Tumor pain grading assessment method based on multi-modal sensing

Abstract

The present application relates to the technical field of pain assessment, in particular to a tumor peace therapy pain grading assessment method based on multi-modal sensing, comprising the following steps: acquiring angular velocity and displacement direction to determine the static state, combining facial and voice features to locate the restricted period, extracting rhythm signals to identify the contradictory state, comparing the label content to extract the uncovered information, analyzing the signal features to divide the dominant channel, and modifying the label to obtain the pain grading verification label. In the present application, a static state recognition mechanism is established by fusing posture changes and signal fluctuation features, the behavior restricted time window is located by combining facial and voice features, the rhythm fluctuation information not covered by the label under the silent state is screened, the dominant channel path is constructed by extracting the signal fluctuation quantity, response sequence and synchronization relationship, the label content is compared and modified, the recognition range of the pain grading for non-extrinsic reaction state is expanded, and the coverage ability and resolution level of the label content for the rhythm behavior state are improved.

Description

Technical Field

[0001] This invention relates to the field of pain assessment technology, and in particular to a method for assessing pain levels in palliative care for tumors based on multimodal sensing. Background Technology

[0002] The field of pain assessment technology primarily involves methods for quantitatively identifying and classifying an individual's pain perception. Core aspects include real-time acquisition of physiological signals, behavioral feature recognition, establishment of pain level classification standards, multidimensional data fusion processing, and assessment model development. This technology typically collects physiological or behavioral signals such as changes in facial expressions, skin conductance, electromyography, heart rate changes, and blood pressure fluctuations, and compares these signals with preset level standards for data comparison and pain assessment. It is widely used in clinical monitoring, intensive care, postoperative management, and palliative care, and is particularly useful for patients who cannot describe their pain. The traditional pain rating assessment method for palliative care in cancer refers to a technique that combines manual observation and experience-based judgment to identify the subjective pain level of patients during palliative care for terminally ill cancer patients. This technique addresses the difficulty patients face in expressing pain due to physical weakness or confusion. It typically involves nurses observing facial muscle twitches, frowning, groaning frequency, changes in body position, and respiratory rhythm, classifying the pain level based on behavioral rating scales. Some methods supplement this with single-sensor data collection of skin electrical activity or heart rate, and then manually comparing the data against grading standards to complete the pain assessment.

[0003] Existing technologies rely heavily on the mapping relationship between a single signal input and surface behavioral changes during the assessment process. They cannot effectively identify dynamic changes and contradictory states between multiple signal channels. When patients lack continuous overt movements and have abnormal fluctuations in their internal rhythms, assessment methods struggle to accurately capture potential pain cues. Especially in scenarios where behavioral characteristics and rhythmic signals are out of sync, assessment criteria risk omissions, and labeling may not accurately reflect the patient's current perceptual state. Furthermore, without analyzing the channel response timing and signal synchronization characteristics, the pain level classification results are subject to interference, affecting the suitability and timeliness of subsequent interventions. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a method for assessing pain levels in tumor palliative care based on multimodal sensing; To achieve the above objectives, the present invention adopts the following technical solution: a multimodal sensing-based method for assessing pain levels in palliative care for tumors, comprising the following steps: S1: Obtain the angular velocity and displacement direction recorded by the attitude sensor, determine whether the angular velocity change in adjacent per-second segments is continuously within the static judgment threshold, detect the overlap between the facial muscle movement non-triggered frames and the speech frequency interruption segment, and obtain the time window of restricted behavior performance. S2: Based on the time window of the behavior performance limitation, extract skin conductance, heart rate and respiratory signals, compare the rhythm fluctuation with the previous non-static state, identify the rhythm active state under the behavior static state, and obtain the signal performance contradictory state region. S3: Based on the contradictory state area of ​​the signal performance, compare whether the behavioral silence and rhythm changes are not reflected by the label, extract the time points not covered by the signal deviation, and obtain the label performance marking group; S4: Based on the tag performance marking group, identify the number of channel fluctuations, response order and synchronization characteristics, determine the dominant signal channel, and obtain the main path sequence of the pain judgment channel; S5: Based on the main path sequence of the pain judgment channel, read the number of behavioral fluctuations, rhythm continuity and signal output during the contradictory state period, compare with the pain label state description, replace the original label, and obtain the pain level verification label.

[0005] As a further aspect of the present invention, the behavioral performance-limited time window includes angular velocity change state, body displacement state, number of facial muscle movement-untriggered frames, number of voice main frequency interruption segments, and timeline overlapping segment positions; the signal performance contradiction state area includes skin conductance fluctuation slope, heart rate rhythm fluctuation interval, respiratory rhythm change amplitude, behavioral path output state, and rhythm signal fluctuation state; the label performance marker group includes rhythm channel fluctuation state, behavioral path silent state, label content-indicated state, and label content-response state difference; the pain judgment channel main path sequence includes the number of chest fluctuation sequence segments, skin conductance fluctuation segments, sound pressure change points, response start order, and signal synchronization status; the pain grading verification label includes the number of behavioral fluctuations, rhythm change continuity, signal duration, signal performance range, and alternative label content.

[0006] As a further aspect of the present invention, the static judgment threshold refers to the range of angular velocity and displacement changes used to identify whether an individual is in a static state; The "facial muscle movement not triggered frame" refers to an image frame in which no muscle movement is observed on the face within the time frame.

