A pressure ulcer prevention method and system based on multi-modal perception

Abstract

The application belongs to the field of medical information technology and relates to a pressure sore prevention method and system based on multi-modal perception. The method comprises the following steps: acquiring multi-modal perception data of a bed surface of a long-term bedridden patient and a nursing risk record to obtain regional nursing window data; constructing regional state quantities of each body surface region and building a multi-modal pressure sore risk state atlas; identifying a part-level risk state, a risk source and an adjacent region risk transfer state; generating a part-level nursing action package and configuring a verification window; forming a verification result, a review priority and a structured nursing record, and updating a next nursing window. The technical scheme of the application realizes pressure sore risk state identification, nursing action generation and cross-window review and update collaborative processing through regional multi-modal state expression and verification feedback closed loop, and can improve part-level nursing intervention matching stability and review consistency of a risk state after intervention.

Description

Technical Field

[0001] This application belongs to the field of medical information technology, and specifically relates to a method and system for preventing pressure ulcers based on multimodal perception. Background Technology

[0002] In existing technologies, pressure ulcer prevention in long-term bedridden patients typically relies on mattress pressure monitoring, skin temperature and humidity observation, and regular turning and rounds by caregivers. This approach aims to meet the need for continuous preventative management of high-risk areas such as the sacrum, coccyx, hips, and heels by collecting data on the condition of pressure points, combining this with the patient's past risk information, and issuing nursing prompts. However, existing pressure ulcer prevention methods have some significant shortcomings in terms of coordinating multi-source sensing results with site-level nursing decision-making.

[0003] In practical applications, existing pressure mattresses or nursing monitoring systems often possess the ability to display pressure distribution, provide temperature and humidity alerts, and remind patients of risk levels. However, their handling of the correlation between the sustained pressure status, shear changes, local microenvironmental changes, and individual patient risk factors in different body surface areas within the same nursing cycle remains relatively crude. Nursing staff typically need to rely on experience to determine the specific intervention site, intervention method, and timing of follow-up examinations. Influenced by factors such as changes in patient position, fluctuations in bed surface contact, disturbances during nursing procedures, and load transfer between adjacent pressure areas, the continuity of risk status and the consistency of post-intervention effect assessments are generally weak across different nursing cycles.

[0004] Therefore, existing technologies often suffer from limitations such as limited accuracy in matching multi-source sensory states with site-specific nursing interventions during pressure ulcer prevention in long-term bedridden patients, and weak consistency in risk status verification after nursing interventions. These are the shortcomings of existing technologies.

[0005] In view of this, it is necessary to provide a pressure ulcer prevention method and system based on multimodal perception to address the aforementioned deficiencies in the prior art. Summary of the Invention

[0006] The purpose of this application is to address the shortcomings of the existing technology, such as limited matching accuracy between multi-source perception status and site-level nursing intervention in the prevention of pressure ulcers in long-term bedridden patients, and weak consistency in risk status review after nursing intervention, by providing a pressure ulcer prevention method and system based on multimodal perception to solve the above-mentioned technical problems.

[0007] To achieve the above objectives, this application provides the following technical solution: In a first aspect, this application provides a pressure ulcer prevention method based on multimodal perception, comprising: Step S1: Obtain the bed surface multimodal perception data and nursing risk profile of long-term bedridden patients within the current nursing window, perform stable nursing window filtering and body surface area mapping on the bed surface multimodal perception data to obtain regional nursing window data; Step S2: Construct regional state quantities for each body surface region based on the regional nursing window data, and construct a multimodal pressure ulcer risk status map based on the regional state quantities. The edges of the multimodal pressure ulcer risk status map are determined by the decrease in load of the target body surface region, the increase in load of adjacent body surface regions, and the spatial proximity of the pressure centers of the target body surface region and adjacent body surface regions. Step S3: Based on the multimodal pressure ulcer risk status map, nursing risk records, and historical risk status of the previous nursing window, identify the site-level risk status, risk source, and risk transfer status of adjacent areas for each body surface region; Step S4: Generate a site-level nursing action package based on the site-level risk status, risk source, and risk transfer status of adjacent areas, and configure a verification window for the site-level nursing action package; Step S5: Reacquire the multimodal perception data of the target body surface area and adjacent body surface areas in the verification window. Based on the target decompression state and adjacent pressure state, form the verification results, review priority and structured nursing record of the site-level nursing action package. Update the multimodal pressure ulcer risk status map, review priority and structured nursing record of the next nursing window based on the verification results.

[0008] By adopting the above technical solution, the multi-source status of the bed surface, individual nursing risks, and post-nursing review results of long-term bedridden patients are incorporated into the same nursing window for collaborative processing. This achieves continuous coordination between the expression of body surface status, site-level risk constraints, nursing action generation, and verification feedback. It can maintain a stable correspondence between pressure ulcer risk assessment and nursing intervention even when the patient's position changes, bed surface contact fluctuates, and nursing operations alternate. It also reduces the status fluctuations caused by single experience judgments, meeting the needs of long-term bedridden patients for high accuracy in matching multi-source perceived status with site-level nursing intervention and strong consistency in post-nursing intervention risk status review during pressure ulcer prevention.

[0009] Specifically, windowing constraints and body surface region affixation for the perceived state of the bed surface help eliminate the influence of short-term nursing disturbances on the judgment of pressure sites, giving subsequent analysis a clear temporal scope and regional boundaries. Forming risk status expressions around body surface regions can consolidate scattered pressure, shear, and local environmental changes under the same semantic category, making risk status present a more coherent regional relationship. Combining individual nursing risks and historical risk continuity to conduct site-level identification can distinguish the current source of pressure, the source of continuous risk, and the direction of load transfer in adjacent areas, making nursing interventions more closely aligned with the patient's actual condition. Forming nursing actions and verification arrangements based on the identification results helps to transform risk prompts into nursing treatments with site orientation and verification timing, maintaining a stable connection between nursing actions and risk sources. Verifying the target area and adjacent areas within the verification window and writing the verification results back to the next nursing window helps to form a traceable closed-loop status update, ensuring that the verification priority and nursing records remain consistent throughout the continuous nursing process.

[0010] Preferably, step S1, which involves performing stable nursing window filtering and body surface region mapping on the multimodal sensing data of the bed surface to obtain regional nursing window data, specifically includes: The sampling segments within the current nursing window are subjected to contact stability determination. Sampling segments that meet the conditions for leaving the bed, nursing relocation, or abnormal sensor quality are marked as disturbed segments, and disturbed segments are removed from the current nursing window to obtain stable nursing segments. An initial body surface area mapping is formed based on the pressure profile, contact area edge, and pressure center location in the stable nursing segment; Within a stable care segment, the migration status of the center of pressure and the change status of the contact area are detected. When the migration status of the center of pressure or the change status of the contact area meet the remapping conditions, the initial body surface region mapping is updated. Based on the updated body surface region mapping or the unupdated initial body surface region mapping, stable nursing segments are assigned to regions to obtain regional nursing window data.

[0011] Based on the above scheme, stable nursing segments and body surface area attribution jointly limit the usable boundaries of bed surface data entering the analysis chain, so that short-term fluctuations caused by leaving the bed, moving, and sensor abnormalities are isolated from risk assessment, which can improve the consistency of body surface area attribution and provide a stable data foundation for subsequent site risk comparison.

[0012] Preferably, step S2 specifically includes: Based on the regional nursing window data, pressure dose, shear dose, microenvironmental load, and contact reliability status of each body surface area were calculated. Based on the contact reliability state, gated calculations of pressure dose, shear dose, and microenvironmental load are performed to obtain the regional load state; A multimodal pressure ulcer risk status map is constructed based on the regional load status. The pressure transfer direction between different body surface areas is determined based on the relationship between the regional load status in the current nursing window and the previous nursing window, as well as the spatial proximity of the pressure center between the target body surface area and adjacent body surface areas. The graph edges in the multimodal pressure ulcer risk status map are determined according to the pressure transfer direction.

[0013] In the above scheme, the regional load status and pressure transfer direction organize the state changes between different body surface areas into a map relationship, so that pressure, shear and microenvironment related information no longer remain at a single point, but can form a basis for regional risk transmission across nursing windows, and enhance the ability of subsequent site-level risk identification to express the load changes of adjacent areas, so that the risk map has a more stable regional relationship and a clearer risk transmission boundary in the continuous nursing process.

[0014] Preferably, the steps of gating calculations of pressure dose, shear dose, and microenvironmental load based on the contact reliability state specifically include: Contact stability and sensing quality are jointly mapped to a pressure confidence state, and slip direction stability and headboard posture stability are jointly mapped to a shear confidence state. Skin contact impedance change and humidity recovery trend are jointly mapped to a microenvironment confidence state. The pressure gating quantity is obtained based on the pressure dose and pressure confidence state; the shear gating quantity is obtained based on the shear dose and shear confidence state; and the microenvironment gating quantity is obtained based on the microenvironment load and microenvironment confidence state. Load merging is performed on the same surface area for pressure-gated, shear-gated, and microenvironment-gated variables, and the load merging results are written into the area load status.

[0015] Based on the above processing, the contact reliability state implements itemized gating for different risk sources, which constrains the impact of contact instability, slip fluctuations and local environmental recovery differences on the regional load state. This can improve the reliability of regional state quantities before they enter the map and reduce the interference of abnormal contact states on risk expression.

[0016] Preferably, step S3 specifically includes: Based on the multimodal pressure ulcer risk status map, determine the regional load status and pressure transfer direction of each body surface area; Based on the nursing risk record, each body surface area is classified into a state of reduced tolerance and a state of increased risk, and an individual tolerance correction state is formed based on the state of reduced tolerance and a state of increased risk. Based on the historical risk status of the previous nursing window, the area load status is continuously calculated using pressure memory to obtain the continuous risk status; Based on regional load status, map edge pressure transfer direction, individual tolerance correction status, and continuous risk status, identify location-level risk status, risk sources, and risk transfer status in adjacent areas.

[0017] Based on the above processing logic, individual tolerance correction status and continuous risk status jointly participate in site-level identification, so that a continuous correspondence is formed between the patient's own risk differences, the residual pressure effect of the previous nursing window and the current regional load. This can improve the stability of risk source attribution and provide a clearer basis for judging the risk transfer status of adjacent areas, making the risk basis before the generation of nursing actions closer to the patient's continuous pressure process and site tolerance differences.

[0018] Preferably, the steps for identifying the risk transfer status of adjacent areas specifically include: When the pressure transfer direction on the edge of the graph points to the adjacent body surface area, the regional load state change of the target body surface area and the regional load state change of the adjacent body surface area are compared to obtain a state where the transfer direction is consistent. When the conditions for transfer review are met for both the consistent transfer direction and the persistent risk status, the adjacent body surface regions will be written into the transfer review object. Write the adjacent body surface area corresponding to the transferred review object into the adjacent area risk transfer status, and update the risk source in the adjacent area risk transfer status to the pressure transfer source.

