A hospital patient dangerous behavior intent recognition and restraint control system

Abstract

The application discloses a hospital patient dangerous behavior intention recognition and restraint control system, comprising an electroencephalogram signal acquisition module, a restraint system and a processing unit; the processing unit pre-processes multi-channel electroencephalogram signals containing central area and occipital scalp leads, calculates power spectral density in a sliding time window, extracts relative delta power, spectral slowening ratio, precursor trend items of posterior head dominant rhythm peak frequency and alpha band functional connection strength as state characteristics, fuses the individualized baseline normalized characteristics into delirium precursor index, judges the patient state as sleep, quiet wakefulness, delirium precursor and restlessness in combination with sleep electroencephalogram criteria; the restraint is triggered only when delirium precursor or restlessness occurs; and the restraint is released after a plurality of time windows without risk. The application recognizes behavior intention before dangerous behaviors such as extubation, falling, self-injury and attacking others occur, and intervenes in advance, thereby shortening the restraint time length and taking into account patient safety and comfort.

Description

Technical Field

[0001] This invention relates to the field of medical devices, specifically to a hospital patient dangerous behavior intention recognition and restraint control system. Background Technology

[0002] Hospitalized patients in a delirious or agitated state may engage in dangerous behaviors such as removing life support tubes, falling from bed, self-harm, or even attacking others, directly endangering the safety of themselves and medical staff. For example, unplanned self-extubation in intensive care (ICU) patients can directly endanger their lives; falls and self-harm incidents caused by delirium are also common among elderly, post-operative, and neurological hospitalized patients. Clinically, restraint devices are often used to implement protective restraints for high-risk patients, and patients are often restrained continuously when no one is specifically monitoring them. However, continuous restraint causes patient discomfort, skin damage, and exacerbates agitation, directly conflicting with the principle of minimum restraint. If the patient's behavioral intent can be identified before the dangerous behavior occurs, and restraint is applied only when the intent is present and removed when the intent is absent, both patient safety and comfort can be ensured.

[0003] However, existing technologies struggle to reliably identify a patient's behavioral intent before dangerous behavior occurs: First, dangerous behavior identification schemes based on video monitoring or motion sensing can only detect such behavior after the patient has already begun actions such as raising their hand to grab a tube or getting out of bed, lacking lead time, meaning the dangerous behavior often has already taken place by the time intervention begins; furthermore, the presence of tubes, bedding, and medical staff in the ward severely obstructs the view, making identification susceptible to light and occlusion, resulting in high false alarm and false negative rates. Second, EEG-based consciousness monitoring schemes are mostly used for assessing sedation depth or detecting EEG abnormalities, and their criteria rely on population statistical thresholds. However, hospitalized patients vary greatly in age, sedation depth, and baseline EEG data, making population thresholds difficult to apply to individual patients; moreover, the EEG data of sleeping and comatose patients also show a slowed background with increased delta power, and the slowed-down characteristic alone cannot distinguish between physiological slow waves and pathological slowed-down patterns that precede delirium, directly using this feature to trigger restraints will result in false restraints of sleeping patients. Third, the above monitoring schemes are all independent monitoring or alarm links, and do not form a closed-loop control with the restraint system to automatically implement and release restraint according to the patient's status, thus failing to resolve the conflict between the principle of full restraint and the principle of minimum restraint.

[0004] Therefore, there is a need for a system that can identify a patient's intention to engage in dangerous behavior before it occurs and automatically implement or remove restraints based on the patient's condition. Summary of the Invention

[0005] To address the above problems, the present invention proposes the following technical solution:

[0006] A hospital patient risky behavior intention recognition and restraint control system includes an electroencephalogram (EEG) signal acquisition module, a restraint system, and a processing unit. The processing unit controls the restraint system to restrain the patient based on signals acquired by the EEG signal acquisition module. The control method of the processing unit includes the following:

[0007] S1: Receives raw multi-channel signals from the EEG signal acquisition module. ,right Perform preprocessing to obtain preprocessed multi-channel signals. ;

[0008] in, Including central scalp leads and occipital scalp leads, Indicates the sampling time;

[0009] S2: According to The power spectral density of each lead is calculated within the sliding time window.

