Neural function detection balancing system and method based on artificial intelligence

Abstract

The invention discloses a neural function detection balancing system and method based on artificial intelligence, the system comprises a detection unit, a signal processing unit, an intelligent analysis unit, a treatment unit and a control unit, the detection unit comprises a multi-mode sensor array and is used for collecting neural signals of a human nervous system; the signal processing unit is used for filtering, amplifying and digitalizing the neural signal acquired by the detection module; the intelligent analysis unit is used for establishing a neural function evaluation model and analyzing the abnormal condition of the neural signal to obtain an analysis result; the treatment unit comprises an electromagnetic pulse generator, an ultrasonic generator and a phototherapy device and is used for performing forward stimulation treatment on the ganglion according to an analysis result; and the control unit is used for controlling and adjusting the treatment parameters of the treatment unit. According to the invention, the integration of neural function detection and treatment is realized, and the accuracy and objectivity of neural function evaluation are remarkably improved through multi-modal signal acquisition and artificial intelligence analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of disease assessment and treatment, and in particular to a neural function detection balance system and method based on artificial intelligence. Background Art

[0002] As an important regulatory system of the human body, the nervous system controls the functions of various organs of the human body through nerve roots of different segments. For example, the nerve roots of the 1st to 6th cervical vertebrae (C1-6) are related to heart function, and the nerve roots of the 2nd to 5th lumbar vertebrae (L2-5) are related to the functions of organs such as gonads and bladder. Therefore, by detecting abnormalities in nerve signals emitted by specific nerve roots, it can help determine the functional status and potential lesions of the corresponding organs.

[0003] Traditional neurological function testing methods mainly rely on a single detection method, such as electromyography or nerve conduction velocity testing. Although these methods can detect nerve signals, they have obvious limitations: first, the detection range is limited to a single nerve segment, and it is difficult to monitor the nerve activity of multiple related segments at the same time; second, the signal type obtained is single and cannot fully reflect the state of nerve function; third, the interpretation of the test results mainly depends on the doctor's experience and lacks objective and quantitative evaluation standards. Summary of the invention

[0004] The present invention aims to solve one of the technical problems in the related art at least to a certain extent. To this end, the purpose of the present invention is to propose a neural function detection balance system and method based on artificial intelligence.

[0005] To achieve the above objectives, on the one hand, a neural function detection and balance system based on artificial intelligence according to an embodiment of the present invention includes:

[0006] A detection unit, comprising a multimodal sensor array, for collecting neural signals of a human nervous system, wherein the neural signals include bioelectric signals, pressure signals and temperature signals;

[0007] A signal processing unit, electrically connected to the detection unit, for filtering, amplifying and digitally processing the neural signal collected by the detection module;

[0008] An intelligent analysis unit, connected to the signal processing unit, is used to establish a neural function evaluation model and analyze abnormal neural signals to obtain analysis results;

[0009] a treatment unit, connected to the intelligent analysis unit by signal, the treatment unit comprising an electromagnetic pulse generator, an ultrasonic generator and a phototherapy device, for performing positive stimulation treatment on the nerve segment according to the analysis result;

[0010] The control unit is connected to the detection unit, the signal processing unit, the intelligent analysis unit and the treatment unit by signal, and is used for controlling and adjusting the treatment parameters of the treatment unit.

[0011] In addition, the neural function detection and balance system based on artificial intelligence according to the above embodiment of the present invention may also have the following additional technical features:

[0012] According to one embodiment of the present invention, the multimodal sensor array comprises:

[0013] Surface electrode array and / or needle electrode array for collecting bioelectric signals of the human nervous system, with a sampling frequency of 1kHz-10kHz;

[0014] A pressure sensor array for detecting nerve segment pressure signals with a detection accuracy of 0.01N-1N;

[0015] Infrared temperature sensor array for detecting temperature signals at nerve conduction sites, with a detection accuracy of 0.1°C;

[0016] The multimodal sensor array is arranged in the nerve segment-dense areas around the spine, wrists, and ankles.

[0017] According to one embodiment of the present invention, the intelligent analysis unit includes:

[0018] A deep learning model for analyzing characteristic parameters of neural signals, wherein the characteristic parameters include at least frequency, amplitude and phase;

[0019] The expert system, based on the neuroanatomical database, compares the characteristic parameters of the detected neural signals with the standard database;

[0020] A neurological function assessment module is used to output the location of abnormal nerve segments, the type and degree of abnormality, and obtain assessment results;

[0021] A treatment plan generator automatically generates a personalized treatment plan based on the evaluation results and transmits it to the treatment unit.