[0019] In the above scheme, the consistency of transfer direction and the transfer review object include the load changes of adjacent body surface areas in the risk transfer judgment, so that the pressure transfer caused by turning over or decompression can be identified separately, which can enhance the traceability of risk sources in adjacent areas and reduce the possibility of delayed discovery of new risks after nursing actions.

[0020] Preferably, step S4 specifically includes: The intervention area on the body surface is determined based on the risk status at the site level, and the intervention area is written into the execution area of ​​the site-level nursing action package; Based on the source of risk, action categories are formed among postural adjustment, decompression, skin examination, and microenvironment screening, and these action categories are written into the execution content of the site-level nursing action package. When the risk transfer status of adjacent areas does not include the transfer review object, the adjacent body surface area corresponding to the intervention body surface area is written into the review area of ​​the site-level nursing action package, and the verification object and verification content in the verification window are formed according to the execution area, execution content and review area; When the risk transfer status of an adjacent area includes a transfer review object, the transfer review object is written into the review area of ​​the site-level nursing action package, and the verification object and verification content in the verification window are formed based on the execution area, execution content and review area.

[0021] Based on the risk identification results mentioned above, the execution area, execution content, and review area jointly define the execution and verification boundaries of the site-level nursing action package. This enables nursing actions to be carried out around specific sites, risk sources, and review requirements of adjacent areas, which can enhance the stability of the connection between nursing prompts and on-site operations and improve the adaptability of the verification window.

[0022] Preferably, step S5 involves re-acquiring the multimodal sensing data of the target body surface area and adjacent body surface areas within the verification window, and forming the verification results, review priorities, and structured nursing records for the site-level nursing action package based on the target decompression state and adjacent pressure states. Specifically, this includes: The execution area is taken as the target body surface area, and the review area is taken as the adjacent body surface area; Within the verification window, a target decompression state is formed based on the pressure dose reduction ratio, shear dose reduction ratio, and microenvironment load recovery ratio corresponding to the execution area. Within the verification window, adjacent pressure states are formed based on the regional load state increment corresponding to the verification area, the pressure transfer intensity at the edge of the graph, and the adjacent transfer verification intensity. The verification results and review priority of the site-level nursing action package are determined based on the execution content, target decompression status, and adjacent pressure status. Write the execution area, execution content, review area, target decompression status, adjacent pressure status, verification results, and review priority into the structured nursing record.

[0023] Based on the above verification arrangements, the target decompression state, adjacent pressure state, and structured nursing records together form the basis for review after the execution of nursing actions. This allows changes in pressure in the target area and changes in pressure in the review area to be included in the result judgment simultaneously, which can improve the consistency between verification results and review priorities, and enhance the traceability of nursing records across shifts. This enables nursing staff to obtain more stable review targets and clearer review basis in the continuous nursing window.

[0024] Preferably, step S5, which updates the multimodal pressure ulcer risk status map, review priority, and structured nursing record for the next nursing window based on the verification results, specifically includes: When the target pressure relief state meets the effective pressure relief conditions and the adjacent pressure-bearing state does not meet the transfer review conditions, the historical risk status of the execution area in the next nursing window is written into the relief inheritance status, and the risk source corresponding to the execution area is updated to the relief source mark in the multimodal pressure ulcer risk status map, and the effective nursing action mark is written into the structured nursing record. When the target pressure relief status does not meet the effective pressure relief conditions, the historical risk status of the execution area in the next nursing window will be written into the unrelieved inherited status, and the risk source corresponding to the execution area will be retained in the multimodal pressure ulcer risk status map. When adjacent pressure conditions meet the transfer review conditions, a transfer review map edge pointing from the execution area to the review area is established in the multimodal pressure ulcer risk status map, and the review priority and structured nursing records are updated based on the transfer review map edge.

[0025] Based on the above feedback logic, the relief inheritance state, the non-relief inheritance state, and the transfer review graph edges will transform the verification results into the graph update boundary of the next nursing window. This will enable effective nursing, ineffective nursing, and pressure transfer in adjacent areas to form different state flows, which can enhance the continuity of risk status, the stability of review priority, and the closure of structured records in the continuous nursing process. It will also enable subsequent nursing judgments to form a traceable risk evolution basis along the verification results.

[0026] Secondly, this application also provides a pressure ulcer prevention system based on multimodal perception, comprising: The data acquisition unit is used to acquire bed surface multimodal perception data and nursing risk profiles of long-term bedridden patients within the current nursing window, and to perform stable nursing window filtering and body surface area mapping on the bed surface multimodal perception data to obtain regional nursing window data. The map construction unit is used to construct the regional state quantity of each body surface region based on the regional nursing window data, and to construct a multimodal pressure ulcer risk state map based on the regional state quantity. The graph edges in the multimodal pressure ulcer risk state map are jointly determined based on the decrease in load of the target body surface region, the increase in load of adjacent body surface regions, and the spatial proximity of the pressure center of the target body surface region and adjacent body surface regions. The risk identification unit is used to identify the site-level risk status, risk source, and risk transfer status of each body surface area based on the multimodal pressure ulcer risk status map, nursing risk records, and the historical risk status of the previous nursing window. The action generation unit is used to generate location-level nursing action packages based on location-level risk status, risk source, and risk transfer status of adjacent areas, and to configure a verification window for the location-level nursing action packages. The verification and update unit is used to reacquire the multimodal sensing data of the target body surface area and adjacent body surface areas within the verification window, generate the verification results, review priorities and structured nursing records of the site-level nursing action package based on the target decompression state and adjacent pressure states, and update the multimodal pressure ulcer risk status map, review priorities and structured nursing records of the next nursing window based on the verification results.

[0027] As can be seen from the above technical solutions, this application has the following advantages: This application provides a pressure ulcer prevention method and system based on multimodal perception. By incorporating the multi-source status of the bed surface of long-term bedridden patients, individual nursing risks, and post-nursing review results into the same nursing window for collaborative processing, it achieves continuous coordination between the expression of body surface area status, site-level risk constraints, nursing action generation, and verification feedback. It can maintain a stable correspondence between pressure ulcer risk assessment and nursing intervention under conditions of patient position changes, bed surface contact fluctuations, and alternating nursing operations, and reduce status fluctuations caused by single experience judgments. It meets the needs of long-term bedridden patients for high accuracy in matching multi-source perceived status with site-level nursing intervention and strong consistency in post-nursing intervention risk status review during pressure ulcer prevention. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of this application, the accompanying drawings used in the embodiments are briefly described below. The following drawings only show some embodiments of this application; those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0029] Figure 1 This is a flowchart of a pressure ulcer prevention method based on multimodal perception provided in this application; Figure 2 This is a block diagram illustrating the principle of a pressure ulcer prevention system based on multimodal perception provided in this application.

[0030] The system comprises: 1. Data acquisition unit; 2. Graph construction unit; 3. Risk identification unit; 4. Action generation unit; and 5. Verification and update unit. Detailed Implementation

[0031] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the following embodiments are only used to illustrate the technical solutions of this application and are not intended to limit the scope of protection of this application. Any adjustments, equivalent substitutions, improvements or other optional implementation methods made by those skilled in the art to the embodiments without departing from the concept and scope of protection of this application should fall within the scope of protection of this application.

[0032] It should be noted that in the description of this application, the terms "comprising," "including," "having," and their synonyms are used to indicate the presence of the described features, structures, steps, operations, elements, components, or combinations thereof, but do not exclude the presence of other features, structures, steps, operations, elements, components, or combinations thereof.

[0033] It should be noted in advance that, in order to facilitate understanding of the technical solutions of the embodiments of this application, some terms and related technologies involved in the embodiments of this application will be briefly explained below: 1. Multimodal perception: refers to a technical approach that uses different types of sensor data to comprehensively perceive the same object or scene. Different modalities can come from different data dimensions such as pressure, temperature, humidity, posture, and contact status. By performing time correspondence, spatial correspondence, and state association on different modal data, a more complete description of the object's state can be obtained than that of single sensor data. It is often used in scenarios such as medical monitoring, intelligent nursing, environmental perception, and state recognition.

[0034] 2. Gated computation: This refers to a computational method that limits, amplifies, or suppresses the participation of input data based on credibility, state conditions, or control parameters during data processing. It is usually used to distinguish between valid information and interference information, allowing more high-quality or state-matched data to participate in subsequent calculations, and constraining the impact of abnormal, low-credibility, or unstable data on the results. It is often used in scenarios such as multi-source data fusion, state identification, and time-series risk assessment.

[0035] 3. Graph Edge: A basic concept in graph structure. Graph structure is usually composed of nodes and edges. Nodes are used to represent objects, states or events, and graph edges are used to represent the connection, influence or transmission relationship between nodes. In data analysis, it can be used to express the direction of association, intensity of influence or state transition path between different objects. It is often used in knowledge graphs, relationship modeling, risk propagation analysis and network structure calculation.

[0036] To address the issues of limited matching accuracy between multi-source perception status and site-level nursing interventions during pressure ulcer prevention in long-term bedridden patients, and weak consistency in risk status verification after nursing interventions, which make it difficult to meet the actual needs of site risk assessment, intervention action coordination, and verification result tracking in continuous nursing scenarios, this application discloses a pressure ulcer prevention method and system based on multimodal perception. By forming a regionalized risk analysis and verification feedback processing method oriented towards nursing windows, it can improve the continuity of risk expression of pressure sites, enhance the stability of the correspondence between nursing actions and risk sources, and improve the traceability and cross-nursing window consistency of verification results after nursing interventions.

[0037] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0038] like Figure 1 As shown in the figure, this embodiment provides a pressure ulcer prevention method based on multimodal perception, including: Step S1: Obtain the bed surface multimodal perception data and nursing risk profile of long-term bedridden patients within the current nursing window, perform stable nursing window filtering and body surface area mapping on the bed surface multimodal perception data to obtain regional nursing window data; Step S2: Construct regional state quantities for each body surface region based on the regional nursing window data, and construct a multimodal pressure ulcer risk status map based on the regional state quantities. The edges of the multimodal pressure ulcer risk status map are determined by the decrease in load of the target body surface region, the increase in load of adjacent body surface regions, and the spatial proximity of the pressure centers of the target body surface region and adjacent body surface regions. Step S3: Based on the multimodal pressure ulcer risk status map, nursing risk records, and historical risk status of the previous nursing window, identify the site-level risk status, risk source, and risk transfer status of adjacent areas for each body surface region; Step S4: Generate a site-level nursing action package based on the site-level risk status, risk source, and risk transfer status of adjacent areas, and configure a verification window for the site-level nursing action package; Step S5: Reacquire the multimodal perception data of the target body surface area and adjacent body surface areas in the verification window. Based on the target decompression state and adjacent pressure state, form the verification results, review priority and structured nursing record of the site-level nursing action package. Update the multimodal pressure ulcer risk status map, review priority and structured nursing record of the next nursing window based on the verification results.

