Inpatient falling intelligent early warning and protection method and system based on multi-modal data fusion

Abstract

The invention provides an intelligent early warning and protection method and system for falling of an inpatient based on multi-modal data fusion. According to the method, individual risk data, behavior posture data and equipment state data of an inpatient are collected, a tumble risk index of the patient is calculated from the three aspects of potential risk, behavior risk and environmental risk by using a tumble risk prediction model, the tumble risk level of the patient is comprehensively evaluated, and then corresponding protection measures are taken. Specifically, each risk factor is quantified through a weight coefficient and a risk mapping function, and finally, an S-type normalization function is utilized to obtain a comprehensive risk index, so that accurate and personalized early warning is realized. The system comprises a sensing layer, a data processing layer, an intelligent analysis layer and an intervention execution layer, and can realize full-process automation and intellectualization. According to the invention, the problems of inaccurate evaluation, incapability of personalized early warning, lack of active intervention and the like in the prior art are effectively solved, and the occurrence rate of tumble of the inpatient is remarkably reduced.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare technology, and more specifically, to a method and system for intelligent early warning and protection against falls in hospitalized patients based on multimodal data fusion. Background Technology

[0002] In hospital settings, falls among hospitalized patients are a long-standing and urgent problem. They not only cause secondary injuries such as fractures and intracranial hemorrhage, but also lead to prolonged hospital stays, increased medical costs, and a decline in quality of life. The risk of falls is even higher and the harm more severe for high-risk groups such as elderly patients, post-operative recovery patients, long-term bedridden patients, and patients using sedative medications. According to relevant statistics, the incidence of falls among hospitalized patients varies across different departments and hospitals, but overall, the incidence remains high, and once a fall occurs, the subsequent treatment and rehabilitation become correspondingly more difficult.

[0003] Among existing technologies, some attempts exist to provide early warning and protection against falls in hospitalized patients, but all have varying degrees of limitations. For example, passive alarm systems based on environmental sensors employ methods such as bed pads, infrared beams, or simple video motion detection. An alarm is triggered when a patient leaves the bed or enters a specific area. However, these systems have drawbacks: normal operations by nursing staff or visits from family members can easily lead to false alarms; and a patient slowly sliding off the bed or making only slight movements near the bed may not trigger an alarm, resulting in missed alarms. Alarming only after the patient has completed the "leaving the bed" action is a reactive intervention, unable to predict or provide early warning of fall risks. Furthermore, they cannot distinguish the different risk levels inherent in the same behavior of different patients (such as a healthy young person versus a frail elderly person).

[0004] For example, wearable device-based solutions require patients to wear wristbands with built-in accelerometers and gyroscopes to detect falls by detecting sudden changes in body posture. However, these solutions have drawbacks: many patients (such as those with altered mental status, sensitive skin, or agitation) are unwilling or unable to wear them, hindering clinical implementation. Daily activities (such as sitting down quickly or bending over to pick something up) may be misinterpreted as falls, while slow, gradual falls may not be effectively identified. Furthermore, the devices cannot obtain crucial environmental information such as whether the patient is near the bed or whether the floor is slippery.

[0005] Furthermore, some existing technical solutions have shortcomings in data fusion and analysis. They often focus only on a single type of data, such as the patient's physiological or behavioral data, while neglecting the importance of multimodal data fusion. In reality, a patient's fall risk is influenced by a combination of factors, including the patient's own physiological condition, behavioral habits, and environmental conditions. Only by effectively fusing and analyzing this multimodal data can a more comprehensive and accurate assessment of a patient's fall risk be made.

[0006] Therefore, there is an urgent need for a multimodal data fusion method that can comprehensively consider individual patient risks, behavioral risks, and environmental risks, to achieve accurate, personalized, and prospective prediction of fall risk for hospitalized patients, and to take corresponding proactive intervention measures to effectively reduce the incidence of falls among hospitalized patients and ensure their safety and health. Summary of the Invention

[0007] The present invention aims to solve at least one of the aforementioned technical problems existing in the prior art.

[0008] Therefore, the first aspect of the present invention provides a method for intelligent early warning and protection of hospitalized patients' falls based on multimodal data fusion.

[0009] A second aspect of the present invention provides an intelligent early warning and protection system for hospitalized patients' falls based on multimodal data fusion.

