A risk early warning intervention system and method for aspiration after stroke in the elderly

Abstract

The application relates to the technical field of health management, and discloses an aspiration risk early warning intervention system and method for the elderly after stroke, which comprises a hardware acquisition module, a data processing module, an algorithm calculation module and an application interaction module; the hardware acquisition module acquires dynamic physiological data and static information, the data processing module pre-processes the dynamic physiological data and the static information and obtains unified feature data through multi-modal fusion, the algorithm calculation module calculates an aspiration risk score and determines a risk level through dynamic weight adjustment, the application interaction module provides login and information input functions, outputs early warning information and intervention guidance content, the application realizes cooperative acquisition and processing of subjective and objective data, adapts to various use scenarios, improves the accuracy and adaptability of aspiration risk assessment, realizes efficient early warning and targeted intervention, and helps reduce the incidence of related complications.

Description

Technical Field

[0001] This invention relates to the field of health management technology, specifically to a system and method for early warning and intervention of aspiration risk after stroke in the elderly. Background Technology

[0002] Stroke, a common cerebrovascular disease among the elderly, often leads to nerve damage affecting swallowing function, resulting in swallowing dysfunction such as slowed swallowing reflexes and decreased coordination of swallowing muscles. This significantly increases the risk of aspiration. Aspiration refers to the accidental entry of food, liquids, or secretions into the airway during eating or swallowing. It not only causes immediate symptoms such as choking and airway spasm but also easily induces serious complications such as aspiration pneumonia and respiratory failure. It is a major contributing factor to the worsening of the condition and prolonged rehabilitation period for elderly stroke patients. In severe cases, it can even endanger life due to airway obstruction, posing a great threat to the health and safety of the elderly and becoming a core challenge in the rehabilitation and long-term care of elderly patients with cerebrovascular diseases.

[0003] As the population ages, the care needs of elderly patients after stroke continue to grow. Multiple care models are emerging, including hospital rehabilitation, integrated medical and elderly care, home-based care, and institutional care. The need for accurate assessment, timely warning, and targeted intervention of aspiration risk after stroke in clinical and daily care is becoming increasingly urgent. Currently, there are still many unresolved issues regarding the assessment and prevention of aspiration risk in the elderly after stroke, making it difficult to adapt to the diverse care needs across different scenarios. Summary of the Invention

[0004] In order to solve or at least partially solve the above-mentioned technical problems, the present invention provides a system and method for early warning and intervention of aspiration risk after stroke in the elderly.

[0005] In a first aspect, the present invention provides an early warning and intervention system for the risk of aspiration after stroke in the elderly, comprising: The hardware acquisition module is used to collect dynamic physiological data and static information of the elderly, and transmit the collected dynamic physiological data and static information to the data processing module. The dynamic physiological data includes swallowing physiological data and blood oxygen data, and the static information includes scale assessment information and basic personal information. The data processing module is used to receive the dynamic physiological data and static information collected by the hardware acquisition module, perform preprocessing and multimodal fusion processing to obtain unified feature data, and transmit the unified feature data to the algorithm calculation module. The algorithm calculation module is used to calculate the risk score of accidental inhalation based on the unified feature data through dynamic weight adjustment, determine the corresponding risk level according to the scenario type, and transmit the risk level to the application interaction module. The application interaction module provides user login and information entry functions, and outputs corresponding early warning information and intervention guidance content based on the risk level determined by the algorithm calculation module.

[0006] Optionally, the hardware acquisition module includes: A smart swallowing monitoring spoon is used to collect swallowing physiological data from the dynamic physiological data and transmit it to the data processing module. A portable pulse oximeter is used to collect blood oxygen data from the dynamic physiological data and transmit it to the data processing module. A voice interaction device is used to collect scale assessment information and basic personal information from the static information, and then transmit the collected information to the data processing module after converting it into structured data.

[0007] Optionally, the application interaction module is configured to provide two login modes: a regular user login mode and a medical staff login mode. The regular user login mode is adapted for home and elderly care institution use scenarios and is configured to support the input of basic personal information and scale assessment information of the elderly, and to view the corresponding warning information and intervention guidance content; The login mode for medical staff is adapted to the use scenarios of hospitals and integrated medical and elderly care centers. It is configured to support the connection with medical systems, obtain the patient's historical medical information, and synchronously store the aspiration risk score, risk level and intervention execution record.

[0008] In a second aspect, the present invention also provides a method for early warning and intervention of aspiration risk after stroke in the elderly, applied to the system described in any of the first aspects, the method comprising the following steps: S1. Obtain the dynamic physiological data and static information of the elderly, wherein the dynamic physiological data includes the swallowing physiological data and the blood oxygen data, and the static information includes the scale assessment information and the basic personal information; S2. Preprocess the acquired dynamic physiological data and static information to remove invalid data and unify the data scale; S3. Perform multimodal fusion processing on the preprocessed dynamic physiological data and the static information to obtain the unified feature data; S4. Based on the unified feature data and the corresponding usage scenario type, calculate the aspiration risk score through dynamic weight adjustment. S5. Based on the risk grading standard corresponding to the usage scenario type, determine the corresponding risk level according to the calculated aspiration risk score; S6. Based on the determined risk level, output the corresponding early warning information and intervention guidance content.

[0009] Optionally, step S3 specifically includes the following sub-steps: S31. Classify the preprocessed dynamic physiological data and static information respectively, and extract the corresponding core evaluation features; S32. The classified core evaluation features are mapped to a unified dimensional space through a neural network to form initial feature data; S33. Based on the correlation between each of the core assessment features and the risk of accidental aspiration, assign corresponding weights to the initial feature data, perform weighted fusion, and obtain the unified feature data.

[0010] Optionally, in step S31, the specific method for extracting the core evaluation features is as follows: Swallowing-related features and blood oxygenation-related features are extracted from the dynamic physiological data to form time-series features; Basic attribute features and scale rating features are extracted from the static information to form structured class features; The specific dimensions of the structured features are determined based on the type of use scenario. The structured features in hospital and integrated medical and elderly care center scenarios have more dimensions than those in home and elderly care institution scenarios.

[0011] Optionally, step S4 specifically includes the following sub-steps: S41. Obtain a preset basic weight library, which includes basic weights corresponding to each of the core evaluation features; S42. Calculate the real-time risk correlation factor based on the dynamic physiological characteristics in the unified feature data; S43. Based on the type of use scenario, determine the target feature among the core assessment features that needs to be adjusted in weight, and multiply the basic weight corresponding to the target feature with the real-time risk correlation factor to obtain the dynamic weight; S44. Based on the quantitative values ​​and corresponding basic weights of the non-target features in the core evaluation features, and the quantitative values ​​and corresponding dynamic weights of the target features, the aspiration risk score is calculated by summing them up.

