Tumor patient falling prediction system based on machine learning

Abstract

The invention relates to the technical field of tumble prediction, in particular to a tumor patient tumble prediction system based on machine learning, which comprises a stride frequency recognition module, a physiological fluctuation monitoring module, a gait coupling judgment module, a posture analysis module and a risk output regulation and control module. According to the method, by accurately positioning the wave crests and the wave troughs in the acceleration signal period and analyzing the frequency change direction of the adjacent sections, accurate marking of the mutation sections is achieved, the sensitive capture capacity of behavior state changes is effectively improved, and the synchronous extraction mode of the mutation sections and the physiological signals is improved. Through consistency comparison of attitude angle change and gravity center track direction, accuracy of motion structure abnormity and imbalance state identification is enhanced, pressure distribution and spine compression direction are extracted, corresponding pairing between structure pressure and physiological load is realized, optimization of time sequence integrity is realized, and accuracy of motion structure abnormity and imbalance state identification is improved. And multi-dimensional data fusion analysis has high pertinence and stability.

Description

Technical Field

[0001] The present invention relates to the technical field of fall prediction, and particularly to a fall prediction system for cancer patients based on machine learning. Background Art

[0002] The technical field of fall prediction involves collecting, modeling, and analyzing human behavior data to identify potential fall risks and trends, so as to take preventive measures in advance. The core content of this technical field is to use machine learning algorithms to model the dynamic and physiological data of different populations, extract characteristic variables, and analyze the individual's fall probability by combining health information, movement patterns, and environmental factors. Overall, this technical field provides fall risk level assessment and individualized prediction services through multi-modal data input, data feature selection, and model training, and is widely used in the health monitoring and risk management of the elderly, patients, and other fall-prone populations.

[0003] Among them, the fall prediction system for cancer patients refers to a system that collects the daily activity data, health monitoring data, and pathological status information of cancer patients, trains an individualized fall risk model through machine learning methods, and combines multiple characteristic variables such as gait pattern, physical fitness status, body position change, and physiological indicators for data analysis and risk prediction. The patent theme mainly focuses on the fall risk problem of cancer patients under special physical conditions and diseases, and establishes an individualized fall risk prediction system by integrating motion monitoring data and pathological information, so as to achieve a scientific assessment of the fall risk.

[0004] The existing technology relies on multi-modal data for modeling and evaluation in fall prediction, but there is a lack of a continuous dynamic association mechanism between behavioral signals and physiological data, resulting in a lag in capturing individual state mutations. Although the characteristic variables have an identification effect, their static nature is strong, and it is difficult to cope with mutations or non-linear changes during the movement process. Especially when facing a group with variable pathological states, there is a problem of weak model generalization ability. Most existing systems generate data labels by manually setting thresholds or segmentation methods, resulting in limited data understanding ability and unable to fully express the deep interaction relationship between data. Taking cancer patients as an example, traditional solutions are difficult to effectively extract linkage signals from continuous posture evolution or physiological state fluctuations, unable to pre-judge latent fall risks in advance, easily causing a lag in risk judgment, reducing the warning effect, and affecting the timeliness of intervention and the quality of individualized precision management. Summary of the Invention

[0005] The purpose of the present invention is to solve the disadvantages existing in the prior art, and to propose a fall prediction system for cancer patients based on machine learning.

[0006] In order to achieve the above purpose, the present invention adopts the following technical solution. The fall prediction system for cancer patients based on machine learning includes: A step frequency recognition module, which is used to obtain the acceleration signal and the ground contact time when a tumor patient walks, locate the peak and trough positions in the step frequency cycle, segment the continuous step frequency states and identify the change direction. When the frequency change direction of adjacent paragraphs changes, mark the time period as a mutation section and generate a step frequency mutation stage label sequence; A physiological fluctuation monitoring module, which is used to synchronously extract the patient's heart rate and blood oxygen signals within the marked mutation period based on the step frequency mutation stage label sequence, identify the signal segments according to the change trend within the time period, and generate a step frequency physiological synchronization annotation segment list; A gait coupling judgment module, which is used to extract the corresponding attitude angle signal and the center of gravity movement trajectory according to the step frequency physiological synchronization annotation segment list, compare the direction of the attitude angle change trend with the lateral offset path of the center of gravity, and generate a gait physiological coupling abnormal segment sequence; An attitude analysis module, which is used to extract the plantar pressure map and the position information of the spinal compression area within the coupling segment through the gait physiological coupling abnormal segment sequence, pair the position of the plantar pressure transfer direction and the spinal compression side positioning number, and generate an attitude offset compression segment position identifier.

[0007] As a further solution of the present invention, the step frequency mutation stage label sequence includes a mutation start time, a mutation end time, and a mutation type. The step frequency physiological synchronization annotation segment list includes a synchronization start and end time, a heart rate change trend, a blood oxygen change trend, and a trend matching state. The gait physiological coupling abnormal segment sequence includes a coupling start and end time, an attitude angle trend direction, a center of gravity trajectory direction, and a coupling state identifier. The attitude offset compression segment position identifier includes a plantar pressure offset direction, a spinal compression side number, and an attitude angle offset persistence.

[0008] As a further solution of the present invention, the step frequency recognition module includes: An acceleration extraction sub-module, which is used to obtain the acceleration signal and the ground contact time when a tumor patient walks, detect the contact point time sequence corresponding to the continuous rising and falling sections in the acceleration time series, calculate the time interval value between adjacent contact points, and generate a contact cycle interval; A waveform positioning sub-module, which is used to call the original acceleration signal sequence according to the acceleration sequence range corresponding to the contact cycle interval, identify the position indexes of local peaks and troughs in the cycle interval, analyze the amplitude difference between the peak and the trough, and generate a peak-trough amplitude difference sequence; A state division sub-module, which is used to calculate the peak-trough difference change rate between adjacent cycles based on the cycle interval number sequence corresponding to the peak-trough amplitude difference sequence, judge the length and start and end cycle numbers of the continuous sections with the same change direction in the change rate sequence, mark the start and end positions of each group of continuous change direction sections, and obtain a frequency change continuous interval section; A mutation marking sub-module, which is used to call the change direction symbols of adjacent sections in the continuous frequency change interval section, judge whether there is a symbol inversion relationship between the change directions, screen the cycle interval numbers corresponding to the time points of the change of the direction symbols, mark the time period where the interval number is located as the mutation position, and generate a step frequency mutation stage label sequence.

