Neurological patient fall risk prediction method and system based on time series coupling

Abstract

The present application relates to the technical field of neural medicine and fall risk prediction, and particularly relates to a neurology patient fall risk prediction method and system based on time sequence coupling. The abnormal fluctuation characteristics are extracted from the amygdala EEG signals collected by the electrode array, the posture mutation characteristics are recognized by synchronously collecting the limb acceleration and angular velocity, the neural-motor coupling strength value is obtained by aligning and calculating the time delay and correlation coefficient of the two characteristic sequences, the probability of losing balance control is calculated by combining the threshold time proportion of the value with the accumulation offset of the center of gravity, and the risk level is divided, so that the time sequence correlation evaluation of the neural and motor systems is realized, and the accuracy of the fall risk prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of neuromedicine and fall risk prediction technology, and in particular to a method and system for predicting fall risk in neurology patients based on temporal coupling. Background Technology

[0002] In the field of fall risk assessment for neurological patients, current routine practices largely rely on clinical scales and single-modal sensor monitoring. Clinicians often use subjective assessment tools such as the Morse Fall Assessment Scale and the Berg Balance Scale, observing patients' walking, turning, and other movements, combined with medical history and medication, to classify fall risk levels. While these scales are easy to use, they are highly dependent on the assessor's experience and the patient's immediate cooperation, making it difficult to capture dynamic changes and transient neurological fluctuations in patients' daily activities. Meanwhile, some medical institutions introduce inertial measurement units or pressure pads to analyze gait and postural control by collecting parameters such as acceleration and angular velocity, identifying gait instability or center of gravity shifts. These methods provide objective kinematic data, but typically only focus on limb movement itself, neglecting the real-time influence of the nervous system on motor control, especially lacking simultaneous consideration of abnormal brain discharges or neuromodulation failures.

[0003] One significant drawback of existing methods is their isolation. Conventional approaches either focus on neurophysiological detection (such as localized EEG abnormalities) or on motor parameter analysis (such as gait phases and swing amplitude), but rarely establish a dynamic coupling relationship between the two over time. For example, abnormal electrical activity may occur in the amygdala region before a seizure in an epileptic patient, subsequently triggering limb rigidity or spasticity, leading to a fall. However, traditional scales cannot predict this transient process, and single sensors struggle to promptly link the causal chain between EEG signals and postural changes, resulting in a severe underestimation of the actual risk. Another drawback is the lag in response and insufficient predictive accuracy. Scale assessments are often based on static observations lasting several minutes to tens of minutes, while falls often occur as instantaneous neuromotor decoupling events. While existing motion monitoring can detect postural changes, it cannot distinguish whether these changes are caused by environmental disturbances or central nervous system dysregulation, thus frequently generating false alarms or missed alarms. The essence of these drawbacks lies in the lack of analysis of the temporal characteristics of brain-muscle information interaction, and the inability to quantify the critical time window and intensity correlation between abnormal neural signals and motor instability. Summary of the Invention

[0004] This invention provides a method and system for predicting the risk of falls in neurology patients based on temporal coupling, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for predicting the risk of falls in neurology patients based on temporal coupling, comprising:

[0006] Raw EEG data was obtained by acquiring EEG signals from the amygdala region of the patient using an electrode array.

[0007] The original EEG data is decomposed in the time and frequency domains and frequency and time domain features are extracted. Abnormal fluctuations are detected by sliding time windows, and an abnormal fluctuation feature sequence containing the time, duration and amplitude of abnormal fluctuations is generated.

[0008] Acceleration and angular velocity signals of the patient's limbs are collected and converted into body posture angles and center of gravity offsets. Posture mutations are identified by calculating the rate of change between adjacent time windows, and a posture mutation feature sequence is generated.

[0009] Align the abnormal fluctuation feature sequence and the attitude change feature sequence by timestamp, calculate the time delay between the abnormal fluctuation time and the attitude change time and the correlation coefficient between the fluctuation amplitude and the change amplitude, and calculate the neural-motor coupling strength value based on the time delay and the correlation coefficient.

[0010] The probability of a patient losing balance control is calculated by statistically analyzing the percentage of time during which the neuro-motor coupling strength value exceeds a preset coupling threshold, and by combining this with the cumulative offset of the center of gravity shift in the posture mutation feature sequence. The risk level is then classified based on the probability value.

[0011] The raw EEG data is decomposed into time and frequency domains, and frequency and time domain features are extracted. Abnormal fluctuations are detected using a sliding time window, generating an abnormal fluctuation feature sequence that includes the time, duration, and amplitude of the abnormal fluctuations.

[0012] The original EEG data is decomposed into multiple frequency band components by wavelet transform. The power spectral density of each frequency band component within the sliding time window is calculated as the frequency domain feature, and the peak amplitude and zero-crossing rate of each frequency band component within the sliding time window are calculated as the time domain features.

[0013] EEG data of the patient at rest was collected as baseline EEG data. Baseline frequency domain features and baseline time domain features of the baseline EEG data were extracted. The frequency domain deviation between the frequency domain features and the baseline frequency domain features and the time domain deviation between the time domain features and the baseline time domain features were calculated.

[0014] When the frequency domain deviation or the time domain deviation exceeds the deviation threshold, the center moment of the current sliding time window is marked as the abnormal fluctuation moment, and the time span from the first deviation exceeding the deviation threshold to falling back below the deviation threshold is recorded as the duration.

[0015] Within the duration, the power spectral density change is weighted and fused based on the frequency domain deviation of each frequency band component to obtain the fluctuation amplitude, and an abnormal fluctuation feature sequence containing the abnormal fluctuation time, duration, and fluctuation amplitude is generated.

[0016] Acceleration and angular velocity signals of the patient's limbs are collected and converted into body posture angles and center of gravity offsets. Posture abrupt changes are identified by calculating the rate of change between adjacent time windows, and a posture abrupt change feature sequence is generated, including:

[0017] Acceleration and angular velocity signals of the patient's limbs are collected, and the acceleration signals are separated into static acceleration and dynamic acceleration components by gravity component separation.

[0018] The initial body posture angle is calculated based on the projection direction of the static acceleration component, and the center of gravity offset is calculated based on the dual time integration of the dynamic acceleration component.

[0019] The angular velocity signal is integrated within a preset time window to obtain the attitude rotation angle increment, and the initial body attitude angle is superimposed with the attitude rotation angle increment to obtain the body attitude angle.

[0020] The body posture angle and the center of gravity offset are divided into time windows. The rate of change of the body posture angle and the rate of change of the center of gravity offset between adjacent time windows are calculated. The rate of change of the body posture angle and the rate of change of the center of gravity offset are normalized and then summed to obtain the comprehensive rate of change.

[0021] The moment of attitude change is marked by the peak position of the comprehensive rate of change. The body posture angle and the center of gravity offset are extracted as the attitude change amplitude to generate an attitude change feature sequence containing the moment of attitude change and the attitude change amplitude.

[0022] Align the abnormal fluctuation feature sequence with the attitude change feature sequence by timestamp, calculate the time delay between the abnormal fluctuation time and the attitude change time, and the correlation coefficient between the fluctuation amplitude and the change amplitude. Calculate the neural-motor coupling strength value based on the time delay and correlation coefficient, including:

[0023] The abnormal fluctuation feature sequence and the attitude change feature sequence are synchronously divided into multiple time segment pairs at fixed time intervals. Within each time segment pair, the minimum time difference between the abnormal fluctuation moment and the attitude change moment is searched as the segment time delay.

[0024] All time delay segments are combined to form a time delay sequence, and the standard deviation of the time delay sequence is calculated as the time delay volatility.

[0025] Within each time segment pair, the fluctuation amplitude is extracted to form a fluctuation amplitude subsequence, and the centroid offset is extracted to form a centroid offset subsequence. The fluctuation amplitude subsequence is then shifted along the time axis according to the segment time delay and paired with the centroid offset subsequence. The Pearson correlation coefficient between the paired fluctuation amplitude subsequence and the centroid offset subsequence is calculated as the segment amplitude correlation coefficient.

[0026] The amplitude correlation coefficients of all time segment pairs are used to form an amplitude correlation coefficient sequence, and the mean of the amplitude correlation coefficient sequence is calculated as the average amplitude correlation coefficient.

[0027] A delay stability coefficient is constructed based on the time delay fluctuation, and the delay stability coefficient is fused with the average amplitude correlation coefficient to obtain the neural-motor coupling strength value.

[0028] Pairing the fluctuation amplitude subsequence with the centroid offset subsequence after shifting it along the time axis according to the segment time delay includes:

[0029] The time delay value of the segment is added to the timestamp corresponding to each fluctuation amplitude value in the fluctuation amplitude subsequence to obtain an offset timestamp sequence. The fluctuation amplitude subsequence is then re-marked according to the offset timestamp sequence to obtain the fluctuation amplitude subsequence after time axis offset.

