follows: if the edge gateway receives heartbeat information or data uploaded by the device within a predetermined period, the device is considered online; if no heartbeat information or data is received from the device beyond the predetermined period, the device is considered offline. Sampling anomaly status is determined as follows: when the sampled value exceeds the range of the corresponding indicator of the device, the sampling interval does not match the set sampling period, or the sampled data is missing, it is considered a sampling anomaly. The second type is the sampling window, including the sampling start and end times and sampling frequency, used to identify the sampling time range corresponding to the event node data. The third type is the data source, including device identifier and data channel identifier, used to trace the device and data channel that generated the event node data. The fourth type is the command response relationship, used to record the control commands that trigger device responses and their corresponding relationships.

[0053] Regarding the establishment of command-response relationships, each control command is assigned a unique command identifier when issued by the bedside control terminal. This identifier can be composed of the command issuance time, the bedside control terminal identifier, and the command sequence number. The device carries this command identifier in its response data when responding to a control command. When writing to an event node, the edge gateway establishes a correspondence between the control command and the device response based on the command identifier carried in the device response data, and writes this correspondence into the command-response relationship field of the event node. Thus, each parameter adjustment command and its triggered device response form a traceable association at the event node level, avoiding the problem of command and response becoming disconnected under per-data-stream scheduling.

[0054] like Figure 4 As shown, the directed graph of clinical events starts with a parameter adjustment event as the initial node, sequentially connecting device response events, physiological response events, and alarm result events. Directed edges represent the dependency of subsequent events on preceding event data. Specifically, the parameter adjustment event represents adjustments made to device operating parameters by clinical personnel or control strategies; the device response event is the device's confirmation of the parameter adjustment; the physiological response event is the change in the patient's physiological indicators after the parameter adjustment; and the alarm result event is the alarm output based on the physiological response. Subsequent events are judged and uploaded only after all preceding event data is available, thus maintaining the data dependency order between events at the transmission level.

[0055] In terms of event type identification, the edge gateway identifies event types based on the source and content of the event data. Parameter adjustment events are identified based on parameter adjustment commands issued by the bedside control terminal; device response events are identified based on parameter change confirmation frames or response data returned by the device; physiological response events are identified based on physiological indicator change data uploaded by the vital signs monitor after parameter adjustment; and alarm result events are identified based on alarm signals output by the monitoring equipment. The edge gateway only converts received alarm outputs into associated event records and does not generate or replace native alarms from medical devices. The edge gateway labels each event node with its corresponding event type as a basis for determining preceding dependencies.

[0056] Regarding the determination of preceding dependencies, the edge gateway establishes data dependencies based on a preset event chain template, patient identifiers, instruction identifiers, and time windows. This relationship is used for data scheduling and operational record tracing, and does not represent a causal conclusion in the sense of medical diagnosis. Parameter adjustment events, device response events, physiological response events, and alarm result events are sequentially connected according to the template; the edge gateway updates the determineability and uploadability status of subsequent related events based on the completeness of preceding event data. The status update only applies to the upload of related events generated by the edge gateway and does not change the independent operation of the alarm and control channels of the monitoring equipment, ventilator, infusion pump, and other medical devices.

[0057] For storing the directed graph of clinical events, the edge gateway uses an adjacency list. Each event node corresponds to a vertex, and each directed edge represents the dependency of a subsequent related event on the preceding event data. Vertex attributes include event node identifier, event type, timestamp, and the completeness status of preceding dependencies, while edge attributes include dependency type. When writing a new event node, its location is determined and a directed edge is established based on the event type, patient identifier, instruction identifier, and time window. When the completeness status of preceding dependencies changes, the decision and upload status of subsequent related events are updated.

[0058] Regarding the calculation of the prerequisite data completeness C, the edge gateway calculates the ratio of the number of complete data items in the prerequisite dependencies of each event sub-flow in the directed graph of clinical events to the total number of prerequisite dependencies for that event sub-flow, thus obtaining the prerequisite data completeness C. The prerequisite data completeness C ranges from 0 to 1, with a larger value indicating more complete prerequisite dependency data for that event sub-flow. The prerequisite data completeness C is used in subsequent scheduling priority calculations to prioritize the completion of prerequisite data for event sub-flows, thereby closing the data dependency chain.

[0059] To adapt to different clinical scenarios, the construction of patient-centric event nodes can be implemented in two specific ways. The first is real-time streaming construction, where the edge gateway immediately generates or updates the corresponding event node and writes it into the directed clinical event graph upon receiving device data. This is suitable for latency-sensitive events such as parameter adjustments and device responses. The second is window aggregation construction, where device data with long sampling periods or large volumes is aggregated according to a predetermined sampling window and then written into the event node. This is suitable for data that is not latency-sensitive but has high integrity requirements. Both implementations maintain the correspondence between event nodes and their predecessor dependencies, and after being written into the directed clinical event graph, they participate in a unified scheduling priority calculation. In the directed clinical event graph, latency-sensitive events are constructed using real-time streaming to reduce end-to-end latency, while latency-insensitive events are constructed using window aggregation to reduce the frequency of event node writing. The two construction methods are executed in parallel within the edge gateway.

[0060] The construction of event nodes and the maintenance of the directed event graph are illustrated using a specific event chain as an example. Clinical staff issue a support level downgrade command to the ventilator via the bedside control terminal. This command is assigned a command identifier, for example, composed of the command issuance time, the bedside control terminal identifier, and the command sequence number. The parameter adjustment event generated by the bedside control terminal is written into an event node. This event node records the event type as parameter adjustment, the patient identifier, the device identifier as the bedside control terminal, the sampling window as the command issuance time, and the command response relationship as pending response. After receiving the command, the ventilator sends back a parameter change confirmation frame. The confirmation frame carries the aforementioned command identifier. The edge gateway generates a device response event based on the confirmation frame. This event node records the event type as device response, the device identifier as the ventilator, the sampling window as the confirmation frame timestamp, and the command response relationship as the aforementioned command identifier. Subsequently, the vital signs monitor uploads data showing a decrease in the patient's blood oxygen saturation. The edge gateway generates a physiological response event. This event node records the event type as physiological response, the device identifier as the vital signs monitor, and the sampling window as the blood oxygen data sampling time. When blood oxygen saturation falls below a preset threshold, the monitoring device outputs a low oxygen alarm, and the edge gateway generates an alarm result event. This event node records the event type as alarm result, the device identifier as monitoring device, and the sampling window as the time the alarm occurred. These four event nodes are sequentially connected in the directed clinical event graph, forming an event chain with prerequisite dependencies: parameter adjustment event, device response event, physiological response event, and alarm result event. Before the prerequisite data for the alarm result event is complete, the edge gateway does not mark the associated alarm result event it generates as decidable and uploading-ready. After the data for the parameter adjustment event, device response event, and physiological response event are complete, the status of the associated event is updated. This status update does not affect the native alarm output of the monitoring device.

[0061] In this embodiment, the patient-centric event nodes and the directed graph of clinical events are maintained by a processing unit within the ICU edge gateway. The edge gateway can be configured with an event node construction unit and an event directed graph maintenance unit responsible for event node construction and event directed graph maintenance respectively, or these can be implemented as a software functional module within the same processing unit. The edge gateway stores the directed graph of clinical events locally and synchronously updates the preceding dependency state after writing a new event node, for use in scheduling priority calculation, slice mapping, and joint allocation.

[0062] Example 3 This embodiment details the calculation of scheduling priorities, slice mapping, and joint allocation implementation. The dynamic mapping and joint allocation relationships of event sub-streams among URLLC slices, eMBB slices, and mMTC slices are as follows: Figure 5 As shown.

[0063] like Figure 5As shown, after priority calculation, event sub-flows enter the first priority threshold judgment. Event sub-flows with scheduling priorities not lower than the first priority threshold are mapped to URLLC slices; event sub-flows with priorities lower than the first priority threshold are mapped to eMBB slices or mMTC slices according to the data carrying characteristics of high bandwidth or a large number of low-frequency connections. All three slices are connected to a joint allocation object, which includes radio resource blocks, MEC containers, and redundant paths, wherein the redundant paths use different radio resources than the primary path; in the next scheduling cycle, the scheduling priority is recalculated and the mapping relationship is updated according to whether the first priority threshold is crossed.