[0010] The first time window within the sliding time window The first lead and the first The power spectral density of each lead is expressed as ;when hour Represents the self-power spectral density of the lead, when hour This represents the cross-power spectral density between leads; , Indicates the lead number; Indicates frequency;

[0011] Calculate the average relative power for each frequency band:

[0012]

[0013] in, express Time and frequency band The average relative power is the average of the relative power of each lead;

[0014] , Indicates the frequency band name. Represents a set of frequency bands. , , ;

[0015] Indicates the lead number. Indicates the total number of leads;

[0016] Calculate the spectral slowdown ratio based on the average relative power of each frequency band:

[0017]

[0018] in, express The larger the value of the spectral slowdown ratio at any given time, the slower the background EEG rhythm.

[0019] Calculate the peak frequency of the dominant posterior head rhythm:

[0020]

[0021]

[0022] in, express Peak frequency of the dominant posterior head rhythm at specific moments;

[0023] The power spectral density of the signals in the occipital scalp leads is represented by the average power spectral density of each occipital scalp lead.

[0024] This represents the set of serial numbers for the scalp leads in the occipital region;

[0025] Indicates the number of leads in the occipital scalp;

[0026] Indicates in The frequency at which the power spectral density reaches its maximum value within the frequency band;

[0027] Calculate the precursor trend term based on the peak frequency of the posterior head dominant rhythm:

[0028]

[0029] in, express The larger the value of the leading trend term at any given moment, the faster the peak frequency of the dominant posterior rhythm decreases.

[0030] Indicates the length of the trend estimation window;

[0031] Calculate the functional connection strength of the α band:

[0032]

[0033] in, express The functional connectivity strength of the alpha band at any given time;

[0034] This represents a set of lead number pairs, which consist of central scalp leads and occipital scalp leads;

[0035] Indicates the number of lead pairs;

[0036] , They represent the first The first lead, the first The self-power spectral density of each lead;

[0037] Will , , and All are considered as state features;

[0038] Calculate the individualized normalization deviation for each state feature:

[0039]

[0040] in, Indicates the first Individualized normalized deviation of a state feature;

[0041] Indicates the first Item state features in The value at time;

[0042] , These represent the patient's number 1 and 2, respectively. Individual baseline mean and baseline standard deviation for each state characteristic;

[0043] The individualized normalized deviations of each state characteristic are mapped to a single scalar according to a preset fusion rule to obtain the delirium precursor index. ;

[0044] S3: Determine the patient's condition category, where, using express The patient's state category at any given time, with a set of possible values. Sleep, calm and alert, delirium precursors, agitation The judgment method is as follows:

[0045] When the sleep EEG criteria are met For sleep; when the sleep EEG criterion is not met and hour, restlessness; when the sleep EEG criteria are not met and hour, As a precursor to delirium; when the sleep EEG criterion is not met and hour, To be quiet and clear-headed;

[0046] in, , These are the delirium precursor threshold and the agitation threshold, respectively.

[0047] S4: According to Control and constraint system.

[0048] Preferably, the preprocessing includes power frequency notch filtering, bandpass filtering, and artifact removal.

[0049] Preferred, , Initialized with population statistics, and updated online using time windows categorized as quiet and awake.

[0050] Preferably, the method for calculating the delirium precursor index is as follows:

[0051]

[0052] in, express The delirium prodrome index at any given time; the higher the value, the greater the degree to which the patient is approaching a delirious or agitated state.

[0053] Represents the total number of state features;

[0054] Indicates the bias term;

[0055] Indicates the first The fusion weights of the state features of each item;

[0056] This represents the sigmoid function.

[0057] Preferred, according to The control strategy of the control constraint system is: when When delirium or agitation occurs, a restraint trigger signal is output to control the restraint system to apply restraint to the patient; when a preset number of time windows are continuously observed... Both output constraint release signals when the user is asleep or quietly awake.

[0058] Beneficial effects:

[0059] 1) Early identification of dangerous behavioral intent, with intervention before the dangerous behavior occurs. Delirium and agitation precede dangerous behaviors such as extubation, falls, self-harm, and aggression. This invention uses the spectral slowing ratio and relative delta power to characterize the slowing of the EEG background, the precession trend term of the peak frequency of the dominant head rhythm to characterize the slowing trend of the dominant rhythm, and the alpha band functional connectivity strength to characterize the decrease in brain inter-brain coordination. These multi-mechanism features are all EEG changes in the prodromal phase of delirium. The processing unit outputs a restraint trigger signal before the patient commits a dangerous behavior, providing a lead time for restraint intervention compared to video or motion sensing solutions that only trigger an alarm after the action has occurred.

