Wearable system for evaluating pain

Abstract

The invention belongs to the field of human body detection devices, and provides a wearable system for evaluating pain, which comprises wearable equipment in communication connection with a cloud system. The wearable device is used for continuously collecting physiological signals when a patient wears the wearable device. The cloud system is provided with a hierarchical evaluation module, a correction module and an intervention control module, the hierarchical evaluation module extracts physiological features in a long-term time window, extracts skin electroreaction mutation, a myoelectricity activity peak value and skin temperature rapid change in a short-term time window, and deduces an acute pain level in combination with instant self-evaluation information of a patient; the correction module is used for correcting the acute score based on the mapping relation between the long-term stability characteristic and the acute pain level and an individualized threshold value; the intervention control module generates individualized electrical stimulation control parameters according to the correction result and issues the individualized electrical stimulation control parameters to the wearable device, and the electrical stimulation execution unit outputs a stimulation signal through the percutaneous electrode to relieve pain of the patient and uploads a feedback signal to the cloud end to form a closed loop.

Description

Technical Field

[0001] This invention belongs to the field of human body detection devices, and more specifically relates to a wearable system for assessing pain. Background Technology

[0002] Pain is one of the most common symptoms in clinical practice and a significant factor affecting patients' quality of life. Its mechanisms are complex, involving not only nerve conduction but also endocrine, circulatory, and psychological factors. In traditional medical settings, pain assessment often relies on patients' subjective descriptions, such as numerical rating scales, visual analog scales, or facial expression scales. While these methods are simple to use, they are highly dependent on patients' self-perception and expressive abilities, lacking objectivity and real-time accuracy.

[0003] To improve objectivity, existing technologies attempt to utilize physiological indicators such as heart rate variability, skin conductance, electromyography (EMG), and electroencephalography (EEG) signals as supplementary references for pain assessment. While these methods improve the scientific rigor of pain assessment to some extent, they typically require specialized monitoring equipment in a hospital setting, which is bulky, costly, and cannot support long-term continuous data collection and dynamic intervention. Although some portable devices can monitor local indicators, they provide limited signal types and fail to integrate with an individual's long-term baseline status, making them prone to bias.

[0004] The shortcomings of existing technologies lie in the fact that assessment methods either rely too heavily on subjective reports, making it difficult to exclude the influence of individual differences and psychological states; or they rely too heavily on single or a small number of physiological indicators, lacking comprehensive analysis across time scales. At the same time, most technologies fail to achieve a closed-loop linkage between assessment and intervention, causing pain relief measures to be disconnected from real-time conditions, resulting in delayed and limited intervention effects. Summary of the Invention

[0005] To address the problems in the prior art, the present invention provides a wearable system for assessing pain, comprising a wearable device and a cloud system, wherein the wearable device is communicatively connected to the cloud system. The wearable device is used to continuously collect pain-related physiological signals while the patient is wearing it, convert the physiological signals into a transmittable data stream, and upload it to a cloud server. The cloud-based system includes a tiered assessment module, a correction module, and an intervention control module. The hierarchical assessment module is used to analyze the data stream at different time scales. Within a long time window, it performs statistical analysis on heart rate variability, sleep rhythm stability, and blood oxygen fluctuation amplitude signals to extract stability features reflecting the patient's chronic pain state. Within a short time window, it extracts features of abrupt changes in skin conductance response, peak values ​​of electromyography, and rapid changes in skin temperature. Combined with the patient's real-time pain self-assessment information input through a mobile device, it infers the patient's acute pain level. The correction module is used to calculate a correction factor based on the mapping relationship between the chronic pain stability characteristics and the acute pain level, and to adjust the acute pain score using the correction factor to obtain a corrected acute pain score. The intervention control module is used to generate individualized electrical stimulation control parameters based on the modified acute pain score, and to send the electrical stimulation control parameters to the wearable device; The wearable device is also used to receive electrical stimulation control parameters sent from the cloud, and drive the electrical stimulation execution unit to perform nerve electrical stimulation on the patient based on the electrical stimulation control parameters to relieve acute pain.

[0006] The wearable device is also used to receive electrical stimulation control parameters sent from the cloud, and drive the electrical stimulation execution unit to perform nerve electrical stimulation on the patient based on the electrical stimulation control parameters to relieve acute pain.

[0007] Furthermore, the short-term time window is set to 10 seconds to 5 minutes.

[0008] Furthermore, the stratified assessment module uses the amplitude of abrupt changes, the rate of rise, and the recovery time as characteristics for skin electrical response, the peak power and spectral energy distribution as characteristics for electromyographic activity, and the rate of change of skin temperature as a characteristic.

[0009] Furthermore, the correction module calls the patient's historical self-assessment data and short-term characteristic signals to perform regression modeling, obtains an individualized pain trigger threshold, and classifies patients into sensitive, normal, and insensitive types based on this threshold.

[0010] Furthermore, the correction module performs multi-source consistency verification before performing the correction. The verification includes calculating the similarity of skin conductance response, electromyography activity, and skin temperature, and reducing the correction weight when the trends are inconsistent.

[0011] Furthermore, the electrical stimulation control parameters include intensity, frequency, pulse width, and waveform.

[0012] Furthermore, when generating electrical stimulation control parameters, the intervention control module calls up the patient's historical intervention records to compare the parameter combinations and relief effects under the same previous scores.

[0013] Furthermore, the wearable device collects feedback signals of heart rate, skin conductance response, and local electromyography activity while performing electrical stimulation, and uploads the feedback signals to the cloud.