[0007] As a further aspect of the present invention, the state of active rhythm under behavioral stillness refers to the contradictory state of no behavioral action but continuous active fluctuation of physiological signals; The signal deviation-uncovered time point refers to the moment of pain that is not covered by the pain label and where there are abnormal physiological fluctuations.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Acquire the angular velocity and body displacement direction continuously recorded by the attitude sensor, extract the angular velocity change value of adjacent per second segments, monitor the continuous trend of the numerical sequence, filter the time intervals in which the numerical fluctuation is within the preset error range in the continuous time period, and obtain the set of time intervals in the static state. S102: Based on the set of static state time intervals, extract the video sequence and audio sequence corresponding to the time intervals, identify the image frames in the video where no muscle movement response was detected and the segments in the audio where the main frequency is zero, cross-reference the time tags, extract the segments with time intersection, and obtain the set of non-response segment intersection regions. S103: Based on the set of intersection regions of the non-response segments, analyze the continuity of the time interval between adjacent segments, merge consecutive adjacent time segments, and output the start and end time points of each segment to obtain the time window of restricted behavior performance.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the time period corresponding to the time window of the behavior limitation, collect the original signal sequences of skin conductance, heart rate and respiratory rhythm, extract the numerical change amplitude in the continuous frame segments of the signal, analyze the time interval between adjacent peaks, and store the parameters into the structure mapping set according to the signal channel to obtain the rhythm fluctuation structure parameter set. S202: Based on the rhythmic fluctuation structure parameter set, retrieve the rhythmic signal fluctuation parameters of the same channel in the previous non-static period, read the corresponding change amplitude and peak interval values, compare the channel data of the current and non-static period point by point, extract the fluctuation difference state under the channel as output data, and obtain the rhythmic signal fluctuation comparison result set. S203: Based on the set of rhythm signal fluctuation comparison results, monitor whether there is continuous static behavior in the current time period, and analyze whether the skin conductance, heart rate and respiratory rhythm signals continue to fluctuate to obtain the contradictory state region of signal performance.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the contradictory state area of ​​the signal, extract the pain label content output within the corresponding time period, filter the field describing the intensity, compare with the fluctuation trajectory of the rhythm channel, extract the starting position and amplitude turning point of continuous fluctuation, and obtain the set of rhythm channel fluctuation segments. S302: Call the rhythm channel fluctuation segment set, extract the continuous untriggered muscle movement segments in the behavior path, compare the start and end points of the segments in the sequence according to the time index, and locate the time segment where the behavior is still and the rhythm is active, to obtain the behavior rhythm comparison segment group. S303: Based on the behavioral rhythm comparison segment group, analyze the pain label field and rhythm activity in the time period, extract the segments not covered by the label performance, and obtain the label performance marking group.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the time period in the tag group, extract the chest undulation sequence, skin conductance fluctuation segment and sound pressure change point, and extract the number of fluctuating signal segments and corresponding time index in each channel after splitting by channel to obtain the channel segment number sequence. S402: Based on the channel segment quantity sequence, retrieve the response start index value of the channel fluctuation segment, cross-compare the start order in the differentiated time period, filter the channel identifiers with early response and extract the total number of initiation segments and time coverage to obtain the response start priority sequence. S403: Call the response start priority sequence, determine whether the synchronization segment time coverage remains continuous, compare the response start time order of the channel, locate the response channel that appears for the first time, extract the continuous fluctuation segment information in the channel and perform time-series mapping with the corresponding synchronization segment to obtain the pain judgment channel main path sequence.

[0012] As a further aspect of the present invention, in the process of extracting the number of fluctuating signal segments in each channel of the channel segment number sequence: extracting signal segments with continuous intervals between respiratory peaks in the chest fluctuation sequence, and extracting the number of fluctuations of the corresponding signal segments according to the channel splitting method; extracting segments with continuous fluctuations in conductance change trends in the skin conductance fluctuation segment, and extracting the number of signal segments and time index according to the channel order. In the process of extracting the information of the continuous fluctuation segment: in the region where the signal amplitude change shows a trend of non-zero amplitude difference between adjacent sampling points, continuous segments are extracted according to the time index and mapped to the time range corresponding to the response channel in the synchronization segment.

[0013] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on the main path sequence of the pain judgment channel, extract the number of channel behavior fluctuations, the number of continuous frames of rhythm changes and the interruption interval of signal segments during the contradictory state period, and compare the continuity relationship of signal items in the time period to obtain the signal dynamic performance parameter set. S502: Call the signal dynamic performance parameter set, retrieve the state type features described by the pain label, and parse the range of the description content in the label and the signal performance index. Compare the matching between the label content and the signal range to obtain the label description correspondence table. S503: Based on the label description correspondence table, determine whether the label content covers the indicator combination of the current signal performance range, and filter the label items that do not have the corresponding coverage relationship, and replace them with the label content that completely overlaps with the time interval to obtain the pain level verification label.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a static state recognition mechanism is established by fusing posture changes and signal fluctuation features. By combining facial and voice features to locate the time window of behavior limitation, rhythmic fluctuation information not covered by labels in the silent state is screened. The number of signal fluctuations, response order and synchronization relationship are extracted to construct the dominant channel path. The label content is compared and corrected to expand the recognition range of pain grading for non-overt response states and improve the coverage and discrimination level of the label content for rhythmic behavior states. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides a method for assessing pain levels in palliative care for tumors based on multimodal sensing, comprising the following steps: S1: Obtain the angular velocity and body displacement direction continuously recorded by the posture sensor, determine whether the angular velocity change in adjacent per-second segments is continuously within the static judgment threshold, filter the time period in which no displacement occurs and set it as the candidate interval, detect the number of frames in which facial muscle movements are not triggered and the number of segments in which the main voice frequency is interrupted within the interval, locate the overlapping part of the segments on the timeline, and obtain the time window of restricted behavior performance. S2: Based on the time window corresponding to the behavioral performance limitation period, extract the fluctuation slope and fluctuation interval of skin conductance, heart rate and respiratory rhythm within the segment, compare the current fluctuation amplitude with the performance of the previous non-static period, identify the state where the behavioral path has no output and the rhythm signal continues to fluctuate, and obtain the contradictory state region of signal performance. S3: Based on the contradictory state area of ​​signal performance, call the pain label content output in the corresponding time period, compare the state of the label performance, analyze whether the rhythm channel continues to fluctuate within the time period, compare the continuous silence of the behavior path with the continuous change of the rhythm path, and obtain the label performance marking group according to the difference in the state of the label content. S4: Based on the time periods in the label performance marking group, extract the chest undulation sequence, skin conductance fluctuation segment and sound pressure change point of the channel, identify the number of continuous undulation segments in the channel, the order of response start points and the synchronization between signals, divide the dominant channel according to the continuity of signal performance and the order of start time, and obtain the main path sequence of pain judgment channel. S5: Based on the main path sequence of the pain judgment channel, read the number of behavioral fluctuations, the continuity of rhythm changes and the duration of signals during the contradictory state period, compare the signal performance range with the state description indicated by the pain label, replace the original label content, and obtain the pain level verification label.

[0023] The time window for behavioral performance limitations includes angular velocity change status, body displacement status, number of frames without facial muscle movement triggering, number of interrupted segments in the main voice frequency, and the position of overlapping segments on the timeline. The contradictory state area of ​​signal performance includes skin conductance fluctuation slope, heart rate rhythm fluctuation interval, respiratory rhythm change amplitude, behavioral path output status, and rhythm signal fluctuation status. The label performance marking group includes rhythm channel fluctuation status, behavioral path silent status, the state indicated by the label content, and the difference in the reaction status of the label content. The main path sequence of the pain judgment channel includes the number of chest fluctuation sequence segments, skin conductance fluctuation segments, sound pressure change points, response start order, and synchronization between signals. The pain rating verification label includes the number of behavioral fluctuations, the continuity of rhythm changes, signal persistence status, signal performance range, and alternative label content.