[0030] Using the timestamp corresponding to each centroid offset in the centroid offset subsequence as the reference time, the fluctuation amplitude value with the smallest time difference from the reference time is searched in the fluctuation amplitude subsequence after the time axis offset and paired to generate a paired dataset of fluctuation amplitude value and centroid offset value.

[0031] All fluctuation amplitude values ​​are extracted from the paired dataset to form a paired fluctuation amplitude subsequence, and all centroid offset values ​​are extracted from the paired dataset to form a paired centroid offset subsequence.

[0032] The probability of a patient losing balance control is calculated by statistically analyzing the percentage of time the neuro-motor coupling strength value exceeds a preset coupling threshold, and combining this with the cumulative shift of the center of gravity in the posture change feature sequence. Risk levels are then categorized based on this probability value, including:

[0033] The cumulative duration of the neural-motor coupling strength value exceeding the preset coupling threshold is counted as a percentage of the monitoring period.

[0034] The cumulative offset of the center of gravity offset is obtained by summing the center of gravity offsets along the time axis in the posture change feature sequence.

[0035] At the start and end times of the period when the neural-motor coupling strength value exceeds the preset coupling threshold, the cumulative offset of the center of gravity offset is extracted and the difference is calculated to obtain the offset increment of the coupling abnormal period. The offset increment of the coupling abnormal period is normalized and coupled with the time proportion to obtain the risk value of the coupling period.

[0036] Extract the posture mutation feature sequence within the time period when the neural-motor coupling strength value exceeds the preset coupling threshold, count the number of posture mutation events in the posture mutation feature sequence to obtain the runaway event frequency, and perform time series analysis on the runaway event frequency to extract the change trend to obtain the runaway acceleration trend index;

[0037] The cumulative offset of the center of gravity offset is correlated with the uncontrolled acceleration trend index to obtain the cumulative dynamic risk value. Based on the interaction between the risk value of the coupling period and the cumulative dynamic risk value, the probability value of the patient losing balance control ability is calculated.

[0038] The risk level is classified and output based on the preset probability threshold range in which the probability value falls.

[0039] The cumulative offset of the center of gravity shift is correlated with the runaway acceleration trend index to obtain a cumulative dynamic risk value. Based on the interaction between the risk value during the coupling period and the cumulative dynamic risk value, the probability value of the patient losing balance control is calculated, including:

[0040] The cumulative offset of the center of gravity offset is segmented into time windows to extract the rate of change feature;

[0041] The change rate feature and the runaway acceleration trend indicator are aligned according to the timestamp and the temporal correlation is calculated to obtain the dynamic coupling coefficient. The dynamic coupling coefficient is used as the correlation weight to weight the cumulative offset of the center of gravity offset to obtain the cumulative dynamic risk value.

[0042] The synergistic enhancement coefficient is obtained by multiplying the risk value of the coupling period and the cumulative dynamic risk value. Based on the synergistic enhancement coefficient, the risk value of the coupling period and the cumulative dynamic risk value are amplified to obtain the amplified risk value of the coupling period and the amplified cumulative dynamic risk value.

[0043] The amplified coupling period risk value and the amplified cumulative dynamic risk value are multiplied to obtain an interactive amplification term. The amplified coupling period risk value and the amplified cumulative dynamic risk value are weighted and summed to obtain a linear fusion term. The interactive amplification term and the linear fusion term are superimposed to obtain the probability value of the patient losing balance control ability.

[0044] A second aspect of the present invention provides a fall risk prediction system for neurology patients based on temporal coupling, comprising:

[0045] The EEG acquisition unit is used to acquire EEG signals from the patient's amygdala region via an electrode array to obtain raw EEG data.

[0046] The time-frequency analysis unit is used to decompose the raw EEG data in the time-frequency domain and extract frequency domain features and time domain features. It detects abnormal fluctuations through a sliding time window and generates an abnormal fluctuation feature sequence containing the time, duration and amplitude of the abnormal fluctuations.

[0047] The posture detection unit is used to collect the acceleration and angular velocity signals of the patient's limbs and convert them into body posture angles and center of gravity offsets. It identifies posture changes by calculating the rate of change between adjacent time windows and generates a posture change feature sequence.

[0048] The coupling analysis unit is used to align the abnormal fluctuation feature sequence and the attitude change feature sequence by timestamp, calculate the time delay between the abnormal fluctuation time and the attitude change time and the correlation coefficient between the fluctuation amplitude and the change amplitude, and calculate the neural-motor coupling strength value based on the time delay and the correlation coefficient.

[0049] The risk assessment unit is used to calculate the probability value of the patient losing balance control ability by statistically analyzing the percentage of time when the neuro-motor coupling strength value exceeds a preset coupling threshold, and by combining the cumulative offset of the center of gravity offset in the posture mutation feature sequence, and classifying the risk level according to the probability value.

[0050] A third aspect of the present invention provides an electronic device, comprising:

[0051] processor;

[0052] Memory used to store processor-executable instructions;

[0053] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0054] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0055] This invention constructs a neuro-motor temporal coupling analysis by simultaneously acquiring EEG signals from the amygdala region and limb movement signals from patients. This accurately quantifies the dynamic correlation between abnormal brain fluctuations and loss of body posture control. Compared with prediction methods based on a single signal source, this significantly improves the accuracy and reliability of fall risk prediction, avoiding misjudgments and missed judgments caused by EEG artifacts or limb movement interference. By performing time-frequency domain decomposition on the raw EEG data and extracting frequency and time domain features, combined with a sliding time window to detect abnormal fluctuations, a feature sequence containing the time, duration, and amplitude of abnormal fluctuations is generated. Simultaneously, limb acceleration and angular velocity signals are acquired, converted into posture angles and center of gravity offsets, and posture abrupt changes are identified, generating a posture abrupt change feature sequence. After aligning the two sequences by timestamp, the time delay and correlation coefficient are calculated. The resulting neuro-motor coupling strength value intuitively reflects the time lag relationship and coordination degree between central nervous system signals and limb responses, providing a sensitive indicator for early warning. Furthermore, the percentage of time when the neuro-motor coupling strength value exceeds a preset threshold is statistically analyzed, and combined with the cumulative offset of the center of gravity offset in the posture abrupt change feature sequence, the probability value of the patient losing balance control is calculated. This probability value combines the duration of neurological abnormalities with the cumulative effect of body displacement, effectively distinguishing different risk levels. Healthcare professionals can develop tiered intervention plans based on risk levels to achieve personalized prevention and significantly reduce the incidence of falls among neurology patients. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating a method for predicting fall risk in neurology patients based on temporal coupling.

[0057] Figure 2 Flowchart for risk assessment and grading of patients losing balance control. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0060] Figure 1 This is a flowchart illustrating the fall risk prediction method for neurology patients based on temporal coupling, according to an embodiment of the present invention.

[0061] Methods for predicting fall risk in neurology patients based on temporal coupling include:

[0062] Raw EEG data was obtained by acquiring EEG signals from the amygdala region of the patient using an electrode array.

[0063] The original EEG data is decomposed in the time and frequency domains and frequency and time domain features are extracted. Abnormal fluctuations are detected by sliding time windows, and an abnormal fluctuation feature sequence containing the time, duration and amplitude of abnormal fluctuations is generated.

[0064] Acceleration and angular velocity signals of the patient's limbs are collected and converted into body posture angles and center of gravity offsets. Posture mutations are identified by calculating the rate of change between adjacent time windows, and a posture mutation feature sequence is generated.

[0065] Align the abnormal fluctuation feature sequence and the attitude change feature sequence by timestamp, calculate the time delay between the abnormal fluctuation time and the attitude change time and the correlation coefficient between the fluctuation amplitude and the change amplitude, and calculate the neural-motor coupling strength value based on the time delay and the correlation coefficient.

[0066] The probability of a patient losing balance control is calculated by statistically analyzing the percentage of time during which the neuro-motor coupling strength value exceeds a preset coupling threshold, and by combining this with the cumulative offset of the center of gravity shift in the posture mutation feature sequence. The risk level is then classified based on the probability value.

[0067] The raw EEG data is decomposed into time and frequency domains, and frequency and time domain features are extracted. Abnormal fluctuations are detected using a sliding time window, generating an abnormal fluctuation feature sequence that includes the time, duration, and amplitude of the abnormal fluctuations.

[0068] The original EEG data is decomposed into multiple frequency band components by wavelet transform. The power spectral density of each frequency band component within the sliding time window is calculated as the frequency domain feature, and the peak amplitude and zero-crossing rate of each frequency band component within the sliding time window are calculated as the time domain features.