[0009] As a further solution of the present invention, the physiological fluctuation monitoring module includes: A step frequency mutation recognition sub-module, which is used to obtain the continuous step frequency data sequence and the corresponding time stamps of the labels based on the step frequency mutation stage label sequence, identify the numerical change of the step frequency data in the time dimension, judge whether the step frequency difference in adjacent time periods exceeds the step frequency change threshold, and if the continuous difference directions are the same and the amplitude exceeds the mutation reference value, mark it as a mutation period to obtain a mutation time interval; A heart rate and blood oxygen extraction sub-module, which is used to call the mutation time interval, collect the heart rate signal sequence and the blood oxygen signal sequence, synchronously intercept the time period signals within the corresponding mutation interval range, and respectively count the change curve slopes and trend directions of the heart rate and blood oxygen signals on the time axis within the interval to obtain the heart rate and blood oxygen change trend values; A physiological synchronization annotation sub-module, which is used to judge whether the change trend directions of the heart rate and blood oxygen are respectively consistent with the step frequency trend direction within the mutation period according to the heart rate and blood oxygen change trend values. If they are consistent, mark it as the synchronization state, calculate the physiological synchronization matching degree, mark the time period with the matching degree value exceeding the synchronization matching reference value as the synchronization section, and obtain a list of step frequency physiological synchronization annotation sections.

[0010] As a further solution of the present invention, the gait coupling judgment module includes: An attitude angle extraction sub-module, which is used to extract the three-axis attitude angle signal data within the time period corresponding to the annotation section based on the step frequency physiological synchronization annotation section list, respectively detect the rotation angle sequences around the horizontal axis, vertical axis and vertical axis, perform time series alignment on the sequences and calculate the change rate and trend direction sequences, analyze the angle change orientation at the time node, and generate an attitude angle change trend sequence; A center of gravity trajectory analysis sub-module, which is used to call the data corresponding to the time period of the attitude angle change trend sequence, detect the corresponding horizontal center of gravity movement trajectory sequence, calculate the horizontal offset between adjacent time points and calibrate the movement direction, screen the time periods with the continuous offset direction remaining unchanged, and generate a center of gravity horizontal offset trend sequence; A coupling trend comparison sub-module, which is used to compare the direction flags of the two sequences at different time nodes according to the center of gravity horizontal offset trend sequence, identify whether their directions are consistent within the continuous time period, calculate the average duration and direction coincidence rate for the continuous paragraphs that meet the direction consistency, calculate the gait coupling strength value of the continuous direction consistency, and generate a gait physiological coupling abnormal section sequence in combination with the time stamps of the step frequency annotation sections.

[0011] As a further solution of the present invention, the posture analysis module includes: A plantar pressure pairing sub-module, configured to extract the plantar pressure map and the position information of the spinal compression area based on the abnormal gait physiological coupling segment sequence, call the pressure transfer direction in the plantar pressure map and the lateral positioning number of the spinal compression area for position pairing, and generate a direction pairing matching degree when the plantar pressure transfer direction is consistent with the spinal compression lateral positioning number direction; A spinal compression positioning sub-module, configured to use the direction pairing matching degree to identify the continuous posture angle offset change in the coupling segment, call the continuous posture angle offset and the matching degree value for calculation, analyze the continuous posture angle offset duration interval, and generate a continuous offset interval degree; A posture instability determination sub-module, configured to determine whether the plantar pressure transfer direction and the spinal compression lateral positioning number remain in the same direction during the continuous posture angle offset according to the continuous offset interval degree, calculate the posture direction consistency index, and determine whether the posture direction consistency index exceeds the posture instability reference value to obtain a posture offset compression segment identification.

[0012] As a further solution of the present invention, the system further includes a risk output regulation module: The risk output regulation module is configured to use the posture offset compression segment identification to extract the step frequency, heart rate, and posture angle fluctuation data of the corresponding section, continue to read the time period after the identification section, check the situation of the data fluctuation direction maintenance in the continuous section, and perform a risk level judgment when the original direction remains unchanged to generate a tumor patient fall risk determination level label; The tumor patient fall risk determination level label includes a risk level, a continuous section fluctuation state, and a risk trend type.