[0024] Please see Figure 2 The specific steps of S1 are as follows: S101: Acquire the angular velocity and body displacement direction continuously recorded by the attitude sensor, extract the angular velocity change value of adjacent per second segments, monitor the continuous trend of the numerical sequence, filter the time intervals in which the numerical fluctuation is within the preset error range in the continuous time period, and obtain the set of time intervals in the static state. First, the raw angular velocity data sequence needs to be divided by second. Within each time period, extract the angular velocity value sequence of consecutive frames. For example, if the angular velocity in a certain segment is 0.12, 0.11, and 0.13 within three frames, then calculating the trend per second requires comparing the direction of angular velocity change between consecutive frames. For instance, a decrease from 0.12 to 0.11 and an increase from 0.11 to 0.13 indicates a direction reversal within that second. If the direction of change does not reverse across multiple consecutive second segments (i.e., the positive and negative signs of the angular velocity changes are consistent, such as consecutive positive values ​​of 0.05, 0.04, and 0.03), then that time segment is marked as having a consistent trend. Next, determine if the absolute difference of the angular velocity values ​​corresponding to this continuous trend is within a low amplitude range. If the difference in angular velocity values ​​across multiple time segments is controlled within 0.01, it indicates that the sequence has not shown significant change within that time segment. Then, read the body displacement direction data corresponding to this segment. Cross-discrimination is performed by the changes in displacement direction in a three-axis coordinate system. For example, if the change in displacement in the X-axis direction is detected to be 0.02, the Y-axis to 0.01, and the Z-axis to 0.00 within a continuous angular velocity range, the maximum change in the three axes is compared with the set displacement change limits. The limits are referenced to the upper and lower limits of the sensor's background noise amplitude when the body is at rest, for example, the noise limit is ±0.03. If the changes in each axis do not exceed the limits, it is determined that no detectable posture movement of the body has occurred during this time period, and this time period is then classified as a static state interval. If such a stable angular velocity change trend is detected in multiple adjacent time periods and corresponding time periods in which no obvious displacement change is detected, these are continuously combined to form a static sequence time segment set. For example, if the 12th to 16th seconds and the 24th to 29th seconds in the sequence are identified as stable segments, they are merged and recorded as two independent static state time intervals, and finally a static state time interval set is obtained. Table 1. Trend of Angular Velocity Change Time period Angular velocity (frame 1) Angular velocity (frame 3) Trend of change 0-1 second 0.12 0.13 Reversal of direction 1-2 seconds 0.05 0.03 Consistent trend 2-3 seconds 0.03 0.01 Consistent trend Table 1 shows the trends of angular velocity changes over different time periods. Whether the trend of angular velocity changes reverses within each time period, and whether there is a continuous and consistent direction of change, affects the determination of a static state. When the trends of angular velocity changes are consistent across multiple consecutive time periods and the magnitude of change is small (e.g., the difference is controlled within 0.01), it can be confirmed that the body is in a static state during that period. This analytical method helps ensure accurate identification of static states without significant displacement when assessing pain in cancer patients.

[0025] S102: Based on the set of static state time intervals, extract the video and audio sequences corresponding to the time intervals, identify the image frames in the video where no muscle movement response was detected and the segments in the audio where the main frequency is zero, cross-reference the time tags, extract the segments with time intersection, and obtain the set of non-response segment intersection regions; First, the start and end frame numbers of each static interval are located, and the synchronously acquired video and audio sequences on the corresponding time axis are extracted according to the frame index position. For the video sequence, the RGB image matrix is ​​retrieved by frame number, and facial regions are located for each frame. Muscle movement changes are identified through facial feature point extraction, such as detecting pixel displacement in key areas like the corners of the mouth, eyebrows, and eyelids. If the displacement value is less than 0.5 pixels in consecutive frames, the frame is considered to have no muscle movement response. For the audio sequence, the amplitude and frequency distribution of the sound signal are extracted segment by time axis. The dominant frequency component of each segment is extracted, and after obtaining the frequency domain features through Fast Fourier Transform, the amplitude information is combined to determine whether the signal energy is below a set threshold. If the average amplitude of the dominant frequency component is close to zero or below the energy threshold, the segment is determined to be silent. For example, in the audio segment from frames 235 to 240, if the average amplitude of the dominant frequency component is below the silence energy threshold, it is identified as a silent segment. Then, the timestamp list corresponding to the image frames where no muscle movement was detected is cross-referenced with the timestamp list of silent segments to compare whether there is an overlap between the two lists. For example, if the image still frame is between 120 and 125 seconds and the audio silent segment is between 123 and 128 seconds, then the intersection is between 123 and 125 seconds. Based on this, the time periods of all static image frames and silent audio segments are cross-referenced to extract their intersection time range. The intersection results are then segmented into time periods to form a set of continuous time periods containing the overlapping duration of static images and silent speech, ultimately yielding the set of non-response segment intersection regions.

[0026] S103: Based on the intersection region set of non-response segments, analyze the continuity of time intervals between adjacent segments, merge consecutive adjacent time segments, and output the start and end time points of each segment to obtain the time window of restricted behavior performance. First, the start and end times of each segment in the set are extracted and sorted chronologically to construct a time series list. Then, for each adjacent time interval in this time series, the time interval difference is compared to determine if the interval between the two segments is less than or equal to a set maximum interval threshold. This maximum interval threshold is set to 5 seconds based on the data frame sampling frequency and the visual and auditory reaction periods. For example, if the first segment is 40 to 44 seconds and the second segment is 45 to 48 seconds, the calculated interval is 1 second, indicating that they are consecutive segments. These two segments are then merged into a single continuous segment of 40 to 48 seconds. When the interval between a pair of adjacent segments is detected to be greater than the judgment value, such as the second segment being 45 to 48 seconds and the third segment being 55 to 58 seconds, with an interval of 7 seconds, the segments are not merged and are treated as another independent time segment. This operation is performed sequentially throughout the entire set of non-response intersection segments until all segments have been judged and processed. Then, the start and end times of the merged and unmerged time segments are output respectively. For example, the output results are that the start and end times of segment 1 are 40 to 48 seconds, segment 2 is 55 to 58 seconds, segment 3 is 61 to 67 seconds, etc. This completes the merging of continuous intervals in the behavioral time sequence dimension and finally obtains the behavioral performance-restricted time window.