[0069] EEG data of the patient at rest was collected as baseline EEG data. Baseline frequency domain features and baseline time domain features of the baseline EEG data were extracted. The frequency domain deviation between the frequency domain features and the baseline frequency domain features and the time domain deviation between the time domain features and the baseline time domain features were calculated.

[0070] When the frequency domain deviation or the time domain deviation exceeds the deviation threshold, the center moment of the current sliding time window is marked as the abnormal fluctuation moment, and the time span from the first deviation exceeding the deviation threshold to falling back below the deviation threshold is recorded as the duration.

[0071] Within the duration, the power spectral density change is weighted and fused based on the frequency domain deviation of each frequency band component to obtain the fluctuation amplitude, and an abnormal fluctuation feature sequence containing the abnormal fluctuation time, duration, and fluctuation amplitude is generated.

[0072] After acquiring the raw EEG data, wavelet transform was applied to decompose the signal into multiple frequency band components. Wavelet transform can perform localized analysis of the signal in both the time and frequency domains simultaneously, and compared to traditional Fourier transform, it is more suitable for handling the non-stationary and nonlinear characteristics of EEG signals in the amygdala region. The decomposed frequency band components cover typical EEG frequency bands such as delta waves (0.5-4Hz), theta waves (4-8Hz), alpha waves (8-13Hz), beta waves (13-30Hz), and gamma waves (above 30Hz), each carrying information closely related to the state of neural activity. In the scenario of fall risk assessment for neurological patients, abnormalities in theta and gamma waves in the amygdala region are often associated with emotional stress and motor control disorders; therefore, fine decomposition of multiple frequency bands helps improve the accuracy of subsequent feature extraction.

[0073] For each frequency band component, the power spectral density is calculated within a sliding time window, serving as the frequency domain feature of that band. Power spectral density reflects the distribution of signal energy along the frequency axis, quantifying the activity level of each frequency band within a specific time period. The window length and step size of the sliding time window can be set according to the actual sampling rate and signal resolution requirements; for example, a window length of 2 seconds and a step size of 0.5 seconds can be used to ensure a balance between temporal resolution and statistical stability. Simultaneously, within the same sliding time window, the peak amplitude and zero-crossing rate are calculated for each frequency band component, serving as time domain features. The peak amplitude reflects the intensity of extreme signal fluctuations within the window, while the zero-crossing rate describes the number of times the signal crosses zero per unit time, serving as an important indicator of the frequency of signal oscillations. In EEG signal analysis, an abnormally high zero-crossing rate often indicates a disordered neural discharge pattern, and is associated with epileptiform discharges or fluctuations in consciousness.

[0074] To determine whether the current EEG state deviates from the normal range, a baseline reference needs to be established. While the patient is at rest, a segment of baseline EEG data is collected. This data is then subjected to the same wavelet transform decomposition and feature extraction to obtain the baseline power spectral density, baseline peak amplitude, and baseline zero-crossing rate for each frequency band, constituting the baseline frequency domain features and baseline time domain features, respectively. It is recommended that the baseline data acquisition time be no less than 5 minutes to ensure that the baseline features are statistically representative and can stably reflect the individual patient's neural activity patterns at rest. Because there are significant individual differences in background EEG activity among different patients, using an individualized baseline rather than the population mean as a reference can effectively reduce the interference of individual differences on abnormality detection.

[0075] After acquiring baseline features, the frequency domain and time domain features extracted within each sliding time window are compared with the corresponding baseline frequency domain and time domain features, respectively, to calculate the frequency domain deviation and time domain deviation. The frequency domain deviation can be defined as the normalized difference between the power spectral density of each frequency band within the current window and the baseline power spectral density, while the time domain deviation is quantified based on the degree of change in peak amplitude and zero-crossing rate relative to the baseline values. Let the power spectral density of a certain frequency band within the current window be... The corresponding baseline power spectral density is The frequency domain deviation of this frequency band It can be represented as Let the peak amplitude of the current window be... The baseline peak amplitude is The zero-crossing rate The baseline zero-crossing rate Then the time domain deviation Both can be combined for calculation, for example, taking the weighted average of the normalized differences, i.e. ,in This is the weighting coefficient between peak amplitude deviation and zero-crossing rate deviation. It can be determined by adjusting parameters based on actual data, and its value ranges from 0 to 1.

[0076] The frequency domain deviation calculated within a certain sliding time window or time domain deviation When the deviation exceeds a preset threshold, the center time of the window is marked as an abnormal fluctuation moment. The deviation threshold can be determined through statistical analysis of training samples, for example, by adding twice the standard deviation to the mean of the baseline deviation distribution, thereby controlling the false alarm rate while ensuring detection sensitivity. Marking abnormal fluctuation moments using the center time rather than the window start time helps to more accurately locate the position of abnormal events on the time axis and reduce the time bias introduced by the window boundary effect.

[0077] After marking the moments of abnormal fluctuations, the temporal changes in deviation are continuously tracked. The time span from the moment the deviation first exceeds the deviation threshold to the moment the deviation falls back below the threshold is recorded as the duration of the abnormal fluctuation event. If the time interval between two consecutive deviations exceeding the threshold is less than the set merging interval (e.g., 1 second), they are considered as continuous fluctuations of the same abnormal event, and their durations are merged to avoid the same event being split into multiple independent events due to a brief decline, which would affect the accuracy of subsequent analysis.

[0078] Over a given duration, the frequency domain deviations of each frequency band component are weighted and fused based on their contribution to calculate the final fluctuation amplitude. Specifically, let there be a total of The frequency band, the first The average frequency domain deviation of each frequency band over the duration is The change in power spectral density in this frequency band within the current window is The contribution weight is then calculated as the proportion of the frequency domain deviation of each frequency band to the sum of the deviations of all frequency bands. ,satisfy Fluctuation range It is obtained by weighted fusion of the power spectral density changes in each frequency band, i.e. This contribution-weighted fusion method can highlight the dominant role of frequency bands with larger abnormal deviations in fluctuation amplitude, while retaining auxiliary information from other frequency bands. This makes the fluctuation amplitude calculation results more reflective of the overall abnormality of the EEG signal, rather than relying solely on changes in a single frequency band.

[0079] The abnormal fluctuation time, duration, and amplitude of each detected abnormal fluctuation event are combined into a feature vector and arranged chronologically to form an abnormal fluctuation feature sequence. This sequence fully records the abnormal activity patterns of the patient's electroencephalogram (EEG) signals in the amygdala region during the monitoring period, providing high-quality neural input data for subsequent temporal coupling analysis with the posture change feature sequence. In practical applications, if the patient's monitoring period is long, the abnormal fluctuation feature sequence can be segmented, and the coupling features can be calculated segment by segment using a sliding analysis window, thereby achieving dynamic real-time assessment of fall risk.

[0080] Acceleration and angular velocity signals of the patient's limbs are collected and converted into body posture angles and center of gravity offsets. Posture abrupt changes are identified by calculating the rate of change between adjacent time windows, and a posture abrupt change feature sequence is generated, including:

[0081] Acceleration and angular velocity signals of the patient's limbs are collected, and the acceleration signals are separated into static acceleration components and dynamic acceleration components by gravity component separation.

[0082] The initial body posture angle is calculated based on the projection direction of the static acceleration component, and the center of gravity offset is calculated based on the dual time integration of the dynamic acceleration component.

[0083] The angular velocity signal is integrated within a preset time window to obtain the attitude rotation angle increment, and the initial body attitude angle is superimposed with the attitude rotation angle increment to obtain the body attitude angle.

[0084] The body posture angle and the center of gravity offset are divided into time windows. The rate of change of the body posture angle and the rate of change of the center of gravity offset between adjacent time windows are calculated. The rate of change of the body posture angle and the rate of change of the center of gravity offset are normalized and then summed to obtain the comprehensive rate of change.

[0085] The moment of attitude change is marked by the peak position of the comprehensive rate of change. The body posture angle and the center of gravity offset are extracted as the attitude change amplitude to generate an attitude change feature sequence containing the moment of attitude change and the attitude change amplitude.

[0086] When acquiring acceleration and angular velocity signals of a patient's limbs, inertial measurement units (IMUs) are typically worn on key limb sites such as the patient's waist, ankle, or wrist, with triaxial accelerometers and triaxial gyroscopes as the primary sensors. The raw signal output by the accelerometers contains both static components caused by gravity and dynamic components caused by limb movement. If these two components are not separated, they will directly affect the accuracy of subsequent attitude calculations. Therefore, a low-pass filter is applied to the raw acceleration signal, with the cutoff frequency typically set between 0.5Hz and 1Hz. The low-frequency portion represents the static acceleration component, reflecting the projection of gravity onto each axis. Subtracting the static acceleration component from the raw acceleration signal yields the dynamic acceleration component, which reflects only the actual acceleration of the limb movement, eliminating gravity interference.

