Long-term treatment follow-up and program optimization system for pain patients based on big data

Abstract

The present invention discloses a long-term treatment follow-up and scheme optimization system for pain patients based on big data, including: data acquisition and perception module: obtaining the patient's neural signals and emotional signals; neural emotion interaction module: fusing neural signals and emotional signals to obtain neural emotion interaction features, and modeling the interaction features in a time series diagram; analysis and prediction module: predicting the intensity and occurrence time of pain through time series diagram modeling, and predicting the patient's emotional fluctuations through the interaction of neural signals and emotional signals; intervention module: identifying the peak period of emotional fluctuations; feedback module: regularly feeding back the patient's neural signals and emotional signals. The present invention can comprehensively and objectively reflect the patient's true state.

Description

Technical Field

[0001] The present invention relates to the field of medical monitoring technology, and in particular to a long-term treatment follow-up and program optimization system for pain patients based on big data. Background Art

[0002] With the acceleration of the aging process of the global population, chronic pain has become an important health problem that affects the quality of life of patients. According to statistics, about 1 / 5 of the world's population suffers from chronic pain to varying degrees, including back pain, joint pain, migraine and other types, and the incidence of chronic pain is on the rise with age. Chronic pain not only directly affects the patient's physical health, but is also often accompanied by psychological and emotional problems, such as anxiety and depression, which make the treatment of pain more complicated and difficult. Traditional pain management models usually rely on a single treatment plan, such as drug therapy, physical therapy or psychotherapy. Although these treatments can relieve patients' pain to a certain extent, due to the diversity of pain and individual differences among patients, traditional models are often difficult to achieve accurate and personalized treatment, and there is also great uncertainty in the treatment effect.

[0003] In recent years, with the rapid development of big data, artificial intelligence and Internet of Things technologies, the application of intelligent medical systems in chronic disease management has gradually emerged. Big data technology can provide doctors with more comprehensive and real-time patient information by integrating multi-dimensional data sources of patients (such as physiological signals, behavioral data, environmental data, etc.), while artificial intelligence (AI) technology can provide scientific basis for the formulation and adjustment of treatment plans through data analysis and model prediction. However, the existing intelligent medical system still has many limitations in pain management: on the one hand, the assessment of pain usually relies on the patient's self-report, which is difficult to fully and objectively reflect the patient's true condition; on the other hand, the existing treatment plans often focus on the relief of a single symptom, lacking comprehensive intervention and personalized adjustment for multi-dimensional factors. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a long-term treatment follow-up and program optimization system for pain patients based on big data.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is as follows: A long-term treatment follow-up and program optimization system for pain patients based on big data, including: Data acquisition and perception module: Obtaining the patient's neural and emotional signals: Neural-emotional interaction module: It fuses neural signals and emotional signals to obtain neural-emotional interaction features, and models the interaction features using a time series graph. Analysis and prediction module: predict the intensity and occurrence time of pain through time series graph modeling, and predict the patient's emotional fluctuations through the interaction of neural signals and emotional signals; Intervention module: Identify peak periods of emotional fluctuations; Feedback module: Regularly provide feedback on the patient's neural and emotional signals.

[0006] Furthermore: the neural signals include: electroencephalogram, heart rate variability, galvanic skin response, electromyography signals; the emotional signals include: voice emotion, facial expression recognition.

[0007] Further: the neural emotion interaction module includes: Each signal is regarded as a node, and the interaction relationship between signals is modeled as an edge. The nodes of the graph are set to represent signals, and the edges represent the interaction relationship between different signals. Apply graph convolution operation to transmit information to each node. After receiving the signal from the neighboring node, each node performs weighted summation and processes it through a nonlinear activation function. Through multi-layer graph convolutional networks, the interactive information of neural signals and emotional signals is gradually integrated, so that each node can obtain information from multiple sources and learn complex interactive patterns; The signal interaction features at each moment are regarded as nodes of the graph, the time series information is introduced into the graph convolution, and the time series data is modeled in combination with the recurrent neural network.

[0008] Further: the analysis and prediction module includes: Convert the signals into corresponding feature matrices respectively and ensure that the signals are aligned in the time dimension to obtain a multimodal feature set of time series; The multimodal feature set is used as the input of the LSTM model. The output of the LSTM is the predicted value of the pain intensity at each time t, and the joint loss function is used to predict the time of pain occurrence. The LSTM model is used to process the temporal information of emotional signals and combine neural signals with emotional signals to capture the temporal dependency between signals and thus predict future emotional fluctuations. Output the probability distribution of the emotion state categories.

[0009] Further: the intervention module includes: Use rate-of-change detection to identify changes in affective fluctuations; Set thresholds for emotional fluctuations and use trend analysis to determine peak periods.