[0010] This invention provides a method for intelligent early warning and protection of hospitalized patients' falls based on multimodal data fusion, comprising: Collect fall-related data of hospitalized patients; the fall-related data includes individualized patient risk data, patient behavioral posture data, and equipment status data; wherein, the individualized patient risk data includes the patient's basic information and medical data; the patient behavioral posture data includes the patient's temporal behavioral data; and the equipment status data includes bed status data; Based on fall-related data, a fall risk prediction model is used to calculate the patient's fall risk index. The fall risk is calculated from three levels: potential risk, behavioral risk, and environmental risk. The combined fall risk calculation results from the three levels are used to obtain a comprehensive risk index that represents the patient's current fall risk. Based on the calculated comprehensive risk index, the patient's fall risk level is assessed and corresponding protective measures are taken.

[0011] The intelligent early warning and protection method for hospitalized patients based on multimodal data fusion according to the above-described technical solution of the present invention may also have the following additional technical features: In the above technical solutions, the calculation of patient fall risk from a potential risk perspective includes: A baseline risk score is calculated based on risk factors affecting the probability of falls from individualized patient risk data; the calculation method includes:

[0012] in, represents the baseline risk score; n represents the number of risk factors in the patient's individualized risk data; i represents the index of the risk factor; This represents the weight coefficient of the i-th risk factor; This represents the value of the i-th risk factor; Let represent the factor mapping function corresponding to the i-th risk factor, which is used to quantify the degree of influence of the risk factor on the risk of falling.

[0013] In the above technical solutions, the calculation of patient fall risk from a behavioral risk perspective includes: Monitoring and statistical analysis of abnormal behaviors of patients within a statistical period based on patient behavior and posture data; The behavioral risk score of patients within the statistical period is calculated based on the statistical results of abnormal behavior; the calculation method includes:

[0014] in, The value represents the behavioral risk score; T represents the monitoring time window duration, i.e., the statistical period; t represents the time index; m represents the number of abnormal behavior types; j represents the abnormal behavior index. The weight coefficient represents the j-th type of abnormal behavior; This represents the monitoring result of the patient's type j abnormal behavior at time t; A risk mapping function representing the j-th type of abnormal behavior.

[0015] In the above technical solution, the abnormal behaviors include getting up quickly at night, poor sitting balance, and unsteady gait; Among them, based on whether the rate of change of bed pressure at the current moment is greater than a set threshold, it is determined whether the patient has abnormal behavior of getting up quickly at night; Based on whether the duration of the patient's body posture angular velocity being greater than a set threshold at the current moment is greater than a set duration, it is determined whether the patient has abnormal behavior of poor sitting balance. Based on whether the patient's current step length standard deviation is greater than a set threshold, it is determined whether the patient has abnormal behavior of unsteady gait.

[0016] In the above technical solutions, the calculation of patient fall risk from an environmental risk perspective includes: The environmental risk score is calculated based on the bed status data from the equipment status data. The calculation method includes:

[0017] in, This represents the environmental risk score.

[0018] In the above technical solution, the factor mapping function adopts a step function or a normalization function; And / or, the risk mapping function employs a step function.

[0019] In the above technical solution, the comprehensive risk index representing the patient's current fall risk is obtained by combining the fall risk calculation results from three levels, including:

[0020] in, This represents the overall risk index; Indicates the basic risk weight; Indicates the behavioral risk weight; Indicates environmental risk weights; Represents the sigmoid normalization function;

[0021] Where x is the input variable representation of the sigmoid normalization function; Indicates the steepness parameter; This indicates the risk threshold offset.

[0022] In the above technical solution, the sum of the basic risk weight, behavioral risk weight, and environmental risk weight is equal to 1; Furthermore, different baseline risk weights, behavioral risk weights, and environmental risk weights are selected based on the characteristics of the patient's ward.

[0023] In the above technical solution, based on the patient's comprehensive risk index, the patient's fall risk is divided into low risk, medium risk or high risk, and corresponding protective measures are taken. For patients whose comprehensive risk index is in the low-risk range, record the data and do not take active intervention. For patients whose comprehensive risk index is in the medium-risk range, early warning measures will be taken to provide reminders and assistance; For patients whose comprehensive risk index is in the high-risk range, proactive protective measures are taken, including sound and light warnings and environmental interventions.

