AI prediction-based endometriosis pain emotional pressure data management system

Abstract

The invention relates to the field of medical data processing, and discloses an endometriosis pain emotion pressure data management system based on AI prediction, which comprises a mobile terminal, an Internet of Things treatment monitoring device, an expiration detection device and a data processing server. The expiration detection device is used for detecting volatile organic compounds in expired gas by utilizing a sensor array; the Internet of Things treatment monitoring device collects heartbeat interval and skin conductivity data and executes transcutaneous nerve electrical stimulation; and the server runs the deep learning model. According to the invention, time alignment and feature splicing are carried out on physiological, expiration and behavior data through a multi-source heterogeneous data fusion module, and the pain occurrence probability is predicted by using a long-short-term memory network and an attention mechanism; early warning is generated according to the prediction result, the electrical stimulation frequency and waveform are adjusted in a closed-loop mode based on the real-time skin conductivity change slope, and multi-dimensional objective monitoring, pain risk active prediction and self-adaptive closed-loop intervention of endometriosis are achieved.

Description

Technical Field

[0001] This invention relates to the field of medical data processing, specifically to an AI-based data management system for endometriosis-related pain, emotional stress, and other symptoms. Background Technology

[0002] Endometriosis is a common chronic gynecological disease, causing patients to suffer from long-term pelvic pain and accompanying anxiety and depression. Currently, daily management of endometriosis largely relies on patients' subjective recording of pain scores and the use of analgesics. Existing technologies for pain assessment are mostly limited to visual analog scales (VAS), which are easily influenced by environmental and psychological factors and lack objective physiological and biochemical indicators. Although some monitoring devices can collect physiological parameters such as heart rate variability, these single-dimensional physiological signals cannot comprehensively characterize the complex pathophysiological state of endometriosis, especially neglecting volatile organic compounds in exhaled breath related to inflammatory metabolism, resulting in a limited and inaccurate assessment of the condition.

[0003] In terms of data processing and early warning, traditional medical management systems focus primarily on data collection and storage, lacking the ability to deeply mine and predictively analyze multi-source heterogeneous data. Existing data analysis models cannot effectively capture the nonlinear characteristics and time dependencies present in long-term physiological data, nor can they accurately identify weak precursory signals before pain attacks. The lack of effective predictive mechanisms means that treatment interventions often lag behind the onset of pain, requiring patients to react passively only after the pain has occurred, thus reducing their quality of life.

[0004] In physical therapy interventions, transcutaneous electrical nerve stimulation (TENS) is a commonly used non-pharmacological analgesia method. Existing TNS devices typically employ open-loop output with preset fixed frequency and pulse width parameters. The human nervous system readily adapts to constant stimulation signals, leading to a decline in analgesic efficacy over time. Current devices lack a closed-loop regulation mechanism based on real-time physiological stress feedback, failing to automatically optimize stimulation parameters according to the patient's current pain relief level and changes in skin impedance, thus hindering personalized and precise analgesia while ensuring treatment safety. Existing technologies also have shortcomings in multidimensional objective assessment, proactive predictive warning, and closed-loop adaptive intervention. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an AI-based data management system for pain, emotion, and stress in endometriosis. This system solves the problems of existing endometriosis management relying on subjective assessments and lacking integrated monitoring of exhaled biochemistry and multimodal physiological indicators, failing to predict pain risk based on temporal characteristics leading to delayed intervention, and physical therapy lacking a closed-loop regulatory mechanism based on real-time stress feedback, resulting in analgesic effects diminishing with adaptation.

[0006] To achieve the above objectives, the present invention provides an AI-based data management system for pain, emotion, and stress in endometriosis, which mainly includes a mobile terminal, an IoT treatment monitoring device, an exhalation detection device, a data processing server, and a management terminal.

[0007] At the system architecture level, the mobile terminal serves as the human-computer interaction and data hub, responsible for receiving users' subjective pain scores, pain locations, and emotional self-assessment data, and aggregating physiological parameter data from the IoT-based treatment monitoring device. The IoT-based treatment monitoring device integrates a transcutaneous electrical nerve stimulation unit and a physiological signal acquisition unit, enabling simultaneous acquisition of intercardiac pulse data sequences and skin conductivity data while applying physical therapy. The breath detection device employs a microelectromechanical system (MEMS) sensor array to specifically detect volatile organic compounds in exhaled breath. The data processing server, as the core computing node, runs a deep learning model to fuse and analyze multi-source heterogeneous data, generating prediction results and control commands.

[0008] In terms of physiological and biochemical signal acquisition principles, the IoT-based therapeutic monitoring device utilizes photoplethysmography (PPG) to acquire intercardia data and a Wheatstone bridge circuit to monitor changes in skin conductivity, thereby quantifying the activity level and stress response of the sympathetic nervous system. The microelectromechanical system (MEMS) sensor array in the breath detection device includes metal oxide semiconductor sensors with different gas-sensitive material films. By doping with different proportions of noble metal catalysts, different sensors exhibit differentiated adsorption and desorption characteristics for characteristic volatile organic compounds associated with endometriosis, such as alkanes, aromatic hydrocarbons, and aldehydes. The sensor output response resistance signal, after analog-to-digital conversion, is used by a data processing server to extract its steady-state amplitude characteristics and dynamic response time characteristics to aid in the identification of pathological conditions.

[0009] In terms of multi-source data processing and AI prediction mechanisms, this invention employs multi-source heterogeneous data fusion technology. The time alignment unit in the data processing server performs timestamp alignment and downsampling aggregation of high-frequency physiological time-series data and low-frequency behavioral record data. The semantic encoding unit converts non-numerical emotion text tags into one-hot encoded vectors. The feature concatenation unit constructs a high-dimensional comprehensive state input vector from normalized physiological features, subjective behavioral features, and exhaled biochemical features.

[0010] The deep learning model of this invention employs an architecture combining a Long Short-Term Memory (LSTM) network and a temporal attention mechanism. The LTM unit utilizes forgetting gates, input gates, and output gates to process comprehensive state input vector sequences over long time spans, capturing the time-dependent features of physiological and behavioral data. The temporal attention mechanism unit calculates the weight scores of hidden state vectors at different time steps, and through weighted summation, highlights key time-point features contributing significantly to future pain prediction, generating a global contextual feature vector. The prediction output unit outputs the probability of pain occurrence within a predetermined future time window based on this feature vector.

[0011] Regarding the closed-loop intervention control logic, this invention achieves dynamic adjustment based on physiological feedback. When the predicted probability of pain occurrence exceeds a preset threshold, the system automatically generates an intervention instruction containing recommended parameters. During treatment by the percutaneous electrical nerve stimulation unit, the system continuously monitors the slope of the subject's skin conductivity change. If the slope is positive, it indicates an increase in stress or pain level, and the system automatically adjusts the pulse frequency or waveform modulation mode; if the slope is negative and the absolute value meets the requirements, it indicates a state of relief, and the system maintains the current parameters. Simultaneously, the safety feedback monitoring circuit monitors the output loop impedance and current in real time to ensure the safety of physical therapy.

[0012] This invention combines specific detection of exhaled biochemical indicators with real-time monitoring of multimodal physiological signals and utilizes deep learning models to mine nonlinear correlations between multidimensional data. It can not only provide early warning of pain risk, but also automatically optimize treatment parameters based on real-time feedback from the body, thus realizing intelligent, personalized, and closed-loop management of pain and emotional stress in endometriosis.