[0014] Furthermore, the cloud system updates patient records based on feedback signals and optimizes the selection of electrical stimulation parameters during subsequent interventions.

[0015] This invention organically combines multimodal physiological signal acquisition, hierarchical assessment, correction processing, and individualized intervention to form a closed-loop structure based on wearable devices and a cloud system. This system can continuously acquire long-term and short-term signals in patients' daily life scenarios and obtain more stable and sensitive pain indicators through cross-timescale analysis, thus overcoming the limitations of traditional single-signal or subjective assessments.

[0016] This invention introduces long-term baselines, individualized thresholds, and multi-source consistency verification through a correction module, enabling objective and personalized correction of acute pain scores across different individuals and states. The resulting scores reflect the patient's current experience and demonstrate consistency across individuals, providing more reliable data support for clinical diagnosis and home rehabilitation.

[0017] This invention generates individualized electrical stimulation intervention parameters based on the assessment results, achieving real-time closed-loop linkage between assessment and intervention. The system can dynamically adjust the intensity, frequency, pulse width, and waveform of electrical stimulation according to the patient's pain level, and optimize it by combining historical response data, making the intervention more precise and efficient, and significantly improving the intelligence and individualization of pain management. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the 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.

[0019] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system principle structure of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to explain the present invention and should not be construed as limiting the scope of protection of the present invention.

[0021] The process principle of this invention is as follows: Figure 1 As shown, the wearable device first collects multimodal physiological signals from the patient, including heart rate, blood oxygen saturation, skin temperature, skin conductance, and electromyography (EMG) signals. Preprocessing, such as filtering, noise reduction, and normalization, is performed locally before the data is uploaded to the cloud system. The hierarchical assessment module in the cloud system extracts chronic pain characteristics such as heart rate variability, sleep rhythm, and blood oxygen fluctuations within a long-term time window, while simultaneously analyzing rapid signals such as skin conductance, EMG activity, and temperature changes within a short-term time window. Combined with the patient's immediate self-assessment results, the system comprehensively infers the level of acute pain.

[0022] Based on this, the correction module refines the acute pain score using long-term baselines, individualized thresholds, and multi-source consistency verification, generating a score that better reflects the patient's true condition. The intervention control module then generates individualized electrical stimulation control parameters based on the revised score and sends them to the electrical stimulation execution unit of the wearable device, which outputs corresponding stimulation signals via transcutaneous electrodes to relieve pain. During execution, the device simultaneously collects intervention feedback signals and uploads them to the cloud, creating a closed-loop mechanism of assessment-correction-intervention-feedback. This improves the objectivity of pain assessment and enables real-time and personalized intervention.

[0023] In one embodiment, such as Figure 2 As shown, a wearable system for pain assessment is provided, including a wearable device and a cloud system. The two interact with each other through wireless communication to form an integrated closed-loop structure of continuous data collection, intelligent analysis and personalized intervention.

[0024] The wearable device is used to continuously monitor a target body surface area while worn by a patient, collecting various physiological signals related to pain. The wearable device is preferably a wristband, armband, or patch structure, enabling long-term wear and signal acquisition without affecting the patient's daily activities. The physiological signals include, but are not limited to, heart rate, blood oxygen saturation, skin temperature, skin conductance, and electromyography (EMG). Specifically, heart rate and blood oxygen saturation signals can be acquired using a photoplethysmography (PPG) sensor; skin temperature signals can be acquired using an integrated thermistor or infrared sensor; skin conductance signals can be acquired by detecting changes in skin conductivity using electrode patches; and EMG signals can be acquired by detecting local muscle activity using surface EMG electrodes.

[0025] The wearable device contains a microprocessor unit, which includes an analog-to-digital conversion module, a filtering unit, and a data preprocessing unit. This unit digitizes the acquired analog signals and uses a multi-level filtering algorithm to suppress power frequency interference, eliminate motion artifacts, and correct baseline drift. The preprocessing unit further performs amplitude normalization, feature extraction, and data format conversion on the signal to generate a standardized data stream.

[0026] The wearable device further includes a wireless communication unit, which can be a Bluetooth Low Energy module, a Wi-Fi module, or a cellular communication module, used to upload the data stream to a cloud server in real time. To ensure the stability and continuity of data transmission, the communication unit is preferably configured with a cache management mechanism and a packet loss retransmission mechanism, and uses an end-to-end encryption protocol to securely protect the data stream, preventing information leakage or data tampering during transmission.

[0027] The cloud-based system includes a tiered assessment module, a correction module, and an intervention control module.

[0028] The stratified assessment module is used to analyze the data stream uploaded by the wearable device across time scales, so as to simultaneously characterize the patient's chronic pain state and acute pain level, thereby achieving a comprehensive and stratified pain assessment. The design of this module is based on the differences in the physiological manifestations of pain at different time dimensions: chronic pain often presents a stable and continuous pattern of change in long-term signals, while acute pain manifests as rapid and intense transient fluctuations in short-term signals.