[0087] The initial attitude of the sensor coordinate system relative to the direction of gravity can be determined by the projection directions of the static acceleration components onto the three coordinate axes. Specifically, let the static acceleration components be projected onto the three coordinate axes. axis, axis, The components on the axis are respectively , , Then the initial pitch angle and initial roll angle The results were obtained by calculating using the arctangent function: , The two angles mentioned above together constitute the initial body posture angle, which describes the degree of tilt of the patient's body relative to the vertical direction when at rest or in a near-resting state.

[0088] The center of gravity offset is calculated from the dynamic acceleration components through double time integration. For the dynamic acceleration components... The speed is obtained by performing the first integration. Then, the displacement is obtained by integrating the velocity a second time. This displacement is the offset of the center of gravity in the horizontal plane. To suppress integral drift error, zero-velocity correction (ZUPT) is performed on the integral result at the beginning of each preset time window. That is, when a limb is detected to be in a brief state of stillness, the velocity is forcibly set to zero and the displacement reference is corrected, thereby effectively controlling the growth of accumulated error. Center of gravity offset Recorded in millimeters, reflecting the actual degree of deviation of the patient's center of gravity in the horizontal direction.

[0089] The angular velocity signal is output from the three-axis gyroscope, and the unit is degrees per second (° / s) or radians per second (rad / s). Within a preset time window... Within that time window, the angular velocity signal is numerically integrated to obtain the attitude rotation angle increment. Let the angular velocity components of the three axes be respectively... , , The corresponding rotation angle increment is , , The initial body posture angle , The body pose angle within the current time window is obtained by superimposing the rotation angle increment along the corresponding axis. and Because gyroscopes exhibit zero-bias drift, long-term integration introduces errors. Therefore, complementary filtering or Kalman filtering is used to fuse the attitude angles calculated by the accelerometer with the gyroscope integration results. The low-frequency reliability of the accelerometer is used to correct the high-frequency accuracy of the gyroscope, ultimately obtaining a stable and accurate sequence of body attitude angles.

[0090] The body posture angle sequence and center of gravity offset sequence are segmented according to a unified time window, with the length of adjacent time windows set to 1. (Unit: seconds). Let the first... The average body posture angle within each time window is , No. The average body posture angle within each time window is Then the rate of change of attitude angle Defined as The unit is degrees per second. Similarly, let the first... The average centroid offset within each time window is , No. The average centroid offset within each time window is Then the rate of change of the center of gravity offset Defined as The unit is millimeters per second.

[0091] To ensure comparability between the rate of change of attitude angle and the rate of change of center of gravity offset, both are normalized. Normalization employs a maximum-minimum normalization method, mapping each rate of change sequence to... Interval. Let... The normalization result is , The normalization result is Then the overall rate of change Defined as Its value range is The overall rate of change reflects the intensity of dynamic changes in both posture angle and center of gravity shift, providing a more comprehensive picture of fluctuations in a patient's physical stability. When a patient exhibits instability behaviors such as sudden tilting or significant shifts in the center of gravity, A distinct local maximum, i.e. a peak, will appear.

[0092] In the comprehensive rate of change series In this process, peak detection algorithms are used to identify wave peak locations. Specifically, a minimum peak height threshold is set for peak detection. and minimum peak spacing ,when At some point The local maximum condition is satisfied at and At that time, Marked as the moment of attitude change. Minimum peak spacing. The setting is to avoid duplicate counting in the same pose change event, and is usually set to 0.5 seconds to 1 second based on clinical experience. Each pose change moment... The corresponding attitude change amplitude consists of two parts: one part is the body attitude angle value within the time window at that moment. One part reflects the degree of body tilt; the other part is the shift of the center of gravity within the time window at that moment. This reflects the degree of spatial shift in the center of gravity. Together, they constitute the vector describing the amplitude of posture changes, which is used for subsequent temporal coupling analysis with abnormal EEG fluctuations.

[0093] The final generated attitude change feature sequence is indexed by timestamps, with each record containing the moment of the attitude change. Corresponding body posture angles and center of gravity offset Three fields. This feature sequence fully records all significant postural abrupt changes that occurred in the patient during the observation period, providing reliable motor input data for subsequent calculation of neuro-motor coupling strength. By aligning this sequence with the EEG abnormal fluctuation feature sequence by timestamp, the temporal correlation between abnormal neural activity and limb motor instability can be quantitatively analyzed, thereby assessing the patient's fall risk.

[0094] Align the abnormal fluctuation feature sequence with the attitude change feature sequence by timestamp, calculate the time delay between the abnormal fluctuation time and the attitude change time, and the correlation coefficient between the fluctuation amplitude and the change amplitude. Calculate the neural-motor coupling strength value based on the time delay and correlation coefficient, including:

[0095] The abnormal fluctuation feature sequence and the attitude change feature sequence are synchronously divided into multiple time segment pairs at fixed time intervals. Within each time segment pair, the minimum time difference between the abnormal fluctuation moment and the attitude change moment is searched as the segment time delay.

[0096] All time delay segments are combined to form a time delay sequence, and the standard deviation of the time delay sequence is calculated as the time delay volatility.

[0097] Within each time segment pair, the fluctuation amplitude is extracted to form a fluctuation amplitude subsequence, and the centroid offset is extracted to form a centroid offset subsequence. The fluctuation amplitude subsequence is then shifted along the time axis according to the segment time delay and paired with the centroid offset subsequence. The Pearson correlation coefficient between the paired fluctuation amplitude subsequence and the centroid offset subsequence is calculated as the segment amplitude correlation coefficient.

[0098] The amplitude correlation coefficients of all time segment pairs are used to form an amplitude correlation coefficient sequence, and the mean of the amplitude correlation coefficient sequence is calculated as the average amplitude correlation coefficient.

[0099] A delay stability coefficient is constructed based on the time delay fluctuation, and the delay stability coefficient is fused with the average amplitude correlation coefficient to obtain the neural-motor coupling strength value.

[0100] After generating the abnormal fluctuation feature sequences and the posture change feature sequences, the two feature sequences need to be aligned and correlated on a unified time axis to quantify the coupling relationship between neural electrical activity and motor posture. The core of timestamp alignment is to ensure that the events recorded by the two sequences are comparable under the same absolute time reference frame. Therefore, during the data acquisition phase, the EEG acquisition device and the inertial sensor need to achieve sub-millisecond synchronization through hardware trigger signals or network time protocols to eliminate the interference of device clock drift on subsequent delay calculations.

[0101] After time alignment, the two feature sequences are synchronously divided at fixed time intervals, dividing the entire observation time series into multiple pairs of equal-length time segments. Each pair of time segments consists of an anomalous fluctuation subsequence and a posture change subsequence, which strictly correspond in time range. The selection of segment length needs to balance statistical sufficiency and timeliness. Segments that are too short may not contain enough event pairs, leading to unstable delay estimation; segments that are too long will mask local dynamic changes. Therefore, in practice, the segment length is usually set to the range of several seconds to tens of seconds and is adaptively adjusted according to the frequency of patient events.

[0102] Within each time segment pair, iterate through all pairwise combinations of abnormal fluctuation moments and all attitude change moments within that segment, calculate the absolute time difference between each pair of moments, and take the minimum value as the segment time delay. ,in This serves as a time segment index. The minimum time difference reflects the shortest response time required for a sudden change in motor posture after an abnormal electrical activity occurs within the observed segment; it is a local estimate of the neural-motor conduction delay. When the number of abnormal fluctuation moments or posture change moments within a segment is zero, the segment time delay is marked as a missing value and removed in subsequent statistics.

[0103] Delay the time of all valid segments. Summarized, forming a time-delayed sequence ,in This represents the total number of valid segments. The standard deviation of this sequence is calculated to obtain the time delay volatility. The physical meaning of time delay volatility lies in measuring the stability of the neuromotor response delay: if the patient's neuromodulation function is intact, the delay of each abnormal neuronal discharge triggering a motor response should remain relatively consistent, with low volatility; conversely, if the neuromodulation function is impaired, the variability of the delay increases, and the volatility rises. Therefore, time delay volatility is one of the important indicators reflecting the quality of neuromotor coupling.

[0104] While performing time delay analysis, amplitude-level correlation analysis is also required within each time segment. The amplitude of each abnormal fluctuation event is extracted from the abnormal fluctuation subsequence and arranged according to the event occurrence time to form an amplitude subsequence; the center of gravity shift corresponding to each posture change event is extracted from the posture change subsequence and arranged according to the event occurrence time to form a center of gravity shift subsequence. Because there is a segmental time delay between neural events and motion events... If the correlation between the two subsequences is calculated directly, the correlation coefficient will be underestimated due to the time misalignment. Therefore, the fluctuation amplitude subsequence is shifted backward on the time axis. The time units are used to align the center of gravity offset subsequence with the center of gravity offset subsequence in time, and then the offset fluctuation amplitude subsequence is paired with the center of gravity offset subsequence.