[0012] Optionally, in step S42, the specific method for calculating the real-time risk correlation factor is as follows: If the application scenario is home and elderly care facility scenario, the real-time risk association factor is calculated based on swallowing pressure characteristics and blood oxygen minimum value characteristics; If the application scenario is a hospital or integrated medical and elderly care center, the real-time risk association factor is calculated based on the swallowing pressure characteristic, the lowest blood oxygen value characteristic, and the swallowing function-specific assessment characteristic in the scale assessment information.

[0013] Optionally, step S5 specifically includes the following sub-steps: S51. Based on the usage scenario type, call the corresponding preset risk grading standard, which includes the score range corresponding to different risk levels. S52. Determine the score range in which the aspiration risk score falls, and determine the risk level corresponding to the score range as the final risk level.

[0014] Optionally, step S6 specifically includes the following sub-steps: S61. Based on the determined risk level, retrieve the preset early warning rules and intervention guidance library; S62. Output the corresponding warning information according to the warning rules, wherein the warning information for high-risk levels has a higher warning intensity than that for medium- and low-risk levels; S63. Match the intervention guidance content corresponding to the risk level and the usage scenario type from the intervention guidance library.

[0015] The present invention has the following beneficial effects: The system provided by this invention collects dynamic physiological data and static information of elderly individuals through a hardware acquisition module. This overcomes the limitations of traditional assessments that rely solely on static information, making the data used for assessment more comprehensive, reducing the influence of subjective judgment on the assessment results, and improving the objectivity of the assessment. The data processing module preprocesses and performs multimodal fusion processing on the collected dynamic physiological data and static information, integrating different types of scattered data into unified feature data. This solves the problem of difficult collaborative use of heterogeneous data and provides a reliable data foundation for subsequent risk calculation. The algorithm calculation module calculates the aspiration risk score based on the unified feature data through dynamic weight adjustment and determines the risk level by combining it with the scenario type. This avoids the shortcomings of fixed-weight mode, which cannot adapt to different individuals and usage scenarios, making the risk assessment results more consistent with the actual situation and improving the accuracy of the risk assessment. The application interaction module provides user login and information entry functions, and outputs corresponding warning information and intervention guidance content according to the risk level. This facilitates user operation and allows relevant personnel to obtain timely and targeted risk warnings and intervention suggestions, which helps to effectively prevent and control the aspiration risk after stroke in the elderly and provides support for the health management of the elderly. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a risk warning and intervention system for aspiration in elderly people after stroke, provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of the hardware acquisition module and algorithm calculation module provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the application scenario of the aspiration risk warning and intervention system for elderly people after stroke, provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the intervention method for early warning of aspiration risk in elderly people after stroke, provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0018] See Figure 1 This invention provides an early warning and intervention system for the risk of aspiration after stroke in elderly individuals, comprising: The hardware acquisition module 100 is configured to collect dynamic physiological data and static information of the elderly. The dynamic physiological data includes swallowing physiological data and blood oxygen data, and the static information includes scale assessment information and basic personal information. The data processing module 200 is configured to receive dynamic physiological data and static information collected by the hardware acquisition module, and perform preprocessing and multimodal fusion processing to obtain unified feature data. The algorithm calculation module 300 is configured to calculate the risk score of accidental aspiration based on unified feature data and through dynamic weight adjustment, and determine the corresponding risk level according to the scenario type. The application interaction module 400 is configured to provide user login and information entry functions, and output corresponding early warning information and intervention guidance content based on the risk level determined by the algorithm calculation module.

[0019] Among them, dynamic physiological data refers to real-time physiological indicators related to the swallowing action and blood oxygen status of the elderly. Swallowing physiological data reflects the physiological changes during the swallowing process, while blood oxygen data reflects the body's oxygen supply. These two types of data can capture the dynamic changes during the swallowing process. Static information is relatively stable information. Scale assessment information is related information that reflects swallowing function, cognitive status, etc., obtained through standardized assessment scales. Basic personal information includes age, past medical history, and other basic personal information closely related to risk assessment.

[0020] The hardware acquisition module acquires various types of data required for the assessment, capturing both real-time changes in dynamic physiological data and collecting complete static information. The data processing module is responsible for organizing the different types of data collected, transforming the scattered data into a unified and effective data format through preprocessing and multimodal fusion. The algorithm calculation module, based on the processed effective data and considering the characteristics of different scenarios, derives risk scores and determines risk levels through reasonable calculation methods. The application interaction module provides users with operation entry points and result feedback channels, enabling users to easily use the system and obtain the necessary early warning information and intervention guidance.

[0021] In practical applications, the hardware acquisition module operates using hardware devices adapted to the specific scenario. For dynamic physiological data, specifically swallowing and blood oxygenation data, specialized acquisition equipment is used for real-time capture, ensuring the data accurately reflects the physiological state of the elderly during swallowing. For static information, including scale assessment information and basic personal data, appropriate data entry devices are used. The collected dynamic and static data are promptly transmitted to the data processing module for preprocessing. This module performs multimodal fusion processing, integrating different types and formats of data into unified feature data. After receiving the unified feature data, the algorithm calculation module considers the corresponding usage scenario type and the key factors influencing aspiration risk in different scenarios. Using dynamic weight adjustment, it calculates an aspiration risk score by comprehensively considering the impact of various data types. Based on the judgment criteria corresponding to the scenario type, it determines the calculated aspiration risk score and establishes the corresponding risk level, ensuring the risk assessment results align with the actual usage scenario. The application interaction module provides users with a convenient login function to meet the needs of different users, and also supports users to enter relevant information; after the algorithm calculation module determines the risk level, the application interaction module outputs the corresponding early warning information based on the risk level, reminding users to pay attention to the risk situation, and at the same time pushing targeted intervention guidance content.

[0022] See Figure 3 In some implementations, the hardware acquisition module includes: a smart swallowing monitoring spoon configured to acquire swallowing physiological data from dynamic physiological data; a portable pulse oximeter configured to acquire blood oxygen data from dynamic physiological data; and a voice interaction device configured to acquire scale assessment information and basic personal information from static information, and convert the acquired information into structured data.

[0023] In the aspiration risk warning and intervention system for elderly stroke patients, the hardware acquisition module collects data through the coordinated use of specialized equipment. Its specific components include a smart swallowing monitoring spoon, a portable pulse oximeter, and a voice interaction device. The technical specifications of each device in the hardware acquisition module are shown in Table 1. Table 1

[0024] In some implementations, the application interaction module is configured to provide two login modes: a regular user login mode and a medical staff login mode. The regular user login mode is adapted for home and elderly care institution use scenarios and is configured to support the input of basic personal information and scale assessment information of the elderly, and to view the corresponding warning information and intervention guidance content; The login mode for medical staff is adapted to the use scenarios of hospitals and integrated medical and elderly care centers. It is configured to support the connection with medical systems, obtain the patient's historical medical information, and synchronously store the aspiration risk score, risk level and intervention execution record.

[0025] In the early warning and intervention system for the risk of aspiration after stroke in the elderly, the interactive module uses two differentiated login modes to adapt to the needs of different usage scenarios, allowing both ordinary users and medical staff to operate the system conveniently and fully leverage its early warning and intervention guidance functions.