[0010] Compared with the prior art, the technical progress achieved by the present invention is that: the present invention constructs a multi-level, comprehensive pain management platform, which can fully understand the patient's nervous system activity and emotional fluctuations by collecting the patient's electroencephalogram (EEG), heart rate variability (HRV), galvanic skin response (EDA), electromyography (EMG) neural signal data, and combining the patient's voice emotion analysis, facial expression recognition and other emotional signal data, thereby achieving accurate prediction of pain and personalized intervention.

[0011] Through the deep integration of the Neural Emotional Interaction Model (NEIM) and the Time-series Graph Neural Network (NET-GNN), the present invention can reveal the complex interactive relationship between neural signals and emotional signals, predict the patient's pain occurrence pattern and emotional fluctuation trend, and then provide doctors with personalized treatment suggestions. For example, the system can predict the peak pain that the patient may experience in the next few hours or days, and provide corresponding psychological adjustment and intervention suggestions based on the patient's emotional state fluctuations. Based on the emotional prediction model, the system can also timely push psychological adjustment methods such as meditation and deep breathing when the patient's emotional fluctuations reach a peak, helping patients relieve negative emotions such as anxiety and depression, thereby achieving the effect of pain relief. In addition, by regularly feeding back the patient's neural signals and emotional signals to the hospital, the system enables doctors to follow up and obtain the patient's health status. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0013] In the attached picture: Figure 1 It is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0014] The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0015] like Figure 1 As shown, the present invention discloses a long-term treatment follow-up and program optimization system for pain patients based on big data, including: Data acquisition and perception module: Obtaining the patient's neural and emotional signals: Neural-emotional interaction module: It fuses neural signals and emotional signals to obtain neural-emotional interaction features, and models the interaction features using a time series graph. Analysis and prediction module: predict the intensity and occurrence time of pain through time series graph modeling, and predict the patient's emotional fluctuations through the interaction of neural signals and emotional signals; Intervention module: Identify peak periods of emotional fluctuations; Feedback module: Regularly provide feedback on the patient's neural and emotional signals.

[0016] Specifically, the goal of the data acquisition and perception module is to collect the patient's neural signals and emotional signals. The core task of this module is to obtain the patient's neural and emotional signals through a variety of sensing devices and signal analysis technologies.

[0017] 1.1 Neural signal acquisition Neural signals reflect the patient's brain activity and the state of the autonomic nervous system, and are crucial for the perception of pain, emotional fluctuations, and the evaluation of treatment effects. EEG (electroencephalogram) can record the brain's electrical activity in real time and provide information about the activity of various brain regions. It is especially important for brain regions related to pain perception and emotional fluctuations (such as the prefrontal cortex, cingulate cortex, insula, etc.).

[0018] Brain wave data is collected using wireless portable EEG devices (such as EmotivEpoc, Muse, etc.). By analyzing the spectral features in brain waves (such as alpha waves, beta waves, theta waves, etc.), information related to pain perception (such as slow waves, changes in alpha waves) and emotional fluctuations (such as fluctuations in high-frequency beta waves) can be obtained.

[0019] Heart rate variability (HRV) reflects the activity of the autonomic nervous system, especially the balance between the parasympathetic nervous system (associated with relaxation and emotional stability) and the sympathetic nervous system (associated with stress and anxiety). Decreases in HRV are often associated with increased anxiety, stress, and pain perception.

[0020] Heart rate data is collected in real time through smart watches or electrocardiogram (ECG) monitoring devices, and HRV is evaluated using time domain analysis (such as standard deviation SDNN), frequency domain analysis (such as the ratio of high frequency HF and low frequency LF) and nonlinear analysis (such as Poincaré analysis).

[0021] Galvanic skin response (EDA), galvanic skin response (also known as skin electrode activity, SDA) can reflect the changes in the conductivity of the skin surface, which is often closely related to emotional fluctuations (such as anxiety and stress). Negative emotions such as anxiety, anger or pain will cause the galvanic skin response to increase.

[0022] The skin electrical activity data is collected through portable EDA sensors (such as EmpaticaE4, ShimmerGSR, etc.). EDA signals are often used to monitor patients' mood swings and emotional reactions in real time.

[0023] Electromyographic signals (EMG) reflect muscle activity and tension. Pain is often accompanied by muscle tension, especially in patients with chronic pain. Electromyographic signals (EMG) are an important indicator for assessing pain perception.

[0024] Use EMG sensors (such as Myoware, Delsys, etc.) to monitor the patient's muscle electrical activity. By measuring the changes in muscle potential, the patient's muscle tension can be inferred and then correlated with pain perception and emotional fluctuations.

[0025] 1.2 Emotional signal collection Emotional signals reflect the patient's psychological state and are closely related to pain perception. Real-time collection of emotional signals through a variety of technical means helps to assess the patient's emotional fluctuations and provide data support for subsequent emotional intervention.