[0029] Chronic pain often leads to a long-term imbalance between the sympathetic and parasympathetic nervous systems, manifesting as decreased heart rate variability and abnormal frequency power distribution. Therefore, the module extracts the statistical distribution characteristics of heart rate variability by calculating the mean, standard deviation, and the ratio of low-frequency power to high-frequency power over a time span of several hours to several days. Regarding sleep rhythm, chronic pain often leads to a decrease in the proportion of deep sleep and an increase in the number of nighttime awakenings. The module combines accelerometer motion monitoring with heart rate change trends and uses a sleep staging algorithm to calculate the integrity and continuity indicators of the sleep cycle, thus reflecting the interference of pain on sleep regularity. Regarding blood oxygenation levels, chronic pain is associated with the body's long-term stress response, manifested as increased fluctuations in blood oxygenation amplitude and a higher frequency of abnormal decreases. The module extracts stability features related to chronic hypoxia by calculating the average level, standard deviation, and the number of abnormal decreases within the monitoring period. Through the processing of these long-term signals, a long-term feature set that can stably characterize the patient's chronic pain state can be obtained.

[0030] Within a long-term time window, the stratified assessment module performs statistical analysis on signals reflecting the body's chronic state. This long-term time window is preferably set to 6 to 72 hours to cover circadian rhythms and multi-day trends. Its implementation includes: Heart rate variability (HRV) analysis: By calculating the mean heart rate, standard deviation, and the ratio of low-frequency power to high-frequency power over a long-term window, this analysis reflects the balance between the sympathetic and parasympathetic nervous systems. Patients with chronic pain often exhibit decreased HRV and increased low-frequency components; this indicator can serve as an important reference for long-term pain load.

[0031] Sleep rhythm regularity analysis: Nighttime body movement frequency was monitored using accelerometers and combined with heart rate fluctuations. A sleep staging algorithm was used to obtain the proportion of deep sleep, the number of nighttime awakenings, and sleep cycle integrity. Chronic pain often leads to sleep fragmentation and reduced deep sleep; this indicator can reveal the interference of pain on sleep patterns.

[0032] Blood oxygen level fluctuation analysis: This method detects the long-term trend of blood oxygen saturation using photoplethysmography (PPG) signals, calculating the average blood oxygen value, standard deviation of fluctuation, and frequency of events falling below a set threshold. Chronic pain, accompanied by prolonged stress, may lead to increased fluctuations in blood oxygen levels; long-term monitoring helps identify potential hypoxic conditions.

[0033] Through the above three aspects of calculation and analysis, the output features within the long-term time window constitute a set of stable features of chronic pain, providing a long-term anchoring baseline for subsequent score correction.

[0034] Acute pain often triggers a strong, transient physiological response within seconds to minutes. For example, the skin conductance response exhibits a significant abrupt change during acute pain, characterized by a rapid increase in conductivity within a short period. The module quantifies the transient stress level of the autonomic nervous system by detecting the amplitude, rate of increase, and recovery time of this abrupt change. Electromyography (EMG) signals during acute pain are often accompanied by involuntary muscle contractions or abnormal discharges. The module extracts the power peak and spectral energy distribution of the sampled signals using methods such as short-time Fourier transform and wavelet analysis, thereby identifying the correspondence between short-term EMG activity peaks and pain triggers. Skin temperature may fluctuate rapidly during acute pain due to local vasoconstriction or vasodilation. The module captures microcirculatory change patterns associated with pain attacks by calculating the rate of temperature change within a sliding window and detecting anomalies.

[0035] Within a short time window, the hierarchical evaluation module extracts features from signals that sensitively reflect the body's instantaneous responses. This short time window is preferably set to 10 seconds to 5 minutes to ensure the capture of rapid changes. Its implementation includes: Detection of GSR mutations: This method monitors transient changes in skin conductivity at high sampling rates, extracting the mutation amplitude, rise rate, and recovery time. Acute pain leads to sympathetic nerve excitation, resulting in a significant GSR peak, which directly reflects the autonomic nervous response induced by pain.

[0036] Electromyographic (EMG) peak extraction: Time-frequency analysis of EMG signals is performed using short-time Fourier transform or wavelet transform to capture abnormal peaks, frequency components, and energy distributions occurring within a range of several seconds to tens of seconds. Acute pain may trigger involuntary muscle contractions or defensive movements; this characteristic can serve as a dynamic manifestation of pain triggering.

[0037] Rapid Skin Temperature Change Detection: This technology monitors the rate of temperature change on the skin surface using a high-resolution temperature sensor, identifying rapid increases or decreases in temperature occurring within a short period. Acute pain is often accompanied by peripheral vasoconstriction or vasodilation, resulting in rapid changes in local temperature.

[0038] While extracting the aforementioned short-term signal features, the hierarchical assessment module also receives real-time pain self-assessment information input by patients through a mobile application. This self-assessment information can be input using a numeric rating scale (NRS), visual analog scale (VAS), or expression scale, etc. This subjective information, together with the objective short-term signal features, constitutes a multimodal dataset of short-term pain.

[0039] In terms of data processing methods, the hierarchical assessment module can employ machine learning-based feature fusion strategies, such as support vector machines, random forests, or deep neural network models, to nonlinearly integrate multimodal features. Through the trained classification or regression model, the module can comprehensively infer the patient's acute pain level on a short-term scale.

[0040] In one specific implementation, the hierarchical assessment module employs a machine learning-based feature fusion strategy in short-term pain analysis to more accurately infer the patient's acute pain level. The system first simultaneously collects skin conductance, electromyography (EMG), and skin temperature signals over a time window of tens of seconds to several minutes using a wearable device, combining this with the patient's real-time self-reported pain score input on their mobile device. This data is then processed to form training samples, where multimodal physiological features serve as input vectors and patient self-reports as output labels, used for model training and validation.