[0105] The Pearson correlation coefficient is calculated for the two paired subsequences to obtain the segment amplitude correlation coefficient. The Pearson correlation coefficient measures the degree of linear correlation between two numerical sequences of equal length, and its value ranges from [value missing]. . A value close to 1 indicates that within this segment, there is a strong positive correlation between the amplitude of abnormal neural fluctuations and the subsequent shift in the center of gravity of motion, meaning that larger abnormal neural discharges tend to trigger larger motor instability. Values ​​close to 0 or negative indicate a weak or inconsistent magnitude correlation between the two. When calculating the magnitude correlation coefficient of segments, if the number of paired data points within a segment is insufficient to support statistical significance, the segment is also marked as invalid and removed in subsequent processing.

[0106] The correlation coefficient of segment amplitude of all valid segments The summaries form a sequence of amplitude correlation coefficients. The arithmetic mean of the sequence is calculated to obtain the average magnitude correlation coefficient. The average amplitude correlation coefficient reflects the strength of the linear correlation between the amplitude of abnormal neural discharges and the amplitude of motion instability over the entire observation period, and is one of the core quantitative indicators for assessing coupling strength.

[0107] In obtaining time delay fluctuation Correlation coefficient with average amplitude Next, the two need to be merged into a single neuro-motor coupling strength value. Delay stability coefficient Due to time delay fluctuation The structure reflects the contribution of delay stability to coupling quality. The smaller the value, the more stable the time delay, the stronger the regularity of the neural-motor coupling, and the higher the corresponding delay stability coefficient. The higher. Specifically, Using negative exponential transformation The mapping yields, i.e. ,in The regularization parameter, used to control the decay rate, is calibrated based on the distribution characteristics of historical patient data. When When approaching zero, A value approaching 1 indicates a highly stable latency; when When it is large, A value close to 0 indicates that the latency is highly unstable.

[0108] Delay stability coefficient Correlation coefficient with average amplitude Weighted fusion is performed to obtain the neural-motor coupling strength value. The calculation method is as follows ,in The fusion weight is used to balance the relative contributions of delay stability and amplitude correlation to the coupling strength. The value of was determined through regression analysis of patient data on known fall events, making It has the best predictive ability for fall risk. (Neuro-motor coupling strength value) By combining response stability in the time dimension with the intensity of coordinated changes in the amplitude dimension, it can more comprehensively characterize the degree of coordination between the patient's neuromodulation and motor control, providing a reliable quantitative basis for subsequent risk level classification.

[0109] In practical applications, The values ​​are continuously updated in a sliding manner throughout the entire monitoring period. The time segments are re-divided and recalculated at fixed intervals to track the dynamic changes in the neuro-motor coupling state in real time, providing continuous risk warning information for clinical nurses.

[0110] Pairing the fluctuation amplitude subsequence with the centroid offset subsequence after shifting it along the time axis according to the segment time delay includes:

[0111] The time delay value of the segment is added to the timestamp corresponding to each fluctuation amplitude value in the fluctuation amplitude subsequence to obtain an offset timestamp sequence. The fluctuation amplitude subsequence is then re-marked according to the offset timestamp sequence to obtain the fluctuation amplitude subsequence after time axis offset.

[0112] Using the timestamp corresponding to each centroid offset in the centroid offset subsequence as the reference time, the fluctuation amplitude value with the smallest time difference from the reference time is searched in the fluctuation amplitude subsequence after the time axis offset and paired to generate a paired dataset of fluctuation amplitude value and centroid offset value.

[0113] All fluctuation amplitude values ​​are extracted from the paired dataset to form a paired fluctuation amplitude subsequence, and all centroid offset values ​​are extracted from the paired dataset to form a paired centroid offset subsequence.

[0114] After obtaining the fluctuation amplitude subsequence and the centroid shift subsequence, they need to be aligned in the time dimension through a time axis offset operation to eliminate the inherent time delay between abnormal fluctuations in neural signals and changes in limb posture. This delay originates from the physiological time required for nerve impulses to travel from the amygdala region to the motor execution system. Directly using the original timestamps for pairing will result in a severe underestimation of the amplitude correlation between the two sequences. Therefore, the time axis of the fluctuation amplitude subsequence must be offset and corrected before performing the pairing operation.

[0115] The specific operation of time axis offset is as follows: Each fluctuation amplitude value in the fluctuation amplitude subsequence corresponds to an original timestamp, denoted as the first... The fluctuation amplitude subsequence within the time segment of the th time segment The original timestamps of each sample point are Add the time delay of the segment corresponding to the current time segment. That is, to obtain the offset timestamp of the sample point. ,satisfy After performing the above operations on all sample points in the fluctuation amplitude subsequence, a complete offset timestamp sequence is obtained. Then, the timestamp of each fluctuation amplitude value in the fluctuation amplitude subsequence is replaced with the corresponding offset timestamp, completing the re-marking and obtaining the time-axis-offset fluctuation amplitude subsequence. This subsequence is numerically identical to the original fluctuation amplitude subsequence; the only difference is that the time coordinate corresponding to each data point is shifted backward as a whole. This allows the time axis of neural signals to be logically aligned with the time axis of motion signals.

[0116] After completing the time axis offset, the pairing search phase begins. Using the timestamp corresponding to each centroid offset in the centroid offset subsequence as the base time, let's denote the... The timestamp of each centroid offset sample point is In the fluctuation amplitude subsequence after the time axis offset, each offset timestamp is calculated one by one. Compared with the current reference time The absolute value of the time difference between From all candidate fluctuation amplitude values, the sample point that minimizes the difference is selected as the pairing object. Its fluctuation amplitude value is then paired with the centroid offset value corresponding to the current reference time to form a paired data pair, which is added to the paired dataset. The above nearest neighbor search operation is performed on each sample point in the centroid offset subsequence, and finally, a complete paired dataset of fluctuation amplitude values ​​and centroid offset values ​​is obtained.

[0117] When performing nearest neighbor search in practice, it is necessary to consider setting a tolerance range for time difference. If the sampling rates of the fluctuation amplitude subsequence after time axis offset and the centroid offset subsequence are different, or if the timestamps of the two sequences are slightly misaligned due to clock drift of the signal acquisition equipment, the minimum time difference may not be zero. To ensure pairing quality, a maximum allowable time difference threshold can be preset. When the minimum time difference found exceeds When the center of gravity offset sample points are not paired, this avoids forcibly pairing sample points with large time differences and introducing noise. The value of is usually determined by the sampling periods of the two signal sequences. It is generally set to about 1.5 times the larger of the two sampling periods, so as to filter out abnormal pairings with significantly large time differences while covering a reasonable pairing range.

[0118] After the paired dataset is constructed, it is unpacked and extracted. The fluctuation amplitude values ​​of each paired data pair are read one by one from the paired dataset and arranged in the pairing order to form a paired fluctuation amplitude subsequence. Similarly, the centroid offset values ​​are extracted from each paired data pair and arranged in the same order to form a paired centroid offset subsequence. After these operations, the two subsequences are strictly equal in length and correspond one-to-one in their index positions. That is, elements with the same index in the two sequences are the closest matching pairs in time after delay correction, satisfying the subsequent calculation of the segment amplitude correlation coefficient. The required element-by-element correspondence.

[0119] It is important to note that the time axis offset operation only modifies the timestamps of each sample point in the fluctuation amplitude subsequence, without performing any filtering, interpolation, or scaling on the fluctuation amplitude values ​​themselves, thus preserving the original amplitude information of the abnormal fluctuations in the neural signal. Similarly, the centroid offset subsequence is not numerically transformed, maintaining the original centroid offset obtained from sensor acquisition and integration calculations. This is done to ensure the accuracy of subsequent calculations of segment amplitude correlation coefficients. In this way, the data used can truly reflect the amplitude covariance between neural activity and changes in body posture, avoiding artificial correlation bias introduced by numerical transformation.

[0120] At the engineering implementation level, the fluctuation amplitude subsequence and the centroid offset subsequence are typically stored in the form of timestamp-value pairs. Time axis offset operations can be performed by uniformly adding a value to the timestamp column of the fluctuation amplitude subsequence. The implementation has a computational complexity of O(n). ,in This represents the number of sample points in the fluctuation amplitude subsequence. If the nearest neighbor search operation uses a linear scan, the computational complexity is O(n). ,in This represents the number of sample points in the centroid offset subsequence. If the offset timestamp sequence is pre-sorted, binary search can be used to reduce the complexity of each search to [missing information]. The complexity of the overall pairing process is reduced to It has good computational efficiency in real-time or near-real-time clinical monitoring scenarios.