[0026] The standard user login mode is primarily designed for both home and elderly care facility scenarios. Users are mainly family members of the elderly, home caregivers, or general caregivers in elderly care facilities. Therefore, the mode's operation is simple, easy to understand, and aligns with daily usage habits. In the home scenario, caregivers open the system terminal. The system homepage displays two entry points: "Standard User Login" and "Medical Staff Login." Caregivers click "Standard User Login," enter their pre-registered account and password, and complete the login. After logging in, the interface presents two core sections: "Information Entry" and "Risk View." Caregivers click "Information Entry" to fill in the elderly person's basic personal information, and can also supplement the information with assessment scales. After completing the entry and clicking submit, the information is automatically transmitted to the data processing module. Once the system completes the risk calculation and determines the risk level, caregivers click "Risk View" to see the corresponding warning information and intervention guidance content tailored to the home scenario.

[0027] The healthcare worker login mode is adapted for hospital and integrated medical and elderly care center scenarios. It is designed for healthcare workers in hospital rehabilitation departments and professional healthcare workers in integrated medical and elderly care centers, aligning with the actual needs of clinical diagnosis and rehabilitation care. In a hospital setting, healthcare workers open the system terminal, click the "Healthcare Worker Login" entry, and enter their hospital-assigned employee ID and password. After logging in, the system will display a "Link Medical System" prompt. After confirming, the healthcare worker enters the patient's hospital number or ID number to quickly link to the hospital's internal medical system. This allows access to the patient's stroke history, previous swallowing function assessment records, medication history, and other historical medical information. After collecting the patient's dynamic physiological data and static information, the system calculates the aspiration risk score and risk level. Healthcare workers can view the detailed calculation process and risk analysis on the system interface. Simultaneously, the system automatically stores the patient's aspiration risk score, risk level, and records of subsequent interventions implemented by the healthcare worker.

[0028] See Figure 4 This invention also provides a method for early warning and intervention of aspiration risk after stroke in the elderly, applied to any of the systems described in the above system embodiments, comprising the following steps: S1. Obtain dynamic physiological data and static information of the elderly. Dynamic physiological data includes swallowing physiological data and blood oxygen data, while static information includes scale assessment information and basic personal information. S2. Preprocess the acquired dynamic physiological data and static information to remove invalid data and standardize the data scale; S3. Perform multimodal fusion processing on the preprocessed dynamic physiological data and static information to obtain unified feature data; S4. Based on unified feature data and combined with the corresponding usage scenario type, calculate the risk score of accidental aspiration through dynamic weight adjustment. S5. Determine the corresponding risk level based on the risk grading standard corresponding to the type of use scenario and the calculated aspiration risk score. S6. Based on the determined risk level, output the corresponding early warning information and intervention guidance.

[0029] Specifically, the process begins with data acquisition, where various types of data transmitted by the hardware acquisition module are obtained. Dynamic physiological data includes swallowing physiological data generated during the swallowing process and blood oxygen data related to the body's oxygen supply. Static information includes relevant information obtained through scale assessments and basic personal information of the elderly. All received data is initially categorized to distinguish between dynamic physiological data and static information.

[0030] After data acquisition, the data preprocessing stage begins. For the received dynamic physiological data and static information, invalid data that does not meet the requirements is first removed using preset filtering rules. Then, data of different formats and magnitudes undergo scale unification processing. After preprocessing, multimodal fusion processing is performed to integrate the two different modalities of dynamic physiological data and static information. Through preset fusion logic, the scattered and different types of data are transformed into unified feature data.

[0031] Based on unified feature data and combined with the current corresponding usage scenario type, the calculation is carried out through dynamic weight adjustment. The influence weight of different data features can be adjusted according to the characteristics of the scenario, and the risk score of accidental aspiration can be calculated by comprehensively considering the role of various features.

[0032] After the aspiration risk score is calculated, the preset risk grading standard is invoked according to the corresponding usage scenario type. The calculated aspiration risk score is matched with the level range in the standard to determine the current risk level of the elderly person.

[0033] Finally, based on the determined risk level, the corresponding early warning information and intervention guidance content are retrieved from the internally stored content library to complete the output operation. The early warning information is used to indicate the current risk status, while the intervention guidance content provides specific directions for risk prevention and control.

[0034] In some implementations, S3 specifically includes the following sub-steps: S31. Classify the preprocessed dynamic physiological data and static information respectively, and extract the corresponding core evaluation features; S32. The core evaluation features after classification are mapped to a unified dimensional space through a neural network to form initial feature data; S33. Based on the correlation between each core assessment feature and the risk of accidental aspiration, assign corresponding weights to the initial feature data, perform weighted fusion, and obtain unified feature data.

[0035] The preprocessed dynamic physiological data and static information are synchronously transmitted to the multimodal fusion processing module. This module first classifies and organizes the two types of data and extracts core assessment features. During the classification process, dynamic physiological data and static information are distinguished according to preset classification rules. Dynamic physiological data is real-time changing time-series data, from which core assessment features closely related to the risk of aspiration are extracted.

[0036] After the core evaluation features are extracted and classified, a neural network maps these features to a unified dimensional space, forming initial feature data. The neural network used here is a pre-defined feature mapping network. The extracted temporal and structured features are input into different input layers of the neural network. Temporal features are processed by the temporal feature processing layer to serialize them, while structured features are normalized by the feature standardization layer. Subsequently, feature transformation is performed through the hidden layers of the neural network, gradually mapping the two types of features with different dimensions to the same dimensional space. During this process, the neural network automatically adjusts its feature parameters according to the mapping rules formed during training, ensuring that the two types of features have the same data dimension and a unified expression form after mapping. Finally, the integrated initial feature data is output, which retains the real-time nature of the temporal features while also including the fundamental nature of the structured features.

[0037] After the initial feature data is generated, weights are assigned to the initial feature data based on the correlation between each core assessment feature and the risk of aspiration, and then weighted and fused to obtain unified feature data. The correlation assessment combines clinical care experience and historical risk assessment data, and a pre-defined feature-risk correlation judgment rule is established. According to this rule, the degree of correlation between each core assessment feature in the initial feature data and the risk of aspiration is analyzed one by one. The higher the correlation, the greater the weight assigned; the lower the correlation, the smaller the weight assigned. After the weight assignment is completed, the initial feature data is fused using a weighted summation method. The initial data corresponding to each core assessment feature is multiplied by the assigned weight, and all products are summed to obtain the final unified feature data.

[0038] In some implementations, the specific method for extracting the core evaluation features in S31 is as follows: Swallowing-related features and blood oxygenation-related features are extracted from dynamic physiological data to form time-series features; Basic attribute features and scale scoring features are extracted from static information to form structured class features; Based on the type of use case, the specific dimensions of the structured features are determined. The structured features in hospital and integrated medical and elderly care center scenarios have more dimensions than those in home and elderly care institution scenarios.