[0064] In terms of scheduling priority calculation, scheduling priority P is calculated according to... Calculation. In the formula, Δ is the normalized remaining time ratio between 0 and 1, C is the pre-data integrity between 0 and 1, R is the event risk between 0 and 1, U is the control interaction urgency between 0 and 1, α, β, γ, and δ are dimensionless non-negative weights, and ε is a positive dimensionless smoothing constant. The first term is the time urgency term, the second term is the pre-data incompleteness gain term, the third term is the event risk term, and the fourth term is the control interaction urgency term. The higher the scheduling priority P, the earlier the event sub-stream appears in slice mapping and resource allocation.

[0065] The normalized remaining time ratio Δ is obtained by dividing the difference between the deadline of the event corresponding to the sub-stream and the current time by the total time of the event, i.e. The values ​​are limited to between 0 and 1. Δ is 0 when the deadline has passed and 1 when the time window has not yet been entered. Since Δ and ε are dimensionless, all terms in the priority formula have consistent dimensions.

[0066] In real-time scheduling, event templates configure clinically permissible time limits for different event types. The edge gateway first obtains the estimated time based on the recent acquisition queuing time, wireless transmission time, and MEC processing time within the sliding window, and then subtracts the estimated time from the clinical cutoff time to form the effective cutoff time. The normalized remaining time ratio Δ is determined by dividing the difference between the effective cutoff time and the current time by the total event time limit. When historical samples are insufficient, the estimated time uses a configured conservative upper bound; Δ is set to 0 when the calculated result is less than 0, and Δ is set to 1 when the result is greater than 1. The estimated time, statistical window, and effective cutoff time are all written to the scheduling record so that Δ can be recalculated with the same input.

[0067] The completeness C of the preceding data is determined by the proportion of complete data in the preceding dependencies of the event sub-flow in the directed graph of clinical events. For each event sub-flow in the directed graph of clinical events, the edge gateway calculates the ratio of the number of complete data in the preceding dependencies of that event sub-flow to the total number of preceding dependencies of that event sub-flow, obtaining the preceding data completeness C, which ranges from 0 to 1. A higher preceding data completeness C indicates more complete preceding dependency data for that event sub-flow, and the closer the subsequent event is to a decisionable and uploadable state. The preceding data completeness C incorporates a preceding incompleteness penalty term β(1 - C), prioritizing the completion of preceding data for event sub-flows with incomplete preceding data, thereby closing the data dependency chain at the transmission layer.

[0068] The event risk quantity R is determined by a weighted average of the degree of physiological abnormality, alarm level, and hazard extent. The degree of physiological abnormality, alarm level, and hazard extent are normalized to X1, X2, and X3, respectively. The event risk quantity is calculated according to... Calculate, where w1, w2, and w3 are dimensionless non-negative weights and satisfy... The event risk value R is limited to between 0 and 1. This event risk value is used for communication scheduling and ordering, and does not output medical diagnostic conclusions.

[0069] The physiological abnormality level X1 is normalized according to the deviation of the physiological indicator from the boundary of the device's preset reference range; the alarm level X2 is normalized according to the alarm level mapping table output by the device; the hazard surface X3 is mapped according to the number of data channels affected by the same event, whether it involves control interactions waiting for response, and whether it triggers the device's native alarm. Each mapping table has its version number and effective time stored in the event template. When an input is missing or marked as a sampling anomaly, the edge gateway does not use the missing value to infer the medical status, but instead uses the conservative communication scheduling value preset by the event template and synchronously writes an invalid input flag.

[0070] The urgency U of the control interaction is determined by the waiting time for a response to a device command or parameter adjustment, and the clinical criticality. Specifically, the edge gateway records the waiting time for a response from the moment the device command or parameter adjustment is issued. The longer the waiting time, the larger the corresponding normalized value of the waiting time. Simultaneously, a normalized value of clinical criticality is determined based on the clinical criticality of the control interaction; the higher the clinical criticality, the larger this normalized value. The urgency U of the control interaction is the weighted sum of the normalized waiting time value and the normalized clinical criticality value, normalized to the range of 0 to 1. The urgency U of the control interaction is used to prioritize the transmission of control interaction data awaiting a response, preventing disconnection between device commands and responses due to transmission lag.

[0071] The normalized waiting time value Y1 is defined as the ratio of the waiting time to the response time of the device command, and is limited to between 0 and 1. The normalized clinical criticality value Y2 is mapped from the control interaction level in the event template to the range of 0 to 1. The urgency of the control interaction is determined according to... Calculate, where v1 and v2 are dimensionless non-negative weights and For monitoring events that do not include control commands or parameter adjustments, U is set to 0; for events that have not received a response after the response time limit has expired, Y1 is set to 1 and the response status is marked as timed out.

[0072] The weighting coefficients α, β, γ, and δ are all not less than 0 and satisfy the following conditions: Each weight is calibrated offline based on a predetermined scenario configuration or confirmed historical operation records, and remains unchanged within a scheduling operation cycle; the smoothing constant ε is a positive dimensionless constant. The specific weights and ε used in this embodiment are only for illustrating the calculation process.

[0073] Weights and the first priority threshold can be calibrated using anonymized historical operation records or offline replay records. During calibration, under the same event set, wireless resource budget, and MEC computing power configuration, the number of times the event deadline exceeds the limit, the number of times the preceding dependencies are not closed, and the waiting time for high-priority events to obtain resources are calculated. Under the constraint that the weights are non-negative and their sum is 1, a parameter set that makes the above indicators meet the preset upper limit is selected. The parameter set is distributed to the edge gateway through the configuration version and remains unchanged within one operation cycle; during recalibration, the original parameters, replay window, indicator calculation results, and effective version are retained. The first priority threshold is determined based on the available capacity of URLLC and the offline replay results, and it can be required that the scheduling priority be on the same side of the threshold for two consecutive scheduling cycles before switching mappings to avoid repeated switching near the critical value.

[0074] The calculation of scheduling priority is illustrated using several specific examples. The weighting coefficients α, β, γ, and δ are set to 0.30, 0.20, 0.30, and 0.20 respectively, and the smoothing constant ε is set to 0.05. For the alarm result event sub-stream, the normalized remaining time ratio Δ is 0.5, the preceding data integrity C is 0.7, the event risk R is 0.9, and the control interaction urgency U is 0.8. The calculated scheduling priority P is 1.035. For the monitoring data sub-stream that only carries routine monitoring data during the same period, with Δ = 0.9, C = 1.0, R = 0.2, and U = 0.1, the scheduling priority P is calculated to be 0.396 using the same formula. For the parameter adjustment event sub-stream waiting for device response, with Δ = 0.7, C = 1.0, R = 0.4, and U = 0.9, the scheduling priority P is calculated to be 0.700 using the same formula. The scheduling priorities of the three event sub-streams, from highest to lowest, are: alarm result event sub-stream, parameter adjustment event sub-stream, and monitoring data sub-stream. The edge gateway uses this to determine the order of slice mapping and resource allocation. The above example is used to illustrate the calculation process of scheduling priority and does not constitute a limitation on the weighting coefficients, normalization methods, or specific values.

[0075] In terms of the specific implementation of slice mapping, the edge gateway presets a first priority threshold. Event sub-streams with scheduling priorities not lower than the first priority threshold are mapped to URLLC slices, while event sub-streams with scheduling priorities lower than the first priority threshold are mapped to eMBB slices or mMTC slices according to their data carrying characteristics. When the scheduling priority of an event sub-stream increases and crosses the first priority threshold, it is mapped to a URLLC slice; when the scheduling priority decreases and falls below the first priority threshold, it is remapped to an eMBB slice or mMTC slice according to its data carrying characteristics.