[0026] Voice emotion analysis: Voice features (such as pitch, speaking speed, tone, pauses, etc.) can reflect the speaker's emotional state. Rapid speaking speed and higher pitch usually reflect emotional states such as anxiety and tension, while lower pitch may indicate depression and powerlessness.

[0027] The patient's voice features are extracted through intelligent speech analysis tools (such as Google Cloud Speech-to-Text, IBM Watson Speech to Text, etc.), and emotional state recognition is performed using sentiment classification algorithms (such as support vector machine SVM, deep neural network CNN, etc.).

[0028] Facial expression recognition: Facial expressions can directly reflect emotional fluctuations. Expressions such as frowning, wide eyes, and drooping mouth corners are often associated with negative emotional states such as anxiety, anger, and sadness. Facial recognition technology (such as Open Face, FaceReader, etc.) is used to analyze the patient's facial expressions, and the facial action coding system (FACS) is used to quantitatively analyze the expressions. Facial expressions and physiological signals are combined to perform multi-dimensional emotional scoring, helping doctors to fully understand the patient's emotional state.

[0029] By using EEG, HRV, EDA, EMG and other neural signal acquisition technologies in combination with speech analysis, facial expression recognition, and self-rating scales and other emotional signal acquisition methods, patients' neural and emotional data can be collected in all directions and in real time. These data will provide strong support for subsequent neural emotional interaction modeling, pain prediction, and emotional intervention. Through highly integrated sensors and intelligent analysis tools, the system will be able to accurately monitor the health status of patients and provide a basis for the formulation of personalized treatment plans.

[0030] Specifically, the goal of the neural-emotional interaction module is to fuse neural signals with emotional signals to obtain neural-emotional interaction features, and to model these features using time series graphs to further explore the complex interactive relationship between pain and emotional fluctuations.

[0031] The core of this module is how to effectively fuse signals from different sensors (EEG, EDA, facial expressions, voice, etc.) through advanced algorithms and modeling techniques, and achieve deep analysis and prediction through time series graph neural networks (NET-GNN). Specifically, this process includes several key steps such as feature fusion, interaction layer modeling, and time series modeling.

[0032] Step 1: Initial fusion of neural signals and emotional signals 1.1 Multi-dimensional feature concatenation of neural signals and emotional signals First, this embodiment needs to integrate the neural signals and emotional signals collected from different sources into a unified feature space to facilitate subsequent interaction modeling. This process is called feature-level fusion, and its goal is to convert EEG signals, galvanic skin response (EDA), facial expression analysis results, and speech emotion analysis results into a multidimensional feature vector.

[0033] EEG signal: First, the EEG signal is subjected to time-frequency analysis (such as short-time Fourier transform (STFT) or wavelet transform (WT)) to extract frequency domain features, such as Wave, Wave, Waves, etc., use multi-channel EEG to record the electrical activities of different brain regions.

[0034] Galvanic skin response (EDA): Galvanic skin response can reflect emotional fluctuations, so this embodiment extracts skin conductance changes (changes in galvanic skin response) from EDA signals and represents them through dynamic change features such as response amplitude and response time window.

[0035] Facial expression analysis: Use computer vision methods, such as Open Face and other libraries, to extract facial feature points, identify emotional states, and convert them into corresponding emotional labels (e.g., anger, happiness, sadness, etc.) through the Facial Action Coding System (FACS).

[0036] Speech emotion analysis: By extracting the acoustic features of the audio signal (such as pitch, speaking rate, formant frequency, etc.), the emotional components in the speech are analyzed using acoustic models (such as Mel spectrogram).

[0037] 1.2 Feature stitching In this step, the EEG signal, EDA signal, facial expression features and speech emotion features are spliced ​​after time alignment to generate a high-dimensional feature vector .

[0038] Step 2: Neural Emotion Interaction Feature Modeling 2.1 Interaction layer modeling In order to capture the complex interactive relationship between neural signals and emotional signals, this embodiment adopts a graph convolutional network (GCN) or a graph neural network (GNN) to embed the interactive features of neural signals and emotional signals into a graph structure by constructing a relationship graph between the signals.

[0039] Constructing a graph structure: each signal (EEG, EDA, facial expression, speech, etc.) is considered as a node, and the interaction between signals is modeled as edges. For example, there may be some nonlinear relationship between EDA signals and EEG signals. Emotional fluctuations may affect the pattern of EEG activity, and vice versa.

[0040] Assume that the nodes of the graph represent signals, and the edges represent the interactions between different signals:

[0041] in, is a node collection (such as EEG nodes, EDA nodes, etc.), is the set of edges (i.e., interactions between different signals).