[0039] Specifically, the process begins with extracting core assessment features from dynamic physiological data. Since dynamic physiological data is time-series data reflecting the real-time physiological state of the elderly, pre-defined dynamic data classification rules are invoked to divide the preprocessed dynamic physiological data into two main categories: swallowing physiological data and blood oxygenation data. Then, corresponding core assessment features are extracted from each category, and finally integrated to form time-series features. During the extraction of core assessment features from swallowing physiological data, a pre-defined feature recognition algorithm can be used to filter out features that directly reflect the state of swallowing function and are highly correlated with the risk of aspiration. These features include the initiation time of swallowing movements, the duration of swallowing, the peak swallowing pressure, and the stability of swallowing rhythm. These features can accurately capture physiological changes during the swallowing process in the elderly, such as whether the swallowing movement is slow, whether the swallowing pressure is sufficient, and whether the rhythm is regular. These changes directly affect the level of aspiration risk. Filtering out irrelevant physiological fluctuation data during swallowing makes the extracted swallowing-related features more targeted. In the process of extracting core assessment features from blood oxygen data, feature recognition algorithms are also used to extract key features related to aspiration risk from blood oxygen data. These features include real-time change rate of blood oxygen level, minimum blood oxygen value, duration of blood oxygen stabilization, and amplitude of blood oxygen fluctuation. Aspiration is often accompanied by an abnormal drop in blood oxygen level. Therefore, these blood oxygen-related features can indirectly reflect whether there is a risk of aspiration in the elderly. The extracted swallowing-related features and blood oxygen-related features are then sequenced and integrated into time-series features according to time order.

[0040] After extracting the core assessment features from dynamic physiological data, the core assessment features from static information are extracted. Static information, as relatively stable basic data, can be categorized into two main types based on pre-defined static data classification rules: basic personal information and scale assessment information. Core assessment features are then extracted from each category and integrated to form structured features. During the extraction of core assessment features from basic personal information, fixed features related to aspiration risk can be selected based on pre-defined extraction dimensions. These include age of the elderly, duration of stroke onset, past medical history, previous swallowing function impairment, and eating habits. During the extraction of core assessment features from scale assessment information, relevant scoring features reflecting swallowing function, cognitive function, and eating ability are extracted based on the scoring rules of standardized assessment scales. These include swallowing function scores, cognitive function scores, self-care eating ability scores, and choking frequency scores. Each scoring data item is extracted, and scoring items irrelevant to aspiration risk are removed. The basic attribute features and scale scoring features are then integrated into structured features, forming a standardized structured data format.

[0041] After extracting temporal and structured features, the specific dimensions of the structured features are adjusted based on the current usage scenario. The login mode of the application interaction module is used to determine whether the current usage scenario is a home-based or elderly care facility scenario, or a hospital or integrated medical and elderly care center scenario. Then, the extraction dimensions of the structured features are adjusted according to preset scenario adaptation rules. When the current scenario is identified as a home-based or elderly care facility scenario, since caregivers in this scenario are mostly non-professionals and the complexity of information collection is relatively low, the core dimensions of the structured features are retained. These include age and eating habits from basic personal information, as well as swallowing function scores and choking frequency scores from scale assessment information. This simplifies the dimensions of the structured features and reduces the difficulty of information collection and processing. When the current scenario is identified as a hospital or integrated medical and elderly care center scenario, additional dimensions of the structured features are added to the core dimensions. Specifically, new features are added, including past aspiration history and clinical treatment records from basic personal information, as well as detailed cognitive function scores, swallowing function grading assessments, and rehabilitation training progress scores from scale assessment information.

[0042] In some implementations, in S32, a feature mapping neural network is constructed to map the classified core evaluation features to a unified dimensional space to form initial feature data.

[0043] This embodiment employs a multilayer perceptron architecture to construct a dedicated feature mapping network, addressing the heterogeneous dimensionality alignment problem between temporal and structured features. The specific implementation method is as follows: (1) Network input layer design Input 1 is the temporal feature branch: dimension = T × F1; T represents the length of the time window, which defaults to 10 swallowing cycles; F1 represents the dimension of blood oxygenation characteristics during swallowing at a single moment, which defaults to 8 dimensions, including peak swallowing pressure, swallowing duration, mean blood oxygenation, and minimum blood oxygenation. Actual input shape: batchsize, T, F1; Input 2 is the structured feature branch: dimension = F2; F2 represents static feature dimensions, with 12 dimensions for home scenarios and 18 dimensions for hospital scenarios, including age, scale scores, frequency of coughing, etc. Actual input shape: batchsize, F2; (2) Network structure definition The system employs a dual-branch heterogeneous input, shared hidden layer, and unified output structure, as shown in Table 2. Table 2

[0044] In some implementations, S4 specifically includes the following sub-steps: S41. Obtain the preset basic weight library, which includes the basic weights corresponding to each core evaluation feature. S42. Calculate real-time risk correlation factors based on dynamic physiological characteristics in unified feature data; S43. Based on the type of use scenario, determine the target features among the core assessment features that need to be adjusted in weight, and multiply the basic weights corresponding to the target features by the real-time risk correlation factors to obtain the dynamic weights. S44. Based on the quantitative values ​​and corresponding basic weights of the non-target features in the core assessment features, as well as the quantitative values ​​and corresponding dynamic weights of the target features, the aspiration risk score is calculated by summing them up.

[0045] Specifically, the system first obtains a pre-defined basic weight library. This library is constructed in advance based on a large amount of clinical care data, aspiration risk assessment cases, and relevant expert experience, and is stored in the system's database. The basic weight library is constructed using a scientific approach combining clinical big data statistics with the Delphi method, and includes data sources, expert selection criteria, and the weight iteration determination process. The basic weight library assigns a corresponding basic weight to each core assessment feature. These core assessment features cover the temporal and structured features extracted in previous steps, including swallowing-related features, blood oxygenation-related features, basic attribute features, scale scoring features, and all other core features related to aspiration risk. The allocation of basic weights follows the principle that the higher the correlation with aspiration risk, the greater the basic weight, and these are assigned the highest basic weight. Age and the duration of stroke onset among the basic attribute features, and the swallowing function score among the scale scoring features, have a moderate correlation with aspiration risk and are assigned a medium basic weight. Features with a low correlation with aspiration risk, such as eating habits, are assigned a lower basic weight.

[0046] Taking a specific implementation example, the basic weight library is constructed based on big data from clinical studies on the risk of aspiration after stroke involving ≥5000 elderly stroke patients. It is constructed using the Delphi method, and an expert pool of 15 associate chief physicians and above from neurology, geriatrics, and rehabilitation departments is selected. The basic weights of each core assessment feature are determined through three rounds of weight scoring iterations. In home and elderly care facility scenarios, the basic weights of the core assessment features are: peak swallowing pressure 20, swallowing duration 8, minimum blood oxygen saturation 18, blood oxygen fluctuation amplitude 7, age 10, stroke duration 6, swallowing function score 15, and choking frequency score 12. The remaining basic attributes / scale scores have a total weight of 4, and the sum of all feature weights is 100. In hospital and integrated medical and elderly care center scenarios, a history of aspiration 8 and a dysphagia grading score 6 are added to the above features. After weight redistribution, the sum of all features remains 100. The update rules for the basic weight library are as follows: clinical application data are collected every 6 months. If the risk correlation deviation of a certain feature is ≥10%, the expert library will be activated to re-score. After the update, it will be validated through 300 clinical samples. Only if the validation accuracy is ≥85% can it be used.