[0041] During feature extraction, the electrodermal response signal was processed into abrupt change amplitude, rise rate, and recovery time. The electromyographic signal was transformed into power density, dominant frequency, and energy concentration through time-frequency conversion. Skin temperature was converted into rate of change, extreme point spacing, and directionality indicators. All features underwent noise filtering and standardization before being uploaded to the cloud to ensure comparability and stability between different signals.

[0042] In the cloud system, the module accesses stored sample data and divides the dataset into training, validation, and test sets. It then trains the dataset using different machine learning models, including Support Vector Machines (SVMs), Random Forests, and Deep Neural Networks (DNNs). SVMs establish a mapping between multimodal features and pain scores using non-linear kernel functions; Random Forests learn from subsets of features using multiple decision trees and use the average result as the prediction value; and Deep Neural Networks achieve non-linear interactions between complex features through multi-layer fully connected structures. During training, the system continuously adjusts hyperparameters and prevents overfitting through cross-validation and early stopping mechanisms.

[0043] When the wearable device uploads new short-term signals, the hierarchical assessment module extracts their features in real time and inputs them into the trained model. The model outputs an acute pain score based on the comprehensive performance of the multimodal signals. If the patient's immediate self-assessment score is also obtained, the system will weight and fuse the model's output with the self-assessment result to form a prediction value that better reflects the patient's actual experience. The weighting ratio can be dynamically adjusted based on the patient's historical consistency; for example, when the model and self-assessment results maintain a high degree of consistency over a long period, the weight of the model's prediction is increased; conversely, the weight of the self-assessment input is appropriately increased.

[0044] In a specific example, a patient exhibited a skin conductance response amplitude of 0.28 microsiemens, an electromyography power density of 0.14 mV² / Hz, and a skin temperature decrease of 0.4 degrees Celsius within a 30-second time window. The hierarchical assessment module input these features into a trained random forest model, which outputs an acute pain score of 7.3. Simultaneously, the patient inputs an immediate self-assessment score of 7 via a mobile device. After fusion, the system outputs a final acute pain level of 7.2 and passes this result to the correction module for further adjustment.

[0045] Through the above implementation, the hierarchical assessment module can leverage the complementarity of multimodal physiological signals in the short term and combine it with the nonlinear integration capabilities of machine learning models to output a score that more closely reflects the patient's true pain state. This method avoids misjudgments caused by a single signal and improves the accuracy and robustness of acute pain assessment.

[0046] In summary, the stratified assessment module extracts stable features reflecting chronic pain status through a long-term time window, and infers acute pain levels by combining subjective and objective features through a short-term time window; the two complement each other. Long-term features provide relatively stable baseline and trend information, while short-term features provide sensitive real-time responses. The joint analysis across time scales can balance the accuracy, sensitivity, and individualized adaptability of pain assessment, providing a solid foundation for subsequent score adjustments in the correction module.

[0047] The correction module is used to jointly model and establish a mapping relationship between long-term pain stability characteristics and short-term acute pain levels output by the hierarchical assessment module, thereby calibrating the acute pain score. Its design principle is that the long-term physiological characteristics formed by chronic pain can reflect the individual patient's baseline pain sensitivity and pain threshold differences, while short-term acute pain signals are easily affected by fluctuations in individual subjective experience, psychological state, or environmental interference. Therefore, by using long-term characteristics to correct short-term scores, subjective bias can be effectively reduced, resulting in pain score results with stronger cross-individual comparability.

[0048] In its specific implementation, the correction module first performs feature quantification on long-term stability characteristics. This preferably includes: calculating the mean and standard deviation of heart rate variability indicators to reflect the long-term balance between the sympathetic and parasympathetic nervous systems; normalizing the continuity indicators of sleep rhythms to quantify the long-term impact of sleep quality on pain; and counting and quantifying the amplitude and intensity of blood oxygen fluctuations and decline events to characterize chronic hypoxia and stress levels. These long-term characteristics are integrated into a baseline vector reflecting the patient's chronic pain sensitivity.

[0049] Subsequently, the correction module incorporates the calculation results of short-term acute pain levels. These short-term results consist of abrupt changes in skin conductance response characteristics, peak characteristics of electromyographic activity, rapid changes in skin temperature, and immediate self-assessment scores. The module establishes a mapping relationship between long-term characteristics and short-term scores, which can be achieved through various algorithms. Regression modeling: Establish a regression equation between long-term features and short-term scores through linear regression, ridge regression, or nonlinear regression, and output a set of coefficients for correction.

[0050] Weighting factor calculation: The weights of different short-term signals are determined based on the stability of long-term characteristics. For example, when long-term indicators show that the patient is in a highly sensitive state, the correction coefficient of the short-term score is increased; conversely, when long-term indicators show that the patient is in a low-sensitivity state, the correction magnitude of the short-term score is decreased.

[0051] Pattern recognition algorithms: Using algorithms such as support vector machines, random forests, or neural networks, the algorithms classify or cluster the combined patterns of long-term and short-term features and output the corresponding correction factors.

[0052] After obtaining the correction factor, the correction module applies it to the short-term acute pain score to form the corrected acute pain score. This correction process is preferably weighted correction or residual correction. Weighted correction refers to multiplying short-term scores by weighting factors generated from long-term characteristics to make the scores more closely reflect the patient's baseline pain sensitivity.

[0053] Residual correction refers to calculating the difference between short-term scores and long-term feature predictions, and using this difference to adjust the short-term scores, thereby reducing bias.