[0042] Graph convolution layer: Apply graph convolution operation to transfer information between nodes. After receiving the signal from neighboring nodes, each node performs weighted summation and processes it through a nonlinear activation function (such as ReLU).

[0043]

[0044] in, is the normalized adjacency matrix, is the output of the graph convolutional network at layer l, is the learned weight matrix, is the activation function (such as ReLU).

[0045] Learning interactive features: Through a multi-layer graph convolutional network, the interactive information of neural signals and emotional signals is gradually integrated, so that each node can obtain information from multiple sources and learn complex interactive patterns.

[0046] 2.2 Nonlinear Modeling of Interaction Information In order to further enhance the expressiveness of the model, this embodiment introduces nonlinear modeling techniques, such as the self-attention mechanism, so that the interactive influence of each signal can be dynamically weighted, thereby better capturing the nonlinear relationship between signals.

[0047]

[0048] Among them, Q, K, V are query, key and value matrices respectively. is a normalization constant used to prevent values ​​from being too large.

[0049] By adding the self-attention mechanism, the model can more flexibly learn the contribution of each signal to the overall interaction features.

[0050] Step 3: Timing diagram modeling 3.1 Introducing timing information: NET-GNN model On the basis of neural emotion interaction modeling, the dimension of time series is added, and the time series graph neural network (NET-GNN) is used to model the dynamic characteristics of signals changing over time. NET-GNN combines graph neural networks and time series modeling methods, and can effectively capture the evolution of different signals in the time dimension.

[0051] Time series graph modeling: The signal interaction features at each moment are regarded as nodes of the graph, the time series information is introduced into the graph convolution, and the time series data is modeled in combination with a recurrent neural network (such as LSTM or GRU).

[0052]

[0053] in, is the hidden state at time t, is the hidden state at time t1, is the feature vector at the current moment (output from graph convolution).

[0054] 3.2 Dynamic Relationship Learning Through the time-series graph neural network, this embodiment can learn the interaction pattern of neural signals and emotional signals in the time dimension. The network predicts future pain intensity, emotional fluctuations, etc. based on past signals and emotional data, and reveals the temporal correlation between pain and emotion.

[0055] Dynamic weight learning: Through the self-attention mechanism or weighted operations in graph neural networks, the temporal correlation of signals is dynamically learned to achieve time series prediction.

[0056]

[0057] in, is the predicted pain intensity or emotional fluctuation, It is the fusion weight of the temporal model and the output of the graph convolutional network.

[0058] Through the multi-dimensional feature fusion of neural signals and emotional signals, the interactive layer modeling of graph convolutional networks (GCN) and self-attention mechanisms, and the temporal modeling of temporal graph neural networks (NET-GNN), this embodiment can deeply explore the dynamic interactive relationship between pain and emotional fluctuations. This method of combining temporal information and graph structure modeling can provide strong data support for pain prediction and personalized treatment plan optimization, providing patients with more accurate and timely intervention.

[0059] Specifically, the goal of the analysis and prediction module is to predict pain intensity and occurrence time based on the interactive features of neural signals and emotional signals constructed through the time-series graph neural network (NET-GNN), combined with multimodal learning methods, and accurately predict the patient's emotional fluctuations.

[0060] Step 1: Multimodal signal input preprocessing and feature extraction 1.1 Signal preprocessing and feature extraction In order to achieve accurate prediction, it is first necessary to extract high-quality features from the collected neural signals (such as EEG, HRV, etc.) and emotional signals (such as facial expressions, voice, EDA, etc.). These features will be used as input for subsequent model learning.

[0061] EEG signal processing: Use time-frequency analysis (such as short-time Fourier transform STFT or wavelet transform WT) to process EEG signals and extract features in the time-frequency domain, such as Wave, Wave, Bo et al.

[0062] HRV signal processing: Perform periodic analysis on heart rate variability (HRV) and calculate its time domain features (such as SDNN, RMSSD) and frequency domain features (such as LF / HF ratio).

[0063] Emotional signal processing: For facial expression analysis, facial features (such as FACS coding) are extracted through image processing technology and converted into emotional states; for speech signals, acoustic features (such as pitch, speaking speed, etc.) are used to extract emotional signs; and galvanic skin response (EDA) can quantify emotional fluctuations by calculating the amplitude and duration of its changes.

[0064] These signals will be converted into corresponding feature matrices respectively, and ensure that each signal is aligned in the time dimension to obtain a multimodal feature set of time series. .

[0065] Step 2: Pain intensity prediction model 2.1 Regression Model: LSTM / GRU Model In terms of pain intensity prediction, traditional regression analysis methods cannot handle high-dimensional, nonlinear and time-series data. Therefore, this embodiment uses a long short-term memory network (LSTM) or a gated recurrent unit network (GRU) for time series regression analysis. LSTM and GRU can effectively capture long-term dependencies in signals and are particularly suitable for processing complex phenomena such as pain intensity that changes over time.