[0027] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the time period corresponding to the time window of behavioral performance limitation, the original signal sequences of skin conductance, heart rate and respiratory rhythm are collected, the numerical change amplitude in the continuous frame segment of the signal is extracted, and the time interval between adjacent peaks is analyzed. The parameters are stored into the structure mapping set according to the signal channel to obtain the rhythm fluctuation structure parameter set. First, the skin conductance sequence is read frame by frame. The numerical difference between adjacent frames is extracted, and the direction of change (positive, negative, or zero) is determined. The absolute value of the difference in each frame is compared with a set conductance change benchmark to determine if it constitutes a valid change. For example, if the collected conductance values ​​are 3, 5, 4, the difference between adjacent frames (2 and 1) is compared, and difference 1 is considered a segment lower than the conductance change benchmark 2 and marked accordingly. Then, the same frame-level reading operation is performed on the heart rate signal. The numerical difference between each two adjacent heartbeats in the heart rate sequence is defined as a heart rate fluctuation amplitude. The difference is compared with the heart rate fluctuation benchmark to determine if it constitutes a valid heart rate change. For example, if a heart rate sequence is 80, 88, 91, then the difference 8 between 8 and 3 can be determined as high. The effective variation segment of the heart rate fluctuation baseline value 5 is defined, while the difference value 3 is marked as an insignificant segment below the baseline value. Then, peak and valley localization is performed on the respiratory rhythm. By detecting the alternation of rising and falling segments in the respiratory waveform, the position of the peak point is statistically analyzed and the time interval between two adjacent peak points is read. For example, when the peak position is located at sample point index 10 and 18 respectively, the inter-peak interval can be calculated as 8 sampling point lengths and used as the interval parameter of the respiratory rhythm segment. After extracting the skin conductance change amplitude, heart rate change amplitude and respiratory peak interval, the three types of parameters are stored in the structure mapping set according to the channel and each type of parameter is consistent with its corresponding channel index so that it can directly correspond to the change characteristics of different signals within the time window of behavioral performance limitation in subsequent association judgment, and finally obtain the rhythm fluctuation structure parameter set. Table 2. Rhythmic Signal Fluctuation Parameters signal type Time period Variation range (unit) Crest interval (unit) Volatility Trend Skin conductance 0-1 second 3→5→4 Fluctuation Heart rate 1-2 seconds 80→88→91 8、3 Increased volatility respiratory rhythm 2-3 seconds 10、18 Continued fluctuations Table 2 summarizes the fluctuation parameters of skin conductance, heart rate, and respiratory rhythm signals, demonstrating the changes in these physiological signals over different time periods. By analyzing the amplitude and peak intervals of these signals, abnormal rhythmic fluctuations can be identified in patients at rest, which is crucial for pain assessment. The trends in these fluctuations can provide additional physiological support for determining the intensity and type of pain, especially when patients do not exhibit significant behavioral changes; in such cases, rhythmic signal analysis can effectively supplement the accuracy of pain assessment.

[0028] S202: Based on the rhythmic fluctuation structure parameter set, retrieve the rhythmic signal fluctuation parameters of the same channel in the previous non-static period, read the corresponding change amplitude and peak interval values, compare the channel data of the current and non-static period point by point, extract the fluctuation difference state under the channel as the output data, and obtain the rhythmic signal fluctuation comparison result set. First, based on the amplitude and peak interval parameters of skin conductance, heart rate, and respiratory rhythm signals collected within the current time window of restricted behavior, and according to the channel type correspondence, historical rhythm signal data sampled from the same channel in the previous non-stationary time period are located. The amplitude and peak interval parameters recorded in the corresponding channel within that time period are then extracted. After channel pairing, the current channel data is compared one-to-one with the historical channel data. The difference between the amplitude of change in the current time period and the amplitude of change in the previous non-stationary time period is compared, and the difference is used to determine whether there is a change in fluctuation trend. For example, if the peak interval of the current respiratory channel is 6 sampling points, while the peak interval of the corresponding channel in the previous non-stationary time period was 10 sampling points, then this channel shows [a certain trend] in the current time period. If the fluctuation frequency shows an upward trend, and the change in skin conductance is 1 in the current period and 5 in the previous period, then the change in conductance has slowed down significantly, thus indicating that the fluctuation characteristics of this channel have weakened. After performing this kind of comparison for each channel, the channel number and fluctuation status of the channel with a difference greater than the set benchmark difference are extracted and output as separate channel information. The channel number is used for subsequent path positioning, and the fluctuation status is used to describe whether there is an increasing or decreasing trend in the rhythm performance during the signal performance change. For example, if the change in heart rate parameters of a certain detected channel decreases from 88 to 65 and the interpeak interval increases from 4 to 7, under the premise that the set fluctuation difference benchmark is 15 and the interpeak difference benchmark is 2, the channel status is identified as a channel with weakened fluctuation and participates in the subsequent rhythm path determination step, finally obtaining the rhythm signal fluctuation comparison result set.