[0013] As a further solution of the present invention, the risk output regulation module includes: A posture fluctuation detection sub-module, configured to use the posture offset compression segment identification to extract the posture angle fluctuation data in the corresponding section, combine the posture angle fluctuation data to detect the change of the fluctuation direction, identify the change of the posture direction, and generate a posture direction change state value; A step frequency and heart rate extraction sub-module, configured to call the posture direction change state value, continue to read the time period after the identification section, extract the step frequency data, heart rate data, and posture angle fluctuation data in the continuous section, and jointly judge the change trend of the step frequency data and the heart rate data and the amplitude change of the posture angle fluctuation data to obtain a data fluctuation direction maintenance record; The risk level determination submodule is used to keep records according to the data fluctuation direction, determine whether the original direction remains unchanged, and if so, extract the fluctuation amplitude range of the step frequency and heart rate data in the continuation segment, analyze the fall risk level, and generate a fall risk determination level label.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by accurately locating the peaks and troughs in the acceleration signal cycle and analyzing the frequency change direction of adjacent sections, the mutation section is accurately marked, and the sensitive capture capability of behavioral state changes is effectively improved. The synchronous extraction method of mutation sections and physiological signals opens up the real-time correlation path between the motion mode and physiological fluctuations such as heart rate and blood oxygen, so that the dynamic coupling relationship between the data can be clearly expressed. By comparing the consistency of the posture angle change with the center of gravity trajectory direction, the accuracy of the recognition of abnormal motion structure and imbalance state is enhanced, and the abnormal state combination with correlation is screened out. The pressure distribution and the direction of spinal compression are extracted to achieve the corresponding pairing between the structural pressure and the physiological load, and the directional locking is combined with the continuous deviation trend of the posture angle to improve the recognition depth of potential structural fall risks. The data fluctuation trends from multiple sources are checked for direction, and the risk level is judged by the continuity of the fluctuation direction. A composite fall risk assessment system that runs through the three levels of behavior, physiology, and structure is constructed, which not only optimizes the time series integrity, but also makes the multi-dimensional data fusion analysis highly targeted and stable. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a system flow chart of the present invention; Figure 2 This is a flow chart of the step frequency recognition module in the present invention; Figure 3 This is a flow chart of the physiological fluctuation monitoring module in the present invention; Figure 4 This is a flow chart of the gait coupling judgment module in the present invention; Figure 5 This is a flow chart of the posture analysis module in the present invention; Figure 6 This is a flow chart of the risk output control module in the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0018] Please refer to Figure 1 , the tumor patient fall prediction system based on machine learning includes: A step frequency recognition module, which is used to obtain the acceleration signal and the ground contact time when a tumor patient walks, locate the peak and valley positions in the step frequency cycle, segment the continuous step frequency states and identify the change direction. When the frequency change direction of adjacent paragraphs changes, mark the time period as a mutation section and generate a step frequency mutation stage label sequence; A physiological fluctuation monitoring module, which is used to synchronously extract the patient's heart rate and blood oxygen signals within the marked mutation period based on the step frequency mutation stage label sequence, identify the signal segments according to the change trend within the time period, and mark them as matching segments when the signal trend is consistent with the step frequency mutation direction, and generate a step frequency physiological synchronization annotation segment list; A gait coupling judgment module, which is used to extract the corresponding attitude angle signal and the center of gravity movement trajectory according to the step frequency physiological synchronization annotation segment list, compare the direction of the attitude angle change trend with the lateral offset path of the center of gravity. When the two directions are consistent within a continuous time period, assign a coupling identifier to the synchronization segment as a combined abnormal judgment segment and generate a gait physiological coupling abnormal segment sequence; An attitude analysis module, which is used to extract the plantar pressure map and the position information of the spinal compression area within the coupling segment through the gait physiological coupling abnormal segment sequence, pair the position numbers of the plantar pressure transfer direction and the spinal compression side. When the two directions are consistent and maintain a continuous attitude angle offset, mark it as an attitude unstable section and generate an attitude offset compression section identifier; A risk output regulation module, which is used to utilize the attitude offset compression section identifier to extract the step frequency, heart rate and attitude angle fluctuation data of the corresponding section, continue to read the time period after the marked section, check the maintenance of the data fluctuation direction within the extended section. When the original direction remains unchanged, perform a risk level judgment and generate a tumor patient fall risk determination level label; The label sequence of the cadence mutation stage includes the mutation start time, mutation end time, and mutation type. The cadence physiological synchronization annotation segment list includes the synchronization start and end time, heart rate change trend, blood oxygen change trend, and trend matching status. The gait physiological coupling abnormal segment sequence includes the coupling start and end time, posture angle trend direction, center of gravity trajectory direction, and coupling status identification. The posture deviation and compression segment identification includes the plantar pressure deviation direction, spinal compression side number, and posture angle deviation continuity. The fall risk assessment level label for tumor patients includes the risk level, continuation segment fluctuation status, and risk trend type.

[0019] See also Figure 2 , the step frequency recognition module includes: The acceleration extraction submodule is used to obtain the contact time between the acceleration signal and the ground when the tumor patient walks, detect the contact point time series corresponding to the continuous rising and falling sections in the acceleration time series, calculate the time interval between adjacent contact points, and generate the contact cycle interval; The acceleration signals of tumor patients in their natural walking state are collected by wearing three-axis acceleration sensors on specific parts of their bodies. When patients walk from the ward to the treatment area, the acceleration change process is continuously recorded by the data collector. In the acquired time series, the signal is initially denoised, such as using low-pass filtering to shield high-frequency noise, retaining the main gait-related components, and dividing the acceleration time series into equal intervals according to the sampling frequency. The acceleration change trend in each data segment is extracted through the sliding window mechanism, and the positive and negative changes of its derivative value are judged to identify continuous rise or fall. The data segment is taken as the candidate interval of the initial contact or air event, and a continuous rising segment is set to reflect the lifting of the sole of the foot, and a continuous falling segment represents the falling of the sole of the foot. Then the local minimum is determined in the segment and used as the time point of actual contact with the ground. The time difference between two consecutive contact points is a gait cycle. A complete contact point sequence is formed by recording the time points. For example, if a patient walks 18 steps in 10 seconds, 18 contact points will be generated, and 17 time interval cycles are constructed. Each cycle uses the start and end time to mark the start and end time of the gait cycle to generate a contact cycle interval sequence.

[0020] The waveform positioning submodule is used to call the original acceleration signal sequence according to the acceleration sequence range corresponding to the contact period interval, identify the position index of the local peak and trough in the period interval, analyze the amplitude difference between the peak and the trough, and generate the peak-trough amplitude difference sequence; Extract the acceleration signal segments corresponding to the contact cycle interval range and perform refined analysis. Set the time corresponding to a certain cycle to be from 1.5 seconds to 2.0 seconds. Then, intercept the original acceleration signal sequence within this range. Based on the duration and fluctuation characteristics of the cycle interval, find the local maximum and minimum values in the extracted signal as the reference positions of the wave peaks and wave valleys. Specifically, through the method of point-by-point comparison, combined with the amplitude difference of extreme values within the set window, identify obvious wave peaks and wave valleys. If there are multiple extreme points in a cycle, the wave peak with the largest amplitude and the wave valley with the lowest amplitude can be selected according to the extreme value intensity to prevent interference signals from affecting the analysis results. After identifying the wave peaks and wave valleys, perform a difference operation on the acceleration values of the two points to form the amplitude difference of a single cycle. Set the wave peak in this cycle to be 1.2g and the wave valley to be 0.4g, then the difference is 0.8g. Apply this processing method to the contact cycle to reflect the difference in the action force of each step of the patient, which has a fundamental role in identifying the gait change trend and serves as an important basis for subsequent state recognition and mutation detection, generating a wave peak and wave valley amplitude difference sequence.

[0021] The state division sub-module is used to calculate the difference change rate of the wave peak and wave valley between adjacent cycles based on the cycle interval number sequence corresponding to the wave peak and wave valley amplitude difference sequence, judge the length and start and end cycle numbers of the continuous section with the same change direction in the change rate sequence, mark the start and end positions of each continuous change direction section, and obtain the continuous interval section of frequency change; Arrange the differences in chronological order, extract the amplitude change trend between adjacent cycles, and mark the change direction in the form of symbols. Set that when the amplitude difference of a certain cycle is 0.9 and the next cycle is 1.1, it is marked as the positive direction. If the next cycle is 0.7, it is marked as the negative direction. Continuously process the entire difference sequence to generate a set of change direction sequences containing positive and negative symbols. Analyze the length of the continuous section with the same change direction in this symbol sequence, and record the start and end cycle numbers of each continuous change section in turn. Set that the patient continuously shows an increasing phenomenon of the wave peak and wave valley difference in the first few steps of walking, then it is recorded as a positive continuous section. This section starts from the 1st cycle and ends at the 4th cycle. Then, further analysis reveals that the change direction changes, so a new next section is created. Each continuous change section records its start and end positions. Such sections can characterize the action regularity of the patient in different gait stages. If it is found that the length of a certain continuous section is significantly too long or too short, it indicates the probability that the patient has unstable gait. This type of information will be used for subsequent mutation marking processing to obtain the continuous interval section of frequency change.