[0060] 2) Restraint is applied only when the patient is in a state of delirium or agitation, and remains unrestrained during the majority of the hospital stay, which is a risk-free period. Compared with the approach that requires full restraint when unattended, this reduces patient discomfort and skin damage, avoids the restraint itself from exacerbating agitation, and conforms to the principle of minimum restraint.

[0061] 3) Individualized normalization adapts state assessment to individual patient differences. Each state characteristic is normalized relative to the patient's individual baseline mean and baseline standard deviation. The individual baseline is initialized with population statistics and updated online using quiet awake time window data. It does not require individual calibration for each patient and can adapt to differences in hospitalized patients in terms of age, disease condition, sedation depth, and baseline EEG, making the delirium prodrome index comparable among different patients.

[0062] 4) Sleep EEG criteria first exclude physiological slow waves to avoid mis-restraint of sleeping patients. The EEG patterns of sleeping and comatose patients are similar to those of the pathological slowing of delirium precursors. This invention first excludes the sleep state using sleep EEG criteria based on central scalp leads, and then performs dual-threshold determination of the delirium precursor index to avoid misjudging physiological slow waves as dangerous behavioral risks and reduce the incidence of mis-restraint.

[0063] 5) Stable constraint control without frequent tightening or loosening. Continuous delirium precursor indices are mapped to well-defined discrete state categories through delirium precursor thresholds and agitation thresholds. Constraint release signals are only output when the state is in a risk-free state for a consecutive preset number of time windows. This avoids frequent tightening or loosening of the constraint system due to fluctuations in state categories near the thresholds, ensuring the stability of constraint control. Detailed Implementation

[0064] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions in the embodiments of this invention will be clearly and completely described below. Obviously, the embodiments described below are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0065] This embodiment uses the protection against unplanned self-extubation by ICU patients as a typical application scenario to illustrate the system and control method of the present invention; the system is also applicable to the protection against other dangerous behaviors such as falls, self-harm, and aggression against others caused by delirium and agitation in patients in general wards.

[0066] A hospital patient risky behavior intention recognition and restraint control system includes an electroencephalogram (EEG) signal acquisition module, a restraint system, and a processing unit. The processing unit controls the restraint system to restrain the patient based on signals acquired by the EEG signal acquisition module. The control method of the processing unit includes the following:

[0067] S1: Receives raw multi-channel signals from the EEG signal acquisition module. ,right Perform preprocessing to obtain preprocessed multi-channel signals. ;

[0068] in, Including central scalp leads and occipital scalp leads, Indicates the sampling time;

[0069] Preprocessing includes power frequency notch filtering, bandpass filtering, and artifact removal; specific preprocessing methods are standard practice in this field and will not be elaborated here.

[0070] In this embodiment, The sampling rate includes central scalp leads (Cz, C3, C4) and posterior scalp leads (Oz), as well as Fpz and FCz. The sampling rate is 500 Hz, and the bandpass range is 0.05–45 Hz. As a further implementation, Fpz and FCz are additional leads that exceed the necessary lead range. Their signals, along with those of the necessary leads, participate in the multi-lead averaging of relative power to reduce the impact of single-lead artifacts on relative power and spectral slowing ratio. They also provide redundancy when individual leads are lost or artifacts are severe, ensuring the continuity of feature calculation. The additional leads do not participate in the detection of sleep spindles, the extraction of dominant posterior head rhythms, or the composition of lead pairs.

[0071] The inclusion of signals from both the central and occipital scalp leads is crucial because extubation originates from voluntary motor control of the sensorimotor cortex. The central scalp leads, covering the sensorimotor cortex, serve two irreplaceable purposes in subsequent steps: first, sleep spindles exhibit maximum amplitude in the central leads, and sleep EEG criteria rely on their detection to exclude physiological slow-wave states; second, the central and occipital leads form a lead pair used to calculate alpha band functional connectivity, and the decreased brain region coordination during the prodromal phase of delirium is reflected in this reduced connectivity. The posterior head dominant rhythm, defined in the occipital region, slows down and disappears, representing one of the earliest EEG changes during the prodromal phase of delirium, making the occipital leads essential for extracting this rhythm. Limiting the necessary leads to these leads reduces the number of electrodes while meeting signal requirements, making it suitable for long-term wear by ICU patients.

[0072] S2: According to The power spectral density of each lead is calculated within the sliding time window.