[0121] The two paired subsequences will be used to calculate the segment amplitude correlation coefficient. The input, and thus participates in the neural-motor coupling strength value The accuracy of the time axis offset pairing operation directly affects the calculation. The accuracy of the estimation affects the reliability of the final fall risk level classification. Therefore, the time delay of the segment... The estimation accuracy is a key prerequisite for the entire pairing process. Sufficient time resolution must be ensured during the time delay calculation stage so that the offset fluctuation amplitude subsequence can be accurately aligned with the centroid offset subsequence in the time dimension.

[0122] The probability of a patient losing balance control is calculated by statistically analyzing the percentage of time the neuro-motor coupling strength value exceeds a preset coupling threshold, and combining this with the cumulative shift of the center of gravity in the posture change feature sequence. Risk levels are then categorized based on this probability value, including:

[0123] The cumulative duration of the neural-motor coupling strength value exceeding the preset coupling threshold is counted as a percentage of the monitoring period.

[0124] The cumulative offset of the center of gravity offset is obtained by summing the center of gravity offsets along the time axis in the posture change feature sequence.

[0125] At the start and end times of the period when the neural-motor coupling strength value exceeds the preset coupling threshold, the cumulative offset of the center of gravity offset is extracted and the difference is calculated to obtain the offset increment of the coupling abnormal period. The offset increment of the coupling abnormal period is normalized and coupled with the time proportion to obtain the risk value of the coupling period.

[0126] Extract the posture mutation feature sequence within the time period when the neural-motor coupling strength value exceeds the preset coupling threshold, count the number of posture mutation events in the posture mutation feature sequence to obtain the runaway event frequency, and perform time series analysis on the runaway event frequency to extract the change trend to obtain the runaway acceleration trend index;

[0127] The cumulative offset of the center of gravity offset is correlated with the uncontrolled acceleration trend index to obtain the cumulative dynamic risk value. Based on the interaction between the risk value of the coupling period and the cumulative dynamic risk value, the probability value of the patient losing balance control ability is calculated.

[0128] The risk level is classified and output based on the preset probability threshold range in which the probability value falls.

[0129] After obtaining the sequence of neural-motor coupling strength values, it is necessary to further transform them into quantifiable probabilities of loss of balance control capability, and finally map them to the risk level output. The entire calculation process revolves around two parallel analysis paths: one focuses on the center of gravity shift behavior during periods of coupling anomalies, and the other focuses on the temporal evolution trend of abrupt posture events. The two paths are eventually interacted and fused to obtain a comprehensive probability value.

[0130] The statistical neural-motor coupling strength value exceeds a preset coupling threshold. The cumulative duration is a fundamental step in quantifying the degree of coupling anomalies. Let the total duration of the monitoring period be... During this period, all values ​​satisfying the coupling strength value exceeding The sum of the durations of the time periods is denoted as Then the time percentage Defined as:

[0131]

[0132] The larger the value, the longer the patient remains in a state of abnormal neuro-motor coupling during the monitoring period, and the higher the likelihood that their balance regulation mechanism is continuously disturbed. This indicator only reflects the degree of abnormality in the time dimension and does not include information on the amplitude of motion; therefore, it needs to be combined with the cumulative changes in the center of gravity shift for joint analysis.

[0133] The centroid offsets in the attitude change feature sequence are accumulated along the time axis to obtain the cumulative offset. Its expression is the centroid offset within each time window. The gradual accumulation. At the beginning of each coupling anomaly period. and termination time Extract the corresponding cumulative offsets respectively. and The difference between the two is the offset increment during the coupling anomaly period in that time period. :

[0134]

[0135] If there are multiple [conditions] within the monitoring period If there are several coupling anomaly periods, the total coupling anomaly period offset increment is obtained by summing the offset increments of each period. To eliminate the influence of differences in body size and range of motion among different patients, [the following was performed]: Normalization is performed, and the normalized reference value is the maximum value of all center of gravity offsets within the monitoring period. The normalized offset increment is denoted as :

[0136]

[0137] normalized offset increment With time percentage Perform coupling operations to obtain the risk value during the coupling period. The coupling operation uses a weighted product form to simultaneously reflect the contributions of both the anomaly duration and the magnitude of the centroid shift during the anomaly:

[0138]

[0139] in The fusion weighting coefficient between the time percentage and the normalized offset increment, with a value range of [value range missing]. It can be adjusted based on prior clinical knowledge. When When the magnitude is large, the degree of anomaly in the time dimension contributes more to the risk value of the coupling period; when When the magnitude is small, the contribution of the center of gravity offset is more prominent.

[0140] During the period of abnormal coupling, the corresponding attitude change feature sequences are extracted, and the number of attitude change events occurring in these sequences is counted and recorded as the frequency of out-of-control events. . This reflects the absolute number of sudden postural instability events in a patient's limbs under abnormal neuro-motor coupling conditions. However, simple frequency statistics cannot reflect whether these instability events show an accelerating trend over time. Therefore, a time-series analysis was performed on the occurrence time sequences of instability events within each period of abnormal coupling. The change in the frequency of instability events between adjacent periods of abnormal coupling was calculated, and a linear trend was fitted to the change sequence. The sign and magnitude of the slope reflect the accelerating or decelerating trend of the frequency of instability events. After standardization, the instability acceleration trend index was obtained. ,when The frequency of loss-of-control events increased over time, indicating a continued deterioration in the patient's balance control ability; when This indicates that the frequency of out-of-control events is stabilizing or decreasing.

[0141] Cumulative offset during the monitoring period With the trend of out-of-control acceleration indicator Correlation analysis was performed to obtain the cumulative dynamic risk value. The core of correlation analysis lies in the fact that if the cumulative shift of the center of gravity continues to increase, while the frequency of out-of-control events also accelerates, the synergistic effect of the two will significantly amplify the patient's risk of losing balance control. Specifically, for Normalization is performed, and the normalization reference value is the preset maximum allowable cumulative offset. The normalization result is denoted as :

[0142]

[0143] Cumulative dynamic risk value Defined as:

[0144]

[0145] in The amplification factor of the runaway acceleration trend indicator on the cumulative dynamic risk value, when hour, The cumulative dynamic risk value degenerates into a normalized value of the pure cumulative offset, unaffected by the trend acceleration effect.

[0146] Risk value based on coupling period With cumulative dynamic risk value The interaction relationship is used to calculate the probability of a patient losing balance control. The interaction relationship employs a non-linear fusion approach to prevent a single indicator from having an excessively dominant effect on the probability value.

[0147]

[0148] in This is a non-linear adjustment index used to control the curve shape when mapping risk values ​​to probability values. When, the above equation degenerates into a linear fusion form; when At that time, the probability mapping of the low-risk value range is more sensitive, which is helpful for early warning; when At this time, the probability mapping of the high-risk value range is more prominent, which is suitable for emphasizing extreme risk states. The range of values ​​is constrained to The values ​​between these values ​​directly correspond to the probability estimate of a patient losing balance control during the current monitoring period.

[0149] according to The patient's risk level is determined by the preset probability threshold range. The interval is divided into several sub-intervals, each corresponding to a different risk level, such as low risk, medium risk, high risk, and very high risk. The threshold boundaries for each level can be configured based on clinical data and expert experience. When the risk falls within the corresponding range, the corresponding risk level is output, along with key intermediate indicators supporting this level judgment, including the percentage of time spent in the risk assessment. Risk values ​​during coupling period Frequency of out-of-control events Uncontrolled acceleration trend indicator and cumulative dynamic risk value This allows clinicians to trace and verify the prediction results and make intervention decisions. This output mechanism not only provides a final risk assessment but also preserves the complete intermediate calculation chain, enhancing the interpretability and clinical credibility of the prediction results.

[0150] The cumulative offset of the center of gravity shift is correlated with the runaway acceleration trend index to obtain a cumulative dynamic risk value. Based on the interaction between the risk value during the coupling period and the cumulative dynamic risk value, the probability value of the patient losing balance control is calculated, including:

[0151] The cumulative offset of the center of gravity offset is segmented into time windows to extract the rate of change feature;

[0152] The change rate feature and the runaway acceleration trend indicator are aligned according to the timestamp and the temporal correlation is calculated to obtain the dynamic coupling coefficient. The dynamic coupling coefficient is used as the correlation weight to weight the cumulative offset of the center of gravity offset to obtain the cumulative dynamic risk value.

[0153] The synergistic enhancement coefficient is obtained by multiplying the risk value of the coupling period and the cumulative dynamic risk value. Based on the synergistic enhancement coefficient, the risk value of the coupling period and the cumulative dynamic risk value are amplified to obtain the amplified risk value of the coupling period and the amplified cumulative dynamic risk value.