[0076] In the specific implementation of joint allocation, the edge gateway first determines the allocation order of radio resource blocks based on the scheduling priority of each event sub-stream. That is, event sub-streams with higher scheduling priorities obtain radio resource blocks first, and radio resource blocks are allocated in units of predetermined resource block granularity. Next, MEC containers are configured for high-priority event sub-streams, enabling their computational tasks to be processed near the edge nodes close to the data source. The selection of MEC containers is determined based on the resource requirements of the computational tasks and the remaining computing power of the edge nodes. Simultaneously, redundant paths are configured for high-priority event sub-streams. These redundant paths use different radio resources than the primary path; for example, the primary path and redundant paths may use different frequency domain resources or different transmission nodes to improve transmission reliability. When the resources required by a high-priority event sub-stream are insufficient, occupied resources are released in ascending order of scheduling priority, allowing critical event sub-streams to obtain resources first. The joint allocation of radio resource blocks, MEC containers, and redundant paths is uniformly scheduled by the edge gateway, ensuring that radio resources, edge computing power, and transmission paths are organized around the clinical event chain, rather than allocated independently.

[0077] The joint configuration request must include at least the event substream identifier, the current slice mapping, the effective deadline, the number of requested radio resource blocks, MEC processing resource requirements, primary and redundant path constraints, and the resource lease expiration time. The slice management unit returns the radio resource block and bearer configuration results, and the MEC orchestration unit returns the container instance, available computing power, and task reception status. The edge gateway marks the joint configuration as effective only if both types of return results are valid. If either unit refuses or fails to return within the timeout period, the edge gateway releases the partially occupied temporary resources, selects the next available MEC container based on the remaining time limit, or maintains the current mapping, and writes the rejection reason and fallback result into the runtime traceability data packet.

[0078] Regarding the specific implementation of out-of-bounds resource reallocation, the edge gateway monitors the transmission latency and reliability of allocated event sub-streams. When the transmission latency of an event sub-stream exceeds the latency limit corresponding to its clinical event, or its transmission reliability falls below the reliability limit, an out-of-bounds situation is determined. Upon out-of-bounds occurrence, within the URLLC slice, low-priority event sub-streams are identified in ascending order of scheduling priority, their occupied radio resource blocks are released, and the released radio resource blocks are allocated to the out-of-bounds high-priority event sub-streams; alternatively, the computation and forwarding tasks of the out-of-bounds event sub-streams are migrated to a backup MEC container or adjacent edge nodes. The selection of backup MEC containers or adjacent edge nodes is based on their remaining computing power and network location. Before migration, a snapshot of the computation state of the out-of-bounds event sub-streams is taken; after migration, the computation state is restored in the backup MEC container or adjacent edge nodes, and redundant paths are updated synchronously.

[0079] The migration status includes the event sub-stream identifier, the maximum processed sequence number, the set of incomplete prerequisite dependencies, intermediate priority calculation values, the cumulative status of the source data digest, and configuration requests for which no confirmation result has been received. The edge gateway first sends the migration status to the target MEC container and receives confirmation of readiness, then stops the original container from receiving new tasks and switches the forwarding path. After the target container recovers, it returns the first processed sequence number, which the edge gateway uses to remove duplicate tasks. If the target container fails to prepare, the confirmation times out, or the recovered sequence number is discontinuous, the edge gateway cancels the migration and continues to use the original container or main path, while retaining the medical device's native alarm channel and bedside control channel to prevent network scheduling failures from interfering with the device's own safety functions.

[0080] In this embodiment, scheduling priority calculation, slice mapping, and joint allocation are performed by the priority calculation unit and the slice mapping and joint allocation unit within the ICU edge gateway. The priority calculation unit calculates the scheduling priority based on the normalized remaining time ratio of each event subflow in the directed clinical event graph, the completeness of preceding data, the event risk, and the urgency of control interactions. The slice mapping and joint allocation unit performs slice mapping, wireless resource block allocation, MEC container configuration, and redundant path establishment based on the scheduling priority. The two units communicate scheduling results through scheduling priority, ensuring that priority calculation and resource allocation are time-coordinated and logically consistent.

[0081] Example 4 This embodiment describes a 5G slice scheduling device for ICU multi-source equipment based on clinical event chains. The device includes an event node construction module, an event directed graph maintenance module, a priority calculation module, a slice mapping and joint allocation module, an out-of-bounds emergency module, and a congestion emergency module. The structural block diagram of the device is shown below. Figure 2 As shown.

[0082] like Figure 2As shown, the device in this embodiment includes an ICU multi-source device, an ICU edge gateway, a 5G network and edge computing unit, and remote applications. The ICU multi-source device provides device data to the event node construction module; the event node construction module, the event directed graph maintenance module, the priority calculation module, and the slice mapping and joint allocation module are connected sequentially. The event directed graph maintenance module also provides a clinical event directed graph to the congestion emergency module, and the priority calculation module provides scheduling priorities to the boundary outage emergency module; the processing results of the slice mapping and joint allocation module, the boundary outage emergency module, and the congestion emergency module are output through the communication interface. The communication interface sends a joint configuration request to the slice management unit and the MEC orchestration unit. The slice management unit and the MEC orchestration unit complete the configuration of URLLC, eMBB, mMTC slices, as well as wireless resource blocks, MEC containers, and redundant paths. The configured slices carry data to the ICU central station, remote consultation terminal, and electronic medical record system.

[0083] In this embodiment, the ICU edge gateway is configured with a processor, memory, and communication interfaces. The processor runs software programs for the event node construction module, event directed graph maintenance module, priority calculation module, slice mapping and joint allocation module, boundary crossing emergency module, and congestion emergency module; it can be a general-purpose embedded processor or an industrial control processor. The memory stores patient-center event nodes, clinical event directed graphs, scheduling priorities, wireless resource blocks and MEC container allocation results, redundant path information, and runtime traceability data packets. The communication interface interacts with multi-source devices, the 5G network's slice management unit, and the MEC orchestration unit. The communication interface may include a short-range communication interface for the device side, a mobile communication module for the 5G network, and an Ethernet interface for the hospital network. The edge gateway can be deployed near the ICU ward or in the hospital's computer room, close to the data source to obtain lower data access latency and nearby computing power.

[0084] The event node construction module is used to collect and standardize monitoring and treatment data from multiple devices within the ICU. The standardized data is aggregated into patient-centric event nodes, recording device status, sampling window, data source, and command response relationships. In terms of implementation, the event node construction module is implemented by the edge gateway's processor running corresponding software programs. It configures protocol parsers for each device, unifying the communication protocols used by vital sign monitors, ventilators, infusion pumps, blood gas analyzers, and bedside control terminals into structured fields. It also standardizes measurements such as heart rate to beats per minute, blood oxygen saturation to percentage, blood pressure to millimeters of mercury, ventilation parameters to centimeters of water or milliliters, and infusion rate to milliliters per hour. The event node construction module records a unified timestamp for the structured data through a mechanism that synchronizes with the clocks of each device. Clock synchronization can be achieved using a network time protocol or a precise time protocol, or the edge gateway can add a local timestamp to the data at the moment of data reception. Using the patient identifier as an index, the event node construction module aligns the multi-source device data for the same patient with the unified timestamp and writes it into the corresponding patient-centric event node. Alignment can be achieved using an alignment window approach, grouping data from devices within the same preset alignment window into the same time slice; alternatively, alignment can be achieved using the most recent timestamp matching method. The event node construction module records the device online status and sampling anomaly status, sampling start and end times and sampling frequency, device identifier and data channel identifier, as well as the control commands that trigger device responses and their corresponding relationships within the event nodes. Device online status is determined by whether heartbeat information or data is received within a predetermined period. Sampling anomaly status is determined by whether the sampled value exceeds the range, whether the sampling interval matches the set period, and whether sampled data is missing. The command-response relationship is established by matching the unique command identifier assigned when the control command is issued with the command identifier carried in the device response data. The patient center event nodes output by the event node construction module are written to the edge gateway's memory for use by the event directed graph maintenance module.