[0073] The first time window within the sliding time window The first lead and the first The power spectral density of each lead is expressed as ;when hour Represents the self-power spectral density of the lead, when hour This represents the cross-power spectral density between leads; , Indicates the lead number; Indicates frequency;

[0074] Calculate the average relative power for each frequency band:

[0075]

[0076] in, express Time and frequency band The average relative power is the average of the relative power of each lead;

[0077] , Indicates the frequency band name. Represents a set of frequency bands. , , ;

[0078] Indicates the lead number. Indicates the total number of leads;

[0079] In this field, The frequency bands have specific meanings, representing frequency bands with different functions. A common set of values ​​is as follows: 0.5–4Hz 4–8Hz 8–13Hz 13–30Hz.

[0080] Calculate the spectral slowdown ratio based on the average relative power of each frequency band:

[0081]

[0082] in, express The larger the value of the spectral slowdown ratio at any given time, the slower the background EEG rhythm.

[0083] Calculate the peak frequency of the dominant posterior head rhythm:

[0084]

[0085]

[0086] in, express Peak frequency of the dominant posterior head rhythm at specific moments;

[0087] The power spectral density of the signals in the occipital scalp leads is represented by the average power spectral density of each occipital scalp lead.

[0088] This represents the set of serial numbers for the scalp leads in the occipital region;

[0089] Indicates the number of leads in the occipital scalp;

[0090] Indicates in The frequency at which the power spectral density reaches its maximum value within the frequency band;

[0091] In this embodiment, only the Oz lead in the occipital scalp is present, therefore ,at this time This refers to the power spectral density of a single occipital scalp lead. When there are multiple occipital scalp leads, the dominant rhythm in the posterior head is the same rhythm throughout the entire occipital region. The peak frequencies of each occipital scalp lead are consistent, but the amplitudes are different. Taking the average does not change the peak frequency position and can reduce the impact of spectral estimation fluctuations and single-lead artifacts on peak extraction.

[0092] Calculate the precursor trend term based on the peak frequency of the posterior head dominant rhythm:

[0093]

[0094] in, express The larger the value of the leading trend term at any given moment, the faster the peak frequency of the dominant posterior rhythm decreases.

[0095] Indicates the length of the trend estimation window;

[0096] Calculate the functional connection strength of the α band:

[0097]

[0098] in, express The functional connectivity strength of the alpha band at any given time;

[0099] This represents a set of lead number pairs, which consist of central scalp leads and occipital scalp leads;

[0100] Indicates the number of lead pairs;

[0101] , They represent the first The first lead, the first The self-power spectral density of each lead;

[0102] Will , , and All of these are considered as state features; in some implementations, other content may be added to the state features as needed.

[0103] Calculate the individualized normalization deviation for each state feature:

[0104]

[0105] in, Indicates the first Individualized normalized deviation of a state feature;

[0106] Indicates the first Item state features in The value at time;

[0107] , These represent the patient's number 1 and 2, respectively. The individual baseline mean and baseline standard deviation of the state characteristics are initialized with population statistics and updated online using time window data with the state category of quiet and awake.

[0108] The individualized normalized deviations of each state characteristic are mapped to a single scalar according to a preset fusion rule to obtain the delirium precursor index. ;

[0109] In this embodiment, the method for calculating the delirium precursor index is as follows:

[0110]

[0111] in, express The delirium prodrome index at any given time; the higher the value, the greater the degree to which the patient is approaching a delirious or agitated state.

[0112] Represents the total number of state features;

[0113] Indicates the bias term;

[0114] Indicates the first The fusion weights of the state features of each item;

[0115] Represents the sigmoid function;

[0116] S3: Determine the patient's condition category, where, using express The patient's state category at any given time, with a set of possible values. Sleep, calm and alert, delirium precursors, agitation The judgment method is as follows:

[0117] When the sleep EEG criteria are met For sleep; when the sleep EEG criterion is not met and hour, restlessness; when the sleep EEG criteria are not met and hour, As a precursor to delirium; when the sleep EEG criterion is not met and hour, To be quiet and clear-headed;

[0118] in, , These are the delirium precursor threshold and the agitation threshold, respectively.