[0039] This embodiment constrains the nursing window and assigns body surface regions to the multi-source states of the bed surface, ensuring that the pressure state has stable time and location boundaries before entering the analysis process. This reduces the impact of short-term disturbances such as getting out of bed, turning over, and nursing care movements on region judgment and provides a unified standard for risk comparison in continuous nursing scenarios. Utilizing regionalized state expressions to represent the risk status of each body surface region helps to transform scattered bed surface information into risk expressions that can be transmitted along regional relationships, ensuring that changes in pressure, local environment, and regional correlations remain connected within the same analytical framework. Combining individual risk data and historical risk continuity relationships to conduct location-level judgments allows the current state judgment to take into account the patient's own tolerance differences and the residual effects of the previous nursing window, and enhances the understanding of risk sources and adjacent areas. Verifiability of regional load transfer direction; forming nursing actions and verification windows around site-level risk results can transform risk judgment into nursing treatments with site-specific, source-specific, and verification timing, promoting a stable connection between nursing prompts and on-site operations; verifying the status of the target area and adjacent areas within the verification window and feeding the verification results back to the next nursing window helps to transform single nursing actions into a basis for continuous status updates, ensuring consistency between verification priority and structured nursing records in cross-window nursing processes; overall, the closed-loop coordination between continuous nursing windows, regional risk expression, site-level action generation, and verification feedback can improve the stability of status judgment, the adaptability of nursing interventions, and the reliability of verification and tracking in the prevention of pressure ulcers in long-term bedridden patients.

[0040] The above steps will be specifically described below based on the embodiments of this application.

[0041] In step S1, the core task is to obtain the effective bed surface status of long-term bedridden patients from the current nursing window and exclude short-term disturbances caused by getting out of bed, nursing care, and sensor abnormalities, so that subsequent analysis can be based on stable nursing segments and clear body surface area boundaries as input.

[0042] In this embodiment, after a long-term bedridden patient is in a bedridden care state, multimodal sensing data of the bed surface and a nursing risk profile can be continuously acquired within the current nursing window. The multimodal sensing data of the bed surface comes from a flexible multimodal sensing pad laid on the bed surface and bed posture acquisition data, mainly including pressure distribution, shear slip, local temperature, local humidity, skin contact resistance, contact stability, headboard elevation angle, overall body position, and frequency of micro-movements. The nursing risk profile is formed from nursing assessment records and includes the patient's weight, height, mobility, nutritional status, edema, urinary and fecal incontinence, use of vasoactive drugs, diabetes or circulatory disorders, sedation status, history of pressure ulcers, and current dressing usage. The current nursing window can be defined according to the continuous pressure time under the same nursing posture; for example, one nursing window can capture... - The sampling segments are composed of consecutive sampling points. The above data are aligned with a unified clock in time and mapped to the sensing grid in spatial terms using bed coordinates, providing the original pressure profile, contact area edge, and pressure center location for subsequent contact stability determination.

[0043] Based on this, stable nursing window filtering and body surface region mapping are performed on the multimodal sensing data of the bed surface. This allows the current nursing window to be divided into multiple sampling segments, and usable data can be filtered at the segment level. Furthermore, the length of the sampling segments can be matched to the time scale of nursing action recognition and changes in bed surface contact. In some embodiments of this application, contact stability can be determined for the sampling segments within the current nursing window, forming a contact stability coefficient based on four aspects: contact area continuity, pressure center drift, effective sensor point ratio, and skin contact impedance continuity. Contact area continuity is used to exclude the overall pressure drop caused by the body leaving the bed surface; pressure center drift is used to distinguish between turning over and natural micro-movements; the effective sensor point ratio is used to identify sensor pad wrinkles or local failures; and skin contact impedance continuity is used to determine whether true contact is maintained between the skin and the bed surface. For example, the contact stability coefficient can be: , in, Indicates the first The first nursing window The contact stability coefficient of the sampling segment corresponding to the individual surface region; , and They represent the first The first nursing window Individual table region in the first The pressure center location vector, high pressure connection area, and skin contact impedance characteristics corresponding to each sampling time; , and These represent the allowable dimensions for pressure center drift, contact area variation, and skin contact resistance variation, respectively. This indicates the proportion of missing effective sensing points. The closer this contact stability coefficient is to 1, the closer the pressure center, contact area, skin contact impedance, and effective sensing point status of the sampled segment are to a stable bedridden contact state. For example, the allowable scale for pressure center drift can be set according to the bed grid span corresponding to the patient's shoulder width, the allowable scale for contact area change can be set according to the percentage of stable supine contact area, and the allowable scale for skin contact impedance change can be set according to the impedance fluctuation range of the same body surface area under stable contact state.

[0044] During this process, sampling segments that meet the conditions for leaving the bed, nursing movement, or abnormal sensor quality are marked as disturbed segments. The contact stability coefficient can be jointly judged with the total pressure, effective contact area, skin contact impedance mutation amplitude, and sensor quality: the condition for leaving the bed corresponds to an effective contact lower limit where the contact area is below the noise level of an empty bed and the total pressure is close to the empty bed baseline; the condition for nursing movement corresponds to a pressure center crossing multiple body surface area boundaries in a short period of time, accompanied by a sudden change in the high-pressure connectivity area; the condition for abnormal sensor quality corresponds to a decrease in the proportion of effective sensor points, a sudden change in local contact impedance, or a lack of continuous data. Furthermore, disturbed segments can be removed from the current nursing window, removing only the disturbed segment itself and the transitional sampling points adjacent to the disturbed segment, avoiding abnormal values ​​caused by nurses briefly tidying up the sheets or adjusting the tubing from being included in the risk calculation. Finally, stable nursing segments are obtained, which still retain the original sampling time, bed surface coordinates, pressure profile, temperature and humidity, and impedance data, so as to maintain temporal continuity when assigning body surface areas later.

[0045] In some embodiments of this application, an initial body surface region mapping can be formed based on the pressure profile, contact area edge, and pressure center location in a stable nursing segment. During this process, the pressure peak location in the patient's supine or confirmed position can be used as a reference point for bony prominences, and body surface region boundaries can be established by combining height, weight, and bed surface coordinates. The pressure profile is used to locate high-pressure distribution areas such as the sacrum, coccyx, scapula, greater trochanter, and heel; the contact area edge is used to identify the contact boundaries of the trunk and lower limbs; and the pressure center location is used to determine the orientation of the body and the left / right region affiliation. The initial body surface region mapping can be saved as a relationship between region identifier, bed surface grid range, reference pressure center vector, reference contact area, and corresponding nursing risk profile fields.

[0046] Furthermore, by detecting the migration status of the pressure center and the change status of the contact area within stable nursing segments, the drift direction, drift distance, and percentage change of the high-pressure connectivity area of ​​the pressure center can be continuously calculated for each stable nursing segment. When the pressure center migration status meets the remapping condition, the initial body surface region mapping is updated. The remapping condition can be triggered by the body surface region mapping offset exceeding a preset region offset threshold and continuously covering multiple stable sampling segments. When the contact area change status meets the remapping condition, the initial body surface region mapping is updated. The remapping condition can be triggered by a continuous change in the high-pressure connectivity area relative to the reference contact area accompanied by a stable change in body posture. During the update, the original mapping is retained as a reference for the previous nursing window, and the new pressure center and new contact area are written into the region boundary of the current nursing window, so that the bed coordinates of the sacral-coccygeal region, greater trochanter region, and heel region can be realigned after turning over. For example, the body surface region mapping offset used for remapping... It can be written as: , in, The median vector representing the location of the pressure center within a stable nursing segment; This represents the reference pressure center vector stored in the current body surface region mapping; The median area represents the high-voltage connectivity area within a stable nursing segment; This represents the reference contact area currently stored in the body surface region mapping. This offset reflects both the shift of the pressure center and changes in the contact area, and is subsequently used to determine whether the initial body surface region mapping needs to be updated.

[0047] Furthermore, stable care segments are assigned to regions based on either the updated body surface region mapping or the unupdated initial body surface region mapping. The former can be performed on stable care segments where turning or changes in contact boundaries have occurred, while the latter can be performed on stable care segments where neither the pressure center migration state nor the contact area change state exceeds the remapping condition. During the assignment process, the pressure, shear slip, local temperature, local humidity, and skin contact impedance of each sensing grid can be written into the corresponding body surface region, and the sampling time, care window number, and segment stability marker are retained. The region assignment results form regional care window data, which includes body surface region identifiers, stable sampling sequences, regional pressure sequences, shear slip sequences, microenvironment sequences, contact reliability sequences, body position identifiers, and patient individual fields corresponding to the care risk profile, for step S2 to construct regional state quantities.

[0048] Thus far, step S1, through perturbation removal and body surface region mapping, forms regional nursing window data with stable nursing segments as time boundaries and body surface regions as location boundaries, providing reliable input for subsequent regional state quantity calculations.

[0049] In step S2, the core task is to convert the regional care window data into comparable load states of the body surface area, and to form a multimodal pressure ulcer risk status map by utilizing the change relationship between the current care window and the previous care window.

[0050] In this embodiment of the application, the regional state quantity of each body surface region is constructed based on the regional nursing window data. This can be done on the basis of the stable sampling sequence formed in step S1. The regional state quantity of each body surface region is constructed around three types of pressure ulcer-related factors: continuous pressure, tangential slippage, and local microenvironment. At the same time, the contact reliability state is retained to avoid sensor abnormalities or short-term nursing disturbances being misjudged as real risks.

[0051] Specifically, pressure dose, shear dose, microenvironmental load, and contact reliability status can be generated for each body surface area based on regional nursing window data. Pressure dose reflects the cumulative pressure exceeding the individual's baseline; shear dose reflects the tangential load generated by head-up elevation and relative movement of the contact center; microenvironmental load reflects decreased skin tolerance caused by local moisture, increased temperature, and abnormal skin contact impedance; contact reliability status is formed based on contact stability, sensing quality, sampling continuity, and local impedance reliability, and participates in the gating of the three branches: pressure, shear, and microenvironment. For example, the first... The first nursing window Pressure dose corresponding to individual surface area It can be written as: , in, This represents the set of stable sampling times for this body surface region; Indicates the number of stable sampling moments; This indicates the regional pressure at the corresponding sampling time. This represents the individual pressure baseline formed by combining the patient's weight and body surface markings for this body surface area; This represents the pressure reliability gating coefficient obtained from the pressure reliability state mapping; This represents the sampling time interval between adjacent sampling moments. Pressure dose increases when pressure exceeds the individual pressure baseline and high-pressure connectivity persists, and provides pressure gating in subsequent gating merging. Meanwhile, shear dose... Possible methods: , in, This represents the contact center displacement vector at the corresponding sampling time. This indicates the headboard elevation angle at the corresponding sampling time. Indicates the stability coefficient in the slip direction; This represents the shear confidence gating coefficient obtained from the shear confidence state mapping. This calculation ensures a higher shear dose to the body surface area undergoing continuous unidirectional sliding under head-up conditions, while transient, directionless nursing movements are inhibited due to the reduced sliding direction stability coefficient. Furthermore, microenvironmental load... It can be written as: , in, Indicates local humidity; This represents the individual's baseline humidity level for that area of ​​the body surface; This indicates the amount of local temperature rise relative to the starting point of the nursing window; This represents the microenvironment reliability gating coefficient obtained from the microenvironment reliability state mapping. When local humidity is consistently higher than the individual humidity baseline and temperature rises significantly, the microenvironment load enters the subsequent regional load state and affects the action categories of skin examination and microenvironment screening.