[0085] The event-directed graph maintenance module is used to construct and maintain a directed graph of clinical events with preceding dependencies, using patient-centric event nodes as vertices and the data dependencies between parameter adjustment events, device response events, physiological response events, and alarm result events as directed edges. The module determines the data dependencies between nodes based on patient identifiers, instruction identifiers, and time windows, and updates the determineability and uploadability status of subsequent related events based on the completeness of preceding event data. This status update only constrains the upload of related events generated by the edge gateway and does not change the independent operation of the medical device's own alarm and control channels.

[0086] The priority calculation module is used to calculate the scheduling priority for each event sub-flow in the directed graph of clinical events, based on the normalized remaining time ratio, the completeness of preceding data, the event risk, and the urgency of control interactions. The priority calculation module follows... The calculation is performed, where Δ, C, R, and U are dimensionless quantities between 0 and 1, α, β, γ, and δ are dimensionless non-negative weights, and ε is a positive dimensionless smoothing constant. The priority calculation module outputs the scheduling priority to the slice mapping and joint allocation module and the boundary violation emergency module.

[0087] The slice mapping and joint allocation module dynamically updates the mapping relationship between event sub-streams in URLLC slices, eMBB slices, and mMTC slices according to scheduling priority and the first priority threshold. This module generates joint configuration requests for radio resource blocks, MEC containers, and redundant paths, and sends the requests to the 5G network's slice management unit and MEC orchestration unit respectively through the communication interface; after the slice management unit and MEC orchestration unit perform resource configuration, they return the configuration results to the edge gateway.

[0088] The boundary violation emergency module is used to release the radio resource blocks occupied by low-priority event sub-streams and allocate them to high-priority event sub-streams that have exceeded their limits when the latency or reliability of an allocated event sub-stream exceeds its limits. Alternatively, it can migrate the computation and forwarding tasks of the event sub-stream that has exceeded its limits to a backup MEC container or an adjacent edge node and update the redundant paths. After the boundary violation is handled, the boundary violation emergency module returns to continuous monitoring status.

[0089] The congestion emergency module is used to backtrack along the preceding dependencies of the directed graph of clinical events, starting from the target event to be uploaded, when network congestion occurs. It removes duplicate event nodes and retains necessary preceding event nodes reachable from the target event, forming a closed dataset of preceding dependencies. This module generates a source data summary and, together with device status, scheduling intermediate quantities, and their calculation results, forms a runtime traceability data packet.

[0090] In this embodiment, the patient-centered event nodes output by the event node construction module enter the event directed graph maintenance module; the clinical event directed graphs output by the event directed graph maintenance module enter the priority calculation module and the congestion emergency module, respectively; the scheduling priorities output by the priority calculation module enter the slice mapping and joint allocation module and the boundary overflow emergency module, respectively; the slice mapping and joint allocation module and the boundary overflow emergency module interact with the slice management unit and the MEC orchestration unit via communication interfaces, and the congestion emergency module outputs runtime traceability data packets. Each module can be implemented by different processing units, or the corresponding program can be executed by the same processor.

[0091] The modules exchange data objects with version numbers. The event node construction module outputs the event node identifier, patient association identifier, unified timestamp, and data validity marker; the event directed graph maintenance module returns the set of preceding dependencies and the status of being able to be determined and submitted; the priority calculation module outputs the scheduling priority, intermediate calculation values, and configuration version; the slice mapping and joint allocation module returns the slice identifier, resource lease, and configuration confirmation; the boundary crossing emergency module and the congestion emergency module return the handling status and the running trace data packet identifier, respectively. After receiving data, adjacent modules verify the version, sequence number, and required status. If verification fails, the last valid configuration is retained and the exception is recorded. Therefore, the module functions can form a closed processing chain through explicit interfaces.

[0092] Example 5 This embodiment illustrates the device configuration, parameter settings, execution process, and evaluation methods in the application scenario of multi-source equipment monitoring and remote diagnosis and treatment in an ICU ward. The application scenario is as follows: Figure 7 As shown.

[0093] like Figure 7 As shown, the application scenario in this embodiment consists of four parts: an ICU ward, an ICU edge gateway, a 5G network, and remote applications. The ICU ward has multiple beds, each equipped with a vital signs monitor, ventilator, infusion pump, and blood gas analyzer, and a bedside control terminal. Monitoring and treatment data generated by each device are aggregated to the ICU edge gateway. The ICU edge gateway performs event node construction, event directed graph maintenance, priority calculation, slice mapping and joint allocation, and boundary crossing and congestion emergency handling. It interacts with remote applications through URLLC slices, eMBB slices, and mMTC slices of the 5G network. Remote applications include an ICU central station, a remote consultation terminal, and an electronic medical record system, used to receive monitoring data, support remote consultations, and retain medical records.

[0094] In order to make Figure 7 The multi-source devices in the system can form an event chain under the same time base. When a bed is activated, the bedside control terminal writes the patient association identifier, bed identifier, and device identifier into the bed binding table of the edge gateway. The patient association identifier is a scheduling identifier obtained by one-way mapping from the hospital's master index. The remote scheduling interface does not directly carry the patient's name. The edge gateway synchronizes with devices equipped with network clock interfaces using a precise time protocol or a network time protocol. For devices that cannot be actively synchronized, the edge gateway adds a receiving timestamp when receiving data and records the difference between the device timestamp and the receiving timestamp. Protocol parsing, unit unification, timestamp correction, event node writing, and scheduling calculation are executed by independent processes or containers, and data objects with sequence numbers are passed between processes through bounded message queues.

[0095] In terms of technical requirements, this embodiment exhibits varying requirements for latency, reliability, bandwidth, and connection density for data from different devices within the ICU ward. Parameter adjustment events, device response events, physiological response events, and alarm result events are critical data within the clinical event chain, requiring low latency and high reliability transmission. Specifically, the latency limit for alarm result events can be 50 milliseconds, for physiological response events 100 milliseconds, and for parameter adjustment and device response events 200 milliseconds. These event sub-streams have high requirements for transmission reliability. High-bandwidth monitoring data requires high transmission rates to accommodate the transmission of images, waveforms, and other data. Massive amounts of low-speed device data require a large number of connections and low power consumption to accommodate the access of numerous low-frequency monitoring devices. These differentiated technical requirements are met by URLLC slices, eMBB slices, and mMTC slices, respectively. Furthermore, this embodiment requires the scheduling process to be interpretable and verifiable, meaning that at any given time, the basis for scheduling priority can be traced back to each event sub-stream to meet the ICU's requirement for traceable and verifiable scheduling criteria.

[0096] In terms of hardware selection, this embodiment uses an industrial-grade edge gateway that supports multi-core processors, has memory, and multiple communication interfaces. The edge gateway is configured with a data communication interface for the device side, a mobile communication module for the 5G network, and an Ethernet interface for the hospital network, and is deployed near the ICU ward or in the hospital's computer room. Vital signs monitors, ventilators, infusion pumps, and blood gas analyzers are medical devices with data communication interfaces, which upload monitoring data, treatment data, and device response status to the edge gateway through their respective communication interfaces. The bedside control terminal assigns instruction identifiers to parameter adjustment commands and receives device confirmation information. The edge gateway only forms event nodes and scheduling requests based on the received data; it does not replace the alarm logic of the medical devices themselves, nor does it change the independent control channel between the bedside control terminal and the medical devices. The 5G network is configured with URLLC slices, eMBB slices, and mMTC slices, and MEC containers are deployed near the edge gateway. The edge gateway's memory is used to store patient-centric event nodes, directed graphs of clinical events, scheduling priorities, slice mapping and joint configuration results, redundant path information, and runtime traceability data packets.

[0097] The edge gateway's software environment includes device protocol adaptation service, event graph service, scheduling service, MEC task service, and traceability storage service. The device protocol adaptation service maintains the received sequence number according to the device identifier and performs duplicate frame filtering; the event graph service partitions the graph storage space according to the patient association identifier to avoid misassociating event nodes from different patients; the scheduling service reads updated nodes and calculates priorities according to a fixed scheduling cycle; the MEC task service maintains the task status of the main container and the backup container; and the traceability storage service partitions and saves the original data summary and structured scheduling records. If a service exits abnormally, it is restarted by the edge gateway's process management program, resuming processing from the most recently confirmed sequence number upon restart.