[0013] This invention provides an AI-based data management system for predicting pain, emotions, and stress in patients with endometriosis. It offers the following beneficial effects: 1. This invention integrates an exhalation detection device with an Internet of Things (IoT) treatment monitoring device. It utilizes a microelectromechanical system (MEMS) sensor array to specifically detect certain volatile organic compounds in exhaled gas. In conjunction with multimodal physiological parameters such as intercardiac interval and skin conductivity, as well as patient subjective scoring data, it achieves a comprehensive quantification of the endometriosis status from multiple dimensions, including biochemical metabolism, autonomic nervous function, and subjective feelings. This improves the comprehensiveness and accuracy of pain and emotional stress assessment.

[0014] 2. This invention deploys a deep learning model containing a long short-term memory network and a time attention mechanism in a data processing server to extract features and perform time-series analysis on time-aligned multi-source heterogeneous data sequences. By using attention weight coefficients to focus on key time step features that have an important impact on future states, it achieves early prediction of the probability of pain occurrence within a set future time window, providing patients with a time window for preventive intervention.

[0015] 3. This invention constructs a closed-loop intervention control logic based on skin conductivity feedback, calculates the slope of physiological index changes in real time during percutaneous electrical nerve stimulation treatment, and dynamically adjusts the pulse frequency and waveform parameters according to the rising and falling trends of the stress response. This achieves adaptive optimization of treatment intensity, avoids the problem of reduced adaptability caused by fixed parameter stimulation, and ensures the comfort and safety of the physical therapy process. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the AI-based predictive data management system for endometriosis pain, emotions, and stress according to the present invention. Figure 2 This is a schematic diagram of the system hardware circuit principle and module composition of the present invention; Figure 3 This is a schematic diagram of the data processing algorithm and neural network architecture of the present invention; Figure 4 This is a schematic diagram of the functional logic architecture of the mobile terminal of the present invention; Figure 5 This is a schematic diagram of the non-invasive assisted diagnostic process of the present invention; Figure 6 This is a schematic diagram of the closed-loop intervention and daily pain prediction process of the present invention.

[0017] Among them, 100, AI-based predictive data management system for endometriosis pain, emotion, and stress; 110, mobile terminal; 120, IoT treatment monitoring device; 121, transcutaneous electrical nerve stimulation unit; 122, physiological signal acquisition unit; 130, exhalation detection device; 131, gas flow path assembly; 132, microelectromechanical system sensor array assembly; 133, signal acquisition circuit assembly; 134, temperature control drive assembly; 140, data processing server; 150, management terminal; 310, main control microcontroller circuit; 320, DC-DC boost converter circuit; 330, bidirectional waveform drive circuit; 340, safety feedback monitoring circuit; 350, heart rate variability acquisition submodule; 36 0. Skin Conductivity Acquisition Submodule; 410. Data Fusion Module; 411. Time Alignment Unit; 412. Data Standardization Unit; 413. Semantic Encoding Unit; 414. Feature Concatenation Unit; 420. Pain Pattern Recognition and Prediction Module; 421. Sequence Input Mapping Unit; 422. Long Short-Term Memory Network Unit; 423. Temporal Attention Mechanism Unit; 424. Prediction Output Unit; 510. User Input Interface Module; 520. Communication Protocol Management Module; 530. Data Encapsulation and Encryption Module; 540. Visualization Rendering Module; 550. Device Control Command Module; 560. Closed-Loop Control Module; 561. Decision Logic Unit; 562. Dynamic Feedback Adjustment Unit. Detailed Implementation

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See attached document Figure 1 The present invention provides an AI-based prediction-based endometriosis pain, emotion and stress data management system 100, which includes: a mobile terminal 110, an Internet of Things treatment monitoring device 120, an exhalation detection device 130, a data processing server 140 and a management terminal 150.

[0020] Mobile terminal 110 is equipped with a first wireless communication unit and a human-computer interaction interface. The human-computer interaction interface is configured to receive pain score data, pain location data, pain duration data, and emotional self-assessment data input by the user. Mobile terminal 110 establishes a bidirectional communication connection with IoT treatment monitoring device 120 through the first wireless communication unit. Mobile terminal 110 is configured to package the pain score data, pain location data, pain duration data, emotional self-assessment data, and physiological parameter data received from IoT treatment monitoring device 120 into a first data packet. Mobile terminal 110 sends the first data packet to data processing server 140.

[0021] The IoT-based therapeutic monitoring device 120 includes a transcutaneous electrical nerve stimulation (TENS) unit 121 and a physiological signal acquisition unit 122. The TENS unit 121 includes a pulse generation circuit and an electrode output interface. The pulse generation circuit is configured to generate an electrical stimulation signal with a corresponding frequency, pulse width, and intensity based on received control commands. The physiological signal acquisition unit 122 includes a heart rate variability sensor and a skin conductance sensor. The heart rate variability sensor is configured to acquire heart rate interval data sequences. The skin conductance sensor is configured to acquire skin conductivity data. The IoT-based therapeutic monitoring device 120 is configured to transmit the heart rate interval data sequences and skin conductivity data to a mobile terminal 110.

[0022] The breath detection device 130 includes a gas sampling path, a microelectromechanical system (MEMS) sensor array, an analog-to-digital converter (ADC), and a second wireless communication unit. The MEMS sensor array includes multiple metal-oxide-semiconductor (MOS) gas sensors configured to generate a response resistance signal to volatile organic compounds in exhaled gas. The ADC is configured to convert the response resistance signal into a digital voltage signal. The breath detection device 130 transmits the digital voltage signal to a data processing server 140 via the second wireless communication unit.

[0023] The data processing server 140 includes a data receiving interface, a storage unit, and a processor. The data receiving interface is configured to receive a first data packet from the mobile terminal 110 and a digital voltage signal from the breath detection device 130. The storage unit stores a trained deep learning model. The processor is configured to run the deep learning model, perform feature extraction and fusion analysis on the received data, and generate pain risk prediction results and intervention control instructions. The data processing server 140 is configured to send the pain risk prediction results and intervention control instructions to the mobile terminal 110 and send long-term monitoring report data to the management terminal 150.

[0024] The management terminal 150 is equipped with a display unit configured to present long-term monitoring report data. This data includes curves showing changes in pain scores, mood indicators, and stress levels, as well as results from breath volatile organic compound (VOC) testing. The management terminal 150 is also configured to receive treatment adjustment instructions input by the doctor and synchronize these instructions to the data processing server 140.

[0025] After receiving the intervention control command from the data processing server 140, the mobile terminal 110 is configured to generate prompt information and display it on the human-computer interaction interface. The mobile terminal 110 is also configured to generate an adjustment signal for the transcutaneous electrical nerve stimulation unit 121 according to the intervention control command, and send the adjustment signal to the Internet of Things treatment monitoring device 120, thereby controlling the output parameters of the transcutaneous electrical nerve stimulation unit 121.