[0119] The sleep EEG criterion precedes The determination is executed to exclude physiological slow-wave states: EEG in both sleep and coma patients shows a slowing background with increased delta power, similar in spectral morphology to the pathological slowing of delirium precursors, and cannot be determined solely by... Since the two cannot be distinguished, they can be ruled out for now;

[0120] Determining sleep status using sleep EEG criteria is standard practice in this field. Here's a brief overview: sleep EEG criteria are defined as follows: when sleep spindle waves, or characteristic sleep waves such as K-complexes, are detected within a sliding time window based on signals from the central scalp leads, or... Continuously above the sleep power threshold At that time, the patient is determined to be in an unconscious state such as sleep or coma, which meets the sleep EEG criteria; the amplitude of sleep spindle waves is largest in the central scalp leads, so their detection is based on the central scalp leads; among them, The sleep power threshold is determined by statistical analysis of group sleep EEG data and is higher than the common delirium relative power level in the prodromal phase of delirium. The detection method of sleep characteristic waves is a routine operation in this field and will not be described in detail here.

[0121] In some possible implementations, , Determined offline using group-labeled data: Using clinical delirium assessment results as labels, the EEG samples of the group patients were divided into three categories: quiet and awake, delirium prodrome, and agitation. The data for each category were then statistically analyzed. Distribution, taking the threshold that distinguishes between quiet lucidity and delirium precursors as... The threshold for distinguishing delirium precursors from agitation is taken as... ; The selection bias ensures high sensitivity in detecting delirium precursors, allowing for intervention before the risk of extubation develops. The selection bias ensures the specificity of agitation identification; the threshold remains fixed during operation after being determined offline; there is no risk of spontaneous extubation for both sleeping and comatose patients, and the two are grouped into the same state category;

[0122] In this embodiment, the sliding time window is 30 seconds long and the step size is 2 seconds. The lead pairs are Cz, C3, and C4, which are respectively paired with Oz to form three lead pairs. , Trend estimation window length min, The weights of the four parameters—spectral slowdown ratio, rate of decline of peak frequency of the dominant posterior rhythm, relative delta power, and functional connectivity strength in the alpha band—are 0.8, 0.6, 0.5, and 0.4, respectively. (This corresponds to the inflection point of the sigmoid function, i.e., the decision boundary where the weighted sum of the features of each state deviates from zero.) , ;

[0123] Each state characteristic is normalized relative to the patient's individual baseline, rather than directly using a population threshold. This is because ICU patients vary greatly in age, sedation depth, and baseline EEG, making it difficult for population thresholds to be applicable to all patients. Individual baselines are initialized with population statistics and updated online, eliminating the need for individual calibration for each patient and ensuring comparability of individualized normalization deviations across different patients. The four state characteristics respectively characterize different pathological changes in the prodromal phase of delirium. The relative delta power and spectral slowing ratio characterize the slowing of the EEG background, the prodromal trend term characterizes the slowing trend of the dominant head rhythm, and the alpha band functional connectivity strength characterizes the decrease in brain inter-brain coordination. The multiple mechanisms of these characteristics corroborate each other, and short-term fluctuations of a single characteristic are insufficient to drive the delirium prodromal index, thus improving the robustness of state assessment.

[0124] The state category is determined by a dual threshold of the sleep EEG criterion and the delirium precursor index. Coma and sleep are grouped into the same state category because continuous delirium precursor indices are... , The two decision thresholds are mapped to discrete state categories with well-defined boundaries, making the constraint control have unambiguous triggering conditions. Comatose and sleeping patients have no risk of spontaneous extubation and their corresponding control actions are the same. After merging, the number of states and disputes over classification boundaries are reduced, thus lowering the risk of state misjudgment.

[0125] S4: According to Control constraint system; in this embodiment, according to The control strategy of the control constraint system is: when When delirium or agitation occurs, a restraint trigger signal is output to control the restraint system to apply restraint to the patient; when a preset number of time windows are continuously observed... When the system is asleep or quietly awake, it outputs a constraint release signal to avoid frequent release and release of the constraint system due to critical jitter.

[0126] In this embodiment, the preset number is 5 time windows;

[0127] The restraint system does not apply restraint throughout the entire hospital stay, but only when the state category is delirium precursor or agitation. This is because delirium precursor occurs before the extubation action, and restraint is applied before the patient performs the extubation action. During the risk-free period, which accounts for most of the hospital stay, the patient is not restrained. Compared with existing technologies that require restraint to be set throughout the entire hospital stay when there is no caregiver, this significantly shortens the restraint duration and improves the patient experience. The restraint release signal is only output when the patient is in a risk-free state for a consecutive preset number of time windows, avoiding frequent opening and closing of the restraint system due to fluctuations in the state category near the threshold, thus ensuring the stability of restraint control.