[0029] S203: Based on the comparison result set of rhythm signal fluctuations, monitor whether there are continuous static behaviors in the current time period, and at the same time analyze whether the skin conductance, heart rate and respiratory rhythm signals continue to fluctuate, and obtain the contradictory state area of ​​signal performance. First, a continuous search is performed on the static behavioral expressions obtained from the posture monitoring channel and facial expression sequence within the current time period. Static behavioral expressions are considered segments where the angular velocity change is near zero and the body displacement direction remains unchanged for several consecutive frames. Then, a fluctuation feature extraction step is performed on the continuous sampling data of three channels—skin conductance, heart rate, and respiratory rhythm—within the current time period. Skin conductance data is analyzed by calculating the amplitude sequence of changes between adjacent sampling points. For example, if the derivative of two adjacent frames in the current continuous sampling increases from 1 to 3 and then decreases to 2, the segment is identified as being in a continuous fluctuation state. For heart rate signals, the heart rate interval sequence is extracted to determine whether it maintains a regular fluctuation. For example, if the heart rate interval shortens from 8 frames to 6 frames and then expands to 9 frames in several consecutive cycles, the sequence is considered to be within a fluctuating range. For respiratory rhythm signals, the trend of chest displacement changes is used to determine whether there is repeated inhalation and exhalation. For example, if the chest displacement increases from 15 to 27 and then decreases to 16 within a continuous time period, it indicates that the respiratory cycle has not terminated. After completing the fluctuation description of the three types of signals, they are compared with the static behavioral expressions. Parallel comparison is used to classify a state as contradictory when the behavior is static and exhibits continuous characteristics in the current time period, and at least two of the three types of rhythmic signals simultaneously show fluctuation continuity. In this process, it is necessary to compare the difference markers in the rhythmic signal fluctuation comparison result set. When the difference marker is consistent with the fluctuation trend collected in the current time period, it is considered a valid fluctuation segment. If a channel is marked as having weakened fluctuation in the difference marker, but still shows fluctuations in the current time period, it is necessary to compare the fluctuation continuity by comparing the intervals of three consecutive peaks of that channel. For example, if the peak intervals are 6, 7, and 5 in sequence, it is considered to still be within the fluctuation range. Then, the behavior static sequence and the rhythmic fluctuation sequence are merged into the same time axis for segment mapping. When the behavior static segment completely covers the start and end range of the rhythmic fluctuation segment, it is judged as a state of coexistence of behavior static and rhythmic activity. If it only partially covers, the length of the covered segment needs to be compared. When the covered length exceeds two-thirds of the length of the fluctuation segment, it is also judged as a contradictory state segment. After completing the above retrieval, all time segments that satisfy the coexistence relationship of static and fluctuation are merged into the same set in sequence, and finally the contradictory state region of signal performance is obtained.

[0030] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the contradictory state area of ​​signal performance, extract the pain label content output within the corresponding time period, filter the fields describing the intensity, compare with the fluctuation trajectory of the rhythm channel, extract the starting position and amplitude turning point of continuous fluctuation changes, and obtain the set of rhythm channel fluctuation segments. First, the pain label content output within the corresponding time period is extracted. Pain labels are a set of text fields describing pain levels. Each label typically includes dimensions such as pain level, location, duration, and perceived nature. At this stage, fields representing pain level or intensity need to be filtered out. Word fragments indicating perceived intensity are read from the label fields, and marker phrases indicating changes in level are identified. For example, when "severe," "mild," or "worsening" appears in the label text, it is mapped to an internally set intensity value. Then, based on the time labels marked in the labels, an intersection operation is performed with the time periods of contradictory signal manifestations to confirm whether the pain labels completely cover the corresponding signal manifestation segments. If there are discrepancies, only the label segments within the intersection range are taken as valid descriptions. Subsequently, for this time range, the signal sequence in the rhythm channel is read. This signal sequence consists of electrophysiological fluctuation values ​​recorded according to time frames, forming a continuously floating curve on the time axis. When processing this type of curve data... First, the starting position of the continuous fluctuation segment needs to be identified. By monitoring the trend of the signal slope change, the point where the signal turns from stable to rising is extracted as the starting position. At the same time, the turning point of amplitude change is monitored in the fluctuation trajectory. When the signal changes from rapid rise to stability or decline, or from decline to rise again, the turning point is taken as the feature node of amplitude change. For example, when the heart rate channel increases from 80 beats per minute to 110 beats per minute in multiple consecutive sampling frames, and then stabilizes at 100 beats per minute, this turning point is the peak amplitude position. Continue to perform the same operation in adjacent channels. Extract the rising turning point and the falling turning point for the skin conduction channel and the respiratory rhythm channel respectively, and align them according to the sampling frame time axis. Map the continuous fluctuation segment recorded in each channel into a unified fluctuation segment sequence. Each segment includes the starting frame position, the ending frame position and the extreme value coordinates, and also marks the fluctuation direction and the rate of change. Finally, extract the segments that conform to the fluctuation pattern in each rhythm channel and include them in the output to obtain the rhythm channel fluctuation segment set. Table 3 Comparison of Pain Labels and Signals Tag Description Signal segment range (time period) Corresponding fluctuation characteristics Coverage Mild discomfort 0-1 second Skin conductance fluctuations Full coverage Moderate oppression 1-2 seconds Heart rate fluctuations Partial coverage Severe colic 2-3 seconds respiratory fluctuations Full coverage Table 3 illustrates the correspondence between pain labels and physiological signal fluctuation characteristics. Each pain label typically includes information such as pain intensity, location, and duration. By comparing the characteristics of signal fluctuations (e.g., the number of fluctuations, duration of rhythmic frames, etc.), it can be determined whether the label is consistent with the actual physiological signal. In this comparison, if the pain label fails to fully cover certain periods of signal fluctuation, the label content needs to be adjusted to ensure a more accurate reflection of the patient's actual pain experience. Through this process, the assessment system can correct the labels, enhancing the accuracy and adaptability of pain rating.

[0031] S302: Call the rhythm channel fluctuation fragment set, extract the continuous untriggered muscle movement fragments in the behavior path, compare the start and end points of the fragments in the sequence according to the time index, and locate the time segment where the behavior is still and the rhythm is active, to obtain the behavior rhythm comparison segment group. First, a behavioral path channel is selected in the time series, namely the muscle movement response channel obtained based on video image analysis. Within this channel, a set of image frame indices representing continuous static or non-active behavioral path sequence segments is extracted. Then, the start and end frame numbers of each segment in this sequence are used as time segment identifiers and compared on the time axis with the corresponding time frame indices of each segment in the rhythmic channel's fluctuating segment set. Specifically, for each muscle movement static segment, the start and end points of the fluctuating segments in the rhythmic signal are checked to see if they fall within the muscle movement time range. If both the start and end frames of the rhythmic segment are within a certain static behavioral segment, it is considered that there is a static behavioral path and an active rhythmic signal during that time period. All static behavioral segments are matched segment by segment through a traversal method, and their corresponding time segments are recorded. The overlapping rhythmic fluctuation channels and their signal characteristics, such as fluctuation amplitude range and number of peak intervals, are analyzed. At the same time, it is determined whether there is a phenomenon of multiple channels overlapping simultaneously. In this process, the start and end frames need to be precisely aligned with millisecond-level indexing precision to ensure the rigor of time crossover determination. Short segments with only partial overlap of start and end ranges or less than two sampling points overlap are further eliminated to ensure that the complete and continuous characteristics of the rhythmic signal are reflected in the judgment. For example, if there is a static segment with a length of 250 frames in the muscle movement channel and an undulating segment in the skin conductance channel, with start and end frames of the 60th and 200th frames respectively, it is considered that the two have an intersection in this time segment and the rhythm is in an active state. Finally, all time segments that meet the conditions of static and active overlap are merged into the output set to obtain the behavioral rhythm comparison segment group.