[0022] The mutation marking sub-module is used to call the change direction symbols of adjacent sections in the continuous interval section of frequency change, judge whether there is a symbol reversal relationship between the change directions, screen the cycle interval numbers corresponding to the time points of the change of the direction symbols, mark the time period where the interval number is located as the mutation position, and generate a step frequency mutation stage label sequence; Read the change direction symbols of two adjacent segments in sequence and determine whether they are reversed. For example, if the previous segment has a continuous positive change and the next segment has a continuous negative change, that is, a change direction reversal occurs. Record the intersection point between the end period of the previous segment and the start period of the next segment where this reversal occurs. Set the cycle number to be between the 7th and the 8th. At this time, it indicates that a mutation in the step frequency feature occurs between the 7th and the 8th steps of the patient. Then, map this change point to the time axis of the original acceleration signal according to the start and end time information of the corresponding cycle. For example, if the time of cycle 7 is from 4.6 seconds to 5.1 seconds, the mutation point is marked near 5.1 seconds. Further expand a short-time window and set it as the mutation stage area within the range of ±0.25 seconds. By scanning the symbol changes between all pairs of continuous frequency change segments, filter out the reversed points, record their time intervals in sequence, which can be used for further medical analysis or joint modeling with pathological features, reflecting the potential discontinuous step frequency change process during the patient's walking, and generating a step frequency mutation stage label sequence.

[0023] Please refer to Figure 3 , the physiological fluctuation monitoring module includes: A step frequency mutation recognition sub-module, which is used to obtain the continuous step frequency data sequence and the corresponding timestamps of the labels based on the step frequency mutation stage label sequence, identify the numerical change of the step frequency data in the time dimension, and judge whether the step frequency difference between adjacent time periods exceeds the step frequency change threshold. If the directions of the continuous differences are the same and the amplitude exceeds the mutation reference value, it is marked as a mutation period to obtain the mutation time interval; Extract continuous step frequency signal data, which is collected by a foot acceleration sensor and generates 60 step frequency samples at a sampling frequency of 1 second. For the step frequency data with each timestamp within 5 seconds before and after the label annotation, perform the calculation of the step frequency value difference to distinguish mutation behaviors. By comparing the absolute values of the step frequency differences between adjacent time points, filter out the point segments with a step frequency difference greater than the step frequency change threshold. The step frequency change threshold is set to 10 beats per minute. When the step frequency differences of 3 consecutive sampling points have the same direction and are all greater than this threshold, it is determined as a step frequency mutation period. For example, when the step frequency data of three segments T1 = 0s, T2 = 1s, and T3 = 2s are 85, 96, and 108 beats per minute respectively, their differences are 11 and 12 groups, which meet the mutation judgment conditions. Mark this segment as a mutation segment and extract its start and end time ranges to obtain the mutation time interval.

[0024] A heart rate and blood oxygen extraction sub-module, which is used to call the mutation time interval, collect the heart rate signal sequence and the blood oxygen signal sequence, synchronously intercept the time segment signals within the corresponding mutation interval range, and respectively count the change curve slopes and trend directions of the heart rate and blood oxygen signals on the time axis within the interval to obtain the heart rate and blood oxygen change trend values; Retrieve the heart rate signal and blood oxygen signal sequence data corresponding to the time stamps of the mutation periods. This data is sourced from a PPG sensor worn on the wrist, with a sampling frequency of 1 Hz. Within each mutation period, the heart rate and blood oxygen values are intercepted by the second to construct a time series array. For the heart rate data, calculate its rate of change per unit time through first-order difference. For example, if the data within a certain section is 78, 81, 85, 88, 91 beats per minute, then the difference sequence is 3, 4, 3, 3, and the change trend is upward. For the blood oxygen data, if it is 96%, 96%, 97%, 98%, 99%, the difference is 0, 1, 1, 1, also showing an upward trend. Based on the difference sequence, the average rate of change can be further determined. The average heart rate difference is 3.25 beats per minute, and the blood oxygen is 0.75%, obtaining the heart rate and blood oxygen change trend values.

[0025] The physiological synchronization annotation sub-module is used to determine whether the change trend directions of the heart rate and blood oxygen are respectively consistent with the step frequency trend direction within the mutation period according to the heart rate and blood oxygen change trend values. If they are consistent, it is marked as the synchronization state, using the formula: ; Calculate the physiological synchronization matching degree, mark the periods with matching degree values exceeding the synchronization matching benchmark value as synchronization sections, and obtain the step frequency physiological synchronization annotation section list; Among them, represents the physiological synchronization matching degree of the th mutation period, represents the step frequency difference at the th time point within the th mutation period, represents the heart rate difference at the th time point within the th mutation period, represents the blood oxygen difference at the th time point within the th mutation period, is the total number of sampling points within the mutation period; The formula is used to calculate the synchronization matching degree of the step frequency, heart rate, and blood oxygen signals in terms of change trend within the mutation period. The specific process is as follows: Calculate the change amount of the step frequency, heart rate change amount, and blood oxygen change amount at each time point item by item, and then perform product summation, square summation, and square root operations respectively to integrate the change information of these three signals; Add the absolute product sum of the step frequency and heart rate change amounts to the square root of the sum of the squares of the blood oxygen changes as the numerator part of the matching degree, reflecting the overall amplitude of the change trend coupling; Take the absolute value of the sum of the step frequency change values within the mutation segment and add 1 as the denominator for normalization to prevent abnormal amplification caused by weak signal variations, obtaining the physiological synchronization matching degree; The benefit of the formula is that it integrates the change data of cadence, heart rate and blood oxygen at the same time, integrates the product term and square root index, and comprehensively measures the dynamic synchronization between mutation behavior and physiological changes. See Table 1, which lists the signal difference and baseline value settings for each mutation period. This calculation structure improves the evaluation accuracy of the linkage degree between multi-source signals. Table 1 Physiological signals during mutation period ; Determine whether the trend direction of heart rate change is consistent with the direction of cadence mutation. If the two meet the trend criteria of rising or falling in the same direction, they are counted as matching segments. Substitute each mutation segment into the formula: ; in, Indicates The physiological synchronization matching degree of each mutation period; is the step frequency difference sequence Item, unit is times / minute; is the heart rate difference sequence Item, unit is times / minute; The blood oxygen difference sequence Item, unit is %; is the number of sampling points in this period, and the overall index is synthesized by absolute value, product and square sum operation. The total difference of step frequency is added by 1 to avoid the divisor being zero. The matching benchmark value is set to 30 to 35 based on the previous calibration experiment; Taking the T1 segment as an example, the step frequency difference sequence is [10, 12, 15], the heart rate difference is [5, 6, 7], and the blood oxygen is [1, 2, 2]. Substituting into the formula: Part I: ; Part II: ; Part 3 (denominator): ; Substitute into the formula to calculate: ; The result shows that the physiological synchronization matching degree is 6.05. The physiological synchronization matching degree indicates whether the cadence change is consistent with the changing trend of heart rate and blood oxygen signal. The higher the matching degree, the tighter the coupling between mutation behavior and physiological state. The result exceeds the matching benchmark value of T1 of 30, indicating that the T1 segment is a synchronization segment. The list of cadence physiological synchronization annotation segments is obtained.