[0128] Specific control strategies can be configured as needed. For example, the restraint system can use different restraint measures depending on whether the patient is in a state of delirium or agitation; it can also send alerts to medical staff when the patient is calm and conscious. In general ward settings, the restraint system can also be replaced with intervention methods such as bed exit alarms, bed rail linkage, or calling for medical staff to prevent dangerous behaviors such as falls, self-harm, and aggression towards others.

[0129] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A hospital patient risky behavior intention recognition and restraint control system, characterized in that, The system includes an electroencephalogram (EEG) signal acquisition module, a restraint system, and a processing unit. The processing unit controls the restraint system to restrain the patient based on the signals acquired by the EEG signal acquisition module. The control method of the processing unit includes the following: S1: Receives raw multi-channel signals from the EEG signal acquisition module. ,right Perform preprocessing to obtain preprocessed multi-channel signals. ; in, Including central scalp leads and occipital scalp leads, Indicates the sampling time; S2: According to The power spectral density of each lead is calculated within the sliding time window. Within the sliding time window The first lead and the first The power spectral density of each lead is expressed as ;when hour Represents the self-power spectral density of the lead, when hour This represents the cross-power spectral density between leads; , Indicates the lead number; Indicates frequency; Calculate the average relative power for each frequency band: in, express Time and frequency band The average relative power is the average of the relative power of each lead; , Indicates the frequency band name. Represents a set of frequency bands. , , ; Indicates the lead number. Indicates the total number of leads; Calculate the spectral slowdown ratio based on the average relative power of each frequency band: in, express The larger the value of the spectral slowdown ratio at any given time, the slower the background EEG rhythm. Calculate the peak frequency of the dominant posterior head rhythm: in, express Peak frequency of the dominant posterior head rhythm at specific moments; The power spectral density of the signals in the occipital scalp leads is represented by the average power spectral density of each occipital scalp lead. This represents the set of serial numbers for the scalp leads in the occipital region; Indicates the number of leads in the occipital scalp; Indicates in The frequency at which the power spectral density reaches its maximum value within the frequency band; Calculate the precursor trend term based on the peak frequency of the posterior head dominant rhythm: in, express The larger the value of the leading trend term at any given moment, the faster the peak frequency of the dominant posterior rhythm decreases. Indicates the length of the trend estimation window; Calculate the functional connectivity strength of the α band: in, express The functional connectivity strength of the alpha band at any given time; This represents a set of lead number pairs, which consist of central scalp leads and occipital scalp leads; Indicates the number of lead pairs; , They represent the first The first lead, the first The self-power spectral density of each lead; Will , , and All are considered as state features; Calculate the individualized normalization deviation for each state feature: in, Indicates the first Individualized normalized deviation of a state feature; Indicates the first Item state features in The value at time; , These represent the patient's number Individual baseline mean and baseline standard deviation for each state characteristic; The individualized normalized deviations of each state characteristic are mapped to a single scalar according to a preset fusion rule to obtain the delirium precursor index. ; S3: Determine the patient's condition category, where, using express The patient's state category at any given time, with a set of possible values. Sleep, calm and alert, delirium precursors, agitation The judgment method is as follows: When the sleep EEG criteria are met For sleep; when the sleep EEG criterion is not met and hour, restlessness; when the sleep EEG criteria are not met and hour, As a precursor to delirium; when the sleep EEG criterion is not met and hour, To be quiet and clear-headed; in, , These are the delirium precursor threshold and the agitation threshold, respectively. S4: According to Control and constraint system.

2. The hospital patient dangerous behavior intention recognition and restraint control system according to claim 1, characterized in that, The preprocessing includes power frequency notch filtering, bandpass filtering, and artifact removal.

3. The hospital patient dangerous behavior intention recognition and restraint control system according to claim 1, characterized in that, , It is initialized with population statistics and updated online using time windows with the state category of quiet and awake.

4. The hospital patient dangerous behavior intention recognition and restraint control system according to claim 1, characterized in that, The method for calculating the delirium precursor index is as follows: in, express The delirium prodrome index at any given time; the higher the value, the greater the degree to which the patient is approaching a delirious or agitated state. Represents the total number of state features; Indicates the bias term; Indicates the first The fusion weights of the state features of each item; This represents the sigmoid function.

5. A hospital patient dangerous behavior intention recognition and restraint control system according to claim 1, characterized in that, according to The control strategy of the control constraint system is: when When delirium or agitation occurs, a restraint trigger signal is output to control the restraint system to apply restraint to the patient; when a preset number of time windows are continuously observed... Both output constraint release signals when the user is asleep or quietly awake.

Images