[0026] Each sensor in the microelectromechanical system (MEMS) sensor array has a different gas-sensitive material membrane layer, configured to exhibit different adsorption and desorption characteristics for volatile organic compounds (VOCs) such as alkanes, aromatic hydrocarbons, and aldehydes. The deep learning model includes a VOC classification sub-model and a pain time-series prediction sub-model. The VOC classification sub-model is configured to determine the presence of endometriosis characteristic fingerprints based on digital voltage signals. The pain time-series prediction sub-model is configured to predict the probability of pain occurrence within a future set time window based on pain score data, self-rating emotion data, heart rate interval data sequences, and skin conductivity data.

[0027] Reference Appendix Figure 1 This invention provides a data interaction and processing flow, which mainly involves the transmission and processing of physiological parameter data streams, exhalation detection data streams, control command data streams, and diagnosis and treatment management data streams.

[0028] The physiological signal acquisition unit 122 in the IoT-based treatment monitoring device 120 periodically acquires the subject's heart rate interval signal and skin conductivity signal. The microprocessor inside the IoT-based treatment monitoring device 120 performs analog-to-digital conversion and filtering on the acquired analog signals to generate a digitized physiological parameter sequence. The IoT-based treatment monitoring device 120 transmits the physiological parameter sequence to the mobile terminal 110 in real time via Bluetooth communication protocol.

[0029] The human-computer interaction interface of mobile terminal 110 receives user-triggered pain score input, pain location selection, and emotional state text input. The processor of mobile terminal 110 timestamps the received pain score value, pain location coordinate code, and emotional state text data with the physiological parameter sequence received from IoT treatment monitoring device 120. Mobile terminal 110 encapsulates the aligned multi-source heterogeneous data into an encrypted first upload data packet and sends the first upload data packet to data processing server 140 via cellular mobile network or wireless local area network.

[0030] When performing a detection task, the exhalation detection device 130's internal microelectromechanical system (MEMS) sensor array adsorbs the sample gas and outputs a multi-channel resistance response value. The processor of the exhalation detection device 130 extracts the steady-state amplitude characteristics and dynamic response time characteristics of the resistance response value to generate an exhalation feature vector. The exhalation detection device 130 encapsulates the exhalation feature vector into a second upload data packet and sends it directly to the data processing server 140.

[0031] Data processing server 140 receives and parses the first and second uploaded data packets. Data processing server 140 inputs the parsed data into a pre-set deep learning model. The deep learning model outputs a pain risk probability value for a future set of time periods and a recommended set of transcutaneous electrical nerve stimulation (TENS) parameters. The TENS parameter set includes pulse frequency, pulse width, and stimulation intensity levels.

[0032] The data processing server 140 determines whether a preset alarm threshold condition is met based on the pain risk probability value. When the condition is met, the data processing server 140 generates a downlink control command packet containing alarm flags and a set of transcutaneous electrical nerve stimulation parameters. The data processing server 140 then sends the downlink control command packet to the mobile terminal 110.

[0033] Mobile terminal 110 receives and parses downlink control command packets. In response to the detection of an alarm flag, mobile terminal 110 displays a prompt window on its screen. In response to user confirmation, mobile terminal 110 sends hardware driver commands to IoT treatment monitoring device 120 via Bluetooth.

[0034] The transcutaneous electrical nerve stimulation (TENS) unit 121 of the IoT-based treatment monitoring device 120 receives hardware driving commands. Based on the frequency, pulse width, and intensity parameters contained in the hardware driving commands, the TENS unit 121 adjusts the output voltage waveform of the pulse generation circuit to apply electrical stimulation intervention to the subject. During the intervention, the physiological signal acquisition unit 122 continuously collects physiological data and transmits it back to the mobile terminal 110, forming a closed-loop feedback.

[0035] The data processing server 140 stores all received historical data, model output results, and intervention records into the database. The data processing server 140 generates statistical report data streams according to a preset cycle and sends them to the management terminal 150. The management terminal 150 parses the statistical report data streams and displays trend charts on the monitor for medical staff to view.

[0036] See attached document Figure 2 The present invention provides a breath detection device 130, which includes: a gas flow path assembly 131, a microelectromechanical system sensor array assembly 132, a signal acquisition circuit assembly 133, and a temperature control drive assembly 134.

[0037] The gas flow path assembly 131 includes an air inlet, a dehumidifying filter column, a miniature vacuum pump, a detection chamber, and a one-way exhaust valve. The air inlet is configured to mechanically connect to a disposable mouthpiece. The dehumidifying filter column, located between the air inlet and the miniature vacuum pump, is filled with silica gel desiccant and activated carbon particles to filter out water vapor and large dust particles from exhaled air. The inlet of the miniature vacuum pump is connected to the dehumidifying filter column via a PTFE conduit, and the outlet of the miniature vacuum pump is connected to the inlet of the detection chamber. The detection chamber is made of stainless steel or PTFE and has a fixed internal volume. The outlet of the detection chamber is connected to the one-way exhaust valve. The miniature vacuum pump is configured to generate negative pressure when energized, introducing filtered gas into the detection chamber at a constant flow rate, and discharging residual gas through the one-way exhaust valve after detection.

[0038] The microelectromechanical system (MEMS) sensor array assembly 132 is disposed within the internal space of the detection chamber and soldered onto a printed circuit board. The MEMS sensor array assembly 132 contains multiple independently operating metal-oxide-semiconductor (MOS) gas sensors. Each MOS gas sensor includes a silicon substrate, an insulating dielectric layer, a heating resistor layer, and a gas-sensitive material layer. The gas-sensitive material layer is coated on the insulating dielectric layer and is made of tin dioxide or tungsten trioxide nanoparticles. Different sensors in the MEMS sensor array assembly 132 have different proportions of noble metal catalysts, including platinum, palladium, or gold, doped into the gas-sensitive material layer, thereby enabling different sensors to exhibit specific resistance change responses to alkanes, aromatic hydrocarbons, aldehydes, and ketones volatile organic compounds in exhaled breath.

[0039] The signal acquisition circuit assembly 133 includes a reference voltage source, a load resistor array, a multiplexer analog switch, a signal amplifier, and an analog-to-digital converter. Each metal-oxide-semiconductor gas sensor's gas-sensitive material layer acts as a variable resistor, connected in series with a fixed-value resistor in the load resistor array to form a resistive voltage divider circuit. The reference voltage source provides a stable DC bias voltage to the resistive voltage divider circuit. When a target volatile organic compound undergoes a redox reaction with the gas-sensitive material layer, the resistance value of the gas-sensitive material layer changes, causing a shift in the output voltage of the intermediate node of the resistive voltage divider circuit. The multiplexer analog switch is configured to receive a selection command from the microprocessor, sequentially connecting the output voltage of each sensor's intermediate node to the input of the signal amplifier. The signal amplifier is configured to perform impedance transformation and amplitude amplification on the input voltage signal. The analog-to-digital converter is connected to the output of the signal amplifier and configured to convert the analog voltage signal into a digital quantized sequence and transmit it to the main control chip.

[0040] The temperature control drive assembly 134 includes a pulse width modulation generator and a power drive transistor. The output of the power drive transistor is connected to the heating resistor layer of each metal-oxide-semiconductor gas sensor. The pulse width modulation generator is configured to output a square wave signal with a variable duty cycle to the control electrode of the power drive transistor. By adjusting the duty cycle of the square wave signal, the average current flowing through the heating resistor layer is controlled, thereby maintaining the operating temperature of the gas-sensitive material layer within a preset sensitive temperature range. The signal acquisition circuit assembly 133 is also equipped with an ambient temperature sensor, which is installed near the detection chamber to monitor the ambient temperature and provide feedback signals to the temperature control drive assembly 134 to achieve closed-loop temperature control.

