CN-121983231-A - Intelligent analgesia control model building method and system

CN121983231ACN 121983231 ACN121983231 ACN 121983231ACN-121983231-A

Abstract

The invention relates to the technical field of medical artificial intelligence, and discloses an intelligent analgesia control model establishment method and system, wherein the system extracts the thermal state characteristics of tissue through a multi-mode data fusion module, realizes the rolling optimization of temperature tracks based on a recurrent neural network predictor and a model prediction controller, and the predictor weight is updated on line by means of the dynamic calibration module, so that the control model is self-adaptive to real-time treatment data change, and the long-term accuracy, stability and individual analgesic effect of temperature control are improved.

Inventors

CUI JINJIANG
XU JIANGEN
XU JIE
YUAN CHUNHUI

Assignees

苏州国科盈睿医疗科技有限公司
中国科学院苏州生物医学工程技术研究所

Dates

Publication Date: 20260505
Application Date: 20251230

Claims (10)

1. An intelligent analgesia control system, comprising: the physiological signal acquisition module is used for acquiring temperature signals, blood flow perfusion rate signals and body surface electromyographic signals from a target treatment area in real time; The multi-mode data fusion processing module is connected to the physiological signal acquisition module and is used for carrying out time alignment, noise reduction and feature extraction on the received multi-source heterogeneous physiological signals and constructing nonlinear mapping from the original signals to the tissue thermal state feature vectors based on the deep confidence network; the self-adaptive prediction control module is connected with the multi-modal data fusion processing module, and a recurrent neural network predictor with online learning capability and a model prediction controller are integrated in the self-adaptive prediction control module; the model prediction controller generates an optimization control instruction which enables the predicted temperature track to be closest to a preset treatment temperature interval by solving a rolling optimization problem with constraint; The treatment execution and dynamic calibration module is connected with the self-adaptive prediction control module and the physiological signal acquisition module and comprises a high-precision refrigerant flow regulating valve and a microprocessor, wherein the module executes an optimized control instruction from a model prediction controller to drive the flow regulating valve to precisely control refrigerant output, and meanwhile continuously compares errors between a predicted temperature track and an actual measured temperature, and when the error accumulation exceeds a preset self-adaptive trigger threshold value, online increment updating of the internal weight of the recurrent neural network predictor is automatically started.
2. The intelligent analgesia control system according to claim 1, wherein the specific operation flow of the multi-mode data fusion processing module comprises unified timestamp synchronization of temperature signals, blood perfusion rate signals and electromyographic signals transmitted by the physiological signal acquisition module, denoising of the synchronized multipath signals by adopting a wavelet transform-based joint denoising algorithm, extracting time domain features, frequency domain features and nonlinear dynamics features from each path of denoised signals to form a high-dimensional original feature set, inputting the high-dimensional original feature set into a deep confidence network which is trained offline in advance, and compressing and mapping the high-dimensional original feature set into a low-dimensional tissue thermal state feature vector through multi-layer nonlinear transformation of the network.
3. The intelligent analgesia control system according to claim 2, wherein the off-line training process of the deep belief network comprises collecting a great number of historical physiological signal data covering different individuals and different treatment positions and corresponding tissue temperature data as a training sample set, constructing a deep belief network structure comprising 3 visible layers and 5 hidden layers, performing unsupervised layer-by-layer pre-training on the network by adopting a contrast divergence algorithm to initialize network weights, and performing supervised fine tuning training on the network by using labeled tissue temperature data after the pre-training is completed to minimize reconstruction errors between tissue thermal state feature vectors and real tissue temperature data output by the network.
4. The intelligent analgesia control system according to claim 1, wherein the recurrent neural network predictor adopts a long-short-term memory network structure, an input layer of the recurrent neural network predictor receives a current tissue thermal state characteristic vector output by a multi-mode data fusion processing module and a control instruction sequence of 5 time steps in the past, a network hiding layer comprises 128 long-short-term memory units for capturing long-term time dependency in tissue thermodynamics, and an output layer predicts tissue temperature values of 3 time steps in the future.
5. The intelligent analgesia control system according to claim 4, wherein the rolling optimization problem of the model predictive controller is mathematically described by finding a set of future control command sequences at each control cycle such that the variance of the predicted temperature output given by the recurrent neural network predictor from the reference trajectory of the preset treatment temperature interval is minimized while satisfying the physical constraints of control command rate of change and absolute value of cryogen flow.
6. The intelligent analgesia control system according to claim 1, wherein the execution mechanism of online incremental update in the treatment execution and dynamic calibration module comprises a microprocessor continuously calculating the root mean square error between the temperature predicted value and the actual measured value of the recurrent neural network predictor for 3 time steps in the future, presetting an adaptive trigger threshold value to be 0.5 ℃, automatically triggering an online learning program by the microprocessor when the calculated root mean square error in 10 continuous control periods exceeds 0.5 ℃, forming an incremental training data set by the online learning program by using the actual control command and the measured temperature data of the last 50 time steps, and performing small adjustment on the weight of the recurrent neural network predictor by using a back propagation algorithm with time, wherein the learning rate is set to be 0.001.
7. The intelligent analgesia control system of claim 6, wherein said weight updating process is completed within 1 control cycle, ensuring that the real-time nature of the control loop is not affected.
8. The intelligent analgesia control system according to claim 1, further comprising a safety monitoring and intervention module, wherein the safety monitoring and intervention module monitors the tissue thermal state characteristic vector output by the multi-mode data fusion processing module in real time and calculates a mutation index of a characteristic value of the tissue thermal state characteristic vector, the safety monitoring and intervention module pre-stores a characteristic template library of normal treatment and abnormal states, and when the calculated mutation index exceeds a safety threshold value of 2.0 for 3 continuous periods and the matching degree of the characteristic vector and the pre-stored abnormal state template reaches 90%, the module sends a coverage instruction with the highest priority to the self-adaptive prediction control module, and the forced model prediction controller outputs a preset safety control mode.
9. The intelligent analgesia control system of claim 8, wherein said safety control mode comprises reducing cryogen flow to a minimum level of maintenance temperature while simultaneously sending an alarm signal to an external monitoring terminal.
10. An intelligent analgesia control model building method, which is characterized in that the intelligent analgesia control system of any one of claims 1-9 is used for realizing analgesia control.