[0154] The amplified coupling period risk value and the amplified cumulative dynamic risk value are multiplied to obtain an interactive amplification term. The amplified coupling period risk value and the amplified cumulative dynamic risk value are weighted and summed to obtain a linear fusion term. The interactive amplification term and the linear fusion term are superimposed to obtain the probability value of the patient losing balance control ability.

[0155] After calculating the risk value during the coupling period, it is necessary to further explore the deeper correlation between the cumulative change in the center of gravity offset and the runaway acceleration trend in order to construct a more discriminative dynamic risk assessment indicator. Regarding the cumulative offset... Processing is divided into segments according to fixed time windows, with the length of each segment window consistent with the previous steps. Extract the rate of change of cumulative offset within each segmented window. The calculation method is the difference between the cumulative offset at the end of the current window and the cumulative offset at the end of the previous window, divided by the window duration. ,in For the first At the end of the window, For the first At the end of the window, The time window length. Rate of change characteristic. It reflects the acceleration or deceleration trend of the center of gravity shift within a local time period. When the patient's balance regulation ability declines, this value will show a continuous increasing trend.

[0156] Extract the rate of change feature sequence from each segmented window With the trend of out-of-control acceleration indicator Alignment is performed according to timestamps. After alignment, calculations are performed within each window. and The Pearson correlation coefficient between corresponding components is used to measure the degree of co-change between them in the time dimension, thus obtaining the dynamic coupling coefficient. . The range of values ​​is ,when When the value approaches 1, it indicates that the acceleration of the center of gravity shift and the trend of loss of control are highly synchronized, suggesting a significantly increased risk of loss of balance control for the patient during this period; when A value close to 0 or negative indicates a weak correlation between the two factors and a relatively low risk level. The dynamic coupling coefficient... As the correlation weight, the cumulative offset within the corresponding window is weighted to obtain the weighted segmented cumulative contribution value. The weighted contribution values ​​of all segmented windows are summed and normalized to obtain the cumulative dynamic risk value. The normalized reference value uses the preset maximum allowable cumulative offset. The normalized cumulative dynamic risk value can eliminate the difference in the absolute dimension of the center of gravity shift between different patients, making the subsequent fusion calculation comparable across patients.

[0157] Obtaining risk values ​​during the coupling period With cumulative dynamic risk value Subsequently, a synergistic enhancement coefficient was constructed to capture the interactive amplification effect between the two. Synergistic enhancement coefficient Defined as the product of the two, i.e. .when and When both are relatively high, This will significantly amplify the risk, consistent with clinical observations showing a sharp increase in fall risk when neurological abnormalities and motor instability worsen simultaneously. (Using the synergistic enhancement coefficient...) Based on this, respectively and Perform amplification calculations to determine the risk value during the amplified coupling period. Compared with the amplified cumulative dynamic risk value The calculation method is as follows: , ,in It is the amplification factor of the runaway acceleration trend indicator on the cumulative dynamic risk value, used to control the strength of the synergistic enhancement effect. The value was determined through statistical analysis of historical data. In the neurology patient population, the typical value range is [value range missing]. to The specific values ​​need to be calibrated based on clinical validation results. The physical significance of the amplification operation is that when both risk dimensions are in a high-risk state, considering either dimension alone will underestimate the actual fall risk. However, by introducing a synergistic enhancement coefficient, the nonlinear effect of the two risk factors superimposing and reinforcing each other can be simulated at the computational level.

[0158] Based on the two amplified risk values, an interactive amplification term and a linear fusion term are constructed, and the two are superimposed to obtain the probability value of the patient losing balance control ability. Interactive zoom item Defined as the product of the amplified risk value during the coupling period and the amplified cumulative dynamic risk value, i.e. This term is specifically used to characterize the synergistic explosive effect of two types of risk factors under high-risk conditions. When both are close to their upper limits, the contribution of the interaction amplification term will far exceed the magnitude of linear superposition. Linear fusion term. The basic contributions of the two types of risks are then combined through a weighted summation method, namely... ,in The fusion weighting coefficient between the time percentage and the normalized offset increment is used to adjust the relative importance of the two types of risks in the linear fusion. In the scenario of fall risk assessment for neurological patients, abnormal neurological fluctuations have a stronger leading effect on loss of motor control, therefore... The value is usually slightly greater than This is to assign a higher base weight to the risk value during the coupling period.

[0159] Finally, the interactive amplification term and the linear fusion term are superimposed, and the exponent is adjusted nonlinearly. Perform a power transformation on the superposition results to obtain the probability values. The calculation formula is: Nonlinear adjustment index The introduction of this is to map the superposition result to The probability space is calculated, and the intermediate risk range is stretched or compressed so that the probability value of the low-risk range remains at a low level and the probability value of the high-risk range quickly approaches 1, thereby improving the sensitivity of risk warning. The typical range of values ​​is to ,when When the curve is convex upwards, it is more sensitive to high-risk conditions; when At this time, the curve bulges downwards, demonstrating a stronger ability to distinguish between low-risk states. In actual deployment, The optimal value was determined by analyzing the receiver operating characteristic curve on a historical dataset of labeled fall events, with the optimization objective being to maximize the area under the curve.

[0160] Calculated Then, based on the pre-set risk grading threshold, it is mapped to the corresponding risk level, so that clinical medical staff can take timely intervention measures, thereby achieving dynamic and accurate prediction of the fall risk of neurology patients.

[0161] A second aspect of the present invention provides a fall risk prediction system for neurology patients based on temporal coupling, comprising:

[0162] The EEG acquisition unit is used to acquire EEG signals from the patient's amygdala region via an electrode array to obtain raw EEG data.

[0163] The time-frequency analysis unit is used to decompose the raw EEG data in the time-frequency domain and extract frequency domain features and time domain features. It detects abnormal fluctuations through a sliding time window and generates an abnormal fluctuation feature sequence containing the time, duration and amplitude of the abnormal fluctuations.

[0164] The posture detection unit is used to collect the acceleration and angular velocity signals of the patient's limbs and convert them into body posture angles and center of gravity offsets. It identifies posture changes by calculating the rate of change between adjacent time windows and generates a posture change feature sequence.

[0165] The coupling analysis unit is used to align the abnormal fluctuation feature sequence and the attitude change feature sequence by timestamp, calculate the time delay between the abnormal fluctuation time and the attitude change time and the correlation coefficient between the fluctuation amplitude and the change amplitude, and calculate the neural-motor coupling strength value based on the time delay and the correlation coefficient.

[0166] The risk assessment unit is used to calculate the probability value of the patient losing balance control ability by statistically analyzing the percentage of time when the neuro-motor coupling strength value exceeds a preset coupling threshold, and by combining the cumulative offset of the center of gravity offset in the posture mutation feature sequence, and classifying the risk level according to the probability value.

[0167] A third aspect of the present invention provides an electronic device, comprising:

[0168] processor;

[0169] Memory used to store processor-executable instructions;

[0170] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0171] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0172] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting fall risk in neurology patients based on temporal coupling, characterized in that, include: Raw EEG data was obtained by acquiring EEG signals from the amygdala region of the patient using an electrode array. The original EEG data is decomposed in the time and frequency domains and frequency and time domain features are extracted. Abnormal fluctuations are detected by sliding time windows, and an abnormal fluctuation feature sequence containing the time, duration and amplitude of abnormal fluctuations is generated. Acceleration and angular velocity signals of the patient's limbs are collected and converted into body posture angles and center of gravity offsets. Posture mutations are identified by calculating the rate of change between adjacent time windows, and a posture mutation feature sequence is generated. Align the abnormal fluctuation feature sequence and the attitude change feature sequence by timestamp, calculate the time delay between the abnormal fluctuation time and the attitude change time and the correlation coefficient between the fluctuation amplitude and the change amplitude, and calculate the neural-motor coupling strength value based on the time delay and the correlation coefficient. The probability of a patient losing balance control is calculated by statistically analyzing the percentage of time during which the neuro-motor coupling strength value exceeds a preset coupling threshold, and by combining this with the cumulative offset of the center of gravity shift in the posture mutation feature sequence. The risk level is then classified based on the probability value.