[0054] In one specific embodiment, after receiving the data output by the hierarchical evaluation module, the correction module first quantifies the long-term stability characteristics. The system calculates the following long-term indicators over the past 24-hour monitoring period: Heart rate variability (HRV): mean 42 ms, standard deviation 18 ms. Based on the established thresholds (mean < 50 ms and standard deviation < 20 ms), the system determined that the patient's autonomic nervous system was in a state of long-term sympathetic dominance.

[0055] Sleep rhythm continuity: The proportion of deep sleep was only 15%, and the number of awakenings during the night was 8. The continuity index, calculated after normalization, was 0.35 (out of 1.0), indicating a significant decrease in sleep quality.

[0056] Blood oxygen fluctuation amplitude: The average blood oxygen level was 94%. There were 6 instances of blood oxygen drop below 90% within 24 hours. The fluctuation amplitude was quantified as 0.25, indicating unstable tissue oxygen supply.

[0057] The above three indicators were integrated into a long-term baseline vector [HRV=low, sleep=poor, blood oxygen=large fluctuations], and the overall characteristics indicate that the patient is in a state of "high sensitivity" pain.

[0058] Within a short time window (1 minute), the stratified assessment module extracted the following features: a skin conductance response amplitude of 0.32 microsiemens, a peak electromyographic activity of 0.18 mV², and a skin temperature decrease of 0.5 degrees Celsius. The patient's immediate self-assessed pain score was 6 points, and the system, after integrating the above features, output an acute pain score of 6.5 points.

[0059] The correction module then proceeds to the correction step: The module calls a pre-built feature-rating mapping model (which can be logistic regression or random forest) to determine whether a patient is "highly sensitive" based on the long-term baseline vector. Under this classification, the correction coefficient is preferably set in the range of 0.80–0.90; in this example, it is set to 0.85.

[0060] The module directly adjusts the short-term acute score according to the correction factor, resulting in a corrected score of 6.5 × 0.85 = 5.5. This process is equivalent to reducing the impact of short-term fluctuations, making the results more consistent with the actual level of long-term sensitivity in patients.

[0061] Simultaneously, the module invokes a prediction function based on long-term features. Based on HRV, sleep patterns, and blood oxygenation, the predicted baseline pain score for the patient is 5.2. The residual between the short-term outcome and the predicted value is then calculated to be 1.3. The module sets the residual correction ratio to 0.25, retaining only 25% of the bias, resulting in a corrected value of 5.2 + (1.3 × 0.25) = 5.5. This result is consistent with the weighted correction method, thus enhancing the reliability of the correction.

[0062] The revised acute pain score of 5.5 was output as the final result to the intervention control module, and was archived along with the short-term score of 6.5 and the long-term baseline vector for subsequent trend analysis and model updates.

[0063] Through the aforementioned correction operations, the correction module can significantly reduce the risk of misjudgment caused by individual differences, psychological state, or momentary environmental interference. For example, for the same electromyographic peak value, some patients may subjectively perceive it as mild pain, while others may perceive it as severe pain. By introducing long-term stability characteristics as a baseline, the correction module can smooth and correct for this difference, ensuring that the final output pain score is more objective and consistent.

[0064] The intervention control module is used to determine electrical stimulation control parameters that match the individual patient's condition based on the corrected acute pain score, after obtaining the score, in order to achieve targeted neuro-electrical stimulation intervention. The control parameters specifically include four aspects: electrical stimulation intensity, frequency, pulse width, and waveform.

[0065] Among them, the intensity parameter is used to adjust the output amplitude of current or voltage to control the degree of activation of the target nerve fiber by electrical stimulation; the frequency parameter is used to set the number of repetitions of electrical stimulation pulses per unit time to affect the excitation rhythm of the nerve; the pulse width parameter is used to determine the duration of a single electrical stimulation pulse, thereby adjusting the conduction depth and range of the current in the tissue; and the waveform parameter is used to select the specific form of the electrical stimulation signal, such as square wave, biphasic wave or gradient wave, to adapt to the sensitivity of different nerve tissues to the stimulation waveform.

[0066] To avoid parameter fluctuations caused by a single scoring result, the intervention control module preferably combines the patient's historical response data for comprehensive judgment. Specifically, when generating control parameters, the module calls upon stored patient intervention records, compares the electrical stimulation parameters used in the past with the corresponding relief effects under the same or similar pain scores, and dynamically optimizes the current output parameter set using a parameter adjustment strategy. For example, when historical data shows that the patient responds better to high-frequency, short-pulse-width stimulation, the module will tend to recommend similar stimulation patterns under the same scoring conditions; conversely, when historical records show that a certain parameter combination is prone to causing adverse reactions, the module will automatically reduce the priority of that combination.

[0067] After the parameters are generated, the intervention control module sends the electrical stimulation control parameters to the wearable device via a communication link. Upon receiving the control command, the electrical stimulation execution unit in the wearable device drives the electrodes to output corresponding electrical stimulation signals according to the set intensity, frequency, pulse width, and waveform, thereby achieving personalized intervention on the patient's target nerve region. Through this closed-loop operation, the system can achieve real-time relief and continuous management of acute pain while ensuring safety.