[0047] After acquiring the basic weight library, a real-time risk correlation factor is calculated based on the dynamic physiological characteristics in the unified feature data. The real-time risk correlation factor is a parameter reflecting the degree of abnormality in the current dynamic physiological characteristics. Its function is to adjust the basic weights so that they align with the current real-time physiological state of the elderly. During implementation, all dynamic physiological characteristics are first selected from the unified feature data, including swallowing-related characteristics and blood oxygenation-related characteristics in the time-series features. Then, a preset correlation factor calculation algorithm is invoked, combining the real-time data of the dynamic physiological characteristics with the baseline values ​​of the dynamic physiological characteristics of normal elderly individuals. The degree of abnormality of the current dynamic physiological characteristics is analyzed. The smaller the deviation between the dynamic physiological characteristics and the baseline value, the lower the real-time aspiration risk, and the closer the real-time risk correlation factor is to 1. Conversely, the larger the deviation, the higher the real-time aspiration risk, and the greater the real-time risk correlation factor is to 1. The greater the deviation, the larger the correlation factor value.

[0048] After the real-time risk correlation factor calculation is completed, the target features requiring weight adjustment in the core assessment features are determined according to the usage scenario type, and dynamic weights are calculated. During implementation, the current usage scenario type is first determined by identifying the login mode of the application interaction module. Different scenarios have different key features affecting aspiration risk, therefore the target features requiring weight adjustment also differ. Based on preset rules for the correspondence between scenarios and target features, when the current scenario is identified as home or elderly care facility, dynamic physiological features such as swallowing-related features and blood oxygen-related features are identified as target features requiring weight adjustment, while structured features are not adjusted and retain their basic weights. When the current scenario is identified as hospital or integrated medical and elderly care center, in addition to using dynamic physiological features as target features, scale rating features with a high correlation to aspiration risk among structured features are also identified as target features and their weights are adjusted. Once the target features are determined, the basic weight corresponding to each target feature is multiplied by the real-time risk correlation factor obtained through calculation. The result is the dynamic weight of that target feature. Non-target features continue to use the basic weights in the basic weight library without adjustment, ensuring the targeting and rationality of weight adjustments.

[0049] Finally, based on the quantified values ​​and corresponding weights of various core assessment features, the aspiration risk score is calculated by summing them up. During implementation, the quantified value of each core assessment feature is first obtained. The quantified values ​​of target features come from the corresponding dynamic physiological features and specified structured features in the unified feature data, while the quantified values ​​of non-target features come from other structured features in the unified feature data. Then, following the calculation method of "quantified value × corresponding weight," the weighted score of each core assessment feature is calculated one by one. That is, the quantified value of the target feature is multiplied by its corresponding dynamic weight, and the quantified value of the non-target feature is multiplied by its corresponding basic weight, yielding the contribution score of each feature to the aspiration risk. After the weighted scores of all features are calculated, they are summed through an accumulation operation; the summation result is the final aspiration risk score.

[0050] In some implementations, in S42, the specific method for calculating the real-time risk correlation factor is as follows: If the application scenario is home or elderly care facility, the real-time risk association factor is calculated based on swallowing pressure characteristics and minimum blood oxygen saturation characteristics. Specifically, the calculation of the real-time risk association factor is highly dependent on the application scenario: in home and elderly care facility scenarios, it is calculated based on swallowing pressure characteristics and minimum blood oxygen saturation characteristics; while in hospital and integrated medical and elderly care center scenarios, swallowing function-specific assessment characteristics from the scale assessment information are added on top of this. During calculation, preset baseline values ​​for swallowing pressure characteristics and minimum blood oxygen saturation values ​​of normal elderly people are retrieved, and the level is determined and a deviation coefficient is assigned based on the magnitude of the deviation using a relative deviation method.

[0051] The baseline values ​​for swallowing pressure and minimum blood oxygen saturation in normal elderly individuals are derived from a norm database established through large-scale clinical studies.

[0052] Peak swallowing pressure baseline range: Referring to common ranges in swallowing manometry studies, the peak pharyngeal contraction pressure in healthy adults is likely between 80-120 mmHg. For elderly post-stroke patients, the range needs to be adjusted according to age and health condition, with a baseline range set at 60-100 mmHg.

[0053] The minimum threshold for normal blood oxygen saturation: A resting blood oxygen saturation (SpO2) level of 95% or higher is considered normal. A brief drop to 92%-94% during eating or swallowing is noteworthy, while a sustained level below 92% is generally considered abnormal. Preferably, the baseline value can be differentiated based on subgroups such as age and gender, and can be periodically optimized and updated through system learning.

[0054] The formula for calculating the real-time risk correlation factor R is: Home scenario: R = w1 × Cswallow + w2 × Cspo2 Hospital scenario: R = w1 × Cswallow + w2 × Cspo2 + w3 × Cassessment; Wherein, Cswallow, Cspo2, and Cassessment represent the deviation coefficients of swallowing pressure, minimum blood oxygen value, and swallowing function specific assessment characteristics, respectively, and w1, w2, and w3 are the preset weights of the corresponding characteristics.

[0055] Taking a specific implementation example, the feature weights of the real-time risk association factors are: w1=0.55, w2=0.45 for home and elderly care institution scenarios; w1=0.4, w2=0.35, w3=0.25 for hospital and integrated medical and elderly care center scenarios, and w1+w2(+w3)=1. The deviation coefficient is calculated using the relative deviation method. , Where k is the deviation level coefficient; The characteristic baseline values ​​are: peak swallowing pressure 60-100 mmHg (median 80 mmHg), lowest blood oxygen saturation 98%, and a swallowing function assessment score of 10 (full marks). The technical criteria for determining the deviation level are as follows: relative deviation ≤5%, no deviation, k=0, C=1; 5% < relative deviation ≤15%, slight deviation, k=0.2, C=1.02~1.15; 15% < relative deviation ≤30%, moderate deviation, k=0.5, C=1.15~1.3; relative deviation >30%, severe deviation, k=0.8, C>1.3.

[0056] If the application scenario is a hospital or integrated medical and elderly care center, the real-time risk association factor is calculated based on swallowing pressure characteristics, minimum blood oxygen value characteristics, and swallowing function-specific assessment characteristics in the scale assessment information.

[0057] Specifically, the current usage scenario type is first determined by the login mode of the application interaction module. If the user logs in using the normal user login mode, the current scenario is determined to be home or elderly care institution scenario; if the user logs in using the medical staff login mode, the current scenario is determined to be hospital or integrated medical and elderly care center scenario. After scenario identification, the preset scenario-feature correspondence rules are invoked to select the core features used to calculate real-time risk correlation factors for that scenario. The core features selected differ in different scenarios because the care priorities, information completeness, and assessment needs vary in different scenarios. It is necessary to specifically select the features that best reflect the real-time aspiration risk to ensure the relevance and practicality of the correlation factor calculation.