[0098] Regarding parameter settings, in this embodiment, the vital signs monitor uploads data such as heart rate, blood oxygen saturation, and blood pressure with a sampling cycle of 1 second; the ventilator uploads ventilation parameters with a sampling cycle of 0.5 seconds; the infusion pump uploads infusion rate with a sampling cycle of 2 seconds; and the blood gas analyzer uploads test results every 30 minutes. The edge gateway aggregates multi-source device data by patient identifier and performs time alignment according to a preset alignment window. The weighting coefficients α, β, γ, and δ for scheduling priority calculation are set to 0.30, 0.20, 0.30, and 0.20, respectively, and the smoothing constant ε is set to 0.05. The weighting coefficients w1, w2, and w3 of the event risk quantity R correspond to the degree of physiological abnormality, alarm level, and hazard surface, respectively, and the sum of the three is 1. The first priority threshold for slice mapping is set according to the clinical scenario, and event substreams with a scheduling priority not lower than the first priority threshold are mapped to URLLC slices. The source data digest is generated using a hash digest algorithm. The above parameters are used to illustrate this embodiment and do not constitute a limitation on actual device configuration, sampling cycle, weighting coefficients, and thresholds.

[0099] This embodiment defines a baseline sampling window before the parameter adjustment event configuration command is issued and a response sampling window after the device responds. The baseline sampling window stores the device status and physiological parameters before parameter adjustment, while the response sampling window starts from the moment the device responds and covers subsequent physiological responses and alarm determinations. The time alignment window is no larger than the shortest event determination window among the participating devices. When multiple records appear on the same data channel within the window, the record with the timestamp closest to the target event is retained, while also retaining the duplicate record count. When illustrating slice mapping, the first priority threshold can be set to 0.65; this value only corresponds to the weight and playback conditions given in this embodiment, and the actual deployment will be recalibrated according to the URLLC capacity and offline playback results.

[0100] Regarding the execution process, this embodiment uses a clinical event chain where a decrease in ventilator support level triggers a hypoxia alarm as an example. Clinical staff issue a command to the ventilator to decrease the support level via the bedside control terminal. The bedside control terminal assigns a unique command identifier to this command, which consists of the command issuance time, the bedside control terminal identifier, and the command sequence number. The edge gateway generates a parameter adjustment event and writes it to the patient center event node, recording the event type as parameter adjustment, device status, sampling window, data source, and command response relationship, with the command response relationship marked as pending response. After receiving the command, the ventilator sends back a parameter change confirmation frame carrying the command identifier. The edge gateway generates a device response event and writes it to the event node, establishing a correspondence between the parameter adjustment event and the device response event based on the command identifier. Subsequently, the vital signs monitor uploads data showing a decrease in the patient's blood oxygen saturation, and the edge gateway generates a physiological response event and writes it to the event node. When the blood oxygen saturation falls below a preset threshold, the monitoring device outputs a hypoxia alarm, and the edge gateway generates an alarm result event and writes it to the event node. In the directed graph of clinical events, the edge gateway sequentially connects parameter adjustment events, device response events, physiological response events, and alarm result events to form an event chain with prerequisite dependencies. Alarm result events are marked as identifiable and uploading only after all prerequisite data is complete.

[0101] When establishing an event edge, parameter adjustment events and device response events must have the same patient association identifier and instruction identifier, and the response time must be within the response window of that type of instruction; device response events and physiological response events must have the same patient association identifier, and the sampling time of the physiological response must be within the response sampling window after the device response; alarm result events, in addition to meeting the patient association identifier and time window conditions, must also reference the alarm sequence number output by the monitoring device. Data that is late but still within the allowed supplementary recording window is written to the original event node and the integrity of the preceding data is recalculated; data that exceeds the supplementary recording window is saved separately as a late record without rewriting the already submitted event version; data with the same sequence number is only retained once and the duplicate reception count is increased.

[0102] The edge gateway calculates the scheduling priority of each event sub-stream. Taking the alarm result event sub-stream as an example, its normalized remaining time ratio Δ is 0.5, the preceding data integrity C is 0.7, the event risk R is 0.9, and the control interaction urgency U is 0.8. Calculation. Substituting the above parameters, the calculated scheduling priority P is 1.035. The scheduling priority of the monitoring data sub-stream, which only carries regular monitoring data, is 0.396, lower than that of the alarm result event sub-stream. The scheduling priority of the alarm result event sub-stream is compared with the first priority threshold; when it is not lower than the first priority threshold, it is mapped to a URLLC slice and a radio resource block, MEC container, and redundant path configuration request are generated. Physiological response events, device response events, and parameter adjustment events determine slice mappings according to their respective scheduling priorities and their relationship with the first priority threshold; event sub-streams lower than the first priority threshold are mapped to eMBB slices or mMTC slices according to their data carrying characteristics.

[0103] This example uses necessary data items in the preceding event nodes as preceding dependency units for statistical analysis. For example, the instruction identifier and adjustment time of the parameter adjustment event, the confirmation status and device status of the device response event, and the sampling fragment and validity mark of the physiological response event constitute preceding dependency units. When seven of the ten necessary units are complete, the preceding data completeness C is 0.7. If node-level statistics are used, the C of the alarm result event in the same four-node event chain takes the values ​​of 0, 1 / 3, 2 / 3, and 1 respectively. The edge gateway fixes the statistical granularity used in the event template, and the same event type does not use two different granularities within a configuration version. According to the first priority threshold of 0.65 in this embodiment, the alarm result event sub-stream with a scheduling priority of 1.035 and the parameter adjustment event sub-stream waiting for response with a scheduling priority of 0.700 enter the URLLC slice, and the regular monitoring event sub-stream with a scheduling priority of 0.396 enters the mMTC slice according to its low-frequency connection characteristics.

[0104] When an alarm result event substream experiences a delay exceeding the limit, the edge gateway, within the URLLC slice, releases some radio resource blocks occupied by low-priority event substreams in ascending order of scheduling priority and allocates them to the alarm result event substream, or migrates the computation and forwarding tasks of the alarm result event substream to a backup MEC container or an adjacent edge node. When network congestion occurs, the edge gateway backtracks along the directed graph of clinical events, extracting parameter adjustment event data, device response event data, physiological response event data, and alarm result event data to form a pre-dependency closed dataset, and generates a runtime traceability data packet containing source data summaries, device status, scheduling intermediate quantities, and their computation results. The edge gateway stores the runtime traceability data packet in local memory and, when conditions permit, uploads it to the ICU central station or electronic medical record system for clinical review and runtime record verification.

[0105] After the joint configuration takes effect, the edge gateway assigns a continuously incrementing transmission sequence number to each event substream and carries the same event substream identifier and different path identifiers in the main path and redundant path data. Remote applications deduplicate events based on the event substream identifier and transmission sequence number, and return an acknowledgment message after receiving the target event and its necessary pre-processing data. If the edge gateway does not receive an acknowledgment within the acknowledgment time limit, it first retransmits through a path that has not yet failed. After the maximum number of retransmissions is reached, the event substream is transferred to a local queue for retransmission and a timeout record is generated. The device's native alarm data can still be transmitted along the independent channel configured for the device itself. The edge gateway's associated event waiting and retransmission do not block the medical device's local display, audible and visual alarms, and bedside control.

[0106] To evaluate the processing performance of this embodiment, it can be compared with a control method that uses fixed mapping slices based on device type and independently schedules each data stream, under the same device data sequence, wireless resource budget, MEC computing power configuration, and network load trajectory. During evaluation, the difference between the event occurrence time and the target application's confirmation reception time is recorded to form the end-to-end latency of critical events; the percentage of subsequent related events arriving before necessary preceding events is calculated to form the related data out-of-order rate; the percentage of event chains with complete necessary preceding data during upload is calculated to form the preceding data completeness rate; and the percentage of high-priority event substreams that obtain resource allocation within specified latency and reliability limits is calculated to form the resource assurance rate. All indicators should be calculated within the same statistical window and under the same event set.