[0041] See attached document Figure 2 The present invention provides a hardware circuit design scheme, wherein the transcutaneous electrical nerve stimulation unit 121 includes: a main control microcontroller circuit 310, a DC boost converter circuit 320, a bidirectional waveform drive circuit 330, and a safety feedback monitoring circuit 340.

[0042] The main control microcontroller circuit 310 includes a microprocessor chip and its peripheral crystal oscillator circuit and reset circuit. The microprocessor chip is equipped with a universal asynchronous receiver / transmitter interface (UART), multiple pulse width modulation signal output pins, and high-precision analog-to-digital converter input pins. The UART is connected to a Bluetooth communication module and configured to receive parameter configuration commands from the mobile terminal 110. The microprocessor chip calculates the target duty cycle value and the target frequency value based on the parameter configuration commands.

[0043] The DC-DC boost converter circuit 320 employs a boost topology, with its input connected to a lithium polymer battery pack and its output connected to a high-voltage DC bus. The DC-DC boost converter circuit 320 includes an energy storage inductor, a power switching transistor, a fast recovery diode, and a high-voltage energy storage capacitor. One end of the energy storage inductor is connected to the positive terminal of the battery, and the other end is connected to the drain of the power switching transistor. The gate of the power switching transistor is connected to the pulse-width modulation signal output pin of the microprocessor chip. The anode of the fast recovery diode is connected to the drain of the power switching transistor, and the cathode is connected to the positive terminal of the high-voltage energy storage capacitor.

[0044] When the microprocessor chip outputs a high-level pulse, the power switching transistor turns on, and current flows through the energy storage inductor, converting electrical energy into magnetic field energy for storage. When the microprocessor chip outputs a low-level pulse, the power switching transistor turns off, and the energy storage inductor generates a reverse induced electromotive force (EMF). This reverse EMF, superimposed on the battery voltage, charges the high-voltage energy storage capacitor through a fast recovery diode. The microprocessor chip adjusts the duty cycle of the output pulse to control the voltage amplitude of the high-voltage DC bus, thereby regulating the intensity of the electrical stimulation.

[0045] The bidirectional waveform drive circuit 330 adopts an H-bridge circuit structure and is connected between the high-voltage DC bus and ground. The bidirectional waveform drive circuit 330 includes a first electronic switch, a second electronic switch, a third electronic switch, and a fourth electronic switch. The first and fourth electronic switches form a first diagonal branch, and the second and third electronic switches form a second diagonal branch. The output electrode pair is connected between the midpoints of the two bridge arms. The microprocessor chip alternately turns on the first and second diagonal branches through logic drive signals, generating alternating positive and negative bidirectional asymmetrical square wave pulses at both ends of the output electrode pair. The repetition frequency of the pulses is determined by the switching frequency of the logic drive signals, and the pulse width is determined by the conduction duration of the logic drive signals.

[0046] A safety feedback monitoring circuit 340 is connected in series in the loop of the bidirectional waveform drive circuit 330. The safety feedback monitoring circuit 340 includes a precision sampling resistor and a differential operational amplifier. The precision sampling resistor converts the stimulation current flowing through the human body into a voltage signal. The differential operational amplifier amplifies the voltage signal and transmits the amplified voltage signal to the analog-to-digital converter (ADC) input pin of the microprocessor chip. The microprocessor chip reads the value of the ADC input pin in real time and compares it with a preset safety threshold. When the read value exceeds the safety threshold, the microprocessor chip is configured to immediately stop outputting the pulse width modulation signal and cut off the drive signal of the bidirectional waveform drive circuit 330 to prevent overcurrent injury.

[0047] See attached document Figure 2 The present invention provides a multimodal physiological signal acquisition module, wherein the physiological signal acquisition unit 122 is integrated on the main circuit board of the Internet of Things treatment monitoring device 120, specifically including: heart rate variability acquisition submodule 350 and skin conductivity acquisition submodule 360.

[0048] The heart rate variability acquisition submodule 350 is constructed based on photoplethysmography (PPG). The PPG submodule 350 includes a photoelectric emitter, a photoelectric receiver, a transimpedance amplifier circuit, and a bandpass filter circuit. The photoelectric emitter is a green light-emitting diode with an emission wavelength between 500 and 600 nanometers, configured to illuminate a constant intensity light beam onto the subject's skin surface. The photoelectric receiver is positioned at a predetermined distance from the photoelectric emitter and is configured to receive light reflected from subcutaneous capillaries. When the heart contracts and relaxes, causing periodic changes in vascular volume, the intensity of the light received by the photoelectric receiver fluctuates accordingly, thereby generating a weak photocurrent signal.

[0049] The input of the transimpedance amplifier circuit is connected to the output pin of the photodetector. The transimpedance amplifier circuit is configured to convert the weak photocurrent signal into a voltage signal. A bandpass filter circuit is connected to the output of the transimpedance amplifier circuit. The bandpass filter circuit is configured with a low-frequency cutoff frequency and a high-frequency cutoff frequency to filter out DC baseline drift signals and high-frequency power frequency interference signals, outputting an AC pulse wave analog signal. The AC pulse wave analog signal is transmitted to the analog-to-digital converter interface of the main control microcontroller circuit 310 for digital sampling.

[0050] The processor in the main control microcontroller circuit 310 executes a peak detection algorithm on the digitized pulse wave signal. The processor identifies each systolic peak in the pulse wave signal and records the time interval between two adjacent systolic peaks. The processor defines the time interval as successive heartbeat interval data and uses the continuous successive heartbeat interval data sequence as the basis for heart rate variability analysis.

[0051] The skin conductivity acquisition submodule 360 ​​includes a skin contact electrode pair, a Wheatstone bridge circuit, and an instrumentation amplifier circuit. The skin contact electrode pair consists of two mutually insulated silver-plated electrode pads configured to fit snugly against the subject's skin surface. One arm of the Wheatstone bridge circuit is formed by the equivalent resistance between two points on the subject's skin, while the other three arms are high-precision fixed resistors. The Wheatstone bridge circuit is equipped with a constant DC excitation voltage source.

[0052] When the subject's sympathetic nervous system activity causes changes in sweat gland secretion, the ion concentration on the skin surface changes, leading to a change in the equivalent resistance. This, in turn, causes a differential voltage signal to be generated at the output of the Wheatstone bridge circuit. The two inputs of the instrumentation amplifier circuit are connected to the two output nodes of the Wheatstone bridge circuit, respectively. The instrumentation amplifier circuit is configured to amplify the differential voltage signal with a high common-mode rejection ratio, outputting an analog voltage signal reflecting changes in skin conductivity. The analog voltage signal is transmitted to another analog-to-digital converter interface of the main control microcontroller circuit 310. The processor maps the acquired voltage value to a skin conductivity value to characterize the subject's instantaneous pressure level and the degree of pain stress response.

[0053] See attached document Figure 3 The present invention provides a data processing method including: a data preprocessing step, a feature parameter calculation step, a feature vector construction step, and a classification decision step.