[0058] When the current scenario is identified as home or elderly care facility, swallowing pressure and minimum blood oxygen saturation are selected as the sole criteria for calculating real-time risk correlation factors according to preset rules. The core reason for selecting these two features is that in home and elderly care facility scenarios, caregivers are mostly non-professionals, and the complexity of information collection needs to be controlled within a reasonable range. Swallowing pressure and minimum blood oxygen saturation are the most direct and easily captured real-time indicators of aspiration risk.

[0059] In the specific implementation process, swallowing pressure characteristics and minimum blood oxygen value characteristics are first screened from the unified characteristic data. The swallowing pressure characteristics include sub-characteristics such as peak swallowing pressure, duration of swallowing pressure, and stability of swallowing pressure fluctuations. These sub-characteristics can directly reflect the strength and coordination of the elderly person's current swallowing action. If the swallowing pressure is insufficient, the duration is too short, or the fluctuation is drastic, it will lead to an increased risk of aspiration. The minimum blood oxygen value characteristic refers to the lowest value of the elderly person's blood oxygen level within the current data collection period. When aspiration occurs, food or secretions blocking the airway will cause a sharp drop in blood oxygen level. Therefore, the minimum blood oxygen value is a key indirect indicator for judging the risk of real-time aspiration.

[0060] After identifying two core features, a pre-defined correlation factor calculation algorithm is invoked to quantify and calculate real-time risk correlation factors. First, the system database retrieves pre-defined baseline values ​​for swallowing pressure and minimum blood oxygen saturation in healthy elderly individuals. These baseline values ​​are constructed based on extensive physiological data from healthy elderly individuals and rehabilitation data from stroke patients, adapting to the physiological characteristics of elderly individuals at different ages and allowing for regular optimization and updates based on clinical practice and care experience. Subsequently, the deviation degree of the swallowing pressure and minimum blood oxygen saturation features from their corresponding baseline values ​​is calculated. The deviation degree is calculated using a relative deviation method, determining the deviation level based on the relative deviation magnitude and corresponding threshold range, categorizing it as no deviation, slight deviation, moderate deviation, or severe deviation. A corresponding deviation coefficient is assigned to each deviation level; the higher the deviation degree, the larger the deviation coefficient.

[0061] Specifically, the level is defined by setting a specific relative deviation percentage range: Swallowing pressure peak deviation: relative deviation within ±5% is no deviation (coefficient 1.0), ±5%-15% is slight deviation (coefficient 1.2), ±15%-30% is moderate deviation (coefficient 1.5), and more than ±30% is severe deviation (coefficient 1.8).

[0062] Minimum blood oxygen deviation: With the baseline value set at 98% (typical value at rest for healthy elderly people), a decrease of ≤2% (i.e., SpO2 ≥ 96%) can be considered as no significant deviation; a decrease of 3%-4% (SpO2 94%-95%) is considered as slight deviation; a decrease of 5%-6% (SpO2 92%-93%) is considered as moderate deviation; and a decrease of >6% (SpO2 < 92%) is considered as severe deviation.

[0063] In one specific embodiment, as shown in Table 3, with a baseline SpO2 of 98%, after aspiration following swallowing, the average SpO2 level decreased to 97.28 ± 2.7%, while swallowing without aspiration showed an average SpO2 decrease of 97.54 ± 2.7%. A total of 5 patients showed a SpO2 decrease of ≥2% in the presence of aspiration, while no such decrease was detected in swallowing without aspiration.

[0064] Table 3

[0065] Swallowing function score deviation: An increase of 0 grades from the baseline score (such as the post-stroke stable period score) is considered no deviation; an increase of 1 grade is considered slight deviation; an increase of 2 grades is considered moderate deviation; and an increase of ≥3 grades is considered severe deviation.

[0066] When the current scenario is identified as a hospital or integrated medical and elderly care center, in addition to the features selected for home and elderly care institution scenarios, a swallowing function-specific assessment feature from the scale assessment information is added as the third basis for calculating real-time risk association factors. In the specific implementation process, swallowing pressure and minimum blood oxygen saturation features are first screened from the unified feature data. Simultaneously, swallowing function-specific assessment features are extracted from the scale assessment information. These features specifically include sub-features such as swallowing function-specific scores, swallowing difficulty grading, and choking frequency assessment. These sub-features are obtained by medical staff through professional scale assessments and can accurately reflect the basic impairment and rehabilitation progress of the elderly's swallowing function.

[0067] After the three core features were selected, the preset correlation factor calculation algorithm was used, combined with the adaptation rules for the hospital scenario, to conduct quantitative calculations. First, the preset normal baseline values ​​were retrieved, and the deviations of the swallowing pressure feature and the lowest blood oxygen value feature from the corresponding baseline values ​​were calculated, and corresponding deviation coefficients were assigned. The calculation method was consistent with that for home and elderly care institution scenarios. Subsequently, for the swallowing function-specific assessment features, the preset specific assessment baseline level was retrieved, and the current swallowing function-specific assessment results were compared with the baseline level to determine the degree of deviation and assign deviation coefficients. If the specific assessment result is close to the baseline level, it indicates that the basic swallowing function is good, and a deviation coefficient of 1 is assigned; if the specific assessment result is slightly lower than the baseline level, it indicates that the basic swallowing function is slightly impaired, and a deviation coefficient of 1.2 is assigned; if the specific assessment result is significantly lower than the baseline level, it indicates that the basic swallowing function is severely impaired, and a deviation coefficient of 1.6 is assigned. After the deviation coefficients are assigned, the deviation coefficients of the three features are weighted and summed according to the preset weights of the hospital and medical-nursing center scenarios. The swallowing pressure feature has the highest weight, while the blood oxygen minimum value feature and the swallowing function special assessment feature have similar weights. The weighted summation yields the real-time risk correlation factor for this scenario.

[0068] In some implementations, S5 specifically includes the following sub-steps: S51. Based on the type of use scenario, call the corresponding preset risk grading standard. The preset risk grading standard includes the score range corresponding to different risk levels. S52. Determine the score range of the aspiration risk score, and determine the risk level corresponding to the score range as the final risk level.

[0069] S53. The core of determining the risk level is to call the appropriate preset risk grading standard according to the current usage scenario type, and match the calculated aspiration risk score with the score range in the standard to finally determine the risk level that reflects the current aspiration risk level of the elderly.

[0070] Specifically, firstly, based on the type of use scenario, the corresponding preset risk classification standard is invoked. During implementation, the identified use scenario type is used to directly invoke the preset scenario-classification standard correspondence within the system to quickly locate the preset risk classification standard required for the current scenario.