[0107] During offline playback, the event template, parameter version, and random network load trajectory are first fixed, and then the fixed device type mapping method and the method in this embodiment are run respectively. Both processing methods read events from the same original data copy and use the same retransmission limit, MEC container number, and total number of wireless resource blocks; records of offline devices, timestamp retrogression, or data corruption are excluded according to the same rules, and the number of excluded records is recorded. Each round of playback outputs an event-level log, which can at least recalculate end-to-end latency, associated data out-of-order rate, preceding data completeness rate, and resource guarantee rate.

[0108] Under the above evaluation conditions, dynamic slice mapping based on priority thresholds enables event sub-streams that are close to the deadline or have increased event risk to enter URLLC slices; the joint configuration of radio resource blocks, MEC containers and redundant paths is used to shorten the transmission and processing paths of critical event sub-streams; and the directed graph of clinical events and the availability of preceding data are used to maintain the dependency order of related data.

[0109] Example 6 This embodiment, using a multi-bed ICU ward under high network load and congestion conditions, illustrates the extraction of pre-dependent closed datasets and the generation of runtime traceability data packets. Application scenarios include... Figure 7 As shown, the emergency response procedures for boundary crossings and congestion are as follows: Figure 6 As shown.

[0110] like Figure 7 As shown, the application scenario in this embodiment consists of four parts: an ICU ward, an ICU edge gateway, a 5G network, and remote applications. The ICU ward has multiple beds, each equipped with a vital signs monitor, ventilator, infusion pump, and blood gas analyzer, and a bedside control terminal. Monitoring and treatment data generated by each device are aggregated to the ICU edge gateway. The ICU edge gateway completes event node construction, event directed graph maintenance, priority calculation, slice mapping and joint allocation, and boundary crossing and congestion emergency handling. It interacts with remote applications through URLLC slices, eMBB slices, and mMTC slices of the 5G network. Figure 6 As shown, when the edge gateway detects an out-of-bounds behavior, it implements resource reallocation or edge migration. When network congestion is detected, it extracts the preceding dependency closed dataset, closes the event chain, and generates a runtime traceability data packet.

[0111] In multi-bed concurrent scenarios, the edge gateway maintains a directed event graph that isolates each other based on patient association identifiers. Simultaneously, a global scheduling queue stores the scheduling priority, effective deadline, and current transmission status of each event sub-stream. Slice mapping still operates on an event sub-stream basis, avoiding the fixation of all data from the same bed or device into a single slice. Each cycle, the global scheduling queue only reads changed event nodes, updates the priority of affected event sub-streams, and maintains the basic transmission of low-priority routine monitoring data through minimum resource reservation, preventing sustained high load from making routine status invisible for extended periods.

[0112] In terms of technical requirements, this embodiment involves the simultaneous operation of numerous monitoring devices in multi-bed ICU wards, resulting in high network load and a high risk of network congestion. While ensuring low-latency and highly reliable transmission of critical events, it also requires the preservation of a complete clinical event chain and the formation of traceable and verifiable operational records under network congestion and resource constraints to meet the needs of clinical review and operational record verification. Specifically, it requires the extraction of pre-dependent closed datasets capable of closing the event chain under congestion conditions; that is, retaining only the necessary pre-dependent data set sufficient to determine event chain dependencies, rather than transmitting or discarding the entire dataset. Furthermore, the operational traceability data packet must include a source data summary, device status, and scheduling intermediate quantities and their calculation results, enabling post-event verification of the event chain determination criteria. The source data summary verifies data integrity, the device status traces the online and sampling status of the devices at the time of data generation, and the scheduling intermediate quantities and their calculation results trace the intermediate quantities used for scheduling priority calculation.

[0113] In terms of hardware selection, this embodiment uses an industrial-grade edge gateway that supports multi-core processors, has memory, and multiple communication interfaces. The edge gateway is configured with a short-range communication interface for the device side, a mobile communication module for the 5G network, and an Ethernet interface for the hospital network, and is deployed near the ICU ward or in the hospital's server room. Each bed in the multi-bed ICU ward is equipped with a vital signs monitor, ventilator, infusion pump, and blood gas analyzer, and a bedside control terminal. The 5G network is configured with URLLC slices, eMBB slices, and mMTC slices, and MEC containers are deployed near the edge gateway. The edge gateway is equipped with sufficient memory to temporarily store pre-dependent closed datasets and runtime trace data packets during network congestion, and upload them after network recovery or when conditions permit.

[0114] Regarding parameter settings, in this embodiment, network congestion is determined based on network load or buffer occupancy. For example, network congestion is determined when the uplink transmission buffer occupancy of the edge gateway exceeds a preset occupancy threshold, or when the network load exceeds a preset load threshold, or when the overall transmission latency of the allocated event substreams increases by more than a preset range. The latency and reliability limits corresponding to each clinical event are set according to the clinical treatment time limit. For example, the latency limit for alarm result events can be 50 milliseconds, and the latency limit for physiological response events can be 100 milliseconds. The weighting coefficients α, β, γ, and δ for scheduling priority calculation are 0.30, 0.20, 0.30, and 0.20, respectively, and the smoothing constant ε is 0.05. The source data digest is generated using a hash digest algorithm, such as SM3 or other hash digest algorithms with anti-collision characteristics. The above parameters are used to illustrate this embodiment and do not constitute a limitation on actual equipment configuration, congestion determination thresholds, and algorithm parameters.

[0115] To prevent repeated entry and exit from congestion due to fluctuations in buffer occupancy around the threshold, the edge gateway uses an entry threshold and an exit threshold to form a hysteresis interval. Congestion occurs when buffer occupancy or network load reaches the entry threshold for multiple consecutive monitoring periods, and exits from congestion when it falls below the exit threshold for multiple consecutive monitoring periods and there are no timeout events; the exit threshold is lower than the entry threshold. The monitoring period, number of consecutive periods, the two types of thresholds, and their effective versions are written to the configuration record. When network-side load indicators are temporarily unavailable, the edge gateway uses the locally transmitted buffer occupancy rate and acknowledgment round-trip time as the basis for judgment, and indicates the source of the judgment in the runtime traceability data packet.

[0116] Regarding the implementation process, this embodiment uses the handling of clinical event chains under network congestion as an example. During the normal operation of a multi-bed ICU ward, the monitoring and treatment data generated by each device are standardized by the edge gateway and written into the patient center event node. The edge gateway maintains the prerequisite dependencies of each event chain in the directed graph of clinical events. Taking the clinical event chain of an alarm triggered by a single infusion pump rate adjustment for a patient in a certain bed as an example, this event chain includes parameter adjustment events, device response events, physiological response events, and alarm result events. These four types of events are connected sequentially in the directed graph of clinical events. The prerequisite dependencies of the alarm result event include parameter adjustment event data, device response event data, and physiological response event data.

[0117] When network load increases and the uplink transmission buffer of the edge gateway exceeds a preset threshold, the edge gateway determines that the network is congested. During congestion, the edge gateway, for the target event in the directed graph of clinical events that is awaiting transmission, backtracks along the graph from the target event forward, collecting preceding event data layer by layer along the directed edges until it reaches the starting node of the event chain, i.e., the parameter adjustment event. After deduplicating the event data collected during the backtracking process, only the necessary preceding data set for determining the event chain dependency is retained, i.e., parameter adjustment event data, device response event data, physiological response event data, and alarm result event data, forming a closed dataset of preceding dependencies. The backtracking algorithm can be implemented using depth-first search or breadth-first search, traversing forward along the preceding dependency direction starting from the target event, collecting data corresponding to each preceding event node during the traversal.