[0054] In the data preprocessing step, the processor in the data processing server 140 receives raw voltage time-series data from the breath detection device 130. The raw voltage time-series data contains sampling point values ​​from multiple sensor channels in the microelectromechanical system (MEMS) sensor array. The processor performs a moving average filtering operation on the data from each sensor channel to filter out high-frequency noise interference. The processor selects sampling point values ​​within a preset time period after the sampling start time, calculates the arithmetic mean of the sampling point values, and defines the arithmetic mean as the baseline voltage value. The processor subtracts the baseline voltage value from each sampling point value in the raw voltage time-series data to obtain the corrected response curve data.

[0055] In the characteristic parameter calculation step, the processor calculates multiple quantization metrics for the response curve data of each sensor channel. The processor iterates through the response curve data, searching for and obtaining the maximum voltage amplitude. The processor calculates the difference between the maximum voltage amplitude and the baseline voltage value, and then calculates the ratio of this difference to the baseline voltage value to obtain the relative response intensity metric. The processor calculates the sum of all sampled values ​​in the response curve data, and then multiplies this sum by the sampling time interval to obtain the response area metric. The processor records the time it takes for the response signal to rise from zero to 90% of the maximum voltage amplitude, obtaining the response time metric. The processor records the time it takes for the response signal to fall from the maximum voltage amplitude to 10% of the maximum voltage amplitude, obtaining the recovery time metric.

[0056] In the feature vector construction step, the processor arranges and combines the relative response intensity, response area, response time, and recovery time indices from all sensor channels in the microelectromechanical system (MEMS) sensor array in a preset order to generate a multidimensional feature vector. The processor then performs a standardization operation on each element of the multidimensional feature vector. The standardization operation includes: calculating the difference between each element and the mean of the corresponding training set samples; calculating the ratio of this difference to the standard deviation of the corresponding training set samples; and obtaining the input feature vector.

[0057] In the classification decision step, the processor inputs the input feature vector into a pre-trained support vector machine (SVM) classification model. The SVM model contains a set of weight coefficient vectors and a bias term parameter. The processor calculates the dot product of the input feature vector and the weight coefficient vector, adds the dot product to the bias term parameter, and obtains the decision function value. The processor then inputs the decision function value into a sign function. When the sign function output is positive, the processor determines the detection result as positive for endometriosis; when the sign function output is negative, the processor determines the detection result as negative for endometriosis. The processor generates diagnostic report data from the decision results.

[0058] See attached document Figure 3 This invention provides a method for fusing multi-source heterogeneous data, which is executed by a data fusion module 410 in a data processing server 140. The data fusion module 410 includes a time alignment unit 411, a data standardization unit 412, a semantic encoding unit 413, and a feature concatenation unit 414.

[0059] The time alignment unit 411 is configured to receive high-frequency physiological time-series data from the IoT treatment monitoring device 120 and low-frequency behavioral recording data from the mobile terminal 110. The high-frequency physiological time-series data includes heart rate variability data and skin conductivity data sampled at a first frequency. The low-frequency behavioral recording data includes pain score data, emotion tag data, and menstrual cycle phase data sampled at a second frequency. Since the first frequency is much higher than the second frequency, the time alignment unit 411 performs downsampling aggregation. The time alignment unit 411 sets a fixed-length time window. Within each time window, the time alignment unit 411 calculates statistical characteristic values ​​of the high-frequency physiological time-series data. The statistical characteristic values ​​include the arithmetic mean, standard deviation, and maximum value of the data points within the time window. The time alignment unit 411 uses the calculated statistical characteristic values ​​as the physiological feature vector corresponding to that time window, thereby adjusting the temporal resolution of the high-frequency data to match the length of the time window. For low-frequency behavior recording data, the time alignment unit 411 uses zero-order hold interpolation to fill the discrete recording point values ​​into the corresponding time window, so that all data dimensions have a uniform step size on the time axis.

[0060] The data standardization unit 412 is configured to perform dimensional unification processing on the time-aligned numerical data. The numerical data includes voltage and resistance values ​​in the physiological feature vector, as well as pain score values. For each dimension of numerical data, the data standardization unit 412 iterates through the historical dataset to obtain the global maximum and global minimum values ​​for that dimension. The data standardization unit 412 calculates the difference between the current value and the global minimum value. The data standardization unit 412 then calculates the difference between the global maximum and global minimum values. The data standardization unit 412 uses the ratio of these two differences as the normalized feature value, thereby mapping data with different physical dimensions to a closed interval between zero and one, eliminating the impact of numerical magnitude differences on subsequent algorithm weights.

[0061] The semantic encoding unit 413 is configured to process non-numerical data in low-frequency behavior record data. Non-numerical data mainly consists of text labels describing emotional states. The semantic encoding unit 413 has a pre-set emotion dictionary mapping table containing multiple predefined emotion categories, each corresponding to a unique one-hot encoded vector. The dimension of the one-hot encoded vector is equal to the total number of emotion categories, and only one element in the vector has a value of one, while the rest have values ​​of zero. When a specific emotion label is received from the user, the semantic encoding unit 413 performs a match search in the emotion dictionary mapping table and outputs the corresponding one-hot encoded vector. If the input label is not in the mapping table, a vector of all zeros is output.

[0062] The feature splicing unit 414 is configured to receive normalized feature values ​​from the data standardization unit 412, one-hot encoded vectors from the semantic encoding unit 413, and expiratory feature vectors from the breath detection device 130. The feature splicing unit 414 concatenates all feature data within the same time window according to a preset dimensional order, constructing a high-dimensional comprehensive state input vector. This comprehensive state input vector includes the subject's physiological state information, subjective feelings, and biochemical metabolic information at a specific moment. The feature splicing unit 414 then transmits the comprehensive state input vector to the subsequent prediction model for analysis.

[0063] See attached document Figure 3 The present invention provides a pain pattern recognition and prediction module 420, which resides in the memory of a data processing server 140 and is executed by a processor. The pain pattern recognition and prediction module 420 includes: a sequence input mapping unit 421, a long short-term memory network unit 422, a temporal attention mechanism unit 423, and a prediction output unit 424.

[0064] The sequence input mapping unit 421 is configured to receive a sequence of integrated state input vectors from the multi-source heterogeneous data fusion layer. The sequence of integrated state input vectors consists of integrated state input vectors from multiple consecutive moments within a defined time window. The sequence input mapping unit 421 constructs a three-dimensional tensor, where the three dimensions correspond to the sample batch size, time step, and feature dimension, respectively. The feature dimension corresponds to the total number of elements in each integrated state input vector. The sequence input mapping unit 421 transmits the three-dimensional tensor to the Long Short-Term Memory network unit 422.

[0065] Long Short-Term Memory (LSTM) network unit 422 comprises a multi-layered stacked LTM neuron network. Each LTM neuron is configured with a forget gate, an input gate, and an output gate. At each time step, the LTM neuron receives the current input vector and the hidden state vector from the previous time step. The forget gate calculates a forgetting coefficient based on the current input and the hidden state from the previous time step, which controls the proportion of cell state information retained from the previous time step. The input gate calculates an update coefficient, which controls the proportion of current input information written into the cell state. The output gate calculates the hidden state vector at the current time step based on the updated cell state. LTM network unit 422 processes the sequence of integrated state input vectors sequentially in chronological order and outputs the hidden state vector sequence corresponding to each time step.