[0022] According to one embodiment of the present invention, the electromagnetic pulse generator is used to generate therapeutic pulses with a frequency of 0.1-1000 Hz and a magnetic field strength of 0.1-2.0 T;

[0023] The ultrasonic generator is used to generate 0.5-3MHz frequency, 0.1-2.0W / cm 2 Ultrasonic sound intensity;

[0024] The phototherapy device is used to emit 780-1100nm near-infrared light and 3-100μm far-infrared light, with an energy density range of 1-100J / cm 2 .

[0025] According to one embodiment of the present invention, the treatment unit also includes a feedback module for real-time acquisition of changes in physiological parameters during the treatment process; the control unit is also used to adjust the treatment parameters according to the changes in the physiological parameters, and the physiological parameters include nerve conduction velocity, heart rate variability, skin resistance, and local body temperature.

[0026] According to one embodiment of the present invention, the signal processing unit includes:

[0027] An adaptive filter, wherein the adaptive filter adopts a minimum mean square error criterion to perform noise reduction and interference suppression processing on the neural signal; and automatically adjusts the filter coefficient and bandwidth parameter in real time according to the frequency characteristics, signal-to-noise ratio and interference type of different types of neural signals;

[0028] A wavelet transform processing module, which is used to extract the frequency and amplitude characteristics of neural signals through multi-scale decomposition using a discrete wavelet transform algorithm;

[0029] Independent component analysis module, which is used to separate multi-channel aliased neural signals into several independent components based on the fast ICA algorithm and utilize the statistical independence between signal sources to extract the target neural signals;

[0030] The artifact recognition and elimination module is used to recognize and eliminate the characteristics of electrocardiogram, electromyography and motion artifact interference signals by using pattern recognition methods.

[0031] According to one embodiment of the present invention, the deep learning model includes:

[0032] The time series feature extraction network is used to model the time series of neural signals using a long short-term memory network structure, extract amplitude, frequency, and phase features, and analyze their time dependence;

[0033] A spatial feature extraction network is used to construct a spatial topology map using a graph convolutional neural network with nerve segments as nodes, and analyze the spatial correlation and signal conduction characteristics between multiple nerve segments;

[0034] The attention mechanism module includes two sub-modules: channel attention and spatial attention. It is used to highlight the neural signal features at key time points and key spatial positions by learning the importance weights of signal features.

[0035] The model adaptive update module is used to optimize network parameters based on the online learning algorithm and use the treatment effect as a supervisory signal to improve the generalization ability and robustness of the model.

[0036] According to one embodiment of the present invention, the treatment plan generator adopts a reinforcement learning algorithm, including:

[0037] A state space construction module, used for parameterizing the neural function state into a multi-dimensional feature vector, wherein the neural function state includes characteristic parameters of neural signals, physiological parameters and clinical indicators;

[0038] An action space construction module is used to discretize the treatment parameters into a finite set of actions, including the parameter combinations of three treatment modes: electromagnetic pulse, ultrasound and phototherapy;

[0039] The reward function design module is used to comprehensively evaluate the three evaluation indicators of treatment effect, safety and comfort, and to build a quantitative reward calculation model;

[0040] The policy network is used to output the optimal treatment parameter combination by adopting a deep deterministic policy gradient algorithm through alternating optimization of the value network and the policy network.

[0041] According to one embodiment of the present invention, a neural pathway imaging unit is further included, and the neural pathway imaging unit includes:

[0042] The electrical impedance tomography module is used to reconstruct the conductivity distribution of nerve tissue by using multi-frequency and multi-channel electrical impedance measurement technology and to construct a three-dimensional model of the nerve conduction pathway by solving inverse problems;

[0043] The neural function area labeling module is used to automatically locate and label abnormal neural segments based on deep learning image segmentation algorithms, and display quantitative scores of abnormality levels;

[0044] The treatment process visualization module is used to display the spatial distribution and scope of treatment energy in real time using 3D rendering technology, and to display the dynamic changes of tissue response during treatment using pseudo-color;

[0045] The anatomical structure registration module is used to use a non-rigid registration algorithm to spatially align the three-dimensional model of the nerve conduction pathway obtained by the electrical impedance tomography module and the abnormal nerve segment information marked by the nerve function area marking module with the standard human anatomical atlas to achieve accurate anatomical positioning of the abnormal nerve function area.