[0024] This invention provides an intelligent early warning and protection system for hospitalized patients' falls based on multimodal data fusion, comprising: The perception layer collects fall-related data of hospitalized patients based on the hospital information system, sensors deployed in wards, and smart beds; The data processing layer receives multi-source data from the perception layer and cleans, aligns, and fuses the data. The intelligent analysis layer, based on the output data of the data processing layer, uses the intelligent early warning and protection method for hospitalized patients based on multimodal data fusion, as described in any of the above technical solutions, to assess the fall risk of patients and generate intervention instructions based on the assessment results; The intervention execution layer executes intervention commands based on the intelligent analysis layer to prevent patients from falling.

[0025] In summary, due to the adoption of the above-mentioned technical features, the beneficial effects of the present invention are: First, this invention comprehensively reflects all aspects of patient fall risk by collecting multimodal data, including individualized patient risk data, patient behavioral and postural data, and equipment status data. Individualized patient risk data encompasses basic patient information and medical data, reflecting the patient's physiological condition and health background, providing crucial evidence for assessing basic risk. Patient behavioral and postural data monitors patient behavior in real time, promptly identifying abnormal behaviors that may lead to falls, thus enabling accurate assessment of behavioral risk. Environmental information, such as bed status data, within the equipment status data reflects the patient's surrounding environment, thereby accurately assessing environmental risk. This fusion of multimodal data makes the assessment of patient fall risk more comprehensive and accurate, avoiding the inaccurate assessments caused by single data sets in existing technologies.

[0026] Secondly, this invention employs a fall risk prediction model to calculate patient fall risk from three levels: potential risk, behavioral risk, and environmental risk. The results from these three levels are then combined to obtain a comprehensive risk index. This hierarchical calculation and comprehensive assessment method allows for a more detailed and in-depth analysis of the various factors contributing to patient fall risk, making the assessment results more scientific and reliable. Furthermore, by setting different weighting coefficients and risk mapping functions, each risk factor can be quantified and assessed in a personalized manner based on the specific circumstances of different patients and the characteristics of the ward. This achieves accurate and personalized prediction of patient fall risk, overcoming the limitation of existing technologies that cannot distinguish between different patient risk levels.

[0027] Furthermore, this invention assesses patients' fall risk levels based on a comprehensive risk index and implements corresponding protective measures. Different intervention strategies are adopted for patients with different risk levels, such as no intervention for low-risk patients (data recording), early warning reminders for medium-risk patients, and proactive protection for high-risk patients. This tiered intervention approach can rationally allocate medical resources according to the actual risk situation of patients, improve the targeting and effectiveness of interventions, and effectively reduce the incidence of falls among hospitalized patients.

[0028] In summary, this invention enables full automation and intelligentization of the entire process from data collection to risk assessment and intervention execution. It not only improves work efficiency but also reduces human interference, further enhancing the accuracy and reliability of fall warning and prevention. Through multimodal data fusion and intelligent analysis, this invention achieves accurate, personalized, and forward-looking prediction of fall risk in hospitalized patients and implements corresponding proactive intervention measures, effectively solving the problems existing in current technologies and possessing significant clinical application value and social significance.

[0029] Additional aspects and advantages of the invention will become apparent in the following description or may be learned by practice of the invention. Attached Figure Description

[0030] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart of an embodiment of the present invention: an intelligent early warning and protection method for hospitalized patients' falls based on multimodal data fusion; Figure 2 This is a schematic diagram of the three-layer risk assessment logic in an intelligent early warning and protection method for hospitalized patients based on multimodal data fusion, according to an embodiment of the present invention. Figure 3 This is an architecture diagram of an intelligent early warning and protection system for hospitalized patients based on multimodal data fusion, according to an embodiment of the present invention. Detailed Implementation

[0031] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0032] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0033] The following reference Figures 1 to 3 This invention describes a method and system for intelligent early warning and protection of hospitalized patients based on multimodal data fusion, provided by some embodiments of the present invention.

[0034] Some embodiments of this application provide a method for intelligent early warning and protection of hospitalized patients' falls based on multimodal data fusion.