[0026] Please refer to Figure 4 , the gait coupling judgment module includes: The attitude angle extraction sub-module extracts the triaxial attitude angle signal data within the corresponding time period of the annotation segment based on the step frequency physiological synchronization annotation segment list, respectively detects the rotation angle sequences in the directions of the horizontal axis, vertical axis, and vertical axis, performs time series alignment on the sequences, calculates the change rate and trend direction sequence, analyzes the angle change orientation at the time node, and generates the attitude angle change trend sequence; Locate the start time and end time of each synchronization annotation interval segment by segment. In actual operation, when setting up a device with a data acquisition frequency of 100Hz, if the start and end times of a certain annotation segment are from 12.3 seconds to 14.8 seconds, the corresponding data serial number range is from the 1230th to the 1480th data point. Extract the triaxial attitude angles within this interval, which are the pitch angle, roll angle, and yaw angle respectively, in order to obtain the human body attitude change situation during the corresponding time period. Further, construct the angle change rate sequences for the triaxial attitude angle time series respectively, that is, divide the difference between the adjacent two attitude angle values by the time difference for each time point. When setting the time point t = 12.34s, the pitch angle changes from 2.5° to 3.2°, then the angular rate is (3.2 - 2.5) / 0.01 = 70° / s. Then, perform rate trend judgment on consecutive multiple time points to determine whether the current rate sign remains consistent continuously. If it is positive within 3 seconds, it is defined as a positive trend, otherwise it is a reverse trend or an oscillating trend. At the same time, number each trend section according to the boundary nodes of the trend change for subsequent comparison and matching with the center of gravity path, and obtain the attitude angle change trend sequence.

[0027] The center of gravity trajectory analysis sub-module is used to call the data corresponding to the time period of the attitude angle change trend sequence, detect the corresponding lateral center of gravity movement trajectory sequence, calculate the lateral offset between adjacent time points and calibrate the movement direction, screen the time periods with the same continuous offset direction, and generate the lateral center of gravity offset trend sequence; Extract the lateral center of gravity displacement data within the time period. The data generally comes from an inertial measurement unit (IMU) or a pressure distribution sensor array. The lateral displacement trajectory is obtained by analyzing the integration of the lateral acceleration. If the lateral center of gravity displacements between 12.3 seconds and 14.8 seconds are 0.3, 0.35, 0.4, 0.42, and 0.39 meters in sequence, then the offset direction within each time interval is to the right (the value increases). Calculate the offset values between adjacent sampling points as 0.05, 0.05, 0.02, -0.03, etc., and represent them as a symbol sequence of 1, 1, 1, -1. Screen the longest continuous same-direction offset segment from this symbol sequence. If the first three in the above sequence are continuously shifted to the right, then classify them into one trend segment. At the same time, calculate the path orientation for each trend segment, use 1 to represent a right shift and -1 to represent a left shift, and then correspond the path orientation data with the time index for numbering for subsequent comparison with the attitude trend, and obtain the lateral center of gravity offset trend sequence.

[0028] The coupling trend comparison sub-module is used to compare the direction flags of two sequences at different differential time nodes according to the lateral center-of-gravity offset trend sequence, identify whether the directions of the two are consistent within a continuous time period, calculate the average duration and direction coincidence rate for continuous paragraphs that meet the direction consistency, and use the formula: ; Calculate the gait coupling intensity value of continuous direction consistency, and combine it with the time stamps of the step frequency annotation segments to generate a sequence of abnormal gait physiological coupling segments; Among them, represents the gait coupling intensity value, represents the attitude angle direction value at the th moment, represents the center-of-gravity offset direction value at the th moment, represents the squared value of the duration of consistent direction, is the number of overlapping direction nodes within the time period, is the number of direction deviation nodes within the time period, is the number of moments; Parameter meaning and formula calculation derivation process: This formula is used to measure the coupling intensity between the attitude angle direction and the center-of-gravity offset direction in time-series data. The parameters , , , , represent the attitude angle direction value, the center-of-gravity offset direction value, the squared value of the duration of consistent trend, the number of direction-consistent nodes, and the number of direction-inconsistent nodes respectively. Among them, represents the total number of trend segments. The attitude angle direction value is obtained by collecting the change rates of the pitch angle, roll angle, and yaw angle and encoding them in symbolic form. Positive rotation is encoded as 1, and negative rotation is encoded as -1. The sampling frequency is set to 100 Hz. For the data collected in segments 400 to 440 of the sample, the attitude angle continuously increases. The calculated pitch angle change value is 3.5 to 6.2 degrees, and the corresponding rate is (6.2 - 3.5) / 0.4 = 6.75 degrees per second. The direction is positive and is encoded as 1. The center-of-gravity offset direction value is based on the integral value of the lateral acceleration of the center of gravity in the same time period. When the attitude angle direction is positive, the center-of-gravity displacement value increases from 0.2 meters to 0.35 meters, and the direction is rightward, so it is assigned a value of 1; The continuous time period is counted from continuous segments with consistent directions. The sampling time is 0.4 seconds, and the continuous consistent time is 0.4 seconds, corresponding to a squared value of , and the number of overlapping nodes is the number of continuous positive nodes, with a total of 5 sampling points. The number of direction-inconsistent nodes Is 0, substitute the above values into the formula for calculation: The first set of parameters is , , , , ; Then: ; The second set of sampling time is 0.3 seconds, the attitude angle drops from -4.1 degrees to -5.0 degrees, the angular rate is (−5.0−(−4.1)) / 0.3 = −3 degrees per second, the direction value , the lateral offset of the center of gravity drops from 0.35 to 0.29 meters, the direction moves to the left, , the square value of the duration , the total number of nodes , the inconsistent nodes , substitute into the formula to get: ; In the third set of samplings, the attitude angle direction remains unchanged, the value is 1, the center of gravity direction is -1, the duration is 0.25 seconds, the total number of nodes is 3, the inconsistent nodes are 2, the square value , substitute: ; Substitute the three segmented values into the averaging formula: ; This result indicates that there is a certain proportion of synchronization in the coupling direction in multiple trend segments. The closer the value is to 1, the more stable the synchronization between the attitude change trend and the center of gravity offset direction. This value is in the low to medium range, indicating that the coupling trend is not yet stable. The inconsistent direction segments in the sequence can be further analyzed to extract abnormal segments as the core reference content of the gait physiological coupling abnormal segment sequence. The benefit of the formula is that by introducing the square term of the trend duration, it strengthens the influence of direction consistency on the result, and at the same time includes inconsistent direction nodes in the denominator structure to form a penalty mechanism for abnormal fluctuations, completing the quantitative evaluation of the gait structure stability.