[0068] The wearable device is further configured to, upon receiving the electrical stimulation control parameters transmitted from the cloud, transmit the parameters to the local control circuit and invoke the built-in electrical stimulation execution unit for driving. The electrical stimulation execution unit includes a pulse generator, a current control circuit, and a transcutaneous electrode assembly. The pulse generator generates a corresponding electrical stimulation pulse sequence according to the received control command, and the current control circuit is used to adjust the intensity, frequency, pulse width, and waveform of the output pulse to ensure that the output electrical signal is consistent with the control parameters transmitted from the cloud.

[0069] The percutaneous electrodes are preferably placed in target nerve regions associated with pain, such as the wrist, forearm, or lower back, to ensure that the electrical stimulation signal can effectively act on peripheral nerves or nerve plexuses. During stimulation, the electrodes are in close contact with the skin surface, and conductive gel or flexible media are used to reduce skin contact resistance, thereby improving current conduction efficiency and stimulation comfort. After the electrical stimulation signal acts on the target area, it can change the excitation threshold of local nerve fibers, inhibit abnormal discharges, or regulate the conduction pathway of pain signals, thereby achieving the effect of alleviating the patient's acute pain.

[0070] During electrical stimulation, the monitoring unit in the wearable device continuously collects intervention-related physiological feedback signals, such as changes in heart rate, skin conductance, and local electromyographic activity. These feedback signals, along with execution status data (including actual output current value, electrode contact status, and duration of electrical stimulation), are uploaded to the cloud system. Based on this feedback information, the cloud system records and analyzes the intervention effect, thereby achieving a closed-loop operation mechanism. This allows the system to further optimize parameter selection based on historical execution results during subsequent interventions.

[0071] Through the above structure, the system described in this embodiment can achieve a complete closed loop from continuous acquisition of physiological signals, hierarchical assessment at different time scales, and correction of pain scores to dynamic generation and execution of electrical stimulation intervention parameters, through the synergistic effect of wearable devices and cloud systems. This improves the objectivity and accuracy of pain assessment and enhances the individualization and timeliness of intervention.

[0072] In another embodiment, the correction module, based on the mapping relationship between chronic pain stability characteristics and acute pain levels, further introduces a dynamic baseline calibration and individual threshold adaptation mechanism, and combines a multi-source consistency verification step to improve the individualization and time sensitivity of acute pain score adjustment.

[0073] The correction module first establishes a long-term baseline curve for the patient in the cloud. This baseline curve is formed by accumulating the stability characteristics of multiple long-term time windows and serves as a reference anchor point for short-term acute pain scores. Specifically: The correction module receives continuous heart rate signals from the wearable device, performs noise reduction and peak detection on the signals, and extracts the RR interval. Within a 24-hour monitoring window, it calculates the mean, standard deviation, and frequency domain power ratio of heart rate variability to obtain indicators reflecting the long-term stability of the patient's autonomic nervous system. The baseline heart rate variability is updated daily through a rolling window and forms a time-series curve to characterize the long-term balance between the sympathetic and parasympathetic nervous systems.

[0074] The correction module accesses accelerometer and heart rate monitoring results in the cloud and identifies deep sleep, light sleep, and wakefulness based on a preset sleep staging algorithm. Over a multi-night monitoring period, it extracts parameters such as total deep sleep duration, number of awakenings, and sleep cycle integrity, and averages these parameters to obtain a sleep rhythm characteristic curve. This curve reflects changes in sleep quality caused by chronic pain and serves as another reference indicator for an individual's long-term pain sensitivity.

[0075] The correction module uses photoplethysmography (PPG) signals collected by wearable devices to calculate blood oxygen saturation values. Within a one-week time window, it statistically analyzes the daily average blood oxygen saturation, standard deviation, and the frequency of events where oxygen saturation drops below a set threshold (e.g., 90%). This data generates a blood oxygen fluctuation stability curve, reflecting the patient's tissue oxygenation and stress levels in the context of chronic pain.

[0076] The correction module uses a moving average method to smooth the baseline characteristics mentioned above, ensuring that short-term fluctuations do not significantly affect the long-term trend. Simultaneously, it employs an exponentially weighted moving average (EWMA) method to give recent data a higher weight in the overall trend, thus achieving dynamic updates that reflect both long-term trends and appropriate sensitivity to recent conditions. The update process is executed periodically in the cloud, preferably daily, and retains at least one week's worth of historical data for trend analysis.

[0077] Once the short-term acute pain score is calculated by the stratified assessment module, the correction module compares the score with the long-term baseline curve. The comparison method includes: calculating the difference between the current score and the baseline curve at the same point in time to obtain the deviation magnitude Δ; and determining the direction of change of this difference relative to the baseline curve (increasing or decreasing). Based on this, the correction module generates dynamic weights to adjust the reliability of the acute pain score.

[0078] For example, when the acute score is significantly higher than the baseline curve and Δ exceeds the set threshold, the correction module determines that the score is consistent with the actual trend of pain aggravation in the patient, thereby increasing the weight of the score in the subsequent correction results; conversely, when the acute score is lower than the baseline curve, but there are abnormal fluctuations in the short-term signal, the correction module will reduce the weight of the score to avoid overestimation caused by transient artifacts or emotional interference.

[0079] Through the above methods, the correction module can establish a dynamic balance between long-term trends and short-term fluctuations, making the corrected acute pain score more consistent with the patient's actual pain state.