[0035] like Figure 1As shown, the first embodiment of the present invention proposes an intelligent early warning and protection method for hospitalized patients' falls based on multimodal data fusion, including the following steps S1 to S3.

[0036] S1. Collect fall-related data of hospitalized patients; the fall-related data includes individualized risk data of patients, patient behavior and posture data, and equipment status data; wherein, the individualized risk data of patients includes the patient's own behavior; the patient behavior and posture data includes the patient's temporal behavior data; and the equipment status data includes bed status data.

[0037] Specifically, individualized patient risk data mainly includes basic patient information and medical data obtained from the hospital information system, including age, primary diagnosis, history of falls, and records of medication use that may cause falls, such as sedatives, antihypertensives, and diuretics. Patient behavioral and postural data is acquired non-contactly in real-time using millimeter-wave radar or ToF cameras deployed on the ward ceiling, capturing temporal behavioral data such as patient posture, position, movement speed, and gait stability. Equipment status data includes bed status data, specifically the status of bed rails, bed height, and head and foot angles. This type of data is typically obtained through smart beds configured in the hospital, but can also be acquired through additional sensors.

[0038] S2. Based on fall-related data, a fall risk prediction model will be used to calculate the patient's fall risk index; among which, such as Figure 2 As shown, the patient's fall risk is calculated from three levels: potential risk, behavioral risk, and environmental risk. The combined fall risk calculation results from the three levels are used to obtain a comprehensive risk index representing the patient's current fall risk.

[0039] In some embodiments, calculating the risk of a patient falling from a potential risk perspective includes: A baseline risk score is calculated based on risk factors affecting the probability of falls from individualized patient risk data; the calculation method includes:

[0040] in, represents the baseline risk score; n represents the number of risk factors in the patient's individualized risk data; i represents the index of the risk factor; This represents the weight coefficient of the i-th risk factor. =1; This represents the value of the i-th risk factor; Let represent the factor mapping function corresponding to the i-th risk factor, which is used to quantify the degree of influence of the risk factor on the risk of falling.

[0041] Specifically, in individualized risk data for patients, not all data needs to be calculated. Instead, data related to falls are selected, such as age, fall history, and use of sedative medications.

[0042] The factor mapping function can be a step function or a normalized function. In one specific embodiment, the factor mapping function uses a step function. For example, for the age risk factor, when the patient is older than 65 years old, the mapping result of the factor mapping function is 1, otherwise it is 0, and the corresponding weight coefficient is configured as 0.3; for the fall history risk factor, when the patient has fallen within the past year, the mapping result of the factor mapping function is 1, otherwise it is 0, and the corresponding weight coefficient is configured as 0.4; for the sedation drug use risk factor, when the patient is currently using sedation drugs, the mapping result of the factor mapping function is 1, otherwise it is 0, and the corresponding weight coefficient is configured as 0.3.

[0043] In some embodiments, calculating the patient's fall risk from a behavioral risk perspective includes: Monitoring and statistical analysis of abnormal behaviors of patients within a statistical period based on patient behavior and posture data; The behavioral risk score of patients within the statistical period is calculated based on the statistical results of abnormal behavior; the calculation method includes:

[0044] in, The value represents the behavioral risk score; T represents the monitoring time window duration, i.e., the statistical period; t represents the time index; m represents the number of abnormal behavior types; j represents the abnormal behavior index. The weight coefficient represents the j-th type of abnormal behavior; This represents the monitoring result of the patient's type j abnormal behavior at time t; A risk mapping function representing the j-th type of abnormal behavior.

[0045] In some embodiments, the abnormal behaviors include getting up quickly at night, poor sitting balance, and unsteady gait; of course, the types of abnormal behaviors can be increased or decreased according to the actual situation, such as adding body swaying after standing.

[0046] In this embodiment, the abnormal behavior of a patient rapidly getting up at night is determined based on whether the rate of change of bed pressure at the current moment is greater than a set threshold. For example, based on the rate of change of bed pressure in the patient's behavioral posture data, a rapid nighttime getting-up behavior is counted once when the rate of change of bed pressure is >0.8m / s².