[0029] Please refer to Figure 5 , the attitude analysis module includes: The plantar pressure pairing sub-module is used to extract the position information of the plantar pressure map and the spinal compression area based on the gait physiological coupling abnormal segment sequence, call the pressure transfer direction in the plantar pressure map and the lateral positioning number of the spinal compression area, perform position pairing, and generate a direction pairing matching degree when the plantar pressure transfer direction is consistent with the spinal compression lateral positioning number direction; Table 2 Plantar and Spinal Pairing Parameter Table ; Extract the plantar pressure map and the position information of the spinal compression area. For each frame of data, collect the two-dimensional coordinates of the plantar pressure map. An example is shown in Table 2. The plantar pressure X coordinates are 20.5, 21.0, 20.8, 21.3, 20.9 respectively, and the Y coordinates are 30.1, 30.4, 30.2, 30.5, 30.3 respectively. At the same time, the number of the spinal compression area is uniformly 1. Pair the X and Y coordinates of the plantar pressure map with the number of the spinal compression area, and calculate the consistency between the plantar pressure transfer direction and the spinal compression side positioning number. Use the formula: ; Among them, represents the direction pairing matching degree value, is the value of the plantar pressure transfer direction in the k-th frame, is the value of the spinal compression side positioning number in the k-th frame, is the number of frames in the coupling section. In this example , substitute the data: ; The summation symbol in the formula means to accumulate the absolute values of the differences between the plantar pressure direction values and the spinal compression numbers of each frame. The denominator is the average value of the number of frames, and the obtained direction pairing matching degree value is 0.178, which is lower than the matching degree reference value of 0.3, indicating that the plantar pressure and the spinal compression area direction are highly consistent.

[0030] The spinal compression positioning sub-module is used to identify the continuous attitude angle offset change in the coupling section by using the direction pairing matching degree, call the continuous attitude angle offset and the matching degree value for calculation, analyze the continuous attitude angle offset duration interval, and generate the continuous offset interval degree; Identify the continuous attitude angle offset change in the coupling section, and call the direction pairing matching degree value obtained in Paragraph 1 , and then collect the continuous attitude angle offset amounts, as listed in Table 2, which are 0.12, 0.15, 0.10, 0.11, 0.14 respectively. Perform a persistence identification on the attitude angle offset amounts to determine whether they are higher than the offset threshold. In this embodiment, the attitude angle offset threshold is set to 0.1. If the offset amount is higher than the threshold, it is included in the continuous interval. According to this standard, in the five frames, the frames with offset amounts higher than 0.1 are 0.12, 0.15, 0.11, 0.14, a total of 4 frames. Calculate the continuous offset interval degree by using the formula: ; Among them, represents the continuous offset interval degree, represents the number of continuous offset frames, represents the number of frames in the coupling section; In this example , , substitute the data: ; In the formula, the ratio expression is used to quantify the proportion of continuous offset, reflecting the stability of the coupling section. When the obtained continuous offset range is 0.8, which is higher than the continuous offset reference value of 0.6, it indicates that there is a continuous offset trend in the attitude angle.

[0031] The attitude instability determination sub-module determines whether the plantar pressure transfer direction and the spinal compression side positioning number remain consistent in the continuous attitude angle offset according to the continuous offset range, using the formula: ; Calculate the attitude direction consistency index, and determine whether the attitude direction consistency index exceeds the attitude instability reference value to obtain the attitude offset compression segment identification; Among them, represents the attitude direction consistency index, represents the numerical value of the plantar pressure transfer direction in the th frame, represents the spinal compression side positioning number value in the th frame, represents the attitude angle offset in the th frame, represents the total number of frames in the coupling section, Table 3 Attitude Analysis Parameter Sampling Table ; Determine whether the plantar pressure transfer direction and the spinal compression side positioning number remain consistent in the continuous attitude angle offset, obtain the continuous attitude angle offset continuous interval length identified in the spinal compression positioning sub-module, denoted as , and collect the plantar pressure transfer direction numerical value , spinal compression side number value and attitude angle offset for each frame in the coupling section. As shown in Table 3, substitute the data of each frame into the formula: ; Taking the data as an example, assume that the number of frames in the coupling section is , and the continuous offset interval length is , then: ; ; ; ; In the formula operation, the product term It represents the consistency intensity measure between the plantar pressure of each frame and the spinal compression direction, and the summation term accumulates the results of all frames; the square root of the sum of the squared attitude angle offsets reflects the overall fluctuation amplitude of the offset change, and the denominator combines the number of frames and the total offset interval to reflect the data scale adjustment; The advantage of the formula is that by introducing the sum of the amplitudes of the attitude angle offsets and the sum of the intensities in the consistent direction between the plantar pressure and the spinal compression for joint normalization operations, it can comprehensively reflect the dynamic offset stability and direction consistency within the same framework; comparing the above results with the set attitude instability reference value, the reference value is set to 0.52 through batch experiments. When the calculated result is greater than the reference value, it is determined that the direction is consistent, and combined with the offset interval, it is determined whether the attitude instability condition is met; The result shows that in the current sample, the attitude direction consistency index is approximately 0.5488, which is higher than the set reference value, indicating that there is a consistent direction and continuous attitude angle offset in this section of coupled frames, and the attitude offset compression section identifier is used.