[0052] In some embodiments of this application, pressure dose, shear dose, and microenvironmental load can be gated based on contact reliability states. Specifically, contact stability and sensing quality can be jointly mapped to a pressure reliability state, using the contact stability coefficient, the proportion of effective sensing points, and the empty bed noise offset as pressure reliability inputs; slip direction stability and bed head posture stability can be jointly mapped to a shear reliability state, using the consistency of the contact center displacement direction and the continuous stability of the bed head elevation angle as shear reliability inputs; and skin contact impedance changes and humidity recovery trends can be jointly mapped to a microenvironment reliability state, distinguishing between actual local dampness, dressing soaking, and short-term poor contact. These three reliability states control the intensity of the load state of the corresponding risk sources entering the area, preventing low-quality data from being directly used as a high-risk trigger.

[0053] It should be noted that the skin contact impedance continuity in step S1 is used to determine whether the sampled segment maintains a stable contact state, and the skin contact impedance change in step S2 is used to determine whether the microenvironment load has a real skin contact background. The two correspond to segment screening and microenvironment reliable gating, respectively, and do not change the consistency of the source of the same skin contact impedance data in the preceding and following processes.

[0054] Furthermore, a pressure-gated value is obtained based on the pressure dose and pressure confidence state. This pressure-gated value can be formed by the combined effect of the pressure dose and the pressure confidence gating coefficient at a stable sampling time. A shear-gated value is obtained based on the shear dose and shear confidence state, retaining the tangential risk caused by continuous slippage and head-up elevation. A microenvironment-gated value is obtained based on the microenvironment load and microenvironment confidence state, retaining the persistent skin tolerance reduction signal caused by localized moisture and temperature increases. Based on this, load merging is performed on the pressure-gated value, shear-gated value, and microenvironment-gated value within the same body surface region. Nonlinear compression can be applied to the clinical sensitivity of this body surface region based on different risk sources, allowing for comparable regional loads from long-term pressure accumulation, stable shear slippage, and persistent microenvironmental abnormalities. For example, the correspondence between regional load states can be expressed as: ,in, Indicates the regional load status; , and These represent the pressure dose normalization scale, shear dose normalization scale, and microenvironment load normalization scale for the body surface region, respectively. The normalization scale can be obtained from the quantile statistics of the historical stability window of the body surface region. The shear dose normalization scale for the sacral, coccygeal, and greater trochanter regions can be set to be more sensitive, and the pressure dose normalization scale for the heel region can be set to be more sensitive. Then, the load merging results are written into the regional load status, which simultaneously saves the proportion of pressure sources, shear sources, and microenvironment sources for step S3 to identify risk sources.

[0055] Based on this, gating calculations can avoid directly increasing risk outcomes for segments with abnormal contact, unstable slip direction, or unreliable humidity recovery trends, thus obtaining the regional load status. Each body surface region forms a numerical status comparable to the previous nursing window. Furthermore, a multimodal pressure ulcer risk status map is constructed based on the regional load status. Body surface regions can be used as graph nodes, and pressure transfer relationships between adjacent body surface regions can be used as graph edges. Each node stores the regional load status, pressure dose, shear dose, microenvironment load, contact reliability status, regional recovery status, and historical nursing feedback. This indicates that the main state of the graph node is driven by the regional load status, reducing misjudgments caused by a single pressure threshold. For example, the pressure transfer intensity of the graph edges can be: , in, Indicates the serial number of the body surface region Pointing to the serial number of the adjacent body surface region The intensity of pressure transfer along the edge of the graph; This indicates the regional load status of the target body surface area in the previous nursing window; This indicates the regional load status of adjacent body surface areas in the current nursing window; and These represent the representative pressure center positions formed by the target body surface region and the adjacent body surface region within the stable nursing segment of the current nursing window. The distance between them represents the spatial distance between the pressure centers of the target body surface region and the adjacent body surface region. The smaller the spatial distance between the pressure centers, the greater the pressure transfer intensity of the graph edge. Accordingly, when the load on the target body surface region decreases, the load on the adjacent body surface region increases, and the spatial proximity of the pressure centers of the two regions, the pressure transfer intensity of the graph edge increases. Further, based on the relationship between the regional load status changes between the current nursing window and the previous nursing window, and the proximity of the pressure centers of the target body surface region and the adjacent body surface region, the pressure transfer direction between different body surface regions is determined. This can be achieved by comparing the increase or decrease in the regional load status between the target body surface region and the adjacent body surface region. Finally, the graph edges in the multimodal pressure ulcer risk status map are determined according to the pressure transfer direction. The graph edges store the direction, intensity, generating nursing window, and corresponding body posture identifier for step S3 to identify the risk transfer status of adjacent regions.

[0056] In some embodiments of this application, the spatial proximity of the pressure centers of the target body surface region and adjacent body surface regions is determined as follows: Representative pressure center positions of the target body surface region and adjacent body surface regions are formed based on their pressure center positions at each sampling time within a stable nursing segment of the current nursing window; the spatial distance between the representative pressure center positions of the target body surface region and adjacent body surface regions is calculated to obtain the pressure center spatial distance; when the pressure center spatial distance is not greater than a preset proximity judgment distance threshold, the target body surface region and adjacent body surface regions are determined to be in a pressure center spatial proximity state; when the pressure center spatial distance is greater than the preset proximity judgment distance threshold, the target body surface region and adjacent body surface regions are determined not to be in a pressure center spatial proximity state, and no pressure transfer graph edge from the target body surface region to the adjacent body surface region is established based on the adjacent body surface region. For example, the representative pressure center position can be the median value of the pressure center positions at each sampling time within the stable nursing segment; the proximity judgment distance threshold can be set based on the anatomical spacing of adjacent body surface regions on the bed surface grid, patient body shape parameters, and body surface region mapping scale. It should be noted that the pressure center position at each sampling time is taken from the stable care segment retained after contact stability determination in step S1. Contact stability determination is used to determine the data source for the spatial proximity determination of the pressure center. The spatial proximity of the pressure center is determined by comparing the spatial distance between the pressure center positions with a preset proximity determination distance threshold. Only between regions in the spatial proximity state of the pressure center, a pressure transfer graph edge is established by combining the decrease in load of the target body surface area and the increase in load of adjacent body surface areas. The direction of the graph edge is from the body surface area with decreasing regional load to the body surface area with increasing regional load.

[0057] At this point, step S2 transforms the regional care window data into regional load status and forms a multimodal pressure ulcer risk status map with graph edge pressure transfer direction, providing a graph structure basis for site-level risk identification.

[0058] In step S3, the core task is to combine the multimodal pressure ulcer risk status map, nursing risk records, and historical risk status to identify the current risk level, risk source, and pressure transfer from adjacent areas for each body surface region.

[0059] In this embodiment, the site-level risk status of each body surface region can be identified based on the multimodal pressure ulcer risk status map, nursing risk record, and historical risk status of the previous nursing window. At this point, the regional load status in the graph node is used as the current local risk input, the nursing risk record is used as the individual tolerance correction input, and the historical risk status of the previous nursing window is used as the continuous pressure memory input. The site-level risk status reflects not only the current pressure level but also whether there is continuous pressure, shear slip, long-term microenvironmental abnormalities, and decreased individual patient tolerance at that site. Simultaneously, identifying the risk sources and risk transfer status of adjacent regions for each body surface region allows for the combined reading of the source proportion stored in the regional load status, the pressure transfer direction at the graph edge, and the historical risk status. This enables the risk sources to be categorized into continuous pressure sources, shear slip sources, microenvironmental sources, sources of decreased individual tolerance, continuous pressure memory sources, and pressure transfer sources.

[0060] Specifically, in some embodiments of this application, the regional load status and edge pressure transfer direction of each body surface region are determined based on a multimodal pressure ulcer risk status map. Node and edge fields can be read at the end of each nursing window. Node fields include regional load status, pressure dose, shear dose, microenvironmental load, and contact reliability status. Edge fields include pressure transfer direction, edge pressure transfer intensity, and corresponding body posture identifier. Further, based on the nursing risk profile, tolerance reduction and risk amplification states are formed for each body surface region. Patient activity level, nutritional status, edema, urinary and fecal incontinence, diabetes or circulatory disorders, and sedation status can be mapped to the tolerance dimension of the body surface region. Different sensitivity coefficients can be used for the sacrum, greater trochanter, and heel. Simultaneously, an individual tolerance correction state is formed based on the tolerance reduction and risk amplification states, serving as the profile input for the site-level risk status model in risk identification. For example, the individual tolerance correction state... It can be written as: ,in, The file code indicating a decrease in activity level; The file code indicating nutritional insufficiency; Indicates the file code related to incontinence and dampness; , and These represent the scale parameters for the three types of archive coding. The lower the individual tolerance correction state, the more sensitive that area of ​​the body surface is to the risk of pressure ulcers under the same regional load.

[0061] Furthermore, based on the historical risk status of the previous nursing window, a continuous pressure memory calculation is performed on the regional load status to obtain a persistent risk status, used to distinguish between short-term high pressure and continuous pressure risk. Specifically, the regional risk that was not alleviated in the previous nursing window can be attenuated and inherited, and updated with the current regional load status. If the nursing action verification result shows that the target area has been effectively relieved, the contribution of the historical risk status to the current nursing window is reduced; if the previous nursing window did not alleviate the pressure or pressure shifts occurred in adjacent areas, the historical risk status continues to participate in the current risk assessment. For example, the continuous pressure memory calculation can be performed as follows: ,in, Indicates a state of ongoing risk; This indicates the valid verification result of the corresponding body surface area in the previous nursing window; This indicates the historical risk memory coefficient of the body surface area; This indicates the site-level risk status of the previous nursing window. The higher the valid verification result, the weaker the inheritance of the risk from the previous nursing window to the current persistent risk status; the historical risk memory coefficient can be set according to the recovery speed of the body surface area, and the values ​​for the sacral, coccygeal, and heel areas can be relatively higher.