[0080] While establishing a long-term baseline, the correction module further utilizes the correspondence between the patient's historical self-assessment data and physiological signals to form an individualized pain threshold range, which serves as a reference for adjusting acute pain scores. The formation of this individualized threshold range includes the following steps: The correction module retrieves the patient's historical self-assessment data from the cloud storage unit. This self-assessment data consists of pain scores entered by the patient through a mobile application, preferably using a numeric rating scale (NRS) of 0-10. Simultaneously, the correction module reads short-term characteristic signals corresponding to the scoring time point. These short-term signals include the peak amplitude of the skin conductance response (GSR), the power density of electromyography (EMG), and the rate of change of skin temperature within a short time window.

[0081] The correction module establishes a functional relationship between the self-reported data and short-term signals by performing regression analysis or logistic regression modeling. For example, using a logistic regression model, a pain score greater than a set threshold (e.g., 5 points) is defined as a "pain event," and the corresponding GSR peak amplitude and EMG power density are used as input variables to obtain an individualized "pain trigger threshold." This threshold reflects the minimum physiological signal intensity required to induce a specific pain experience in the patient.

[0082] After obtaining the pain trigger threshold, the correction module further classifies patients, preferably into three categories: sensitive, moderate, and insensitive. For sensitive patients, even when the detected changes in GSR peak amplitude or EMG power density are only low, they still report moderate or higher pain. The correction module sets the threshold range for this type of patient to the lower range and automatically lowers the correction lower limit when correcting acute scores to avoid underestimating minor physiological fluctuations.

[0083] For patients with common symptoms, physiological signals and self-assessment scores are basically linearly related. The correction module keeps the upper and lower limits of the threshold unchanged and adopts the standard correction range.

[0084] For patients with a dulled pain profile, a significant increase in GSR peak or EMG power density must be detected for the patient to report moderate or higher pain. The correction module sets the threshold range for this type of patient to the high range and raises the correction upper limit when correcting acute scores to ensure that the true pain response is not underestimated.

[0085] The correction module incorporates a periodic update mechanism to ensure the long-term adaptability of the threshold range. Specifically, at a preset period (preferably every week or after every ten self-assessment data accumulations), the module re-accesses the latest self-assessment data and corresponding short-term signals, performs regression analysis using the same method as the threshold determination, and updates the individualized pain trigger threshold. When the difference between the newly calculated threshold and the old threshold exceeds a preset tolerance, the system automatically adjusts the current patient's classification and the correction upper and lower limits.

[0086] Meanwhile, the correction module introduces a data weighting mechanism during the update process, preferably using the exponential weighted average (EWMA) method, which gives more weight to recent data when updating the threshold, thereby ensuring that the threshold can be adjusted in a timely manner according to changes in the patient's physiological state, but will not fluctuate drastically due to a single abnormal score.

[0087] To prevent distortion caused by a single abnormal signal, the correction module incorporates a multi-source consistency verification step before performing acute pain score correction. This consistency verification consists of a preprocessing and alignment unit, a similarity measurement unit, and a decision and weighting unit, as detailed below: Preprocessing and Alignment (1) Time window and step size: The skin conductance response (GSR), electromyography (EMG) and skin temperature (Temp) were analyzed within a short time window. The preferred time window was 60s and the step size was 10s.

[0088] (2) Noise reduction and standardization: GSR: Bandpass filtering removes power frequency and low frequency drift and extracts skin conductance derivatives to highlight abrupt changes; EMG: Rectification – Low-pass envelope processing, suppressing high-frequency noise and obtaining a power envelope; Temp: Moving average and first difference, emphasizing rapid rise and fall.

[0089] The above signals are normalized to zero mean and unit variance to obtain comparable dimensionless sequences.

[0090] (3) Time alignment: Each sequence is resampled to a uniform sampling rate (preferably 10Hz), and fine-tuned alignment is performed using a dynamic time warping constraint band with limited bandwidth (preferably a Sakoe-Chiba band with a window length of 10%) to eliminate the difference in response delay between different sensing channels.

[0091] Similarity measurement Linear correlation measure: Calculate the Pearson correlation coefficient for any two channel normalized sequences, and the preferred threshold is denoted as [value missing]. When the correlation coefficient is ≥ When the value is preferably 0.60 to 0.70, the two channels are considered to have the same trend.

[0092] (2) Shape similarity measure: Calculate the DTW distance of the same pair of channels, and the preferred threshold is denoted as When DTW distance ≤ When the range is 0.8 to 1.2 (obtained from equipment calibration, preferably after standardization), the two channels are determined to be similar in shape.

[0093] (3) Event time test: Extract the event time (GSR mutation peak, EMG envelope peak, Temp rapid change extreme value) on the three channels respectively. If the events of at least two channels appear simultaneously within the allowable offset (preferably ±5s), it is recorded as a "multi-source co-occurrence".

[0094] Signal quality and channel validity assessment To avoid poor channel quality affecting decision-making, a Signal Quality Index (SQI) is set: GSRSQI: Electrode contact resistance is within permissible limits (preferably <500kΩ), and baseline drift is below the threshold. EMGSQI: The saturation sample ratio is below the threshold (preferably <2%) and is out of sync with the high-amplitude interference of the motion sensing (IMU); TempSQI: The sensor self-test is normal, sampling is stable, and the packet loss rate is low.

[0095] EMGSQI: The saturation sample ratio is below the threshold (preferably <2%) and is out of sync with the high-amplitude interference of the motion sensing (IMU); TempSQI: The sensor self-test is normal, sampling is stable, and the packet loss rate is low.

[0096] When a channel's SQI is unqualified, that channel will not participate in the current window's consistency voting to avoid misleading judgments.