[0066] The temporal attention mechanism unit 423 is connected to the output of the long short-term memory network unit 422. The temporal attention mechanism unit 423 is configured to calculate the weight score of each hidden state vector in the hidden state vector sequence. The temporal attention mechanism unit 423 passes each hidden state vector through a nonlinear transformation layer to obtain the corresponding energy value. The temporal attention mechanism unit 423 calculates the exponential function value of the energy values ​​for all time steps and calculates the ratio of each exponential function value to the sum of all exponential function values ​​to obtain the normalized attention weight coefficient. The temporal attention mechanism unit 423 multiplies the hidden state vector for each time step with the corresponding attention weight coefficient and adds all the product results vector-wise to generate a global context feature vector. The global context feature vector represents the key time point features in the input sequence that contribute the most to future pain prediction.

[0067] The prediction output unit 424 is a fully connected neural network layer configured to receive a global context feature vector. The prediction output unit 424 multiplies the global context feature vector by the weight matrix and adds a bias vector to obtain a linear output value. The prediction output unit 424 then inputs this linear output value into a sigmoid activation function, compressing the output range to a closed interval between zero and one. The value output by the prediction output unit 424 is defined as the probability of pain occurring within a predetermined prediction window.

[0068] During the model training phase, the data processing server 140 uses a historical dataset to optimize the parameters of the pain pattern recognition and prediction module 420. The historical dataset contains a sequence of historical integrated state input vectors as input and actual pain occurrence records as labels. The processor calculates the binary cross-entropy loss value between the pain occurrence probability value and the actual pain occurrence records. The processor uses the backpropagation algorithm to calculate the gradient of the loss value with respect to all weight parameters in the long short-term memory network unit 422, the temporal attention mechanism unit 423, and the prediction output unit 424. The processor uses an adaptive moment estimation optimization algorithm to update the network parameters based on the gradient until the loss value converges to a preset error range.

[0069] See attached document Figure 4 The present invention provides an application logic for a mobile terminal 110. The mobile terminal 110 runs an application program, which includes: a user input interface module 510, a communication protocol management module 520, a data encapsulation and encryption module 530, a visualization rendering module 540, and a device control instruction module 550.

[0070] User input interface module 510 is configured to generate an interactive graphical interface on the touchscreen display of mobile terminal 110. The interactive graphical interface includes a visual analog scale (VAS) slider area, an anatomical diagram area, and an emotion label selection area. When a user's finger is detected sliding within the VAS slider area and then stops, user input interface module 510 reads the horizontal coordinate value of the stopped position and linearly maps the horizontal coordinate value to an integer between zero and ten, using it as pain score data. When a user's click is detected in the anatomical diagram area, user input interface module 510 identifies the predefined area range to which the click coordinates belong and generates the corresponding location code as pain location data. When a user's click is detected in the emotion label selection area, user input interface module 510 obtains the text identifier of the selected label, using it as emotion self-assessment data.

[0071] The communication protocol management module 520 is configured to manage the first wireless communication channel and the cellular mobile network channel. In the background, the communication protocol management module 520 periodically scans for surrounding Bluetooth broadcast signals and matches them with the unique device identification code of the IoT treatment monitoring device 120. After establishing a connection, the communication protocol management module 520 subscribes to the service characteristic values ​​of the IoT treatment monitoring device 120 and receives a byte stream containing heart rate interval data sequences and skin conductivity data. The communication protocol management module 520 is also configured to maintain a long-term connection session with the data processing server 140, monitor the network connection status, and execute automatic reconnection logic when the connection is lost.

[0072] The data encapsulation and encryption module 530 is configured to perform data time synchronization and packaging operations. The data encapsulation and encryption module 530 obtains the current system timestamp. It associates pain score data, pain location data, self-evaluation of emotions data, buffered heart rate interval data sequences, and skin conductivity data to generate a unified JSON data structure. The data encapsulation and encryption module 530 calls an Advanced Encryption Standard (AES) algorithm, uses a preset key to encrypt the JSON data structure, generates a ciphertext data packet, and sends it to the data processing server 140 via the communication protocol management module 520.

[0073] The visualization rendering module 540 is configured to receive and parse the downlink data stream from the data processing server 140. The downlink data stream contains pain risk prediction results and long-term monitoring report data. Based on the probability values ​​in the pain risk prediction results, the visualization rendering module 540 dynamically adjusts the pointer position and background color of the dashboard controls in the main interface. The visualization rendering module 540 parses the time series points in the long-term monitoring report data and calls the drawing engine to draw line graphs and bar charts on the screen, displaying the historical trends of pain scores and physiological parameters.

[0074] The device control command module 550 is configured to process intervention control logic. When the mobile terminal 110 receives a downlink control command packet containing an alarm flag and a set of transcutaneous electrical nerve stimulation (TENS) parameters, the device control command module 550 triggers a pop-up event, displaying a treatment suggestion prompt and a confirmation button on the interface. In response to detecting a user click on the confirmation button, the device control command module 550 converts the TENS parameter set into hardware control commands conforming to the Bluetooth protocol. The device control command module 550 writes the hardware control commands to the control characteristic values ​​of the IoT treatment monitoring device 120 via the communication protocol management module 520, thereby initiating the physical therapy output.

[0075] See attached document Figure 5 This invention provides a non-invasive assisted diagnostic method, which is executed collaboratively by a breath detection device 130, a mobile terminal 110, and a data processing server 140. The method includes: During system initialization, the application interface of mobile terminal 110 receives the user's start command. Mobile terminal 110 sends a preheating control signal to breath detection device 130 via Bluetooth communication channel. The main control chip inside breath detection device 130 receives the preheating control signal and controls the temperature control drive component 134 to heat the microelectromechanical system sensor array component 132. When the operating temperature of the microelectromechanical system sensor array component 132 reaches the preset sensitive temperature range and remains stable, breath detection device 130 returns a ready status signal to mobile terminal 110.

[0076] During the sample collection phase, the user exhales through a disposable mouthpiece into the gas flow path assembly 131 of the breath detection device 130. A miniature vacuum pump in the gas flow path assembly 131 activates, introducing exhaled gas into the detection chamber at a constant flow rate. The microelectromechanical system (MEMS) sensor array assembly 132 is exposed to the exhaled gas environment, and its gas-sensitive material layer undergoes a redox reaction with volatile organic compounds in the gas, causing a change in resistance. The signal acquisition circuit assembly 133 monitors the change in resistance in real time, converts it into a voltage signal, and performs analog-to-digital conversion to obtain raw voltage time-series data.

[0077] During the data transmission phase, the breath test device 130 packages the raw voltage time series data and transmits it to the mobile terminal 110 via Bluetooth communication. The data encapsulation and encryption module 530 of the mobile terminal 110 receives the raw voltage time series data, appends the current timestamp information and user identification information, and constructs a diagnostic request message. The data encapsulation and encryption module 530 encrypts the diagnostic request message and sends the encrypted message to the data processing server 140 via the cellular mobile network.

[0078] During the feature extraction phase, the data processing server 140 receives and decrypts the diagnostic request message. The processor in the data processing server 140 performs preprocessing on the raw voltage time series data, calculates the baseline voltage value, and performs baseline correction. For each sensor channel's data, the processor calculates the relative response intensity index, response area index, response time index, and recovery time index. The processor combines these indices in a preset order and performs standardization processing to construct the input feature vector.