2. The method according to claim 1, characterized in that, The raw EEG data is decomposed into time and frequency domains, and frequency and time domain features are extracted. Abnormal fluctuations are detected using a sliding time window, generating an abnormal fluctuation feature sequence that includes the time, duration, and amplitude of the abnormal fluctuations. The original EEG data is decomposed into multiple frequency band components by wavelet transform. The power spectral density of each frequency band component within the sliding time window is calculated as the frequency domain feature, and the peak amplitude and zero-crossing rate of each frequency band component within the sliding time window are calculated as the time domain features. EEG data of the patient at rest was collected as baseline EEG data. Baseline frequency domain features and baseline time domain features of the baseline EEG data were extracted. The frequency domain deviation between the frequency domain features and the baseline frequency domain features and the time domain deviation between the time domain features and the baseline time domain features were calculated. When the frequency domain deviation or the time domain deviation exceeds the deviation threshold, the center moment of the current sliding time window is marked as the abnormal fluctuation moment, and the time span from the first deviation exceeding the deviation threshold to falling back below the deviation threshold is recorded as the duration. Within the duration, the power spectral density change is weighted and fused based on the frequency domain deviation of each frequency band component to obtain the fluctuation amplitude, and an abnormal fluctuation feature sequence containing the abnormal fluctuation time, duration, and fluctuation amplitude is generated.

3. The method according to claim 1, characterized in that, Acceleration and angular velocity signals of the patient's limbs are collected and converted into body posture angles and center of gravity offsets. Posture abrupt changes are identified by calculating the rate of change between adjacent time windows, and a posture abrupt change feature sequence is generated, including: Acceleration and angular velocity signals of the patient's limbs are collected, and the acceleration signals are separated into static acceleration and dynamic acceleration components by gravity component separation. The initial body posture angle is calculated based on the projection direction of the static acceleration component, and the center of gravity offset is calculated based on the dual time integration of the dynamic acceleration component. The angular velocity signal is integrated within a preset time window to obtain the attitude rotation angle increment, and the initial body attitude angle is superimposed with the attitude rotation angle increment to obtain the body attitude angle. The body posture angle and the center of gravity offset are divided into time windows. The rate of change of the body posture angle and the rate of change of the center of gravity offset between adjacent time windows are calculated. The rate of change of the body posture angle and the rate of change of the center of gravity offset are normalized and then summed to obtain the comprehensive rate of change. The moment of attitude change is marked by the peak position of the comprehensive rate of change. The body posture angle and the center of gravity offset are extracted as the attitude change amplitude to generate an attitude change feature sequence containing the moment of attitude change and the attitude change amplitude.

4. The method according to claim 1, characterized in that, Align the abnormal fluctuation feature sequence with the attitude change feature sequence by timestamp, calculate the time delay between the abnormal fluctuation time and the attitude change time, and the correlation coefficient between the fluctuation amplitude and the change amplitude. Calculate the neural-motor coupling strength value based on the time delay and correlation coefficient, including: The abnormal fluctuation feature sequence and the attitude change feature sequence are synchronously divided into multiple time segment pairs at fixed time intervals. Within each time segment pair, the minimum time difference between the abnormal fluctuation moment and the attitude change moment is searched as the segment time delay. All time delay segments are combined into a time delay sequence, and the standard deviation of the time delay sequence is calculated as the time delay volatility. Within each time segment pair, the fluctuation amplitude is extracted to form a fluctuation amplitude subsequence, and the centroid offset is extracted to form a centroid offset subsequence. The fluctuation amplitude subsequence is then shifted along the time axis according to the segment time delay and paired with the centroid offset subsequence. The Pearson correlation coefficient between the paired fluctuation amplitude subsequence and the centroid offset subsequence is calculated as the segment amplitude correlation coefficient. The amplitude correlation coefficients of all time segment pairs are used to form an amplitude correlation coefficient sequence, and the mean of the amplitude correlation coefficient sequence is calculated as the average amplitude correlation coefficient. A delay stability coefficient is constructed based on the time delay fluctuation, and the delay stability coefficient is fused with the average amplitude correlation coefficient to obtain the neural-motor coupling strength value.

5. The method according to claim 4, characterized in that, Pairing the fluctuation amplitude subsequence with the centroid offset subsequence after shifting it along the time axis according to the segment time delay includes: The time delay value of the segment is added to the timestamp corresponding to each fluctuation amplitude value in the fluctuation amplitude subsequence to obtain an offset timestamp sequence. The fluctuation amplitude subsequence is then re-marked according to the offset timestamp sequence to obtain the fluctuation amplitude subsequence after time axis offset. Using the timestamp corresponding to each centroid offset in the centroid offset subsequence as the reference time, the fluctuation amplitude value with the smallest time difference from the reference time is searched in the fluctuation amplitude subsequence after the time axis offset and paired to generate a paired dataset of fluctuation amplitude value and centroid offset value. All fluctuation amplitude values ​​are extracted from the paired dataset to form a paired fluctuation amplitude subsequence, and all centroid offset values ​​are extracted from the paired dataset to form a paired centroid offset subsequence.

6. The method according to claim 1, characterized in that, The probability of a patient losing balance control is calculated by statistically analyzing the percentage of time the neuro-motor coupling strength value exceeds a preset coupling threshold, and combining this with the cumulative shift of the center of gravity in the posture change feature sequence. Risk levels are then categorized based on this probability value, including: The cumulative duration of the neural-motor coupling strength value exceeding the preset coupling threshold is counted as a percentage of the monitoring period. The cumulative offset of the center of gravity offset is obtained by summing the center of gravity offsets along the time axis in the posture change feature sequence. At the start and end times of the period when the neural-motor coupling strength value exceeds the preset coupling threshold, the cumulative offset of the center of gravity offset is extracted and the difference is calculated to obtain the offset increment of the coupling abnormal period. The offset increment of the coupling abnormal period is normalized and coupled with the time proportion to obtain the risk value of the coupling period. Extract the posture mutation feature sequence within the time period when the neural-motor coupling strength value exceeds the preset coupling threshold, count the number of posture mutation events in the posture mutation feature sequence to obtain the runaway event frequency, and perform time series analysis on the runaway event frequency to extract the change trend to obtain the runaway acceleration trend index; The cumulative offset of the center of gravity offset is correlated with the uncontrolled acceleration trend index to obtain the cumulative dynamic risk value. Based on the interaction between the risk value of the coupling period and the cumulative dynamic risk value, the probability value of the patient losing balance control ability is calculated. The risk level is classified and output based on the preset probability threshold range in which the probability value falls.

7. The method according to claim 6, characterized in that, The cumulative offset of the center of gravity shift is correlated with the runaway acceleration trend index to obtain a cumulative dynamic risk value. Based on the interaction between the risk value during the coupling period and the cumulative dynamic risk value, the probability value of the patient losing balance control is calculated, including: The cumulative offset of the center of gravity offset is segmented into time windows to extract the rate of change feature; The change rate feature and the runaway acceleration trend indicator are aligned according to the timestamp and the temporal correlation is calculated to obtain the dynamic coupling coefficient. The dynamic coupling coefficient is used as the correlation weight to weight the cumulative offset of the center of gravity offset to obtain the cumulative dynamic risk value. The synergistic enhancement coefficient is obtained by multiplying the risk value of the coupling period and the cumulative dynamic risk value. Based on the synergistic enhancement coefficient, the risk value of the coupling period and the cumulative dynamic risk value are amplified to obtain the amplified risk value of the coupling period and the amplified cumulative dynamic risk value. The amplified coupling period risk value and the amplified cumulative dynamic risk value are multiplied to obtain an interactive amplification term. The amplified coupling period risk value and the amplified cumulative dynamic risk value are weighted and summed to obtain a linear fusion term. The interactive amplification term and the linear fusion term are superimposed to obtain the probability value of the patient losing balance control ability.

8. A fall risk prediction system for neurological patients based on temporal coupling, used to implement the method as described in any one of claims 1-7, characterized in that, include: The EEG acquisition unit is used to acquire EEG signals from the patient's amygdala region via an electrode array to obtain raw EEG data. The time-frequency analysis unit is used to decompose the raw EEG data in the time-frequency domain and extract frequency domain features and time domain features. It detects abnormal fluctuations through a sliding time window and generates an abnormal fluctuation feature sequence containing the time, duration and amplitude of the abnormal fluctuations. The posture detection unit is used to collect the acceleration and angular velocity signals of the patient's limbs and convert them into body posture angles and center of gravity offsets. It identifies posture changes by calculating the rate of change between adjacent time windows and generates a posture change feature sequence. The coupling analysis unit is used to align the abnormal fluctuation feature sequence and the attitude change feature sequence by timestamp, calculate the time delay between the abnormal fluctuation time and the attitude change time and the correlation coefficient between the fluctuation amplitude and the change amplitude, and calculate the neural-motor coupling strength value based on the time delay and the correlation coefficient. The risk assessment unit is used to calculate the probability value of the patient losing balance control ability by statistically analyzing the percentage of time when the neuro-motor coupling strength value exceeds a preset coupling threshold, and by combining the cumulative offset of the center of gravity offset in the posture mutation feature sequence, and classifying the risk level according to the probability value.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images