[0097] Decision-making and demotion (1) Consistency scoring: Within a time window, the logarithm of pairwise combinations of the three channels satisfying "correlation metric passes and shape metric passes" is counted; at the same time, the number of "multi-source co-occurrences" is counted. If any of the following conditions are met, it is determined to be "passing consistency": At least two pairs of combinations simultaneously pass both correlation and shape thresholds; The "multi-source co-occurrence" event appears at least once in this window.

[0098] (2) Trend verification with acute pain scores: Compare the trend (increase / decrease) of the initial acute pain score with the multi-channel composite trend. If the trends are consistent, maintain the correction strength; if the trends are inconsistent and the inconsistency lasts for ≥2 consecutive windows, it is judged as "low consistency".

[0099] (3) Weight reduction and amplitude limiting: When the result is determined to be "low consistency", the correction module reduces the weight of the correction intensity of the current window, preferably reducing the correction weight to 50% to 70% of the original value, and applies amplitude limiting to the corrected acute score (preferably not exceeding ±20% of the score of the previous period). To avoid frequent jitter, a hysteresis mechanism is set: the original correction weight can only be restored after two consecutive windows pass the consistency test.

[0100] Exception and rollback handling (1) Channel missing: When there are fewer than two available channels, this window will only record the information and will not make any enhancement corrections, maintaining the correction weights of the previous period.

[0101] (2) Strong artifact identification: If the IMU indicates violent body movement and the EMG saturation ratio exceeds the limit, and the GSR baseline jumps instantaneously beyond the upper limit, it is marked as an "artifact window". Upward adjustment is prohibited within this window, and downward adjustment or maintenance is only allowed.

[0102] (3) Logs and traceability: Consistency results, pass / fail status of each threshold, final weight and limit value are written to the cloud log for subsequent parameter adjustment and compliance audit.

[0103] After the above steps, the acute pain score output by the correction module is no longer a single short-term signal result, but a weighted correction result based on long-term baseline trend constraints, adaptive correction of individual thresholds, and consistency verification of multi-source signals. This corrected acute pain score better reflects the patient's true pain experience and reduces misjudgments caused by individual differences and abnormal signals.

[0104] For any module structures not specifically defined in this invention, the existing technical descriptions shall prevail. The prior art mentioned in the foregoing background and specific embodiments sections can be considered part of this invention and used to understand the meaning of certain technical features or parameters.

Claims

1. A wearable system for assessing pain, characterized in that, The system comprises a wearable device and a cloud system, and the wearable device is in communication connection with the cloud system: The wearable device is used for continuously collecting physiological signals related to pain in the state of patient wearing, converting the physiological signals into a transmissible data stream, and uploading the data stream to the cloud server; The cloud system comprises a hierarchical evaluation module, a correction module, and an intervention control module; The hierarchical evaluation module is used for analyzing the data stream in different time scales, statistically analyzing heart rate variability, sleep rhythm stability, and blood oxygen fluctuation amplitude signals in a long-term time window to extract stability features reflecting the chronic pain state of the patient, and extracting features of skin electric response mutation, muscle electric activity peak, and skin temperature rapid change in a short-term time window, and combining the instant pain self-evaluation information input by the patient through the mobile terminal to infer the acute pain level of the patient; The correction module is used for calculating a correction factor based on the mapping relationship between the chronic pain stability features and the acute pain level, and adjusting the acute pain score by using the correction factor to obtain a corrected acute pain score; The intervention control module is used for generating individualized electric stimulation control parameters according to the corrected acute pain score, and issuing the electric stimulation control parameters to the wearable device; The wearable device is also used for receiving the electric stimulation control parameters issued by the cloud, and driving the electric stimulation execution unit to perform nerve electric stimulation on the patient to relieve acute pain based on the electric stimulation control parameters.

2. The wearable system for assessing pain of claim 1, wherein, The long-term time window is set to 6 hours to 72 hours.

3. The wearable system for assessing pain of claim 1, wherein, The short-term time window is set to 10 seconds to 5 minutes.

4. The wearable system for assessing pain of claim 1, wherein, The hierarchical evaluation module uses mutation amplitude, rising rate, and recovery time as features for skin electric response, uses power peak and frequency spectrum energy distribution as features for muscle electric activity, and uses change rate as a feature for skin temperature.

5. The wearable system for assessing pain of claim 1, wherein, The correction module calls patient historical self-evaluation data and short-term feature signals for regression modeling to obtain an individualized pain trigger threshold, and divides the patient into sensitive type, normal type, and sluggish type based on the threshold.

6. The wearable system for assessing pain of claim 1, wherein, The correction module performs multi-source consistency verification before performing correction, and the verification comprises calculating the similarity of skin electric response, muscle electric activity, and skin temperature, and reducing the correction weight when the trends are inconsistent.

7. The wearable system for assessing pain of claim 1, wherein, The electric stimulation control parameters include intensity, frequency, pulse width, and waveform.

8. The wearable system for assessing pain of claim 1, wherein, The intervention control module calls patient historical intervention records when generating the electric stimulation control parameters, and compares the parameter combinations and relief effects under the same scores in the past.

9. The wearable system for assessing pain of claim 1, wherein, The wearable device collects heart rate, skin electric response, and local muscle electric activity feedback signals while performing electric stimulation, and uploads the feedback signals to the cloud.

10. The wearable system for assessing pain of claim 1, wherein, The cloud system updates the patient file based on the feedback signals, and optimizes the electric stimulation parameter selection in the subsequent intervention process.

Images