[0066] Model input: The previously extracted multimodal feature vector As the input of the LSTM model, LSTM processes these time series features step by step and learns the temporal dependencies.

[0067]

[0068] Output: The output of LSTM is the predicted value of pain intensity at each time t .

[0069]

[0070] in, Represents the predicted value of pain intensity, function It is the output layer of LSTM, which is used to map the hidden state to a specific pain intensity value.

[0071] Multi-task learning: While predicting pain intensity, this embodiment can design a joint loss function to predict the time of pain occurrence. By setting appropriate loss weights, the model will optimize the prediction of pain intensity and occurrence time at the same time.

[0072]

[0073] in, and are the loss functions of pain intensity and occurrence time, respectively. and is the corresponding loss weight.

[0074] 2.2 Multimodal Learning Multimodal learning can fully tap into the complementary information between different signals. For example, neural signals (such as EEG) and emotional signals (such as facial expressions and voice) may have different expression patterns when reflecting pain states. By simultaneously inputting these signals into the LSTM or GRU model, the model can learn to predict pain intensity more accurately. Multimodal learning improves the accuracy of prediction by fusing the temporal information of different signals.

[0075] Input fusion: splicing the temporal features of neural signals (EEG, HRV, etc.) and emotional signals (facial expressions, voice, etc.) to form multimodal input.

[0076]

[0077] Multi-task learning: Using a joint training strategy, some layers are shared between pain intensity prediction and emotion fluctuation prediction to improve the generalization ability of the model.

[0078] Step 3: Emotional Fluctuation Prediction Model 3.1 Classification Model: LSTM / Transformer The prediction of emotional fluctuations can be regarded as a multi-classification problem, that is, predicting the changing trend of emotional states (such as anxiety, depression, happiness, etc.) at different time points. This embodiment uses LSTM or Transformer model to model emotional fluctuations.

[0079] LSTM modeling: The LSTM network can effectively predict the fluctuations of patients' emotional states by processing the temporal information of emotional signals (such as facial expressions, voice, etc.). By combining neural signals with emotional signals, LSTM can capture the temporal dependencies between signals and thus predict future emotional fluctuations.

[0080]

[0081] in, is the hidden state at time t, combining the emotional signal and the neural signal.

[0082] Output layer: Outputs the probability distribution of emotional state categories (such as anxiety, depression, etc.).

[0083]

[0084] Among them, W and b are the weight and bias of the output layer.

[0085] 3.2 Reinforcement Learning: Dynamic Prediction Strategy Adjustment Reinforcement learning (RL) can optimize the emotion prediction strategy by continuously interacting with the environment in emotion fluctuation prediction. Through RL, the system can continuously adjust the prediction model of emotion fluctuation based on historical data, making the model more personalized and accurate.

[0086] State and action: In the emotion prediction task, the state can be defined as the patient’s current emotional state and physiological indicators (such as HRV, EDA, etc.), while the action is the adjustment of the prediction strategy for the patient’s emotional fluctuations.

[0087] Reward function: The reward function can be set as the accuracy of the model prediction. Higher accuracy corresponds to higher rewards. The system optimizes the reward function through continuous iteration to improve the personalized accuracy of emotion prediction.

[0088] Q-learning or Policy Gradient: Using Q-learning or Policy Gradient methods for policy optimization can enable the model to dynamically adjust the prediction strategy based on the individual patient's response, thereby improving the personalized effect of emotion prediction.

[0089] In this module, this embodiment combines advanced technologies such as LSTM, GRU, Transformer and reinforcement learning to establish a multimodal prediction framework that can simultaneously predict pain intensity, occurrence time and emotional fluctuations. By deeply analyzing the interaction between neural signals and emotional signals and combining time series modeling technology, the system can accurately predict the patient's pain intensity changes and emotional fluctuation trends, and on this basis generate personalized treatment and intervention recommendations, providing strong support for pain management and emotional health.

[0090] Specifically, the goal of the intervention module is to accurately identify the peak of patients' emotional fluctuations based on the prediction and real-time data of emotional fluctuations, and to promptly push appropriate psychological adjustment suggestions according to the nature of the emotional fluctuations. By providing personalized intervention measures at critical moments, it helps patients relieve emotional stress and pain and improves the overall treatment effect.

[0091] Step 1: Identify the peak of emotional fluctuations 1.1 Emotional fluctuation prediction and change rate detection In order to monitor the patient's emotional state in real time, we first rely on the emotional fluctuation prediction model. This model predicts the trend of emotional changes in the future by analyzing historical emotional data. By combining the interaction of neural signals and emotional signals, the model can accurately predict the peak period of emotional fluctuations. The specific implementation is as follows: Emotional fluctuation prediction: Use time series models such as LSTM and GRU (as described in the analysis and prediction module) to predict the emotional state in the short term based on the patient's real-time emotional data (facial expressions, voice, EDA, etc.). For example, the predicted sequence of emotional fluctuations can be obtained through the output of the LSTM model. , where t is the time step.