Description

Intelligent analgesia control model building method and system Technical Field The invention belongs to the technical field of medical artificial intelligence, and particularly relates to an intelligent analgesia control model building method and system. Background In the technical field of medical equipment, a closed-loop automatic control system realizes accurate control of treatment parameters through a real-time monitoring and feedback regulation mechanism, so that the safety and effectiveness of treatment are improved, and the closed-loop automatic control system becomes an important component of modern accurate medical treatment. Among them, a noninvasive analgesic treatment apparatus is an important type of medical apparatus, and is focused on achieving a desired physiological effect by precisely controlling the output of energy or substances. The intelligent analgesic control technology aims at accurately regulating and controlling the output dosage and the output rate of media such as a refrigerant and the like through a closed-loop automatic control mechanism, and realizing and maintaining a preset effective treatment temperature interval in a specific target area of a human body, so that a noninvasive and controllable analgesic treatment effect is achieved. The prior art typically integrates temperature sensing, control algorithms and actuators to form a closed loop from monitoring to regulation. The prior art can realize basic temperature control, but has obvious limitations in practical application, namely the system has insufficient adaptability to individual physiological differences, tissue thermal characteristic changes and environmental interference, so that the temperature control precision and stability are difficult to guarantee, the traditional control model depends on fixed parameters or simple feedback, cannot dynamically learn and optimize according to real-time treatment data, influences individuation and consistency of analgesic effects, and meanwhile, lacks effective fusion and intelligent analysis of multisource physiological signals, and limits accurate perception and early warning of the treatment state and risk of the system. Therefore, there is still a technical problem to be solved in realizing safe, accurate and self-adaptive noninvasive analgesia control. Disclosure of Invention The invention aims to provide an intelligent analgesia control model establishment method and system, which are used for solving the problems that in the prior art, temperature control precision and stability are insufficient due to individual physiological differences, tissue thermal characteristic changes and environmental interference, and a traditional control model cannot be dynamically learned and optimized based on real-time treatment data and lacks of effective fusion and intelligent analysis of multisource physiological signals. In order to achieve the above purpose, the technical scheme adopted by the invention is to construct an intelligent analgesia control system, which comprises a physiological signal acquisition module, a multi-mode data fusion processing module, a self-adaptive prediction control module and a treatment execution and dynamic calibration module. The physiological signal acquisition module is used for acquiring temperature signals, blood flow perfusion rate signals and body surface electromyographic signals from a target treatment area in real time. The multi-mode data fusion processing module is connected to the physiological signal acquisition module and is used for carrying out time alignment, noise reduction and feature extraction on the received multi-source heterogeneous physiological signals and constructing nonlinear mapping from the original signals to the tissue thermal state feature vectors based on the deep confidence network. The self-adaptive prediction control module is connected with the multi-modal data fusion processing module, a recurrent neural network predictor with online learning capability and a model prediction controller are integrated in the self-adaptive prediction control module, the recurrent neural network predictor predicts tissue temperature change tracks of a plurality of time steps in the future based on the current tissue thermal state characteristic vector and the historical control instruction sequence, and the model prediction controller generates an optimization control instruction which enables the predicted temperature track to be closest to a preset treatment temperature interval by solving a constrained rolling optimization problem. The treatment execution and dynamic calibration module is connected with the self-adaptive prediction control module and the physiological signal acquisition module and comprises a high-precision refrigerant flow regulating valve and a microprocessor, wherein the module executes an optimized control instruction from a model prediction controller to drive the flow regulating valve to precisely con