[0062] In some embodiments of this application, a site-level risk status model with graph constraints can be embedded in step S3. This model serves site-level pressure ulcer risk assessment, and its input mainly consists of the multimodal pressure ulcer risk status graph of the current nursing window, the patient's nursing risk profile code, and the historical risk status of the previous nursing window. The model as a whole includes a region node encoding layer, an adjacent region propagation layer, a continuous pressure memory layer, a feedback gating layer, a risk source readout layer, and an action pre-readout layer. The regional node encoding layer organizes the pressure dose, shear dose, microenvironmental load, contact confidence status, and regional load status of each body surface region into node feature vectors. For example, the node feature vectors can be set to 32 dimensions, where the pressure-related dimension, shear-related dimension, microenvironment-related dimension, and contact confidence-related dimension each occupy a fixed subspace. The adjacent region propagation layer reads the pressure transfer direction and intensity at the edges of the graph, expressing pressure transfer after actions such as turning over, adding padding, heel suspension, or adjustment of the headboard angle through message passing between adjacent body surface regions. The continuous pressure memory layer receives the persistent risk status, allowing unrelieved risks from the previous nursing window to enter the current risk assessment. The feedback gating layer receives the valid verification results, relieved inheritance status, and unrelieved inheritance status from the previous nursing window, reducing the intensity of repeated prompts for confirmed effective nursing actions and increasing the review priority for body surface regions that are unrelieved or show adjacent region risk transfer. For example, the attention relationship of the adjacent region propagation layer can be written as: , in, This indicates the intensity of attention propagated from adjacent body surface regions to the target body surface region; The node feature vector representing the target body surface region; The node feature vectors representing adjacent body surface regions; Represents the set of adjacent regions on the surface of the target body; Indicates the number of adjacent regions; Note the scaling parameters; This represents the transformation matrix of the propagation layer between adjacent regions. This attention intensity allows surface regions with higher edge pressure transfer intensity and more significant load changes in adjacent regions to contribute more to risk propagation. For example, location-level risk status... Possible methods: , in, This represents a risk compression function; This represents the transpose of the risk readout parameter vector. The model output of the site-level risk status simultaneously incorporates the node features of the target body surface region, the propagation results of adjacent regions, the individual tolerance correction status, and the persistent risk status, and then proceeds to the identification of site-level risk status, risk source, and risk transfer status in adjacent regions.

[0063] It should be noted that the model training samples are constructed on a per-nursing-window basis. Each training sample includes a multimodal pressure ulcer risk status map, nursing risk profile, nursing action record, validation results, and shift skin examination record. The training set, validation set, and test set can be hierarchically divided according to the patient dimension, with an exemplary ratio of 7:2:1 to avoid adjacent nursing windows of the same patient appearing in both the training and test sets. Sample augmentation does not change the medical semantics and can employ window pruning, low-amplitude sensor noise perturbation, simulation of unstable fragment removal, and repeated sampling of a few sites. Weak labels are generated jointly from skin examination records, continuous high-burden status in high-area regions, and validation failure records; label credibility is increased when the three types of information are consistent, and label credibility is decreased and the sample enters the disputed sample branch when information conflicts. During training, a batch size of 32 nursing window samples can be used, with an initial learning rate of 0.001 and 100-200 training rounds; training stops when the site-level risk ranking in the validation set is stable, the action category matching is stable, and the misjudgment rate of risk transfer between adjacent areas is lower than the preset requirement. During the online usage phase, the main parameters of the regional node coding layer, the adjacent regional propagation layer, and the risk readout layer remain frozen. Individual tolerance correction parameters and feedback gating parameters are only updated within a limited scope when stable effective or stable ineffective validation results appear in multiple consecutive care windows. For example, site-level risk identification loss... It can be written as: , in, Indicates the number of training samples; Indicates the first The credibility of the labels corresponding to each training sample; Indicates the first training sample Weak labels in individual table regions; This represents the predicted risk state at the location level corresponding to the training sample. This loss makes high-confidence weak labels contribute more to the risk boundary, thus suppressing the impact of disputed samples on the location-level risk state. For example, the adjacent region transfer consistency loss... It can be written as: , in, Represents the set of edges between adjacent regions in the training samples; Indicates the number of edges in adjacent regions; This represents the predicted risk status of adjacent body surface regions in the training samples; This indicates the location-level risk status of the target body surface area in the previous nursing window; This represents the graph edge pressure transfer intensity from the target surface region to adjacent surface regions in the training samples. It reflects the continuity of risk transfer to adjacent surface regions after the loss constraint is reduced.

[0064] For example, action category matching loss It can be written as: ,in, Indicates the number of action categories; Indicates the first training sample Action tags for each action category; This represents the predicted value for the corresponding action category. The three losses are minimized together during training, and the model output after training is fed into the subsequent generation of site-specific nursing action packages.

[0065] Based on this, according to regional load status, map edge pressure transfer direction, individual tolerance correction status, and persistent risk status, location-level risk status, risk sources, and risk transfer status in adjacent areas can be identified, thus defining location-level risk status. Risks are categorized into four types: low risk, risk requiring attention, risk requiring intervention, and risk requiring review. Risk sources are determined based on the proportions of pressure, shear, and microenvironmental factors in the regional load status, as well as the contributions of individual tolerance correction status and persistent risk status. The risk transfer status between adjacent regions is determined based on the pressure transfer direction along the graph and changes in pressure in adjacent regions. When the pressure transfer direction along the graph points to an adjacent body surface region, the changes in the regional load status of the target body surface region and the adjacent body surface regions are compared. When the regional load status of the target body surface region decreases while the regional load status of the adjacent body surface regions increases, a transfer trend can be considered, resulting in a consistent transfer direction status. This status, along with the persistent risk status, is used to determine whether a review of adjacent body surface regions is necessary. For example, the adjacent transfer review intensity can be written as: , in, This indicates the intensity of adjacent transfer verification from the target body surface area to the adjacent body surface area; this intensity simultaneously requires that the target body surface area is decompressed, the adjacent body surface area is under pressure, and the adjacent body surface area is in a state of continuous risk. Further, when the consistent transfer direction and continuous risk state meet the transfer verification conditions, the adjacent body surface area is written into the transfer verification object. The transfer verification condition can be that the adjacent transfer verification intensity exceeds a preset transfer verification threshold and continuously covers the shortest sampling segment required for verification. Based on this, the adjacent body surface area corresponding to the transfer verification object is written into the adjacent area risk transfer status. The adjacent area risk transfer status saves the adjacent body surface area identifier, transfer source area, edge pressure transfer intensity, and verification reason; then, the risk source in the adjacent area risk transfer status is updated to the pressure transfer source, so that when generating the site-level nursing action package in step S4, the adjacent body surface area can be included in the verification area.

[0066] At this point, step S3 combines graph nodes, graph edges, individual files, and historical memories to form location-level risk status, risk sources, and risk transfer status in adjacent areas, providing a basis for determining nursing actions.

[0067] In step S4, the core task is to transform the site-level risk identification results into an executable and verifiable site-level nursing action package, and to configure verification windows for the target body surface area and the verification area.

[0068] In this embodiment, location-level nursing action packages are generated based on location-level risk status, risk source, and risk transfer status of adjacent areas. Each location-level nursing action package is built around an intervention area on the body surface, and the risk source is converted into a nursing action category. A location-level nursing action package may include an execution area, execution content, review area, verification object, verification content, review time, and secondary adjustment conditions. The execution area specifies the body surface area requiring intervention; the execution content specifies postural adjustment, decompression, skin examination, or microenvironment screening; and the review area specifies adjacent body surface areas requiring simultaneous observation. Subsequently, a verification window is configured for the location-level nursing action package. The start time and duration of the verification window can be determined based on the location-level risk status, action category, and risk transfer status of adjacent areas, ensuring sufficient time to observe decompression changes in the target body surface area and pressure changes in adjacent body surface areas after the nursing action is executed.

[0069] Specifically, determining the intervention area on the body surface based on the site-specific risk status can reduce the site-specific risk status. Body surface areas that meet the risk requirements for intervention or verification are included in the candidate intervention set, and the execution order is determined according to risk status, clarity of risk source, and persistent risk status. Simultaneously, the intervention body surface areas are written into the execution area of ​​the site-level nursing action package. The execution area field stores the body surface area identifier, bed surface coordinate range, risk status of the previous nursing window, current area load status, and risk source identifier. Subsequent verification windows relocate the target body surface area using the execution area field. For example, nursing action priority... It can be written as: The numerator (the number of parameters) indicates that site-specific risk status, shear dose, and microenvironmental load collectively influence the sequence of actions, while the denominator (the number of parameters) indicates that the intensity of repeated cues for surface areas that have been validated in the previous nursing window has decreased. When there is an increased shear dose in the sacral, coccygeal, and greater trochanter regions, the priority of nursing actions is more likely to increase; similarly, when there is a sustained pressure dose and increased microenvironmental load in the heel region, the priority of nursing actions will also increase accordingly.

[0070] Furthermore, by classifying actions based on the source of risk among postural adjustment, decompression, skin examination, and microenvironmental investigation, the risk source and regional status can be mapped to different action categories. Specifically, when a persistent pressure source is dominant, the action category tends to be postural adjustment or decompression; when a shear slip source is dominant and the head of the bed is consistently elevated, the action category tends to be postural adjustment with associated head of the bed angle review; when a microenvironmental source is dominant, the action category tends to be skin examination and microenvironmental investigation; when a pressure transfer source is dominant, the action category should retain the decompression verification of the target body surface area and add the pressure verification of adjacent body surface areas. Based on this, the action category is written into the execution content of the site-level nursing action package. The execution content field stores the action category, target postural suggestion, decompression equipment placement requirements, skin examination focus, moisture source investigation content, and secondary adjustment conditions. Subsequent step S5 uses this to determine whether the target decompression status and adjacent pressure status meet expectations.

[0071] In some embodiments of this application, when the risk transfer status of adjacent areas does not include a transfer review object, the adjacent body surface area corresponding to the intervention area is written into the review area of ​​the site-level nursing action package. The review area can be selected as the area adjacent to the intervention area in the body surface area status map and most likely to be under pressure after the nursing action, such as the ipsilateral or contralateral greater trochanter area after sacral and coccygeal decompression. When the risk transfer status of adjacent areas includes a transfer review object, the transfer review object is written into the review area of ​​the site-level nursing action package. The review area directly adopts the adjacent body surface area identified in step S3 and saves the pressure transfer source field.

[0072] Based on this, the verification objects and verification contents in the verification window are formed according to the execution area, execution content, and review area. The verification objects include the target body surface area corresponding to the execution area and the adjacent body surface area corresponding to the review area. The verification contents include the decrease in pressure dose, decrease in shear dose, reduction in high-pressure connectivity area, and recovery of microenvironment in the target body surface area, as well as the regional load status increment and risk transfer status change in adjacent body surface areas.

[0073] For example, the duration of the verification window Possible methods: ,in, Indicates the shortest verification time; This indicates the longest verification time. The higher the location-level risk status and the intensity of adjacent transfer verification, the closer the verification window duration is to the longest verification time. For example, a shorter verification window can be used for low-risk verification, while the verification window should be extended when there is a source of pressure transfer to observe changes in pressure in adjacent areas.

[0074] At this point, step S4 converts the site-level risk identification results into site-level nursing action packages and establishes a correspondence between the execution area, execution content, review area, and verification window.

[0075] In step S5, the core task is to verify the target decompression and adjacent pressure changes after the nursing action is performed in the verification window, and write the verification results back to the next nursing window.