[0032] S303: Based on the behavioral rhythm comparison segment group, analyze the pain label field and rhythm activity in the time period, extract the segments not covered by the label performance, and obtain the label performance marking group; First, extract the time segments in the group where the rhythm is active while the behavior is still. Obtain the start and end frame index values ​​of each segment as the time index reference for the current processing. Based on this, read the original pain label content within the same time range. These labels are textual field information previously generated by assessors or automated processing methods based on behavioral or physiological responses. The content may contain language expressions describing the pain level, such as "mild discomfort," "moderate pressure," and "severe colic." Perform a lexical scan on each label field to extract key phrases or terms representing intensity, such as "pressure," "twitching," and "stabbing," as the core content of the label expression. Construct a label index set according to the time interval of the corresponding label. Then, compare each behavioral rhythm segment. The time intervals in the group are compared at the frame index level with the time intervals covered by the label index set to determine whether the time of the rhythm segment is completely covered by the time range of any label. If both the start and end frames are within a certain label time interval, it is marked as covered; otherwise, it is considered a label missing region. Extraction is performed on all rhythmic active segments that are not completely covered by the pain label, and the corresponding behavior static channel identifier and rhythm channel identifier information are recorded. For example, in a rhythmic behavior comparison segment from frame 120 to frame 340, no label coverage is detected. The system records this time interval and its corresponding channel as a subsequent alternative label reference area. Finally, all such uncovered segment information is merged into the output result to obtain the label performance label group.

[0033] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the time period in the tag group, extract the chest undulation sequence, skin conductance fluctuation segment and sound pressure change point. After splitting by channel, extract the number of fluctuating signal segments and corresponding time index in each channel to obtain the channel segment number sequence. First, signal sequences corresponding to each time period index were retrieved from chest undulation signals, skin conductance fluctuation data, and sound pressure change data. Sampling frames of the signal sequences were read and a time-axis mapping table was constructed. Each type of signal was split into three channel datasets according to its original channel number. For chest undulation signals, the corresponding frame segments of each channel were traversed to find segments with continuous upward or downward trends. Whenever the trend direction changed, it was considered the end of an undulation segment, and the start and end frames of that segment were recorded. For example, in the chest undulation channel, a continuous rise from frame 200 to frame 248 was recorded as an undulation segment. For skin conductance signals, segments with baseline fluctuations moving up and down were extracted. Each fluctuation was defined as... The definition is a segment whose amplitude exceeds the baseline offset value. For each offset segment, its start and end frames and amplitude change value are extracted. In the sound pressure channel, the abrupt change point is determined by sampling the frequency domain peak value. The waveform change segment exceeding the average baseline amplitude is divided into intervals, and its start and end time points and change direction features are extracted. Then, the total number of segments extracted from each type of channel and the corresponding start and end time index are written into independent data lists. Under the premise that there is no overlap between channels, three lists are output as the chest undulation segment count table, the skin conductance fluctuation segment count table, and the sound pressure change segment count table, respectively. Subsequently, the number of segments of the three types are uniformly encapsulated into a set of indexed structures to obtain the channel segment count sequence.

[0034] S402: Based on the channel segment quantity sequence, retrieve the response start index value of the channel fluctuation segment, cross-compare the start order in the differentiated time period, filter the channel identifiers with early response and extract the total number of initiation segments and time coverage to obtain the response start priority sequence. First, the starting index value of the fluctuation segment corresponding to each channel is retrieved one by one in the sequence. All starting indices of the three channels are extracted and placed into three lists in time order. Based on this, the position of the first fluctuation segment of each channel is determined. For example, the starting index of the first fluctuation segment of the chest fluctuation channel is frame point A, the starting index of the first fluctuation segment of the skin conductance channel is frame point B, and the starting index of the first fluctuation segment of the sound pressure channel is frame point C. The three index values ​​A, B, and C are directly compared according to the frame point size. If A is less than B and C, the chest fluctuation channel is extracted as the early response channel. If B is the smallest, the skin conductance channel is extracted. If C is the smallest, the sound pressure channel is extracted. At the same time, the total number of fluctuation segments in each channel is counted as the total number of initiating segments, and the coverage ratio of the fluctuation segment of the channel to the current time period is calculated. By comparing the coverage ratios of the three channels, their order in the subsequent sequence is determined. Finally, the channel identifiers of the early response, the number of initiating segments, and the coverage ratio are written into the output set in channel order to obtain the response starting point priority sequence.

[0035] S403: Call the response start priority sequence, determine whether the synchronization segment time coverage remains continuous, compare the response start time order of the channel, locate the response channel that appears for the first time, extract the continuous fluctuation segment information in the channel and perform time-series mapping with the corresponding synchronization segment to obtain the pain judgment channel main path sequence. First, the start time index of the response in each channel fluctuation segment is extracted, and a set of time periods corresponding to the response channel is established. Then, the time intervals in each set are compared one by one to check whether each response segment overlaps with the corresponding segment of other channels on the time axis, and to confirm whether the distribution of time points in the overlapping interval remains continuous. For example, if there are multiple adjacent segments in the chest fluctuation channel, and the interval between each segment is no more than three frames, it is defined as a continuous response segment. Then, the time coverage of this channel is compared with the synchronous response segments of other channels. If the continuous segment overlaps with the fluctuation segment of the sound pressure channel or skin conductance channel on the time axis, the segment is considered to have synchronous coverage. Subsequently, all channel segment groups with synchronous time coverage are extracted. The system compares the first response segment in each channel according to their time sequence to locate the channel where the first response action occurred. For example, if the starting frame of the response in channel A is 100, the starting frame of channel B is 120, and the starting frame of channel C is 150, then channel A is the first response channel. Next, the frame index and start and end points of all continuous fluctuating segments in the channel are extracted. These fluctuating segments are compared with the synchronous coverage segments identified in the previous step, and the time intersection area is used as the intersection point mapping. Finally, the correspondence between the channel and its synchronous segment is constructed. Through this mapping relationship, the time index of all fluctuating segments in the channel is matched one by one with the time interval of the coverage segment, and the output content is obtained as the main path sequence of the pain judgment channel.