[0032] Please refer to Figure 6 , the risk output control module includes: An attitude fluctuation detection sub-module, which is used to adopt the attitude offset compression section identifier to extract the attitude angle fluctuation data within the corresponding section, combine the attitude angle fluctuation data to detect the change of the fluctuation direction, identify the change of the attitude direction, and generate an attitude direction change status value; Combined with the time series data obtained by the attitude sensor in the wearable device for preprocessing and analysis. It is set in the daily activity monitoring of the elderly. The IMU sensor continuously collects the trunk attitude angle data. By analyzing the angle change rate and the cumulative change amplitude, the position where the attitude has a sudden change is identified. When an elderly person slowly squats from a standing state, if the pitch angle continuously drops by more than the preset threshold, it can be marked as the starting point of the attitude offset. When the angle maintains or slightly changes after reaching the lowest point, it can be marked as the compression section. During this period, the attitude angle fluctuation data within this section is extracted. The fluctuation data is processed by a sliding time window, and the attitude angle value of each time slice within this section range is retained. At the same time, the change trend is judged by combining these angle values. If the pitch angle first rises and then falls within a certain section and the roll angle synchronously deflects left or right, it can be further judged whether the direction has changed during the attitude change process. The direction change detection is based on the continuity of the angle change. The positive and negative trend conversion situations of the pitch angle and the roll angle in time are statistically analyzed. When a trend reversal occurs and the direction deflection reaches a considerable degree, it is judged that the direction has changed. Through this series of recognition processes, an attitude direction change status value is generated.

[0033] The step frequency and heart rate extraction sub-module is used to call the attitude direction change status value, continue to read the time period after the identification segment, extract the step frequency data, heart rate data, and attitude angle fluctuation data within the continuation segment, jointly judge the change trends of the step frequency data and heart rate data and the amplitude change of the attitude angle fluctuation data, and obtain the record of the data fluctuation direction maintained; After the compression segment ends, a continuation segment with a fixed duration is set, set to continue for 8 seconds, during which the user's step frequency, heart rate, and attitude angle fluctuation data are continuously read. In an actual scenario, for example, when an elderly person wearing the device starts walking after standing, the step frequency data will be continuously obtained and it will be judged whether it is a stable walking mode. The heart rate sensor synchronously records the heart rate, and the attitude angle fluctuation is used to analyze the trunk swing situation. At this time, the sub-module matches and integrates the three data dimensions, observes their trends over time, especially on the premise that the attitude direction has changed, and pays attention to whether the step frequency and heart rate maintain the same trend. It is set that if the user's step frequency starts to increase, the heart rate rises synchronously and the attitude angle continues to fluctuate, it indicates that the user is in motion. If the fluctuation amplitude does not change much during this process, it means that the direction remains consistent, and this data combination can be recorded as "direction maintained". On the contrary, if the attitude angle fluctuates significantly but the step frequency is discontinuous or the heart rate does not change accordingly, it is judged that the direction is not maintained, and the motion data within the continuation segment is classified and recorded, providing key data support on the time axis for subsequent risk level analysis, and obtaining the record of the data fluctuation direction maintained.

[0034] The risk level determination sub-module is used to judge whether the original direction remains unchanged according to the record of the data fluctuation direction maintained. If it remains unchanged, it extracts the fluctuation amplitude range of the step frequency and heart rate data within the continuation segment, analyzes the fall risk level, and generates a fall risk determination level label; Judge whether the user's movement direction continues to be stable. If the direction is continuously maintained, based on this, analyze the fluctuation amplitude range of the step frequency and heart rate within the continuation segment, and conduct level division in combination with the preset classification criteria. Set an elderly person stands up and tries to walk after bending down to pick up something once. If the direction remains stable within a few consecutive seconds during this process, and at the same time the step frequency changes greatly, such as from a slow walk to a fast walk, and the heart rate rises significantly from the normal range, this will be recorded as a data segment with a medium risk level. When specifically dividing, different amplitude threshold intervals will be set to identify the "low", "medium", and "high" risk levels. For example, if the step frequency fluctuation is small and the heart rate remains within the normal value range, it is classified as a low risk. If both values fluctuate significantly, it is a high risk. Each level label can be preset as "Fall Risk Level: Low / Medium / High" and is marked in real-time in the user's physiological data stream. The continuous determination process enables continuous monitoring of the user in a time slice manner. When a potentially high risk level is detected, it is used for relevant early warning or intervention processing, and a fall risk determination level label is generated.

[0035] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the relevant art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A tumor patient fall prediction system based on machine learning, characterized in that, The system includes: A step frequency recognition module, which is used to obtain the acceleration signal and the ground contact time when a tumor patient walks. By locating the peak and valley positions in the step frequency cycle, the continuous step frequency states are segmented and the change direction is identified. When the frequency change direction between adjacent paragraphs changes, the time period is marked as a mutation section, and a step frequency mutation stage label sequence is generated; A physiological fluctuation monitoring module, which is used to synchronously extract the patient's heart rate and blood oxygen signals during the marked mutation period based on the step frequency mutation stage label sequence, identify the signal segments according to the change trend within the time period, and generate a list of step frequency physiological synchronization annotation segments; A gait coupling judgment module, which is used to extract the corresponding attitude angle signal and the center of gravity movement trajectory according to the list of step frequency physiological synchronization annotation segments, compare the direction of the attitude angle change trend with the lateral deviation path of the center of gravity, and generate a sequence of gait physiological coupling abnormal segments; An attitude analysis module, which is used to extract the plantar pressure map and the position information of the spinal compression area within the coupling section through the sequence of gait physiological coupling abnormal segments, pair the position of the plantar pressure transfer direction and the spinal compression side positioning number, and generate an attitude offset compression section identification; 2. The tumor patient fall prediction system based on machine learning according to claim 1, wherein The step frequency mutation stage label sequence includes the mutation start time, the mutation end time, and the mutation type. The list of step frequency physiological synchronization annotation segments includes the synchronization start and end times, the heart rate change trend, the blood oxygen change trend, and the trend matching status. The sequence of gait physiological coupling abnormal segments includes the coupling start and end times, the attitude angle trend direction, the center of gravity trajectory direction, and the coupling status identification. The attitude offset compression section identification includes the plantar pressure offset direction, the spinal compression side number, and the attitude angle offset persistence.