[0076] In this embodiment, multimodal sensing data of the target body surface area and adjacent body surface areas are reacquired within the verification window. The acquisition criteria remain consistent with step S1, but the acquisition range is limited to the execution area and review area of ​​the site-level nursing action package. The target body surface area is used to observe whether the nursing actions truly reduce local pressure, shear, and microenvironmental load; adjacent body surface areas are used to observe whether turning over, adding padding, or suspending the heel creates new pressure risks. Based on this, the verification results, review priorities, and structured nursing records of the site-level nursing action package are formed. During this process, the execution area, execution content, review area, target decompression state, and adjacent pressure state need to be read simultaneously. The verification results can be marked as valid, require review, or invalid; the review priority is updated based on the adjacent pressure state and the unrelieved inherited state; the structured nursing record stores the nursing window number, execution area, execution content, review area, verification result, and the next nursing window read field, with the body surface area as the basic unit.

[0077] Specifically, the execution area can be considered the target body surface area. Within the verification window, the pressure dose, shear dose, microenvironmental load, and regional load status of the target body surface area are recalculated and compared with the corresponding data before the nursing action. The verification area is considered the adjacent body surface area, and within the verification window, the regional load status increment and edge pressure transfer intensity of the adjacent body surface area are calculated simultaneously. This comparison method can simultaneously retain the decompression status of the target body surface area and the pressure status of adjacent body surface areas, reducing the risk of delayed identification of new pressure status. Based on this, within the verification window, the target decompression status is formed according to the pressure dose reduction ratio, shear dose reduction ratio, and microenvironmental load recovery ratio corresponding to the execution area. For example, the target decompression status... It can be written as: ,in, and Indicates the pressure dose to the target body surface area before the nursing action is performed and within the verification window; and These represent the shearing dose before the nursing action is performed and within the verification window, respectively. and These represent the microenvironmental load before the execution of nursing actions and within the verification window, respectively. , and These represent stability constants to prevent the denominator from becoming too small. The higher the target decompression state, the more significant the overall load reduction in the execution region within the verification window.

[0078] Furthermore, within the verification window, adjacent pressure states are formed based on the regional load state increment corresponding to the verification area, the edge pressure transfer intensity, and the adjacent transfer verification intensity. This allows for comparison of the regional load state changes before and after the execution of nursing actions in the verification area, and the adjacent regional risk transfer states identified in step S3 are overlaid. For example, adjacent pressure states can be defined as follows: ,in, This indicates the adjacent pressure status of the execution area pointing to the review area; This indicates the area load status of the area being reviewed after a nursing action has been performed; This indicates the regional load status of the area to be reviewed before the nursing action is performed. The adjacent pressure status combines the load increment of the review area, the pressure transfer intensity at the edge of the graph, and the adjacent transfer review intensity into the same judgment criterion, which is used to verify the results and review priority; if the pressure dose, shear dose, or microenvironment load of the review area increases significantly, and the pressure transfer direction at the edge of the graph points to the review area, then the adjacent pressure status increases.

[0079] Based on this, the verification results and review priorities of site-level nursing action packages are determined according to the execution content, target decompression state, and adjacent pressure ulcer states. This process can be combined with the expected changes corresponding to the execution content for judgment. Position adjustment and decompression treatments require the target decompression state to meet effective decompression conditions and adjacent pressure ulcer states to be below the transfer review conditions; skin examination and microenvironment screening treatments also require the microenvironment load recovery ratio to reach the corresponding boundary. Further, the execution area, execution content, review area, target decompression state, adjacent pressure ulcer states, verification results, and review priorities are written into a structured nursing record. The structured nursing record saves the values ​​of each field, the judgment time, the nursing window number, and the read identifier. After writing, the next nursing window reads the verification results and review priorities from the structured nursing record to update the historical risk status and multimodal pressure ulcer risk status map.

[0080] Furthermore, the multimodal pressure ulcer risk status map, review priority, and structured nursing record of the next nursing window are updated based on the verification results. Specifically, the results of the current verification window can be converted into the initial state of the next nursing window. When the verification result is valid, the historical risk status of the target body surface area enters the mitigation inheritance phase; when the verification result is invalid, the target body surface area continues to retain the unmitigated risk; when adjacent pressure states meet the transfer review conditions, a transfer review map edge is added between the execution area and the review area. For example, the review priority of the next nursing window can be written as: , in, Indicates the review priority of the review area in the next nursing window; This indicates the review priority of the review area in the current nursing window. When the adjacent pressure status is higher than the target decompression status, the review priority of the next nursing window increases; when the target decompression status is sufficient and the adjacent pressure status is low, the review priority remains stable.

[0081] In some embodiments of this application, when the target decompression state meets the effective decompression condition and the adjacent pressure state does not meet the transfer review condition, the historical risk status of the execution area in the next nursing window is written into the relief inheritance status. The effective decompression condition can be triggered by the target decompression state reaching a preset effective decompression threshold and the corresponding indicator of the key risk source decreasing. The adjacent pressure state not meeting the transfer review condition indicates that there is no obvious pressure transfer in the review area. At the same time, in the multimodal pressure ulcer risk status map, the risk source corresponding to the execution area is updated to the relief source mark and the effective nursing action mark is written into the structured nursing record. The relief source mark saves the nursing action category, verification window and target decompression state, so as to reduce the intensity of repeated prompts in the next nursing window. The effective nursing action mark is saved together with the execution content and execution time for easy reading in subsequent shifts.

[0082] It should be noted that when the target decompression state does not meet the effective decompression conditions, the historical risk status of the executed area in the next nursing window is written into the unrelieved inherited status. The unrelieved inherited status saves the reason why the target decompression state did not reach the preset effective decompression threshold, such as insufficient pressure dose reduction, still high shear dose, or unrecovered microenvironmental load. In this case, the risk source corresponding to the executed area is retained in the multimodal pressure ulcer risk status map. The retained risk source continues to participate in site-level risk status identification in the next nursing window and can trigger secondary adjustments. When adjacent pressure states meet the transfer review conditions, a transfer review edge pointing from the executed area to the review area is established in the multimodal pressure ulcer risk status map. The transfer review edge saves the executed area, review area, adjacent pressure states, edge pressure transfer intensity, and review triggering reason. In this case, the review priority and structured nursing record are updated based on the transfer review edge. The next nursing window will focus on reviewing the review area and retain the risk transfer mark of the adjacent area in the structured nursing record.

[0083] At this point, step S5 writes the target pressure relief result and adjacent pressure bearing result after the nursing action into the next nursing window, so that the multimodal pressure ulcer risk status map, review priority and structured nursing record form a closed loop update.

[0084] In summary, this method integrates multimodal perception data of bed surfaces from long-term bedridden patients with their nursing risk profiles into a single nursing window. By combining stable nursing window screening, body surface region mapping, and the construction of a multimodal pressure ulcer risk status map, it achieves continuous expression of pressure, shear, microenvironment, and historical risk status in each body surface region. Furthermore, it utilizes site-level nursing action packages and verification windows to generate post-intervention feedback updates. This approach improves the matching accuracy between multi-source perception status and site-level nursing interventions, enhances the pressure relief effect on target areas after nursing action execution, and reinforces the consistency of risk transfer verification between adjacent areas. It also reduces the impact of short-term disturbances and experience-based judgments on risk identification, thereby improving the continuity, traceability, and stability of nursing interventions in pressure ulcer prevention for long-term bedridden patients.

[0085] It should be noted that, although the embodiments in this application are based on... Figure 1 The steps are explained in turn, but this does not mean that the steps must be performed in a strict order. The step numbers are only used to distinguish different steps and do not constitute a restriction on the order of execution. The specific execution order of each step can be appropriately adjusted according to actual needs, functional requirements and the inherent logic in the actual application scenario.

[0086] In some embodiments of this application, a pressure ulcer prevention method based on multimodal perception is applied to long-term bedridden patients in intensive care settings. This patient has reduced mobility, a nursing risk profile showing a Braden score of 12, and a weight of [weight missing]. There is a risk of mild edema and urinary / fecal incontinence. The sacral and coccygeal regions, the right greater trochanter region, and both heel regions were designated as key surface areas. A flexible multimodal sensing pad collected data on pressure distribution, shear slip, local temperature, local humidity, skin contact impedance, and contact stability. The headboard elevation angle was recorded as 30 degrees. The current nursing window number was set to 12, the nursing window length was set to 20 minutes, and the sampling interval was set to 4 seconds. The complete implementation process may include the following steps: Step 1: Acquire multimodal sensing data and nursing risk profiles for the current nursing window, and form stable nursing segments. After the current nursing window is divided into multiple sampling segments, contact stability is determined for the corresponding sampling segments in the sacral and coccygeal regions. Taking one sampling segment as an example, the drift of the pressure center between adjacent sampling times is 1.2 bed grids, the allowable drift scale of the pressure center is 5.0 bed grids, the change in high-pressure connectivity area is 4 grid areas, the allowable change scale of contact area is 20 grid areas, the change in skin contact impedance characteristics is 0.03, the allowable change scale of skin contact impedance is 0.10, and the proportion of missing effective sensing points is 0.04. Substituting these values ​​into the contact stability coefficient yields 0.458, which is lower than the stable sampling segment threshold of 0.60. Therefore, this sampling segment is marked as a disturbance segment caused by nursing movement; adjacent transitional sampling points are simultaneously removed. After removal, the current care window still retains a 16-minute stable care segment, which preserves the pressure profile, contact area edge, pressure center location, temperature and humidity sequence, and skin contact resistance sequence.

[0087] Step 2 involves mapping the stable nursing segment to the body surface region and generating regional nursing window data. In the stable nursing segment, the sacrococcygeal region forms a sustained high-pressure peak, the right greater trochanter region forms a secondary high-pressure peak, and the heel region forms a small high-pressure point. Based on the pressure profile and contact area edges, the sacrococcygeal region is mapped as the current key target body surface region, and the right greater trochanter region is mapped as the adjacent body surface region. The offset of the pressure center median vector relative to the reference pressure center in the initial body surface region mapping does not exceed the remapping threshold, and the contact area change rate is 8%, which does not meet the remapping condition. Therefore, the initial body surface region mapping is used for this nursing window. In the regional nursing window data, the sacrococcygeal region stores the stable sampling sequence, regional pressure sequence, shear slip sequence, microenvironment sequence, and nursing risk file association fields, while the right greater trochanter region stores similar fields required for adjacent region verification.

[0088] Step 3: Calculate the regional state variables of the sacrococcygeal region and write them into the multimodal pressure ulcer risk status map. Taking three representative stable sampling times of the sacrococcygeal region as examples, the positive pressure differences relative to the individual pressure baseline are 16, 18, and 20, respectively; the high-pressure connectivity areas are 32, 34, and 35, respectively; the pressure confidence gating coefficients are 0.90, 0.92, and 0.94, respectively; and the sampling time interval is 4 seconds.