[0118] During backtracking, the edge gateway pushes the target event node into the set to be visited and sets the set of visited nodes. For each node retrieved, its mandatory preceding dependencies are read according to the event template. Unvisited preceding nodes are added to the set to be visited, and the current node is written to the candidate set; this process continues until the set to be visited is empty or the starting node of the event chain is reached. The candidate set is sorted by event timestamp and dependency level, and records with the same event node identifier and data sequence number are merged into one. Bypass monitoring nodes unrelated to the target event determination, duplicate sample fragments, and old nodes replaced by updated versions are not included in the closed dataset, thus obtaining the minimum set relative to the current event graph and event template.

[0119] If the data corresponding to the mandatory prerequisite dependency has not yet arrived, the edge gateway marks the target event as unclosed and writes the missing node identifier, the waiting start time, and the allowed supplementation deadline into the record to be completed, without mistakenly marking it as a closed dataset of prerequisite dependencies. If the data is still missing after the supplementation deadline, the edge gateway generates an abnormal operation traceability data packet containing the reason for the missing data and a summary of the existing data, and sends the existing data according to the scheduling priority of the target event itself; this processing only applies to the associated records generated by the edge gateway and does not delay the native alarm information of the monitoring or treatment devices.

[0120] The edge gateway calculates a hash digest for each data point in the preceding dependency closed dataset. This hash digest generates a fixed-length digest value for each data point, used to verify data integrity. The edge gateway then combines the hash digests with the device online and sampling statuses recorded by event nodes, as well as intermediate quantities used in scheduling priority calculations, to generate a runtime traceability data packet. In the runtime traceability data packet, the source data digest is the set of hash digests for each data point in the preceding dependency closed dataset; the device status is the device online and sampling statuses recorded by each event node; and the scheduling intermediate quantities and their calculation results include the normalized remaining time ratio, preceding data integrity, event risk, control interaction urgency, and weighting coefficients used in scheduling priority calculations. The runtime traceability data packet is stored in the edge gateway's memory and uploaded to the ICU central station or electronic medical record system when the network recovers or conditions permit, for clinical review and runtime record verification. When no network congestion occurs, the edge gateway transmits data normally according to the mapping and allocation results.

[0121] The data packet for traceability is serialized using a defined field order. The packet header records the data packet version, summary algorithm identifier, patient association identifier, target event node identifier, event template version, and generation time. The packet body records, in order of dependency hierarchy, each event node identifier, data source identifier, sampling window, device status, data validity marker, source data summary, and Δ, C, R, U, weight, and scheduling priority P. The packet tail records missing dependency markers, slice mapping results, resource configuration confirmation, and upload status. The source data summary is calculated based on standardized but uncompressed event data. The summary input order is determined by the event timestamp, device identifier, and data sequence number, enabling the receiver to recalculate in the same order.

[0122] When the edge gateway writes runtime trace data packets to its local circular storage area, it uses the entire event chain as the smallest commit unit: first, it writes to the temporary area and completes digest verification, then atomically updates it to a sendable state. When storage space is insufficient, it prioritizes cleaning up low-priority data packets that have received remote acknowledgments and exceeded the retention period, without splitting unacknowledged event chains; if storage space is still insufficient, it transfers low-priority unclosed records to the delay queue and reserves space for high-priority closed datasets. After network recovery, data packets are sent up according to scheduling priority and generation time, and the remote application returns a data packet identifier and digest value as acknowledgment; duplicate data packets are deduplicated based on the data packet identifier, and if the digest values ​​are inconsistent, they are rejected for overwriting and a verification failure status is returned.

[0123] To evaluate the processing effectiveness under network congestion conditions, this embodiment can be compared with a control method that reduces speed by a fixed proportion or directly discards data during congestion, under the same set of events to be transmitted, buffer capacity, network load trajectory, and statistical window. During evaluation, the proportion of data chains containing the target event and its necessary preceding events out of all target event chains is statistically analyzed to form the event chain integrity rate; the end-to-end delay of critical events is calculated based on the sending time and the acknowledgment reception time; transmission reliability is calculated based on the ratio of the number of acknowledgment received data packets to the number of sent data packets; and the operation traceability data packets are checked to ensure they contain source data digests, device status, and scheduling intermediate quantities and their calculation results.

[0124] The evaluation process also records the ratio of the closed dataset bytes to the original pending data bytes, the time required to clear the pending transmission queue after congestion is relieved, the number of summary verification failures, and the distribution of reasons for unclosed events. The control method and the method in this embodiment use the same event arrival order, the same initial buffer state, and the same remote confirmation delay; when equipment goes offline or source data is corrupted, both methods retain the same anomaly records and are counted or excluded according to the same rules. Event-level logs allow for the verification of target events, necessary preceding events, and data packet confirmation status item by item, avoiding the need to judge the completeness of the clinical event chain solely based on the total bytes.

[0125] Under the above evaluation conditions, backtracking along the directed graph of clinical events and forming a closed dataset of prior dependencies can avoid indiscriminately retaining all data or indiscriminately discarding related data during congestion; running the traceability data package saves the source data summary, device status, and scheduling intermediate quantities and their calculation results, which facilitates the review of data integrity and scheduling basis.

[0126] After congestion is cleared, the edge gateway first stops generating new congestion pruning results, maintains the effective slice mapping, and completes the retransmission of high-priority data packets. Then, it resumes sending regular monitoring data according to the generation time. If congestion re-enters during the retransmission process, data packets that have received remote confirmation will not be retransmitted, while unconfirmed data packets will remain in the local circular storage area. The above-mentioned state transitions of entering, processing, exiting, and re-entering congestion all retain configuration versions and timestamps, so those skilled in the art can reproduce the processing results based on the same event graph, thresholds, and queue states.

[0127] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An ICU multi-source device 5G slice scheduling method based on a clinical event chain, characterized in that, Performed by the ICU edge gateway, including: S1 collects and standardizes multi-source equipment data in the ICU, and aggregates it by patient into a patient-centric event node that records equipment status, sampling window, data source, and command response relationship; S2, construct a directed graph of clinical events with preceding dependencies, with the patient-centered event node as the vertex and the data dependencies between clinical events as the directed edges, and update the decisionable and uploadable status of subsequent events according to the completeness of the preceding data. S3 calculates the scheduling priority based on the normalized remaining time ratio of the event sub-stream, the completeness of the preceding data, the event risk level, and the urgency of the control interaction. S4. Dynamically update the mapping of event sub-streams between URLLC slices, eMBB slices and mMTC slices according to scheduling priority and first priority threshold, and send a joint configuration request for radio resource blocks, MEC containers and redundant paths to the slice management unit and MEC orchestration unit. S5 monitors the transmission status of allocated event sub-streams. When latency or reliability exceeds the limit, it reallocates the radio resource blocks occupied by low-priority event sub-streams to high-priority event sub-streams that have exceeded the limit, or migrates the calculation and forwarding tasks of out-of-limit event sub-streams to backup MEC containers or adjacent edge nodes. S6, when network congestion occurs, backtracks the preceding dependencies from the target event along the directed graph of clinical events, deduplicates and retains necessary preceding nodes to form a closed dataset of preceding dependencies, and generates a runtime traceability data packet containing a summary of source data, device status and scheduling intermediates and their calculation results.

2. The ICU multi-source device 5G slice scheduling method based on clinical event chain according to claim 1, characterized in that, In step S1, the patient-centered event node is associated with the patient identifier. Multi-source device data for the same patient is aligned by sampling time and written to the corresponding event node. The patient-centered event node records device status, sampling window, data source, and command response relationship. In step S2, the data dependency relationship is determined by the patient identifier, command identifier, and time window. The directed graph of clinical events starts with the parameter adjustment event as the starting node, sequentially connecting device response events, physiological response events, and alarm result events. Directed edges represent the dependency of subsequent related events on preceding event data. The determineable and uploadable states are used for uploading related events to the edge gateway, without changing the independent operation of the medical device's alarm and control channels. The alignment uses a preset alignment window or the most recent timestamp matching. When multiple records exist in the same data channel, the record closest to the target event timestamp is retained, and the duplicate reception count is recorded. Data records exceeding the supplementary recording window are late data.