[0071] The pre-defined risk grading standards are pre-built by the system based on a large number of clinical aspiration risk cases and stroke patient rehabilitation care data, combined with care capabilities and risk prevention priorities in different scenarios. These standards are stored in the system database and can be regularly updated and optimized based on clinical practice and care experience. Specifically, the pre-defined risk grading standards adapted for home and elderly care institution scenarios adopt a simplified three-level grading model (low risk, medium risk, high risk) because caregivers in these scenarios are often non-professionals, and risk assessment needs to be simple, easy to understand, and quick to identify. Each risk level corresponds to a clear score range, with relatively broad intervals to avoid overly detailed grading that would increase the difficulty of understanding and operation for caregivers. The pre-defined risk grading standards adapted for hospital and integrated medical and elderly care center scenarios adopt a more detailed four-level grading model (low risk, low-to-medium risk, medium-to-high risk, high risk) because medical staff in these scenarios have professional assessment capabilities and need to accurately grasp the patient's risk level to develop personalized intervention plans. Each risk level corresponds to a more precise score range, with more detailed intervals that can more accurately reflect the patient's real-time aspiration risk fluctuations.

[0072] After invoking the preset risk grading criteria, the system determines the score range within which the aspiration risk score falls, and then identifies the risk level corresponding to that score range as the final risk level. During implementation, the system first retrieves the generated aspiration risk scores and loads all score ranges from the preset risk grading criteria for the current scenario. The aspiration risk scores are then matched against these ranges in ascending or descending order. During the matching process, a preset range matching algorithm compares each aspiration risk score with the corresponding score range for each level to determine if the score falls within a specific range.

[0073] If the aspiration risk score falls within a specific score range, the risk level corresponding to that range is directly determined as the final risk level. If the aspiration risk score is exactly at the boundary between two score ranges, a preset boundary judgment rule can be activated. This rule combines the core characteristics used in calculating the aspiration risk score, such as the degree of abnormality in dynamic physiological characteristics and the magnitude of real-time risk correlation factors, to further determine the final risk level. The boundary score is then classified as a higher-level risk level, ensuring the rigor of risk control and avoiding missed aspiration risks due to boundary judgment deviations.

[0074] In one specific embodiment, the aspiration risk score is based on a 100-point scale, with higher scores indicating a higher risk of aspiration. The score range is set based on the clinical aspiration incidence rate: low risk corresponds to an incidence rate ≤10%, medium-low / medium risk corresponds to an incidence rate ≤40% (10% < incidence rate), medium-high risk corresponds to an incidence rate ≤70% (40% < incidence rate), and high risk corresponds to an incidence rate ≥70%. The three-level grading score ranges for home and elderly care institution scenarios are: low risk 0-30 points, medium risk 31-60 points, and high risk 61-100 points; the four-level grading score ranges for hospital and integrated medical and elderly care center scenarios are: low risk 0-25 points, medium-low risk 26-45 points, medium-high risk 46-70 points, and high risk 71-100 points. Boundary score determination rule: if the aspiration risk score falls exactly at the boundary of the range (30 points or 60 points), it is classified as a higher risk level. Furthermore, if the real-time risk correlation factor R ≥ 1.2, it is directly determined to be a higher risk level.

[0075] Once the risk level is determined, the final risk level information will be temporarily stored.

[0076] In some implementations, S6 specifically includes the following sub-steps: S61. Based on the determined risk level, retrieve the preset early warning rules and intervention guidance database; S62. Output the corresponding early warning information according to the early warning rules, wherein the warning information intensity for high-risk levels is higher than that for medium- and low-risk levels; S63. Match intervention guidance content corresponding to risk level and usage scenario type from the intervention guidance library.

[0077] Specifically, firstly, based on the determined risk level, the pre-set early warning rules and intervention guidance library are retrieved. During implementation, the final determined risk level is retrieved from the S5 judgment results, and the current usage scenario type is reconfirmed. Then, through the system's internal pre-set association logic, the risk level, usage scenario type, and early warning rules and intervention guidance library are associated, and the appropriate content is retrieved.

[0078] The pre-set early warning rules and intervention guidance library are both built in advance by the system based on a large number of clinical care cases, experience in aspiration prevention and control of stroke patients, and care capabilities and needs in different scenarios. They can be regularly updated and optimized based on clinical practice, care feedback, and medical guidelines to ensure the scientific accuracy, practicality, and timeliness of the content. The early warning rules are mainly used to standardize the output format, intensity, and frequency of early warning information under different risk levels and scenarios. The core is to set differentiated early warning modes based on the risk level; the higher the risk level, the stronger the warning intensity and the higher the push frequency. At the same time, the presentation and output channels of early warning information are adjusted according to the characteristics of the usage scenarios to adapt to the usage habits of different users. The intervention guidance library is a clearly categorized collection of content, primarily stored according to risk level and usage scenario dimensions. It covers various guidance contents for aspiration prevention and control, including feeding guidance, swallowing training, body positioning, emergency treatment, and daily monitoring, with different emphases for guidance content corresponding to different risk levels and scenarios.

[0079] Specifically, the early warning rules are divided into two main scenario categories, adapted to home and elderly care institutions, and hospitals and integrated medical and elderly care centers, respectively. In the intervention guidance library, each risk level corresponds to specific guidance content, with optimized wording for each scenario. After retrieving the early warning rules and intervention guidance library, the corresponding early warning information is output according to the rules. The warning information for high-risk levels has a higher intensity than that for medium- and low-risk levels. Furthermore, the output format and content of the warning information are optimized based on the characteristics of the usage scenario. During implementation, firstly, based on the retrieved early warning rules, the current risk level, the corresponding warning intensity, push frequency, and output channel are determined. Then, the early warning information is generated according to a preset format, completing the output operation.

[0080] In home and elderly care facility settings, the main channel for issuing early warning information is the system terminal. The output formats include pop-up notifications, sound alerts, and push notifications. For low-risk levels, only a pop-up notification is issued, with concise content such as "Current aspiration risk is low; please maintain daily monitoring and basic prevention measures," without sound interference, and pushed once daily. For medium-risk levels, a pop-up notification plus a slight sound alert is issued, with content such as "Current aspiration risk is moderate; please strengthen feeding monitoring, adjust feeding position, and closely monitor swallowing status," with a shorter sound alert duration, and pushed twice daily. For high-risk levels, a pop-up notification plus continuous sound alerts plus high-frequency push notifications are issued, with content such as "Current aspiration risk is high; there is a risk of aspiration. Please immediately adjust feeding methods, strengthen monitoring, and contact medical staff if necessary." The sound alert can be manually turned off, and pushed every 3 hours until the risk level decreases. The pop-up information can also briefly indicate the core risk cause, such as "insufficient swallowing pressure" or "abnormal blood oxygen level," to help caregivers quickly locate the risk point.