[0089] The pressure-dose calculation relationship can be written as: Substituting the above values, the pressure dose is 16×32×0.90×4+18×34×0.92×4+20×35×0.94×4=6727.36. Based on the same stable nursing segment, the shear slip sequence shows a downward trend in the same direction when the head of the bed is raised by 30 degrees, and the calculated shear dose is 0.9518; the local humidity is continuously higher than the individual humidity baseline, the local temperature rises by about 0.4-0.5 degrees, and the microenvironmental load is 102.37. The pressure dose normalization scale is taken as 9000, the shear dose normalization scale is taken as 2.5, and the microenvironmental load normalization scale is taken as 140. The correspondence between the regional load status can be expressed as: After substituting the pressure dose, shear dose, and microenvironmental load mentioned above, the regional load status of the sacrococcygeal region is 0.805. In the previous nursing window, the regional load status of the sacrococcygeal region was 0.86. In the current nursing window, the regional load status of the right greater trochanter region has increased from 0.36 to 0.42, indicating that the decompression trend in the sacrococcygeal region is not yet sufficient, and at the same time, an increase in pressure has appeared in the right greater trochanter region. The multimodal pressure ulcer risk status map uses the sacrococcygeal region and the right greater trochanter region as graph nodes, and retains the graph edge pressure transfer relationship between them for subsequent verification.

[0090] Furthermore, the spatial distance between the representative pressure center location of the sacrococcygeal region and the representative pressure center location of the right greater trochanter region is 6.0 bed surface grids, and the proximity judgment distance threshold is set to 8.0 bed surface grids. Since the spatial distance between the pressure centers is less than the proximity judgment distance threshold, the sacrococcygeal region and the right greater trochanter region are determined to be in a spatial proximity state of pressure centers. Combined with the decrease in the regional load state of the sacrococcygeal region from 0.86 to 0.805 and the increase in the regional load state of the right greater trochanter region from 0.36 to 0.42, the pressure transfer direction is determined to be from the sacrococcygeal region to the right greater trochanter region, and the pressure transfer map edge from the sacrococcygeal region to the right greater trochanter region is written into the multimodal pressure ulcer risk state map.

[0091] Step four involves calling the site-level risk status model constrained by the graph to identify the site-level risk status and risk sources in the sacral and coccygeal regions. This model employs a temporal risk structure oriented towards the body surface region graph, and its network morphology differs from ordinary autoencoders, noise-reducing autoencoders, stacked noise-reducing autoencoders, convolutional autoencoders, and variational autoencoders. The model input consists of graph node features of the current nursing window, graph edge pressure transfer intensity, nursing risk profile encoding, and historical risk status of the previous nursing window. The region node encoding layer receives 32-dimensional node features, where the pressure-related subspace stores pressure dose, baseline area, and pressure confidence gating results; the shear-related subspace stores contact center displacement, headboard elevation angle, and slip direction stability results; the microenvironment-related subspace stores local humidity, local temperature rise, and skin contact impedance changes results; and the contact confidence-related subspace stores contact stability and sensing quality results. The regional node encoding layer adopts a two-layer fully connected structure with an input dimension of 32, a hidden dimension of 64, and an output dimension of 32. The activation function is ReLU. The adjacent region propagation layer reads the regional node features of the right large trochanter region and the pressure transfer intensity of the graph edges, and forms adjacent propagation vectors through single-head graph attention. The continuous pressure memory layer synthesizes the site-level risk status of the previous nursing window and the regional load status of the current nursing window into a continuous risk status. The feedback gating layer reads the valid verification results of the previous nursing window. If the verification result is valid, it suppresses repeated prompts; if the verification result is invalid or a source of pressure transfer is found, it increases the review priority. The site-level risk status of the model can be: , Within the patient's current care window, the model inputs the sacrococcygeal region node features, the right greater trochanter region propagation vector, individual tolerance correction state, and persistent risk state into the risk readout layer. The trained risk readout layer outputs a risk compression function input value of 1.52, resulting in a site-level risk state of 0.821 for the sacrococcygeal region, falling within the intervention-required risk range. The risk source readout layer shows that persistent pressure sources contribute the most, with shear slip sources and microenvironment sources jointly amplifying the risk. The adjacent pressure increment corresponding to the right greater trochanter region has not yet reached the transfer verification threshold, but it has been added to the validation window as a candidate for verification.

[0092] Step 5: Train the model and use it online using specific nursing samples. In the offline training phase, nursing windows are used as the sample unit. Each training sample includes a multimodal pressure ulcer risk status map, nursing risk profile, nursing action record, validation results, and shift skin examination record. For example, 4800 nursing window samples are collected. Positive weak labels come from skin erythema records, continuous high regional load states, and nursing action validation failure records; negative weak labels come from windows with decreased regional load states, restored microenvironment, and no abnormalities in shift skin examinations. The training set, validation set, and test set are divided by patient dimension in a 7:2:1 ratio to avoid adjacent nursing windows of the same patient being included in both training and testing simultaneously. Data augmentation is applied only to the sensor data layer, including stable window pruning, low-amplitude pressure noise perturbation, slight local humidity perturbation, and unstable fragment removal simulation; the augmentation process does not change the nursing risk profile and skin examination record. The training batch size is 32 nursing window samples, the initial learning rate is 0.001, and the number of training rounds is 150. The training objective includes part-level risk identification loss, adjacent region transfer consistency loss, and action category matching loss. The part-level risk identification loss can be written as: , During training, the label confidence level for high-confidence weak labels was set to 0.90, while the label confidence level for disputed samples with sensor conflicts or nursing record conflicts dropped to 0.40. The model was also designed to account for whether the risk in the right greater trochanter region increased with the intensity of pressure transfer after decompression following adjacent region transfer consistency constraints. Furthermore, the model was designed to account for the output categories of postural adjustment, decompression, skin examination, and microenvironment screening processes based on action category matching loss constraints. During online use, the main parameters of the region node encoding layer, adjacent region propagation layer, and risk readout layer remained frozen. Individual tolerance correction parameters and feedback gating parameters were only updated within a limited scope when multiple consecutive nursing windows yielded stable valid or stable invalid validation results. This setting allowed the model to adapt to individual patient recovery speeds while preventing a single abnormal nursing record from altering the main risk boundary.

[0093] Step Six: Generate a site-specific nursing action package for the sacrococcygeal region and configure a verification window. Since the site-specific risk status of the sacrococcygeal region is 0.821, with sustained pressure as the primary source, and shear slip and microenvironmental sources contributing to amplification, the site-specific nursing action package includes the sacrococcygeal region in the execution area and the right greater trochanter region in the verification area. The execution includes adjusting the patient to a right-sided tilt position, lowering the head of the bed elevation angle, adding a pressure-reducing pad to the sacrococcygeal region, and checking the local skin color and source of moisture in the sacrococcygeal region.

[0094] The priority of nursing actions can be written as: Substituting the site-level risk status (0.821), shear dose (0.9518), microenvironmental load (102.37), shear dose normalization scale (2.5), microenvironmental load normalization scale (140), and the previous nursing window's valid verification result of 0, the nursing action priority is 1.962, ranking higher than the general attention prompt for the heel area. Subsequently, verification windows are configured based on the site-level risk status and the intensity of adjacent transfer review. The verification window duration can be written as: In this specific embodiment, the shortest verification time is 10 minutes, the longest verification time is 30 minutes, and the adjacent transfer review intensity is 0.04. After substitution, the verification window duration is approximately 19.3 minutes, which can be taken as 19 minutes according to the nursing record. The verification objects of the verification window include the sacrococcygeal region and the right greater trochanter region. The verification content includes the pressure dose, shear dose, and microenvironment load decrease in the sacrococcygeal region, as well as the regional load status increment in the right greater trochanter region.

[0095] Step 7: Re-acquire multimodal sensing data of the target body surface area and adjacent body surface areas within the verification window, and generate closed-loop update results. After the nursing staff performed position adjustment and placed the decompression pad, the pressure dose in the sacral and coccygeal region within the verification window decreased from 6727.36 to 2980.00, the shear dose decreased from 0.9518 to 0.4800, and the microenvironmental load decreased from 102.37 to 55.00. The target decompression state can be written as: After substituting the values ​​before and after the above nursing actions, the target decompression state is 0.877.

[0096] The load condition of the right large rotor region before the nursing action was 0.36, and after the nursing action was performed, it was 0.42. The edge pressure transfer intensity was taken as 0.08, and the adjacent transfer verification intensity was taken as 0.04. The adjacent pressure state can be written as: , After substitution, the adjacent pressure state is 0.132. The target decompression state is higher than the effective decompression threshold of 0.45, and the adjacent pressure state is lower than the transfer review threshold of 0.25. Therefore, the verification result of the site-level nursing action package is valid. The structured nursing record is written with the execution area as the sacral-coccygeal region, the execution content as body position adjustment and decompression treatment, the review area as the right greater trochanter region, the target decompression state as 0.877, the adjacent pressure state as 0.132, and the verification result as valid. The review priority of the next nursing window can be written as: In the current nursing window, the review priority for the right greater trochanter region is set to 0.25. After substituting the adjacent pressure-bearing state (0.132), the target decompression state (0.877), and the adjacent transfer review intensity (0.04), the review priority for the right greater trochanter region in the next nursing window is approximately 0.159, which is lower than the conventional review prompt threshold. This review priority is jointly determined by the target decompression state, the adjacent pressure-bearing state, and the adjacent transfer review intensity. This ensures that the review object in the next nursing window is affected by both the decompression effect of the target body surface area and the changes in pressure in the adjacent body surface areas, thus allowing the risks of adjacent areas after the target area nursing action to enter the subsequent review ranking. The historical risk status of the sacrococcygeal region in the next nursing window is written into the relief inheritance status. The risk source of the sacrococcygeal region in the multimodal pressure ulcer risk status atlas is updated to the relief source marker, and the structured nursing record is simultaneously written with the effective marker of the nursing action.

[0097] Through the complete implementation process described above, this method can connect the multimodal perception data of the bed surface, nursing risk profiles, body surface area mapping, atlas node status, model risk readout, site-level nursing action packages, and verification window feedback in a continuous bed rest nursing task into a complete data stream. This allows for the timely identification of continuous pressure and shear slip risks in the sacral and coccygeal regions. After the nursing action is performed, closed-loop verification is performed using the target decompression state and adjacent pressure states. This reduces the interference of getting out of bed, moving, and sensor abnormalities on risk assessment, reduces inconsistencies caused by relying solely on experience to determine intervention sites and verification timing, and improves the accuracy of site-level nursing intervention matching, the ability to detect risk transfer in adjacent areas, and the continuity of structured records across nursing windows.

[0098] It should be noted that the step numbers in the above embodiments, such as Step 1, Step 2, etc., are only used to distinguish different processing stages and do not limit the execution of each processing stage in that order. The specific execution order of each processing stage can be adjusted according to the arrival status of nursing window data, the timing of nursing action execution, and the triggering conditions of the verification window. The above step numbers should not be construed as limiting the implementation process of the embodiments of this application.

[0099] like Figure 2 As shown, the following is an embodiment of a pressure ulcer prevention system based on multimodal perception provided by this application. This pressure ulcer prevention system based on multimodal perception and the pressure ulcer prevention methods based on multimodal perception in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the pressure ulcer prevention system based on multimodal perception, please refer to the embodiments of the pressure ulcer prevention methods based on multimodal perception described above.