[0036] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the pain judgment channel main path sequence, extract the number of channel behavior fluctuations, the number of continuous frames of rhythm changes and the interruption interval of signal segments during the contradictory state period, and compare the continuity relationship of signal items in the time period to obtain the signal dynamic performance parameter set. First, based on the behavioral fluctuation segment data extracted from the channel, the total number of fluctuation markers in the frame segment is retrieved. For example, in the breathing channel, the upper and lower boundary frame indices of each fluctuation segment are extracted, and the number of adjacent fluctuation segments is calculated as the number of breathing fluctuations. Then, the frame segment index differences between adjacent peaks and troughs in the rhythm trajectory of the channel are read. Based on this, the duration range of each rhythm change at the frame level is calculated. Segments with a frame length greater than 20 are recorded as continuous segments, and the number of segments is accumulated as the number of continuous frame segments of the rhythm change. Next, the time index interval between any two fluctuation segments in the same channel is retrieved, and the portion with a frame difference greater than 50 is taken as an interruption marker. All interruption periods and their corresponding time intervals are counted. The corresponding index interval uses the number of interrupt segments as the interrupt interval parameter. The above three parameters constitute the original dataset of the channel during the contradictory state period. Then, the signal trajectory is retrieved according to the time index, and the fluctuation boundary and interrupt start and end index are searched frame by frame to determine whether each segment is continuous in time or has a breakpoint. If there is an interval of no less than 20 frames between multiple fluctuation segments in the same channel, and there are more than two consecutive interrupt segments, then its continuity relationship is considered broken. Otherwise, it is judged as a continuous state. The above continuity judgment is combined with the number of signal segments, rhythm frame segment persistence and interruption number to output the three numerical parameters of the corresponding channel and the continuity relationship label, and finally obtain the signal dynamic performance parameter set.

[0037] S502: Call the signal dynamic performance parameter set, retrieve the state type features described by the pain label, and parse the range of the description content in the label and the signal performance index. Compare the matching between the label content and the signal range to obtain the label description correspondence table. First, retrieve the corresponding pain label description field from the time period corresponding to each signal group. Check whether the field contains keywords such as "mild," "continuous," "intermittent," and "severe," which indicate pain intensity or frequency. Then, read the three key indicator values ​​from the signal parameter set: number of behavioral fluctuations, number of rhythmic segments, and number of interruptions. Pre-set reference intervals for each type of descriptive word in the field content. For example, "mild" corresponds to 1 to 3 behavioral fluctuations, no more than 2 rhythmic segments, and more than 3 interruptions; "severe" requires more than 6 fluctuation segments, more than 4 rhythmic segments, and 0 or 1 interruptions. When reading parameters, compare the three parameters of the current signal channel within the specified time period with these set intervals. When the number of fluctuations is in the upper-middle range and the number of rhythmic segments is also within the range, the signal is considered to be at its best. When the upper limit and the number of interruptions are zero, the current parameter group matches the "violent" type. Conversely, if the number of fluctuations is low and the number of interruptions is higher than the set threshold, the "slight" or "intermittent" label is matched. Then, the matching results are output item by item for each channel. For cases with unclear matching, it is necessary to read whether there are combined descriptions such as "slight and intermittent" or "continuous but not violent" in the label text. In such compound expressions, the three parameters are respectively matched with their respective expression ranges. It is confirmed one by one whether the parameters fall within the acceptable range. If any parameter is inconsistent with the label semantics, it is marked as "not completely matched" and otherwise marked as "completely matched". Finally, the correspondence between the label fields and the signal parameter ranges is sorted out. By comparing the results, the docking status between each label description statement and the signal performance is summarized to obtain the label description correspondence table.

[0038] S503: Based on the label description correspondence table, determine whether the label content covers the indicator combination of the current signal performance range, and filter the label items that do not have the corresponding coverage relationship, and replace them with the label content that completely overlaps with the time interval to obtain the pain level verification label. First, the ranges of three key parameters corresponding to each tag—the number of behavioral fluctuations, the number of rhythmic frames, and the interval between interruptions—are extracted. Each set of parameter values ​​in the current signal's dynamic performance parameters is read, and compared with the tag's coverage area on the same channel to see if there is a complete intersection. Specifically, each set of signal parameters is compared in numerical order. If any parameter exceeds the boundary of the tag's corresponding interval, it is considered incomplete coverage. Based on this, the relationship between each tag item and the signal combination is determined item by item. If a tag only covers the number of fluctuation segments but not the number of rhythmic frames or the interruption interval, the tag is marked as partial coverage, and further investigation is conducted to check if there are other uncorresponding tags or missing descriptions in the behavioral path during the time period when the tag appears. For tag items that do not have a coverage relationship, the remaining content in the tag description library is compared. Prioritize entries that contain expressions of fluctuation, rhythm, and interruption in their semantic descriptions and have already shown a "complete match" status in the corresponding relationship table. Further extract the parameter coverage range of the candidate entry and re-compare it with the current signal parameters. If it covers the entire signal performance range, replace the original label. During the replacement process, the time index, channel identifier, and sequence number identifier must remain unchanged. Only the label text content is updated. After updating the label, iterate again to confirm whether the replaced description is still consistent with the current parameter combination. If there are still uncovered parameters after replacement, continue to search for the next more complete match in the candidate labels until the position is replaced with the label content that completely overlaps with the time range. Finally, complete the above operations in all signal paths to form a complete set of alternative labels and obtain the pain level verification label.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for assessing pain levels in palliative care for tumors based on multimodal sensing, characterized in that, Includes the following steps: S1: Obtain the angular velocity and displacement direction recorded by the attitude sensor, determine whether the angular velocity change in adjacent per-second segments is continuously within the static judgment threshold, detect the overlap between the facial muscle movement non-triggered frames and the speech frequency interruption segment, and obtain the time window of restricted behavior performance. S2: Based on the time window of the behavior performance limitation, extract skin conductance, heart rate and respiratory signals, compare the rhythm fluctuation with the previous non-static state, identify the rhythm active state under the behavior static state, and obtain the signal performance contradictory state region. S3: Based on the contradictory state area of ​​the signal performance, compare whether the behavioral silence and rhythm changes are not reflected by the label, extract the time points not covered by the signal deviation, and obtain the label performance marking group; S4: Based on the tag performance marking group, identify the number of channel fluctuations, response order and synchronization characteristics, determine the dominant signal channel, and obtain the main path sequence of the pain judgment channel; S5: Based on the main path sequence of the pain judgment channel, read the number of behavioral fluctuations, rhythm continuity and signal output during the contradictory state period, compare with the pain label state description, replace the original label, and obtain the pain level verification label.