3. The tumor patient fall prediction system based on machine learning according to claim 1, wherein The step frequency recognition module includes: An acceleration extraction sub-module, which is used to obtain the acceleration signal and the ground contact time when a tumor patient walks, detect the contact point time sequence corresponding to the continuous rising and falling sections in the acceleration time sequence, calculate the time interval value between adjacent contact points, and generate a contact cycle interval; A waveform positioning sub-module, which is used to call the original acceleration signal sequence according to the acceleration sequence range corresponding to the contact cycle interval, identify the position indexes of local peaks and valleys in the cycle interval, analyze the amplitude difference between the peak and the valley, and generate a peak-valley amplitude difference sequence; A state division sub-module, which is used to calculate the peak-valley difference change rate between adjacent cycles based on the cycle interval number sequence corresponding to the peak-valley amplitude difference sequence, judge the length and start and end cycle numbers of the continuous sections with the same change direction in the change rate sequence, mark the start and end positions of each group of continuous change direction sections, and obtain a continuous interval section of frequency change; A mutation marking sub-module, which is used to call the change direction symbols of adjacent sections in the continuous interval section of frequency change, judge whether there is a sign reversal relationship between the change directions, screen the cycle interval numbers corresponding to the time points of the direction sign change, mark the time period where the interval number is located as the mutation position, and generate a step frequency mutation stage label sequence.

4. The tumor patient fall prediction system based on machine learning according to claim 3, wherein, The physiological fluctuation monitoring module includes: The step frequency mutation recognition sub-module is used to obtain the continuous step frequency data sequence and the corresponding timestamps of the labels based on the step frequency mutation stage label sequence, identify the numerical change of the step frequency data in the time dimension, and judge whether the step frequency difference in adjacent time periods exceeds the step frequency change threshold. If the continuous difference directions are the same and the amplitude exceeds the mutation reference value, it is marked as a mutation period to obtain the mutation time interval; The heart rate and blood oxygen extraction sub-module is used to call the mutation time interval, collect the heart rate signal sequence and the blood oxygen signal sequence, synchronously intercept the time period signals within the corresponding mutation interval range, and respectively count the change curve slopes and trend directions of the heart rate and blood oxygen signals on the time axis to obtain the heart rate and blood oxygen change trend values; The physiological synchronization annotation sub-module is used to judge whether the change trend directions of the heart rate and blood oxygen are respectively consistent with the step frequency trend direction within the mutation period according to the heart rate and blood oxygen change trend values. If they are consistent, it is marked as the synchronization state, calculate the physiological synchronization matching degree, mark the time periods with the matching degree value exceeding the synchronization matching reference value as the synchronization sections, and obtain the step frequency physiological synchronization annotation section list.

5. The tumor patient fall prediction system based on machine learning according to claim 4, characterized in that, The gait coupling judgment module includes: The attitude angle extraction sub-module is used to extract the three-axis attitude angle signal data within the time period corresponding to the step frequency physiological synchronization annotation section list, respectively detect the rotation angle sequences around the horizontal axis, vertical axis and vertical axis, perform time series alignment on the sequences and calculate the change rate and trend direction sequences, and analyze the angle change orientation at the time node to generate the attitude angle change trend sequence; The center of gravity trajectory analysis sub-module is used to call the data corresponding to the time period of the attitude angle change trend sequence, detect the corresponding lateral center of gravity movement trajectory sequence, calculate the lateral offset between adjacent time points and calibrate the movement direction, and screen the time periods with the continuous offset direction remaining unchanged to generate the center of gravity lateral offset trend sequence; The coupling trend comparison sub-module is used to compare the direction flags of the two sequences at different time nodes according to the center of gravity lateral offset trend sequence, identify whether the directions of the two sequences are consistent within the continuous time period, calculate the average duration and direction coincidence rate for the continuous paragraphs satisfying the direction consistency, calculate the gait coupling intensity value of the continuous direction consistency, and generate the gait physiological coupling abnormal section sequence in combination with the step frequency annotation section timestamps.

6. The tumor patient fall prediction system based on machine learning according to claim 5, characterized in that, The attitude analysis module includes: The plantar pressure pairing sub-module is used to extract the plantar pressure map and the position information of the spinal compression area based on the gait physiological coupling abnormal section sequence, call the pressure transfer direction in the plantar pressure map and the side positioning number of the spinal compression area for position pairing, and generate the direction pairing matching degree when the plantar pressure transfer direction is consistent with the spinal compression side positioning number direction; The spinal compression positioning sub-module is used to identify the continuous attitude angle offset change within the coupling section by using the direction pairing matching degree, call the continuous attitude angle offset and the matching degree value for calculation, analyze the continuous attitude angle offset duration interval, and generate the continuous offset interval degree; The posture instability determination sub-module is used to determine whether the plantar pressure transfer direction and the spinal compression side positioning number remain consistent in the continuous posture angle offset according to the continuous offset interval degree, calculate the posture direction consistency index, determine whether the posture direction consistency index exceeds the posture instability reference value, and obtain the posture offset compression segment identification.

7. The tumor patient fall prediction system based on machine learning according to claim 1, characterized in that, The system further includes a risk output regulation module: The risk output regulation module is used to utilize the posture offset compression segment identification to extract the step frequency, heart rate, and posture angle fluctuation data of the corresponding section, continue to read the time period after the identification section, check the retention situation of the data fluctuation direction in the continued section, and perform a risk level judgment when the original direction remains unchanged, and generate a tumor patient fall risk determination level label; The tumor patient fall risk determination level label includes a risk level, a continued section fluctuation state, and a risk trend type.

8. The tumor patient fall prediction system based on machine learning according to claim 7, characterized in that, The risk output regulation module includes: The posture fluctuation detection sub-module is used to adopt the posture offset compression segment identification to extract the posture angle fluctuation data in the corresponding section, combine the posture angle fluctuation data to detect the change of the fluctuation direction, identify the change of the posture direction, and generate a posture direction change state value; The step frequency and heart rate extraction sub-module is used to call the posture direction change state value, continue to read the time period after the identification section, extract the step frequency data, heart rate data, and posture angle fluctuation data in the continued section, and jointly judge the change trend of the step frequency data and the heart rate data and the amplitude change of the posture angle fluctuation data to obtain a data fluctuation direction retention record; The risk level determination sub-module is used to judge whether the original direction remains unchanged according to the data fluctuation direction retention record. If it remains unchanged, the fluctuation amplitude interval of the step frequency and heart rate data in the continued section is extracted, the fall risk level is analyzed, and a fall risk determination level label is generated.

Images