3. The ICU multi-source device 5G slice scheduling method based on clinical event chain according to claim 1, characterized in that, In step S3, the scheduling priority P is determined according to... The calculation is performed where Δ is the normalized remaining time ratio, C is the completeness of the preceding data, R is the event risk, U is the urgency of the control interaction, α, β, γ, and δ are dimensionless non-negative weights, and ε is a positive dimensionless smoothing constant. The normalized remaining time ratio is obtained by dividing the difference between the deadline of the event corresponding to the event substream and the current time by the total time of the event and is limited to between 0 and 1. The completeness of the preceding data is determined by the ratio of the number of complete preceding dependencies to the total number of preceding dependencies. The event risk is determined by weighting the normalized physiological abnormality degree, alarm level, and hazard surface. The urgency of the control interaction is determined by weighting the normalized response waiting time and clinical criticality.

4. The ICU multi-source device 5G slice scheduling method based on clinical event chain according to claim 1, characterized in that, In step S4, event sub-streams with scheduling priorities not lower than the first priority threshold are mapped to URLLC slices, and event sub-streams with scheduling priorities lower than the first priority threshold are mapped to eMBB slices or mMTC slices according to their data carrying characteristics. When the scheduling priority of an event sub-stream crosses the first priority threshold, its slice mapping relationship is updated. The configuration order of radio resource blocks is determined according to the scheduling priority, and MEC container and redundant path configuration requests are generated for event sub-streams mapped to URLLC slices. The redundant path uses different radio resources than the primary path. The joint configuration request carries the event sub-stream identifier, valid deadline, and resource requirements. It takes effect when the configuration results of the slice management unit and the MEC orchestration unit are both valid. If any configuration fails, the temporary resources are released and the original mapping is maintained or the next available MEC container is selected.

5. The ICU multi-source device 5G slice scheduling method based on clinical event chain according to claim 1, characterized in that, In step S5, the transmission latency and transmission reliability of the allocated event sub-streams are monitored. When the transmission latency of an event sub-stream exceeds the latency limit corresponding to its clinical event or the transmission reliability is lower than the reliability limit, it is determined to be out of bounds. When out of bounds, some radio resource blocks occupied by low-priority event sub-streams are released in the URLLC slice in order of scheduling priority from low to high. The released radio resource blocks are allocated to the high-priority event sub-streams that have exceeded the bounds, or the calculation and forwarding tasks of the out-of-bounds event sub-streams are migrated to the backup MEC container or adjacent edge nodes, and the redundant paths are updated synchronously. Before migration, the event sub-stream identifier, the maximum processed sequence number, and the set of prerequisites that are not yet complete are sent to the target MEC container. After the target MEC container confirms that it is ready, it switches the forwarding path and removes duplicate tasks according to the recovery sequence number. If the confirmation fails, the migration is canceled.

6. The ICU multi-source device 5G slice scheduling method based on clinical event chain according to claim 1, characterized in that, In step S6, starting from the target event to be submitted, backtracking is performed along the direction of the preceding dependencies of the directed graph of clinical events to remove duplicate event nodes and retain the necessary preceding event nodes that can be reached by the target event, thus forming a closed dataset of preceding dependencies. A source data summary is generated from the data in the aforementioned pre-dependent closed dataset, and the source data summary, the device status recorded by the corresponding event node, and the intermediate scheduling quantities and their calculation results used for scheduling priority calculation are encapsulated into a runtime traceability data package; If a mandatory prerequisite has not yet been reached, mark the target event as unclosed and wait for it to be completed before the deadline for completion. The runtime traceability data packet is written to the temporary area in the order of event chain as the smallest submission unit, and is updated to the uploadable state after the digest verification is passed.

7. A 5G slicing scheduling device for ICU multi-source equipment based on clinical event chains, characterized in that, It includes an event node construction module, an event directed graph maintenance module, a priority calculation module, a slice mapping and joint allocation module, an out-of-bounds emergency module, and a congestion emergency module, which are deployed on the ICU edge gateway; The event node construction module is used to collect and standardize monitoring and treatment data from multiple sources in the ICU. The standardized data is aggregated into patient-centered event nodes according to the patient, and the device status, sampling window, data source and command response relationship are recorded. The event directed graph maintenance module is used to construct and maintain a clinical event directed graph with preceding dependencies, using the patient-centered event node as the vertex and the data dependencies between parameter adjustment events, device response events, physiological response events and alarm result events as directed edges. It also updates the determineability and uploadability status of subsequent related events based on the completeness of preceding event data. The priority calculation module is used to calculate the scheduling priority for each event sub-flow in the directed graph of clinical events based on the normalized remaining time ratio, the completeness of the preceding data, the event risk, and the urgency of the control interaction. The slice mapping and joint allocation module is used to dynamically update the mapping relationship between event sub-streams in URLLC slices, eMBB slices and mMTC slices according to scheduling priority and first priority threshold, and generate joint configuration requests for radio resource blocks, MEC containers and redundant paths. The boundary crossing emergency module is used to release the radio resource blocks occupied by the event sub-stream with lower scheduling priority when the latency or reliability of the allocated event sub-stream exceeds the limit, allocate the released radio resource blocks to the high-priority event sub-stream that has exceeded the limit, or migrate the calculation and forwarding tasks of the event sub-stream that has exceeded the limit to the backup MEC container or adjacent edge node. The congestion emergency module is used to backtrack the prerequisite dependencies of the target event along the directed graph of clinical events when the network is congested, remove duplicate nodes and retain the necessary prerequisite nodes of the target event to form a closed dataset of prerequisite dependencies, and generate a runtime traceability data packet containing a summary of source data, device status and scheduling intermediate quantities and their calculation results.

8. The ICU multi-source equipment 5G slicing scheduling device based on clinical event chain according to claim 7, characterized in that, The event node construction module associates patient identifiers, aligns multi-source device data for the same patient according to sampling time, and writes it into the corresponding event node, recording device status, sampling window, data source, and command response relationship. The event directed graph maintenance module determines data dependencies based on patient identifiers, command identifiers, and time windows, sequentially connecting device response events, physiological response events, and alarm result events with parameter adjustment events as the starting node. It updates the determineability and uploadability status of subsequent associated events based on the completeness of preceding event data. The status update does not change the independent operation of the medical device's alarm and control channels. The event node construction module and the event directed graph maintenance module interact through data objects with version numbers and serial numbers. When verification fails, the previous valid state is maintained and the anomaly is recorded.

9. The ICU multi-source equipment 5G slicing scheduling device based on clinical event chain according to claim 7, characterized in that, The priority calculation module according to The scheduling priority P is calculated, where Δ is the normalized remaining time ratio, C is the pre-data integrity, R is the event risk, U is the control interaction urgency, α, β, γ, and δ are dimensionless non-negative weights, and ε is a positive dimensionless smoothing constant. The slice mapping and joint allocation module maps event sub-streams with scheduling priorities not lower than the first priority threshold to URLLC slices, and maps event sub-streams with scheduling priorities lower than the first priority threshold to eMBB slices or mMTC slices according to data carrying characteristics. The mapping relationship is updated when the scheduling priority crosses the first priority threshold.

10. The ICU multi-source equipment 5G slicing scheduling device based on clinical event chain according to claim 7, characterized in that, The boundary crossing emergency module determines whether an event sub-stream has crossed the boundary based on transmission delay and transmission reliability. It releases the radio resource blocks occupied by low-priority event sub-streams and allocates them to high-priority event sub-streams that have crossed the boundary, according to the scheduling priority from low to high. Alternatively, it migrates the computation and forwarding tasks of the event sub-streams that have crossed the boundary and updates redundant paths. The congestion emergency module backtracks along the direction of the preceding dependencies, starting from the target event to be sent. It removes duplicate event nodes and retains the necessary preceding event nodes that can be reached by the target event to form a closed dataset of preceding dependencies. It determines the backtracking order according to the event timestamp and dependency level. Records with the same event node identifier and data sequence number are merged and deduplicated. It also marks the missing mandatory preceding dependency records that have not yet arrived.