[0047] Based on whether the duration of the patient's body posture angular velocity being greater than a set threshold at the current moment is greater than the set duration, it is determined whether the patient has abnormal behavior of poor sitting balance; for example, based on the posture angular velocity in the patient's behavioral posture data, when the posture angular velocity is >20° / s and the duration exceeds 3 seconds, it is counted as one instance of poor sitting balance behavior.

[0048] Based on whether the patient's current step length standard deviation is greater than a set threshold, it is determined whether the patient has abnormal behavior of unsteady gait; for example, based on the step length standard deviation in the patient's behavioral posture data, when the step length standard deviation is >0.3m, unsteady gait behavior is counted once.

[0049] In this embodiment, the abnormal behavior monitoring results The value can be 0 or 1, meaning that if there are valid counts of this type of abnormal behavior at the current moment, The value is 1 if it is not 0 otherwise.

[0050] In one specific embodiment, the risk mapping function adopts a step function, that is, the output is 0 when the input is ≤0 and the output is 1 when the input is >0.

[0051] It should be noted that time t is the smallest time step, usually in seconds or minutes.

[0052] In some embodiments, calculating the risk of patient falls from an environmental risk perspective includes: The environmental risk score is calculated based on the bed status data from the equipment status data. The calculation method includes:

[0053] in, This represents the environmental risk score.

[0054] It should be noted that in this embodiment, the environmental risk aspect mainly considers the impact of bed data on the risk of patient falls. In other embodiments, consideration of whether the ground is slippery may also be added.

[0055] In some embodiments, the combined fall risk calculation results from the three levels yield a comprehensive risk index representing the patient's current fall risk, including:

[0056] in, This represents the overall risk index; Indicates the basic risk weight; Indicates the behavioral risk weight; Indicates environmental risk weights; Represents the sigmoid normalization function;

[0057] Where x is the input variable representation of the sigmoid normalization function; Indicates the steepness parameter; This indicates the risk threshold offset.

[0058] Specifically, the sum of the basic risk weight, behavioral risk weight, and environmental risk weight equals 1.

[0059] In some embodiments, different baseline risk weights, behavioral risk weights, and environmental risk weights are selected based on the characteristics of the patient's ward. For example, the behavioral risk weight should be appropriately increased for patients in orthopedic wards. Specifically, the allocation of baseline risk weights, behavioral risk weights, and environmental risk weights based on the characteristics of the patient's ward can be done manually or dynamically adjusted using an online learning algorithm.

[0060] S3. Assess the patient's fall risk level based on the calculated comprehensive risk index and take corresponding protective measures.

[0061] In some embodiments, based on the patient's comprehensive risk index, the patient's fall risk is classified as low risk, medium risk, or high risk, and corresponding protective measures are taken.

[0062] Specifically, in the above embodiments, the comprehensive risk index is quantified as a percentage score to characterize the patient's fall risk. This allows for the division of risk ranges: a comprehensive risk index of 0-40 corresponds to low risk, 41-70 to medium risk, and 71-100 to high risk. The assessment results can be displayed using different colors to clearly distinguish different risk levels; for example, green for low risk, yellow for medium risk, and red for high risk.

[0063] For patients whose comprehensive risk index is in the low-risk range, record the data and do not take active intervention. For patients whose comprehensive risk index is in the medium-risk range, early warning measures are taken to provide reminders and assistance. The reminders and assistance methods include the system automatically sending reminder information to the nurse's mobile terminal, such as "Patient in bed 301 is getting up and moving around, need to be monitored"; at the same time, the soft guide lights on the ward floor can be automatically turned on.

[0064] For patients whose comprehensive risk index falls within the high-risk range, proactive protective measures are implemented, including audible and visual warnings and environmental interventions. These proactive measures include: issuing gentle voice reminders in the ward, such as "Please slow down and hold on tight before walking," while adjusting the lighting appropriately; automatically lowering the bed height to the lowest safe position and raising the bed rail on the patient's side (if clinically permissible); and sending a high-priority alarm to the nursing station, specifying the cause of the risk (e.g., "unsteady gait, high risk of fall").

[0065] In one specific embodiment, let's take "an elderly patient (>65 years old) taking sedative medication getting up at night" as an example: Millimeter-wave radar detected the patient's change from a lying to a sitting position and calculated that the speed of getting up was relatively fast. The smart bed reported, "The bed rails are in the lowered state." The server retrieved the patient's information from the HIS: "Age 85, taking estazolam."