[0081] In hospitals and integrated medical and elderly care centers, the output channels for early warning information include system terminals, medical staff terminals, and bedside call devices. The output format is more comprehensive. For low-risk levels, only text prompts are output on system terminals and medical staff terminals, such as "The patient's current risk of aspiration is low; it is recommended to maintain routine clinical monitoring and basic prevention and control measures," without any audio prompts. For medium-low and medium-high risk levels, text prompts and slight bedside audio prompts are output, including the risk level and core risk characteristics, such as "abnormal swallowing rhythm, slight fluctuations in blood oxygen," and are pushed to medical staff terminals and bedside devices for easy review by medical staff during rounds. For high-risk levels, text prompts, continuous bedside audio prompts, and emergency push notifications on medical staff terminals are output, with detailed information including the risk level, core risk evidence, and preliminary intervention recommendations. At the same time, the alarm devices at the nurse station are activated to issue a clear alarm, reminding medical staff to immediately respond. The push frequency is once every hour until the risk is controlled.

[0082] In addition, if the risk level changes, the corresponding early warning rules can be automatically retrieved to update the output format and content of the early warning information, ensuring that the early warning information can match the current risk status in real time.

[0083] The above description is merely a preferred embodiment of the present invention and the technical principles employed. The present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention.

Claims

1. A risk warning and intervention system for aspiration in elderly people after stroke, characterized in that, include: The hardware acquisition module is used to collect dynamic physiological data and static information of the elderly and transmit the collected dynamic physiological data and static information to the data processing module. The dynamic physiological data includes swallowing physiological data and blood oxygen data, and the static information includes scale assessment information and basic personal information. The data processing module is used to receive dynamic physiological data and static information collected by the hardware acquisition module, perform preprocessing and multimodal fusion processing to obtain unified feature data, and transmit the unified feature data to the algorithm calculation module. The algorithm calculation module is used to calculate the risk score of accidental inhalation based on unified feature data and through dynamic weight adjustment, and to determine the corresponding risk level according to the scenario type, and transmit the risk level to the application interaction module. The application interaction module provides user login and information entry functions, and outputs corresponding early warning information and intervention guidance content based on the risk level determined by the algorithm calculation module.

2. The aspiration risk warning and intervention system for elderly people after stroke according to claim 1, characterized in that, The hardware acquisition module includes: The intelligent swallowing monitoring spoon is used to collect swallowing physiological data from dynamic physiological data and transmit it to the data processing module. A portable pulse oximeter is used to collect blood oxygen data from dynamic physiological data and transmit it to a data processing module. The voice interaction device is used to collect scale assessment information and basic personal information from static information, and then transmits the collected information to the data processing module after converting it into structured data.

3. The aspiration risk warning and intervention system for elderly people after stroke according to claim 1, characterized in that, The application interaction module is configured to provide two login modes: a regular user login mode and a medical staff login mode. The regular user login mode is adapted for home and elderly care institution use scenarios and is configured to support the input of basic personal information and scale assessment information of the elderly, and to view the corresponding warning information and intervention guidance content; The login mode for medical staff is adapted to the use scenarios of hospitals and integrated medical and elderly care centers. It is configured to support the connection with medical systems, obtain the patient's historical medical information, and synchronously store the aspiration risk score, risk level and intervention execution record.

4. A method for early warning and intervention of aspiration risk in elderly people after stroke, characterized in that, The system for early warning and intervention of aspiration risk after stroke in the elderly, as described in any one of claims 1 to 3, comprises the following steps: S1. Obtain dynamic physiological data and static information of the elderly. Dynamic physiological data includes swallowing physiological data and blood oxygen data, while static information includes scale assessment information and basic personal information. S2. Preprocess the acquired dynamic physiological data and static information to remove invalid data and standardize the data scale; S3. Perform multimodal fusion processing on the preprocessed dynamic physiological data and static information to obtain unified feature data; S4. Based on unified feature data and combined with the corresponding usage scenario type, calculate the risk score of accidental aspiration through dynamic weight adjustment. S5. Determine the corresponding risk level based on the risk grading standard corresponding to the type of use scenario and the calculated aspiration risk score. S6. Based on the determined risk level, output the corresponding early warning information and intervention guidance.

5. The method for early warning and intervention of aspiration risk after stroke in the elderly according to claim 4, characterized in that, Step S3 includes: S31. Classify the preprocessed dynamic physiological data and static information respectively, and extract the corresponding core evaluation features; S32. The core evaluation features after classification are mapped to a unified dimensional space through a neural network to form initial feature data; S33. Based on the correlation between each core assessment feature and the risk of accidental aspiration, assign corresponding weights to the initial feature data, perform weighted fusion, and obtain unified feature data.

6. The method for early warning and intervention of aspiration risk after stroke in the elderly according to claim 5, characterized in that, In step S31, the specific method for extracting the core evaluation features is as follows: Swallowing-related features and blood oxygenation-related features are extracted from dynamic physiological data to form time-series features; Basic attribute features and scale scoring features are extracted from static information to form structured class features; Based on the type of use case, the specific dimensions of the structured features are determined. The structured features in hospital and integrated medical and elderly care center scenarios have more dimensions than those in home and elderly care institution scenarios.

7. The method for early warning and intervention of aspiration risk after stroke in the elderly according to claim 6, characterized in that, Step S4 includes: S41. Obtain the preset basic weight library, which includes the basic weights corresponding to each core evaluation feature. S42. Calculate real-time risk correlation factors based on dynamic physiological characteristics in unified feature data; S43. Based on the type of use scenario, determine the target features among the core assessment features that need to be adjusted in weight, and multiply the basic weights corresponding to the target features by the real-time risk correlation factors to obtain the dynamic weights. S44. Based on the quantitative values ​​and corresponding basic weights of the non-target features in the core assessment features, as well as the quantitative values ​​and corresponding dynamic weights of the target features, the aspiration risk score is calculated by summing them up.

8. The method for early warning and intervention of aspiration risk after stroke in the elderly according to claim 7, characterized in that, In step S42, the specific method for calculating the real-time risk correlation factor is as follows: If the application scenario is home and elderly care facility, the real-time risk association factor is calculated based on swallowing pressure characteristics and blood oxygen minimum value characteristics; If the application scenario is a hospital or integrated medical and elderly care center, the real-time risk association factor is calculated based on swallowing pressure characteristics, minimum blood oxygen value characteristics, and swallowing function-specific assessment characteristics in the scale assessment information.

9. The method for early warning and intervention of aspiration risk after stroke in the elderly according to claim 4, characterized in that, Step S5 includes: S51. Based on the type of use scenario, call the corresponding preset risk grading standard. The preset risk grading standard includes the score range corresponding to different risk levels. S52. Determine the score range of the aspiration risk score, and determine the risk level corresponding to the score range as the final risk level.

10. The method for early warning and intervention of aspiration risk after stroke in the elderly according to claim 4, characterized in that, Step S6 includes: S61. Based on the determined risk level, retrieve the preset early warning rules and intervention guidance database; S62. Output the corresponding early warning information according to the early warning rules, wherein the warning information intensity for high-risk levels is higher than that for medium- and low-risk levels; S63. Match intervention guidance content corresponding to risk level and usage scenario type from the intervention guidance library.

Images