[0092]

[0093] Rate of Change Detection: The Rate of Change Detection (RoCD) method is used to determine whether there is a sharp fluctuation based on the rate of change of the emotion prediction value. A change rate threshold is set, and when the rate of change of the emotion state exceeds a certain value, it is judged as the peak period of emotion fluctuation.

[0094]

[0095] if Greater than the set threshold , then the time point t is marked as a potential peak of emotional fluctuation.

[0096] 1.2 Threshold method and trend analysis On the basis of the change rate detection, in order to further accurately identify the peak period of emotional fluctuation, this embodiment combines the threshold method and the trend analysis method. These methods can determine whether intervention is needed by setting a fixed emotional score threshold, and combine trend analysis to determine the persistence of emotional fluctuation: Threshold method: Set thresholds for multiple emotional indicators (such as anxiety scores, skin electrical response, etc.). When these indicators exceed the preset thresholds, it is judged that the emotional fluctuation has entered the peak period.

[0097]

[0098] in, It is the threshold of emotional fluctuations, which may be set as anxiety score, emotional state score, etc. according to the specific situation.

[0099] Trend analysis: Combine the prediction results of emotional fluctuations with the trend analysis of historical data to identify whether emotional fluctuations are showing a continuous upward trend. If the emotional fluctuation trend continues to rise for more than the set duration , it further confirms that the emotional fluctuation is at its peak.

[0100]

[0101] Through this multi-level, multi-dimensional analysis, the peak periods of emotional fluctuations can be accurately identified.

[0102] 1.3 Multi-dimensional confirmation and marking during peak period Ultimately, through the combined effects of rate of change detection, threshold method, and trend analysis, the system will identify the peak period of emotional fluctuations and mark these time points as trigger points for intervention. Under the real-time monitoring of the system, any time point that meets the peak period conditions will be marked as a critical moment of "needing intervention" and will be fed back to the intervention module in a timely manner.

[0103] Step 2: Push psychological adjustment suggestions 2.1 Meditation and deep breathing training suggestions At the peak of emotional fluctuations, one of the most direct and effective intervention measures is to recommend patients to meditate or practice deep breathing. These methods have been proven to quickly regulate emotional fluctuations, especially for relieving anxiety and stress. The system pushes meditation and deep breathing training suggestions through the following steps: Personalized intervention suggestion generation: Based on the intensity and nature of emotional fluctuations, the system generates personalized psychological adjustment suggestions through preset rules. For example, if the emotional fluctuation index is high and the anxiety level is significant, the system will recommend simple deep breathing training, such as the "478" deep breathing method, or meditation methods, such as meditation guided videos or audios.

[0104] Real-time push notifications: When a peak period is identified, the system will push personalized intervention prompts on the user's device (such as mobile phones, wearable devices). For example, push a reminder that "during the peak period of anxiety, it is recommended to meditate or deep breathe for 10 minutes" or play meditation audio or video.

[0105] 2.2 Cognitive behavioral therapy (CBT) recommendations If the emotional fluctuations are more severe and last longer, the system will recommend cognitive behavioral therapy (CBT) self-help training based on the patient's emotional fluctuation history and predicted results to help the patient identify and adjust negative emotions. The specific steps include: CBT recommendation generation: Based on the patient's emotional state, the system determines whether CBT is appropriate. For patients with more severe emotional fluctuations such as anxiety and depression, the system will recommend online CBT training courses to help patients understand the source of their negative emotions and adjust negative emotions through cognitive restructuring techniques.

[0106] Modular CBT training push: The system will provide a phased CBT training plan based on the patient's emotional state. For example, the initial stage provides emotion recognition and reflection training, the mid-term stage provides emotion regulation skills, and the late stage provides practical methods for dealing with anxiety.

[0107] 2.3 Emotional support and psychological counseling For patients with extremely drastic emotional fluctuations and weak self-regulation ability, the system can recommend emotional support services, such as online psychological counseling, emotional counseling, etc. Through integration with third-party psychological counseling platforms, the system can provide patients with timely emotional counseling services to relieve their emotional pressure.

[0108] Support service push: When the system determines that the patient's emotional fluctuations are extremely serious, it will push a link or push notification for the "emotional support" or "make an appointment for psychological counseling" function to help the patient seek professional emotional support.