[0066] Based on the factors of advanced age, sedation, and a history of falls, a static baseline risk score was calculated. A dynamic behavioral risk score was calculated based on the patient's rapid nighttime arousal. An environmental risk score was calculated by combining this with the observation that the bed rails were not raised. The final model calculated the patient's overall risk index to be 85, classifying them as high-risk.

[0067] The system automatically executed the command, immediately sending the instruction to the smart bed: "Lower the bed height to the lowest level and raise the near-side bed rails." Simultaneously, it triggered a voice reminder: "Please sit on the edge of the bed for a while before getting up," and turned on the floor lamp. The system issued an alarm, sending a red alert to the on-duty nurse's PDA: "Wang, bed 301, high risk! Rapidly getting up at night; automatic protection activated. Please verify immediately!"

[0068] Other embodiments of the present invention provide an intelligent early warning and protection system for hospitalized patients' falls based on multimodal data fusion, such as... Figure 3 As shown, it mainly includes: perception layer, data processing layer, intelligent analysis layer and intervention execution layer.

[0069] The perception layer, based on the hospital information system, sensors deployed in wards, and smart beds, collects fall-related data of hospitalized patients. Specifically, the perception layer includes hardware devices deployed in the wards and the HIS system. Millimeter-wave radar / ToF sensors are used to detect patient behavior and posture. Smart beds provide bed status data. The Hospital Information System (HIS) interface is used to obtain personalized patient medical data.

[0070] The data processing layer receives multi-source data from the perception layer and cleans, aligns, and fuses the data. Specifically, the data processing layer includes an edge computing gateway / data preprocessing module, which receives multi-source data from the perception layer, cleans, aligns, and fuses the data, and packages it into a unified data packet format.

[0071] The intelligent analysis layer, based on the output data of the data processing layer, uses the intelligent early warning and protection method for hospitalized patients based on multimodal data fusion as described in any of the above embodiments to assess the fall risk of patients and generate intervention instructions based on the assessment results; that is, the intelligent analysis layer is equipped with a central server for deploying fall risk prediction models.

[0072] The intervention execution layer executes measures to prevent patient falls based on intervention instructions from the intelligent analysis layer. Specifically, the intervention execution layer includes the nurse station monitoring center, mobile nursing terminal (PDA), ward voice prompts, smart bed controllers, and lighting controllers, which receive intervention instructions issued by the parallel intelligent analysis layer; users can provide manual confirmation / feedback at the nurse station monitoring center and mobile nursing terminal (PDA).

[0073] The data transmission relationship between the various hardware components is as follows: the millimeter-wave radar and smart bed send data to the edge computing gateway via Wi-Fi or Ethernet. The gateway communicates with the central server through the hospital's intranet. After analysis, the server sends intervention commands via the network to the voice player, lighting controller, smart bed controller, and nurses' PDAs in the target ward.

[0074] In this specification, the illustrative expressions of the terms used do not necessarily refer to the same embodiments or examples. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0075] Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this invention shall be included within the scope of protection of this invention.

Claims

1. A method for intelligent early warning and protection of hospitalized patients' falls based on multimodal data fusion, characterized in that, include: Collect fall-related data of hospitalized patients; the fall-related data includes individualized patient risk data, patient behavioral posture data, and equipment status data; wherein, the individualized patient risk data includes the patient's basic information and medical data; the patient behavioral posture data includes the patient's temporal behavioral data; and the equipment status data includes bed status data; Based on fall-related data, a fall risk prediction model is used to calculate the patient's fall risk index. The fall risk is calculated from three levels: potential risk, behavioral risk, and environmental risk. The combined fall risk calculation results from the three levels are used to obtain a comprehensive risk index that represents the patient's current fall risk. Based on the calculated comprehensive risk index, the patient's fall risk level is assessed and corresponding protective measures are taken.

2. The intelligent early warning and protection method for hospitalized patients based on multimodal data fusion according to claim 1, characterized in that, Calculating the risk of patient falls from a potential risk perspective includes: A baseline risk score is calculated based on risk factors affecting the probability of falls from individualized patient risk data; the calculation method includes: in, represents the baseline risk score; n represents the number of risk factors in the patient's individualized risk data; i represents the index of the risk factor; This represents the weight coefficient of the i-th risk factor; This represents the value of the i-th risk factor; Let represent the factor mapping function corresponding to the i-th risk factor, which is used to quantify the degree of influence of the risk factor on the risk of falling.