[0079] During the diagnostic decision phase, the processor invokes a pre-trained support vector machine (SVM) classification model. The processor inputs the feature vector into the SVM model. The SVM model calculates the dot product of the input vector and the weight vector, and performs calculations in conjunction with the bias term, outputting a decision function value. Based on the sign of the decision function value, the processor determines the classification result of the detected sample as either endometriosis positive or endometriosis negative.

[0080] During the results feedback phase, the data processing server 140 generates diagnostic report data containing classification results and confidence scores. The data processing server 140 then sends the diagnostic report data to the mobile terminal 110. The visualization rendering module 540 of the mobile terminal 110 parses the diagnostic report data and presents the diagnostic conclusion on the display screen. When the diagnostic conclusion is positive, the mobile terminal 110 automatically triggers the subsequent pain management module, prompting the user to enter the pain monitoring and intervention process.

[0081] See attached document Figure 6 This invention provides a closed-loop intervention method, which is jointly executed by a closed-loop control module 560 in a data processing server 140, a mobile terminal 110, and an IoT treatment monitoring device 120. The closed-loop control module 560 includes a decision logic unit 561 and a dynamic feedback adjustment unit 562.

[0082] The decision logic unit 561 is configured to receive the probability value of future pain occurrence from the pain pattern recognition and prediction module 420. The decision logic unit 561 reads a preset intervention threshold parameter from the memory. The decision logic unit 561 compares the probability value of future pain occurrence with the intervention threshold parameter. When the probability value of future pain occurrence is greater than or equal to the intervention threshold parameter, the decision logic unit 561 determines that a preventative intervention process is triggered.

[0083] The decision logic unit 561 retrieves the corresponding initial parameter set for transcutaneous electrical nerve stimulation from the parameter database based on the subject's historical data. The initial parameter set includes pulse frequency settings, pulse width settings, and current intensity level settings. The decision logic unit 561 encapsulates the initial parameter set and the start command into a downlink control message and sends it to the mobile terminal 110 via the network interface.

[0084] After receiving the downlink control message, the mobile terminal 110 generates a countdown confirmation window on the display interface. If no cancellation operation is detected before the preset countdown ends, or if a confirmation operation is detected, the mobile terminal 110 sends a hardware drive sequence to the IoT treatment monitoring device 120 via Bluetooth. The main control circuit of the IoT treatment monitoring device 120 parses the hardware drive sequence and controls the high-voltage drive circuit to output electrical pulse waveforms according to the initial parameter set.

[0085] During the electrical stimulation intervention, the dynamic feedback adjustment unit 562 executes real-time monitoring and parameter correction logic. The IoT treatment monitoring device 120 continuously collects the subject's skin conductivity data at a first sampling frequency and transmits it back to the mobile terminal 110 and the data processing server 140 in real time. The dynamic feedback adjustment unit 562 receives the skin conductivity data and calculates the slope of the skin conductivity change over time.

[0086] The dynamic feedback adjustment unit 562 determines the sign and absolute value of the slope of change. When the slope is negative and the absolute value is greater than a preset relaxation threshold, it indicates that the subject is in a stress-relief state, and the dynamic feedback adjustment unit 562 generates a command to maintain the current parameters. When the slope is positive and the duration exceeds a preset time window, it indicates that the subject's pain or stress level is rising, and the dynamic feedback adjustment unit 562 generates a parameter adjustment command. The parameter adjustment command is configured to increase the pulse frequency setting value by a preset step size or switch the waveform modulation mode, and send the new parameter set to the mobile terminal 110 to update the output status of the IoT treatment monitoring device 120.

[0087] The internal safety monitoring circuit of the IoT-based treatment monitoring device 120 synchronously monitors the impedance value of the output circuit. When it detects that an electrode has detached, resulting in an infinite impedance value, or that a short circuit has caused the current value to exceed the hardware safety limit, the IoT-based treatment monitoring device 120 immediately cuts off the output and sends an abnormal interruption signal to the mobile terminal 110. After receiving the abnormal interruption signal, the mobile terminal 110 records the interruption timestamp and uploads it to the data processing server 140. The data processing server 140 marks this intervention event as incomplete.

[0088] See attached document Figure 6 This invention provides a method for predicting and intervening in daily pain, which is executed collaboratively by an Internet of Things (IoT) treatment monitoring device 120, a mobile terminal 110, and a data processing server 140. The method includes: During the continuous data monitoring phase, the IoT-based therapeutic monitoring device 120 is worn on the subject's abdominal area. The heart rate variability acquisition submodule 350 and skin conductivity acquisition submodule 360 ​​within the IoT-based therapeutic monitoring device 120 acquire the subject's photoplethysmography (PPG) wave signal and skin resistance signal at a first preset frequency. The IoT-based therapeutic monitoring device 120 performs preliminary digitization processing on the acquired signals and transmits the processed physiological data packets to the mobile terminal 110 via Bluetooth. Simultaneously, the mobile terminal 110 receives the subject's input emotion tags and pain scores through an application interface at a second preset frequency. The data encapsulation and encryption module 530 of the mobile terminal 110 timestamps the physiological data packets with the subject's input behavioral data, generating an encrypted monitoring data stream, which is then uploaded to the data processing server 140 via a wireless network.

[0089] During the risk assessment and prediction phase, the multi-source heterogeneous data fusion module 410 of the data processing server 140 receives the monitoring data stream. The multi-source heterogeneous data fusion module 410 performs standardization and feature concatenation operations on the monitoring data stream to construct a comprehensive state input vector for the current time window. The data processing server 140 then calls the pain pattern recognition and prediction module 420. The pain pattern recognition and prediction module 420 inputs the comprehensive state input vector into the long short-term memory network unit 422 and the time attention mechanism unit 423 to calculate the probability value of pain occurrence within a preset future time period.

[0090] During the decision-making and early warning phase, the closed-loop control module 560 of the data processing server 140 compares the probability value of pain occurrence with a preset intervention threshold. When the probability value of pain occurrence exceeds the intervention threshold, the closed-loop control module 560 generates an early warning notification and a recommended intervention plan. The intervention plan includes suggested electrical stimulation waveform parameters. The data processing server 140 sends the early warning notification and recommended intervention plan to the mobile terminal 110. The mobile terminal 110 displays a confirmation dialog box on its screen, requesting the subject's authorization to initiate physical therapy.

[0091] During the closed-loop feedback intervention phase, in response to the subject's confirmation operation on the mobile terminal 110, the mobile terminal 110 sends a start command to the IoT treatment monitoring device 120. The transcutaneous electrical nerve stimulation unit 121 of the IoT treatment monitoring device 120 applies electrical pulse stimulation to the subject according to the recommended intervention protocol. During stimulation, the skin conductivity acquisition submodule 360 ​​continuously monitors the changing trend of skin conductivity values. The data processing server 140 receives real-time skin conductivity values ​​and calculates their rate of change over time.