[0109] Through the above steps, the real-time intervention module can accurately identify the peak period of patients' emotional fluctuations based on the prediction model of emotional fluctuations and multi-level analysis methods, and push personalized psychological adjustment suggestions at the right time. Whether it is meditation and deep breathing training, cognitive behavioral therapy recommendations, or the push of emotional support services, the system can make accurate and timely responses based on the patient's specific emotional state and historical data, provide patients with effective emotional intervention, and thus optimize the patient's overall treatment experience.

[0110] Specifically, the goal of the feedback module is to achieve regular feedback on the patient's neural and emotional signals, ensuring that the medical team can understand the patient's health status in a timely manner and make timely adjustments to the treatment plan, thereby improving the personalization and accuracy of treatment. The feedback module will ensure that doctors can obtain patients' health data in real time through automated feedback mechanisms, intelligent decision support systems, and data visualization tools, and make scientific treatment decisions based on data support.

[0111] Step 1: Automated feedback mechanism 1.1 Data collection and real-time upload In order to ensure timely feedback, the system will collect the patient's neural signals (such as EEG, HRV, EDA, EMG, etc.) and emotional signals (such as voice emotion analysis, facial expression recognition, etc.) in real time, and upload these data to the cloud regularly. The process of data collection and uploading is divided into the following steps: Data collection: Through interfaces with patients’ wearable devices (such as electroencephalographs, heart rate monitors, galvanic skin response sensors, etc.) and smart phones / tablets, neural and emotional signals are acquired in real time. These signals include but are not limited to EEG signals, HRV, galvanic skin response (EDA), electromyographic signals (EMG), speech emotion analysis results, etc.

[0112] Real-time data transmission: The collected data is transmitted to the cloud platform in real time through a secure and encrypted communication channel (such as the HTTPS protocol) to ensure the security and real-time nature of the data. The system can set a scheduled upload mechanism (such as uploading once every hour) so that doctors can understand the patient's status in a timely manner.

[0113] 1.2 Data Storage and Management In the cloud, all historical data of patients (including neural and emotional signals) will be stored in a database that complies with data privacy protection standards such as HIPAA (Health Insurance Portability and Accountability Act). The data will be marked with timestamps to ensure that doctors can access the complete history of the patient every time the patient visits.

[0114] Data management: Use efficient database management systems (such as NoSQL databases) to ensure efficient and scalable data access.

[0115] Data backup and recovery: To avoid data loss, the system will perform data backup tasks regularly and provide a fast recovery mechanism to ensure that data can be quickly restored in any failure situation.

[0116] Step 2: Data visualization and diagnostic support 2.1 Design and implementation of data visualization tools In order to help doctors quickly understand and analyze the patient's condition, the system will use data visualization technology to present the patient's real-time health data and prediction information through easy-to-understand charts and graphics. The specific functions are as follows: Signal data visualization: Through visualization tools such as time series graphs and spectrum analysis graphs, doctors can view the real-time changes of patients' neural signals (EEG, HRV, etc.) and emotional fluctuations. For example, a time series graph can be used to display the patient's emotional fluctuations and pain prediction results throughout the day, helping doctors to intuitively understand the patient's health status.

[0117] Emotional fluctuations and pain prediction trend chart: Combines the patient's emotional fluctuations with the pain prediction results (obtained through neuro-emotional interaction modeling) to show the trend of the patient's emotional fluctuations and the intensity changes of the pain prediction. This helps doctors evaluate the treatment effect of patients from a comprehensive perspective.

[0118] Anomaly detection and labeling: Use anomaly detection algorithms (such as the statistical zscore method or machine learning anomaly point detection method) to automatically identify abnormal changes in patients' neural signals or emotional fluctuations, and highlight them on the visual interface to remind doctors to pay special attention. When abnormal fluctuations are detected, the system will automatically mark the abnormal time point and generate a notification to prompt the doctor to take further examinations.

[0119] 2.2 Dynamic real-time update and diagnostic report generation The system will generate dynamically updated diagnostic reports based on regularly collected patient data. Each report contains a detailed description of the patient's current condition and provides personalized treatment recommendations based on pain prediction and emotional fluctuation prediction. The report includes: Patient status summary: This includes a comprehensive picture of neural signals (EEG, HRV, etc.) and emotional fluctuations. Doctors can view trend graphs, prediction results, and abnormality detection of these data.

[0120] Treatment recommendations: Based on the output of the intelligent analysis platform, treatment recommendations are given for the patient's current condition, such as medication adjustment, emotional regulation intervention, rehabilitation training, etc.

[0121] Comparative analysis of historical data: Compare with the patient's historical data to analyze the difference between the current treatment effect and the previous treatment cycle, and help doctors evaluate the patient's treatment progress.