3. The intelligent early warning and protection method for hospitalized patients based on multimodal data fusion according to claim 2, characterized in that, Calculating patient fall risk from a behavioral risk perspective includes: Monitoring and statistical analysis of abnormal behaviors of patients within a statistical period based on patient behavior and posture data; The behavioral risk score of patients within the statistical period is calculated based on the statistical results of abnormal behavior; the calculation method includes: in, The value represents the behavioral risk score; T represents the monitoring time window duration, i.e., the statistical period; t represents the time index; m represents the number of abnormal behavior types; j represents the abnormal behavior index. The weight coefficient represents the j-th type of abnormal behavior; This represents the monitoring result of the patient's type j abnormal behavior at time t; A risk mapping function representing the j-th type of abnormal behavior.

4. The intelligent early warning and protection method for hospitalized patients based on multimodal data fusion according to claim 3, characterized in that, The abnormal behaviors include getting up quickly at night, poor sitting balance, and unsteady gait; Among them, based on whether the rate of change of bed pressure at the current moment is greater than a set threshold, it is determined whether the patient has abnormal behavior of getting up quickly at night; Based on whether the duration of the patient's body posture angular velocity being greater than a set threshold at the current moment is greater than a set duration, it is determined whether the patient has abnormal behavior of poor sitting balance. Based on whether the patient's current step length standard deviation is greater than a set threshold, it is determined whether the patient has abnormal behavior of unsteady gait.

5. The intelligent early warning and protection method for hospitalized patients based on multimodal data fusion according to claim 3, characterized in that, Calculating patient fall risk from an environmental risk perspective includes: The environmental risk score is calculated based on the bed status data from the equipment status data. The calculation method includes: in, This represents the environmental risk score.

6. The intelligent early warning and protection method for hospitalized patients based on multimodal data fusion according to claim 3, characterized in that, The factor mapping function adopts a step function or a normalization function; And / or, the risk mapping function employs a step function.

7. The intelligent early warning and protection method for hospitalized patients based on multimodal data fusion according to claim 5, characterized in that, The combined fall risk calculation results from the three levels yield a comprehensive risk index representing the patient's current fall risk, including: in, This represents the overall risk index; Indicates the basic risk weight; Indicates the behavioral risk weight; Indicates environmental risk weights; Represents the sigmoid normalization function; Where x is the input variable representation of the sigmoid normalization function; Indicates the steepness parameter; This indicates the risk threshold offset.

8. The intelligent early warning and protection method for hospitalized patients based on multimodal data fusion according to claim 7, characterized in that, The sum of the basic risk weight, behavioral risk weight, and environmental risk weight is equal to 1; Furthermore, different baseline risk weights, behavioral risk weights, and environmental risk weights are selected based on the characteristics of the patient's ward.

9. The intelligent early warning and protection method for hospitalized patients based on multimodal data fusion according to claim 7, characterized in that, Based on the patient's comprehensive risk index, the patient's fall risk is classified as low risk, medium risk, or high risk, and corresponding protective measures are taken. For patients whose comprehensive risk index is in the low-risk range, record the data and do not take active intervention. For patients whose comprehensive risk index is in the medium-risk range, early warning measures will be taken to provide reminders and assistance; For patients whose comprehensive risk index is in the high-risk range, proactive protective measures are taken, including sound and light warnings and environmental interventions.

10. A smart early warning and protection system for hospitalized patients' falls based on multimodal data fusion, characterized in that, include: The perception layer collects fall-related data of hospitalized patients based on the hospital information system, sensors deployed in wards, and smart beds; The data processing layer receives multi-source data from the perception layer and cleans, aligns, and fuses the data. The intelligent analysis layer, based on the output data of the data processing layer, uses the intelligent early warning and protection method for hospitalized patients based on multimodal data fusion as described in any one of claims 1 to 9 to assess the fall risk of patients and generate intervention instructions based on the assessment results; The intervention execution layer executes intervention commands based on the intelligent analysis layer to prevent patients from falling.

Images