[0100] Based on the same concept, another embodiment of this application provides a pressure ulcer prevention system based on multimodal perception, comprising: Data acquisition unit 1 is used to acquire bed surface multimodal perception data and nursing risk profiles of long-term bedridden patients within the current nursing window, and to perform stable nursing window filtering and body surface area mapping on the bed surface multimodal perception data to obtain regional nursing window data; The map construction unit 2 is used to construct the regional state quantity of each body surface region based on the regional nursing window data, and to construct a multimodal pressure ulcer risk state map based on the regional state quantity. The graph edges in the multimodal pressure ulcer risk state map are determined by the decrease in load of the target body surface region, the increase in load of the adjacent body surface region, and the spatial proximity of the pressure center of the target body surface region and the adjacent body surface region. Risk identification unit 3 is used to identify the site-level risk status, risk source, and risk transfer status of each body surface area based on the multimodal pressure ulcer risk status map, nursing risk records, and historical risk status of the previous nursing window; Action generation unit 4 is used to generate location-level nursing action packages based on location-level risk status, risk source and risk transfer status of adjacent areas, and to configure a verification window for location-level nursing action packages; The verification update unit 5 is used to reacquire the multimodal perception data of the target body surface area and adjacent body surface areas within the verification window, form the verification results, review priorities and structured nursing records of the site-level nursing action package based on the target decompression state and adjacent pressure states, and update the multimodal pressure ulcer risk status map, review priorities and structured nursing records of the next nursing window based on the verification results.

[0101] The above-disclosed embodiments are merely preferred embodiments of this application, but this application is not limited thereto. Any changes, improvements, and modifications made by those skilled in the art without departing from the principles of this application, without inventive effort, shall fall within the protection scope of this application.

Claims

1. A pressure ulcer prevention method based on multimodal perception, characterized in that, include: Step S1: Obtain the bed surface multimodal perception data and nursing risk profile of long-term bedridden patients within the current nursing window, perform stable nursing window filtering and body surface area mapping on the bed surface multimodal perception data to obtain regional nursing window data; Step S2: Construct regional state quantities for each body surface region based on the regional nursing window data, and construct a multimodal pressure ulcer risk status map based on the regional state quantities. The edges of the multimodal pressure ulcer risk status map are determined by the decrease in load of the target body surface region, the increase in load of adjacent body surface regions, and the spatial proximity of the pressure centers of the target body surface region and adjacent body surface regions. Step S3: Based on the multimodal pressure ulcer risk status map, nursing risk records, and historical risk status of the previous nursing window, identify the site-level risk status, risk source, and risk transfer status of adjacent areas for each body surface region; Step S4: Generate a site-level nursing action package based on the site-level risk status, risk source, and risk transfer status of adjacent areas, and configure a verification window for the site-level nursing action package; Step S5: Reacquire the multimodal perception data of the target body surface area and adjacent body surface areas in the verification window. Based on the target decompression state and adjacent pressure state, form the verification results, review priority and structured nursing record of the site-level nursing action package. Update the multimodal pressure ulcer risk status map, review priority and structured nursing record of the next nursing window based on the verification results.

2. The pressure ulcer prevention method based on multimodal perception as described in claim 1, characterized in that, Step S1, which involves performing stable nursing window filtering and body surface region mapping on the multimodal sensing data of the bed surface to obtain regional nursing window data, specifically includes: The sampling segments within the current nursing window are subjected to contact stability determination. Sampling segments that meet the conditions for leaving the bed, nursing relocation, or abnormal sensor quality are marked as disturbed segments, and disturbed segments are removed from the current nursing window to obtain stable nursing segments. An initial body surface area mapping is formed based on the pressure profile, contact area edge, and pressure center location in the stable nursing segment; Within a stable care segment, the migration status of the center of pressure and the change status of the contact area are detected. When the migration status of the center of pressure or the change status of the contact area meet the remapping conditions, the initial body surface region mapping is updated. Based on the updated body surface region mapping or the unupdated initial body surface region mapping, stable nursing segments are assigned to regions to obtain regional nursing window data.

3. The pressure ulcer prevention method based on multimodal perception as described in claim 1, characterized in that, Step S2 specifically includes: Based on the regional nursing window data, pressure dose, shear dose, microenvironmental load, and contact reliability status of each body surface area were calculated. Based on the contact reliability state, gated calculations of pressure dose, shear dose, and microenvironmental load are performed to obtain the regional load state; A multimodal pressure ulcer risk status map is constructed based on the regional load status. The pressure transfer direction between different body surface areas is determined based on the relationship between the regional load status in the current nursing window and the previous nursing window, as well as the spatial proximity of the pressure center between the target body surface area and adjacent body surface areas. The graph edges in the multimodal pressure ulcer risk status map are determined according to the pressure transfer direction.

4. The pressure ulcer prevention method based on multimodal perception as described in claim 3, characterized in that, The steps for gating calculations of pressure dose, shear dose, and microenvironmental load based on the contact confidence state specifically include: Contact stability and sensing quality are jointly mapped to a pressure confidence state, and slip direction stability and headboard posture stability are jointly mapped to a shear confidence state. Skin contact impedance change and humidity recovery trend are jointly mapped to a microenvironment confidence state. The pressure gating quantity is obtained based on the pressure dose and pressure confidence state; the shear gating quantity is obtained based on the shear dose and shear confidence state; and the microenvironment gating quantity is obtained based on the microenvironment load and microenvironment confidence state. Load merging is performed on the same surface area for pressure-gated, shear-gated, and microenvironment-gated variables, and the load merging results are written into the area load status.

5. The pressure ulcer prevention method based on multimodal perception as described in claim 3, characterized in that, Step S3 specifically includes: Based on the multimodal pressure ulcer risk status map, determine the regional load status and pressure transfer direction of each body surface area; Based on the nursing risk record, each body surface area is classified into a state of reduced tolerance and a state of increased risk, and an individual tolerance correction state is formed based on the state of reduced tolerance and a state of increased risk. Based on the historical risk status of the previous nursing window, the area load status is continuously calculated using pressure memory to obtain the continuous risk status; Based on regional load status, map edge pressure transfer direction, individual tolerance correction status, and continuous risk status, identify location-level risk status, risk sources, and risk transfer status in adjacent areas.

6. The pressure ulcer prevention method based on multimodal perception as described in claim 5, characterized in that, The steps for identifying the risk transfer status of adjacent areas specifically include: When the pressure transfer direction on the edge of the graph points to the adjacent body surface area, the regional load state change of the target body surface area and the regional load state change of the adjacent body surface area are compared to obtain a state where the transfer direction is consistent. When the conditions for transfer review are met for both the consistent transfer direction and the persistent risk status, the adjacent body surface regions will be written into the transfer review object. Write the adjacent body surface area corresponding to the transferred review object into the adjacent area risk transfer status, and update the risk source in the adjacent area risk transfer status to the pressure transfer source.

7. The pressure ulcer prevention method based on multimodal perception as described in claim 6, characterized in that, Step S4 specifically includes: The intervention area on the body surface is determined based on the risk status at the site level, and the intervention area is written into the execution area of ​​the site-level nursing action package; Based on the source of risk, action categories are formed among postural adjustment, decompression, skin examination, and microenvironment screening, and these action categories are written into the execution content of the site-level nursing action package. When the risk transfer status of adjacent areas does not include the transfer review object, the adjacent body surface area corresponding to the intervention body surface area is written into the review area of ​​the site-level nursing action package, and the verification object and verification content in the verification window are formed according to the execution area, execution content and review area; When the risk transfer status of an adjacent area includes a transfer review object, the transfer review object is written into the review area of ​​the site-level nursing action package, and the verification object and verification content in the verification window are formed based on the execution area, execution content and review area.

8. The pressure ulcer prevention method based on multimodal perception as described in claim 7, characterized in that, Step S5 involves re-acquiring the multimodal sensing data of the target body surface area and adjacent body surface areas within the verification window, and forming the verification results, review priorities, and structured nursing records for the site-level nursing action package based on the target decompression state and adjacent pressure states. Specifically, this includes: The execution area is taken as the target body surface area, and the review area is taken as the adjacent body surface area; Within the verification window, a target decompression state is formed based on the pressure dose reduction ratio, shear dose reduction ratio, and microenvironment load recovery ratio corresponding to the execution area. Within the verification window, adjacent pressure states are formed based on the regional load state increment corresponding to the verification area, the pressure transfer intensity at the edge of the graph, and the adjacent transfer verification intensity. The verification results and review priority of the site-level nursing action package are determined based on the execution content, target decompression status, and adjacent pressure status. Write the execution area, execution content, review area, target decompression status, adjacent pressure status, verification results, and review priority into the structured nursing record.

9. The pressure ulcer prevention method based on multimodal perception as described in claim 8, characterized in that, Step S5, which updates the multimodal pressure ulcer risk status map, review priority, and structured nursing record for the next nursing window based on the verification results, specifically includes: When the target pressure relief state meets the effective pressure relief conditions and the adjacent pressure-bearing state does not meet the transfer review conditions, the historical risk status of the execution area in the next nursing window is written into the relief inheritance status, and the risk source corresponding to the execution area is updated to the relief source mark in the multimodal pressure ulcer risk status map, and the effective nursing action mark is written into the structured nursing record. When the target pressure relief status does not meet the effective pressure relief conditions, the historical risk status of the execution area in the next nursing window will be written into the unrelieved inherited status, and the risk source corresponding to the execution area will be retained in the multimodal pressure ulcer risk status map. When adjacent pressure conditions meet the transfer review conditions, a transfer review map edge pointing from the execution area to the review area is established in the multimodal pressure ulcer risk status map, and the review priority and structured nursing records are updated based on the transfer review map edge.

10. A pressure ulcer prevention system based on multimodal perception, characterized in that, include: The data acquisition unit is used to acquire bed surface multimodal perception data and nursing risk profiles of long-term bedridden patients within the current nursing window, and to perform stable nursing window filtering and body surface area mapping on the bed surface multimodal perception data to obtain regional nursing window data. The map construction unit is used to construct the regional state quantity of each body surface region based on the regional nursing window data, and to construct a multimodal pressure ulcer risk state map based on the regional state quantity. The graph edges in the multimodal pressure ulcer risk state map are jointly determined based on the decrease in load of the target body surface region, the increase in load of adjacent body surface regions, and the spatial proximity of the pressure center of the target body surface region and adjacent body surface regions. The risk identification unit is used to identify the site-level risk status, risk source, and risk transfer status of each body surface area based on the multimodal pressure ulcer risk status map, nursing risk records, and the historical risk status of the previous nursing window. The action generation unit is used to generate location-level nursing action packages based on location-level risk status, risk source, and risk transfer status of adjacent areas, and to configure a verification window for the location-level nursing action packages. The verification and update unit is used to reacquire the multimodal sensing data of the target body surface area and adjacent body surface areas within the verification window, form the verification results, review priorities and structured nursing records of the site-level nursing action package based on the target decompression state and adjacent pressure states, and update the multimodal pressure ulcer risk status map, review priorities and structured nursing records of the next nursing window based on the verification results.

Images