[0122] Step 3: Physician Decision Support System 3.1 Decision support based on intelligent analysis The core of the doctor decision support system is to help doctors analyze the patient's condition and recommend optimized treatment plans through integrated artificial intelligence-assisted diagnostic tools. Specific methods include: Machine learning model training: The system uses historical case data to train multiple machine learning models (such as decision trees, support vector machines, etc.) to help identify the patterns of different emotions and pain patterns in order to provide decision support. Through the model output, doctors can learn about the patient's disease risk and treatment adjustment recommendations.

[0123] Personalized treatment recommendations: Based on the patient's personalized data (such as neural signals, emotional fluctuations, and pain predictions), the system generates personalized treatment recommendations. For some patients, medication dosage adjustment recommendations may be provided; for other patients with greater emotional fluctuations, psychological intervention measures may be recommended.

[0124] Treatment adjustment reminder: The system provides doctors with reminders to adjust treatment plans based on the results of AI analysis. For example, when emotional fluctuations exceed the set threshold, the system will remind doctors to increase the number of cognitive behavioral therapy (CBT); when the peak of pain intensity is expected to come, the system will remind doctors to adjust the medication plan in advance.

[0125] 3.2 Automated treatment plan optimization Based on feedback mechanisms and real-time data, the system will make recommendations for optimizing treatment plans through rule-based automation mechanisms. For example: Automatic adjustment of drug dosage: Based on the pain intensity prediction model, the system automatically generates the drug dosage adjustment plan required by the patient. For example, if the pain intensity exceeds the predicted value, the system will remind the doctor to increase the drug dosage in advance.

[0126] Optimization of emotion regulation measures: By predicting emotional fluctuations, the system will recommend increasing the frequency of psychological counseling or cognitive behavioral therapy for patients, and optimize the timing of intervention based on pain prediction results.

[0127] Through automated feedback mechanisms, intelligent decision support systems, and data visualization tools, the feedback module ensures that the patient's neural and emotional signals are fed back to the medical team in real time and accurately. Doctors can view the patient's health status through an integrated intelligent analysis platform, obtain personalized treatment recommendations based on data, and dynamically adjust the treatment plan. The system can not only help doctors make decisions quickly, but also continuously optimize the patient's treatment effect in a data-driven way, ultimately improving the accuracy and personalization of treatment.

[0128] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of protection of the claims of the present invention.

Claims

1. A long-term treatment follow-up and program optimization system for pain patients based on big data, characterized by: include: Data acquisition and perception module: Obtaining the patient's neural and emotional signals: Neural-emotional interaction module: It fuses neural signals and emotional signals to obtain neural-emotional interaction features, and models the interaction features using a time series graph. Analysis and prediction module: predict the intensity and occurrence time of pain through time series graph modeling, and predict the patient's emotional fluctuations through the interaction of neural signals and emotional signals; Intervention module: Identify peak periods of emotional fluctuations; Feedback module: Regularly provide feedback on the patient's neural and emotional signals.

2. The long-term treatment follow-up and program optimization system for pain patients based on big data according to claim 1 is characterized in that: The neural signals include: electroencephalogram, heart rate variability, galvanic skin response, and electromyography signals; the emotional signals include: voice emotion and facial expression recognition.

3. The big data-based long-term treatment follow-up and program optimization system for pain patients according to claim 2 is characterized in that: The neural emotion interaction module includes: Each signal is regarded as a node, and the interaction relationship between signals is modeled as an edge. The nodes of the graph are set to represent signals, and the edges represent the interaction relationship between different signals. Apply graph convolution operation to transmit information to each node. After receiving the signal from the neighboring node, each node performs weighted summation and processes it through a nonlinear activation function. Through multi-layer graph convolutional networks, the interactive information of neural signals and emotional signals is gradually integrated, so that each node can obtain information from multiple sources and learn complex interactive patterns; The signal interaction features at each moment are regarded as nodes of the graph, the time series information is introduced into the graph convolution, and the time series data is modeled in combination with the recurrent neural network.

4. The long-term treatment follow-up and program optimization system for pain patients based on big data according to claim 3 is characterized in that: The analysis and prediction module includes: Convert the signals into corresponding feature matrices respectively and ensure that the signals are aligned in the time dimension to obtain a multimodal feature set of time series; The multimodal feature set is used as the input of the LSTM model. The output of the LSTM is the predicted value of the pain intensity at each time t, and the joint loss function is used to predict the time of pain occurrence. The LSTM model is used to process the temporal information of emotional signals and combine neural signals with emotional signals to capture the temporal dependency between signals and thus predict future emotional fluctuations. Output the probability distribution of the emotion state categories.

5. The big data-based long-term treatment follow-up and program optimization system for pain patients according to claim 4 is characterized in that: The intervention modules include: Use rate-of-change detection to identify changes in affective fluctuations; Set thresholds for emotional fluctuations and use trend analysis to determine peak periods.

Images