[0092] When the rate of change is negative and its absolute value exceeds a preset standard, the subject's stress level is determined to have decreased, and the closed-loop control module 560 instructs the IoT treatment monitoring device 120 to maintain the current output parameters. When the rate of change is positive, the subject's stress level is determined to have increased or pain not relieved, and the closed-loop control module 560 generates a parameter adjustment command, instructing the IoT treatment monitoring device 120 to gradually increase the pulse frequency or adjust the pulse width until the rate of change turns negative or reaches the safe output limit. After treatment, the mobile terminal 110 records the duration of the intervention and the parameter change trajectory, and uploads it to the data processing server 140 for storage in the historical database.

Claims

1. An AI-based data management system for predicting pain, emotions, and stress in patients with endometriosis, characterized in that: include: Mobile terminals, IoT-based treatment monitoring devices, breath detection devices, data processing servers, and management terminals; The mobile terminal is configured to collect subjective user data and receive physiological parameter data from the IoT treatment monitoring device, and send the subjective user data and the physiological parameter data to the data processing server. The IoT-based treatment monitoring device is configured to collect the physiological parameter data and perform transcutaneous electrical nerve stimulation based on the received adjustment signals. The exhalation detection device is configured to collect the digital voltage signal corresponding to the exhaled gas and send it to the data processing server; The data processing server is configured to run a deep learning model to generate pain risk prediction results and intervention control instructions based on the user's subjective data, the physiological parameter data, and the digital voltage signal, and send the intervention control instructions to the mobile terminal. The mobile terminal is configured to generate the adjustment signal according to the intervention control command and send it to the Internet of Things treatment monitoring device.

2. The AI-based predictive data management system for endometriosis-related pain, emotions, and stress as described in claim 1, is characterized in that... The mobile terminal is equipped with a human-computer interaction interface, and the user's subjective data includes pain score data, pain location data, pain duration data, and emotional self-rating data received through the human-computer interaction interface. The IoT-based treatment monitoring device includes a transcutaneous electrical nerve stimulation unit and a physiological signal acquisition unit. The physiological parameter data includes a heart rate interval data sequence and skin conductivity data acquired by the physiological signal acquisition unit. The exhalation detection device includes a microelectromechanical system (MEMS) sensor array, and the digital voltage signal is converted from the response resistance signal generated by the MEMS sensor array to volatile organic compounds in the exhaled gas.

3. The AI-based predictive data management system for endometriosis-related pain, emotions, and stress as described in claim 2, is characterized in that... The transcutaneous electrical nerve stimulation unit includes a main control microcontroller circuit, a DC boost converter circuit, and a bidirectional waveform drive circuit. The DC-DC boost converter circuit is configured to convert the battery voltage into a high-voltage DC bus voltage. The bidirectional waveform driving circuit adopts an H-bridge circuit structure and is configured to generate bidirectional asymmetrical square wave pulses with alternating positive and negative values ​​at both ends of the output electrode pair according to the driving signal of the main control micro-control circuit. The physiological signal acquisition unit includes a photoelectric transmitter and a photoelectric receiver configured to acquire pulse wave signals, and a Wheatstone bridge circuit configured to detect the change in equivalent resistance between skin contact electrode pairs.

4. The AI-based predictive data management system for endometriosis-related pain, emotions, and stress as described in claim 2, is characterized in that... The breath detection device also includes a gas flow path assembly and a temperature control drive assembly; the gas flow path assembly includes a miniature vacuum pump and a detection chamber, configured to introduce filtered gas into the detection chamber at a preset flow rate; The microelectromechanical system sensor array is disposed in the detection gas chamber and includes multiple metal oxide semiconductor gas sensors. Each metal oxide semiconductor gas sensor has a different gas-sensitive material film layer, and the gas-sensitive material film layer is doped with a noble metal catalyst selected from platinum, palladium and gold. The temperature control drive component is configured to adjust the average current of the heating resistor layer in the metal oxide semiconductor gas sensor.

5. The AI-based predictive data management system for endometriosis-related pain, emotions, and stress as described in claim 2, is characterized in that... The data processing server is configured to perform multi-source heterogeneous data fusion; The processor in the data processing server is configured to perform downsampling and aggregation operations on the heartbeat interval data sequence and the skin conductivity data, and calculate the statistical feature values ​​within a preset time window as physiological feature vectors. The pain score data is filled into the corresponding time window using zero-order hold interpolation. The self-evaluation data of emotion is converted into a one-hot encoded vector using a preset emotion dictionary mapping table. The normalized physiological feature vector, the pain score data and the one-hot encoded vector are then concatenated to construct a comprehensive state input vector.

6. The AI-based predictive data management system for endometriosis-related pain, emotions, and stress as described in claim 5, is characterized in that... The deep learning model includes a pain time-series prediction sub-model, which includes a long short-term memory network unit, a time attention mechanism unit, and a prediction output unit. The long short-term memory network unit is configured to receive the time series of the comprehensive state input vector and output the hidden state vector sequence corresponding to each time step. The temporal attention mechanism unit is configured to calculate the attention weight coefficient of each hidden state vector in the hidden state vector sequence to generate a global context feature vector; The prediction output unit is configured to output the probability value of pain occurrence within a future preset time window based on the global context feature vector.

7. The AI-based predictive data management system for endometriosis-related pain, emotions, and stress as described in claim 1, characterized in that, The deep learning model also includes a sub-model for classifying volatile organic compounds; The data processing server is configured to perform baseline correction on the digital voltage signal, calculate the relative response intensity index, response area index, response time index, and recovery time index, and construct an expiratory feature vector. The volatile organic compound classification sub-model is configured to receive the exhaled feature vector, use a support vector machine classification algorithm to determine whether there is endometriosis feature fingerprint, and generate diagnostic report data.

8. The AI-based predictive data management system for endometriosis pain, emotions, and stress according to claim 1, characterized in that, The intervention control command includes a set of transcutaneous electrical nerve stimulation parameters; The data processing server is configured to determine whether the pain risk prediction result meets the preset alarm threshold condition. When the condition is met, it generates a downlink control command packet containing an alarm flag bit and the set of transcutaneous electrical nerve stimulation parameters and sends it to the mobile terminal. The mobile terminal is configured to, in response to the recognition of the alarm flag, pop up a prompt window on the human-computer interaction interface, and in response to the confirmation operation, send a hardware drive command containing the set of transcutaneous electrical nerve stimulation parameters to the Internet of Things treatment monitoring device.

9. The AI-based predictive data management system for endometriosis-related pain, emotions, and stress as described in claim 2, is characterized in that... The data processing server also includes a dynamic feedback adjustment unit; The dynamic feedback adjustment unit is configured to receive real-time skin conductivity data and calculate the slope of the skin conductivity data over time during the intervention process performed by the transcutaneous electrical nerve stimulation unit. When the slope of the change is positive and the duration exceeds a preset time window, the dynamic feedback adjustment unit generates a parameter adjustment command, which is configured to increase the pulse frequency and pulse width. When the slope of the change is negative and the absolute value is greater than the preset relaxation threshold, the dynamic feedback adjustment unit generates a command to maintain the current parameters.

10. The AI-based predictive data management system for endometriosis pain, emotions, and stress according to claim 1, characterized in that, The management terminal is configured to receive treatment adjustment instructions and synchronize the treatment adjustment instructions to the data processing server; The data processing server is configured to generate long-term monitoring report data, which includes pain score change curves, mood index change curves, stress index change curves, and breath volatile organic compound detection results. The management terminal is configured to present the long-term monitoring report data in the form of trend charts.

Images