2. The method for assessing pain levels in palliative care for tumors based on multimodal sensing according to claim 1, characterized in that, The behavioral performance-restricted time window includes angular velocity change state, body displacement state, number of frames without facial muscle movement triggering, number of interrupted speech frequency segments, and timeline overlapping segment positions. The signal performance contradiction state area includes skin conductance fluctuation slope, heart rate rhythm fluctuation interval, respiratory rhythm change amplitude, behavioral path output state, and rhythm signal fluctuation state. The label performance marker group includes rhythm channel fluctuation state, behavioral path silent state, label content indicating state, and label content response state difference. The pain judgment channel main path sequence includes the number of chest fluctuation sequence segments, skin conductance fluctuation segments, sound pressure change points, response start order, and signal synchronization status. The pain rating verification label includes the number of behavioral fluctuations, rhythm change continuity, signal duration, signal performance range, and alternative label content.

3. The method for assessing pain levels in palliative care for tumors based on multimodal sensing according to claim 1, characterized in that, The static judgment threshold refers to the range of angular velocity and displacement changes used to identify whether an individual is in a static state; The "facial muscle movement not triggered frame" refers to an image frame in which no muscle movement is observed on the face within the time frame.

4. The method for assessing pain levels in palliative care for tumors based on multimodal sensing according to claim 1, characterized in that, The term "rhythmic active state under behavioral stillness" refers to the contradictory state in which there is no behavioral movement but physiological signals continue to fluctuate actively. The signal deviation-uncovered time point refers to the moment of pain that is not covered by the pain label and where there are abnormal physiological fluctuations.

5. The method for assessing pain levels in palliative care for tumors based on multimodal sensing according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire the angular velocity and body displacement direction continuously recorded by the attitude sensor, extract the angular velocity change value of adjacent per second segments, monitor the continuous trend of the numerical sequence, filter the time intervals in which the numerical fluctuation is within the preset error range in the continuous time period, and obtain the set of time intervals in the static state. S102: Based on the set of static state time intervals, extract the video sequence and audio sequence corresponding to the time intervals, identify the image frames in the video where no muscle movement response was detected and the segments in the audio where the main frequency is zero, cross-reference the time tags, extract the segments with time intersection, and obtain the set of non-response segment intersection regions. S103: Based on the set of intersection regions of the non-response segments, analyze the continuity of the time interval between adjacent segments, merge consecutive adjacent time segments, and output the start and end time points of each segment to obtain the time window of restricted behavior performance.

6. The method for assessing pain levels in palliative care for tumors based on multimodal sensing according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the time period corresponding to the time window of the behavior limitation, collect the original signal sequences of skin conductance, heart rate and respiratory rhythm, extract the numerical change amplitude in the continuous frame segments of the signal, analyze the time interval between adjacent peaks, and store the parameters into the structure mapping set according to the signal channel to obtain the rhythm fluctuation structure parameter set. S202: Based on the rhythmic fluctuation structure parameter set, retrieve the rhythmic signal fluctuation parameters of the same channel in the previous non-static period, read the corresponding change amplitude and peak interval values, compare the channel data of the current and non-static period point by point, extract the fluctuation difference state under the channel as output data, and obtain the rhythmic signal fluctuation comparison result set. S203: Based on the set of rhythm signal fluctuation comparison results, monitor whether there is continuous static behavior in the current time period, and analyze whether the skin conductance, heart rate and respiratory rhythm signals continue to fluctuate to obtain the contradictory state region of signal performance.

7. The method for assessing pain levels in palliative care for tumors based on multimodal sensing according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the contradictory state area of ​​the signal, extract the pain label content output within the corresponding time period, filter the field describing the intensity, compare with the fluctuation trajectory of the rhythm channel, extract the starting position and amplitude turning point of continuous fluctuation, and obtain the set of rhythm channel fluctuation segments. S302: Call the rhythm channel fluctuation segment set, extract the continuous untriggered muscle movement segments in the behavior path, compare the start and end points of the segments in the sequence according to the time index, and locate the time segment where the behavior is still and the rhythm is active, to obtain the behavior rhythm comparison segment group. S303: Based on the behavioral rhythm comparison segment group, analyze the pain label field and rhythm activity in the time period, extract the segments not covered by the label performance, and obtain the label performance marking group.

8. The method for assessing pain levels in palliative care for tumors based on multimodal sensing according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the time period in the tag group, extract the chest undulation sequence, skin conductance fluctuation segment and sound pressure change point, and extract the number of fluctuating signal segments and corresponding time index in each channel after splitting by channel to obtain the channel segment number sequence. S402: Based on the channel segment quantity sequence, retrieve the response start index value of the channel fluctuation segment, cross-compare the start order in the differentiated time period, filter the channel identifiers with early response and extract the total number of initiation segments and time coverage to obtain the response start priority sequence. S403: Call the response start priority sequence, determine whether the synchronization segment time coverage remains continuous, compare the response start time order of the channel, locate the response channel that appears for the first time, extract the continuous fluctuation segment information in the channel and perform time-series mapping with the corresponding synchronization segment to obtain the pain judgment channel main path sequence.

9. The method for assessing pain levels in palliative care for tumors based on multimodal sensing according to claim 8, characterized in that, In the process of extracting the number of fluctuating signal segments in each channel of the channel segment number sequence: extract signal segments with continuous intervals between respiratory peaks in the chest fluctuation sequence, and extract the number of fluctuations of the corresponding signal segments according to the channel splitting method; extract segments with continuous fluctuations in conductance change trend in the skin conductance fluctuation segment, and extract the number of signal segments and time index according to the channel order. In the process of extracting the information of the continuous fluctuation segment: in the region where the signal amplitude change shows a trend of non-zero amplitude difference between adjacent sampling points, continuous segments are extracted according to the time index and mapped to the time range corresponding to the response channel in the synchronization segment.

10. The method for assessing pain levels in palliative care for tumors based on multimodal sensing according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Based on the main path sequence of the pain judgment channel, extract the number of channel behavior fluctuations, the number of continuous frames of rhythm changes and the interruption interval of signal segments during the contradictory state period, and compare the continuity relationship of signal items in the time period to obtain the signal dynamic performance parameter set. S502: Call the signal dynamic performance parameter set, retrieve the state type features described by the pain label, and parse the range of the description content in the label and the signal performance index. Compare the matching between the label content and the signal range to obtain the label description correspondence table. S503: Based on the label description correspondence table, determine whether the label content covers the indicator combination of the current signal performance range, and filter the label items that do not have the corresponding coverage relationship, and replace them with the label content that completely overlaps with the time interval to obtain the pain level verification label.

Images