[0046] On the other hand, the neural function detection balance method based on artificial intelligence according to an embodiment of the present invention is applied to the system as described above, including:

[0047] The detection unit collects neural signals of the human nervous system, wherein the neural signals include bioelectric signals, pressure signals and temperature signals;

[0048] The signal processing unit filters, amplifies and digitizes the neural signals collected by the detection module;

[0049] The intelligent analysis unit establishes a neural function assessment model and analyzes abnormal neural signals to obtain analysis results;

[0050] The treatment unit performs positive stimulation treatment on the nerve segment according to the analysis result, and the positive stimulation treatment includes electromagnetic pulse treatment, ultrasonic treatment, and infrared light treatment;

[0051] The control unit controls and adjusts the treatment parameters of the treatment unit.

[0052] According to the artificial intelligence-based neural function detection and balance system provided by the embodiment of the present invention, a comprehensive collection of bioelectric signals, pressure signals and temperature signals of the nervous system is realized through a multimodal sensor array, and high-quality neural signal features are effectively extracted with the help of advanced signal processing technologies such as adaptive filtering, wavelet transform and independent component analysis. The system uses a deep learning model to analyze the time series and spatial characteristics of neural signals, and combines the expert system to achieve accurate positioning and evaluation of abnormal nerve segments. In terms of treatment, through the coordinated application of various treatment methods such as electromagnetic pulses, ultrasound and phototherapy, and the reinforcement learning algorithm automatically generates personalized treatment plans, the intelligent optimization of treatment parameters is achieved. At the same time, the system also has a real-time feedback adjustment function, which can dynamically adjust the treatment strategy according to the changes in physiological parameters during the treatment process. Through electrical impedance tomography technology, the system visualizes the three-dimensional model of the neural conduction pathway and realizes real-time monitoring and precise positioning of the treatment process. This integrated and intelligent detection and treatment closed-loop system significantly improves the diagnosis and treatment accuracy and treatment effect of neurological dysfunction.

[0053] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0055] Figure 1 It is a block diagram of a neural function detection and balance system based on artificial intelligence according to an embodiment of the present invention;

[0056] Figure 2 It is a block diagram of an intelligent analysis unit in a neural function detection and balance system based on artificial intelligence according to an embodiment of the present invention;

[0057] Figure 3It is a block diagram of a signal processing unit in a neural function detection and balance system based on artificial intelligence according to an embodiment of the present invention;

[0058] Figure 4 It is a block diagram of a deep learning model in a neural function detection and balance system based on artificial intelligence according to an embodiment of the present invention;

[0059] Figure 5 It is a block diagram of a treatment plan generator in a neural function detection and balance system based on artificial intelligence according to an embodiment of the present invention;

[0060] Figure 6 It is a block diagram of a neural pathway imaging unit in a neural function detection and balance system based on artificial intelligence according to an embodiment of the present invention;

[0061] Figure 7 It is a flow chart of a neural function detection balance method based on artificial intelligence according to an embodiment of the present invention.

[0062] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0063] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but are not to be construed as limiting the present invention. All other embodiments obtained by ordinary technicians in the field without creative work based on the embodiments of the present invention are within the scope of protection of the present invention.

[0064] Reference Figures 1 to 5 As shown, the artificial intelligence-based neural function detection and balance system 100 provided according to an embodiment of the present invention includes a detection unit 101, a signal processing unit 102, an intelligent analysis unit 103, a treatment unit 104 and a control unit 105.

[0065] Specifically, the detection unit 101 includes a multimodal sensor array for collecting neural signals of the human nervous system, wherein the neural signals include bioelectric signals, pressure signals, and temperature signals. That is, the detection unit 101 uses a multimodal sensor array to simultaneously collect multiple types of signals from multiple nerve segments. In specific applications, the multimodal sensor array is accurately positioned at important nerve segments to achieve multi-point synchronous collection, such as C1-6, L2-5 and other nerve stages.

[0066] The signal processing unit 102 is electrically connected to the detection unit 101, and is used to filter, amplify and digitally process the neural signals collected by the detection module. That is, the signal processing unit 102 is used to pre-process the collected neural signals to ensure the accuracy and reliability of the signals.

[0067] The intelligent analysis unit 103 is data-connected to the signal processing unit 102, and is used to establish a neural function evaluation model, analyze neural signal abnormalities, and obtain analysis results. Preferably, the neural function evaluation model is trained with a large amount of clinical data to establish a mapping relationship between neural signal characteristics and organ function status, and can analyze the characteristic parameters of neural signals in real time, such as frequency characteristics, time domain characteristics, etc., and compare them with the standard pattern library to identify abnormalities. The model can also have adaptive learning capabilities, and can continuously optimize the evaluation accuracy based on new data.

[0068] The treatment unit 104 is connected to the intelligent analysis unit 103 by signal. The treatment unit 104 includes an electromagnetic pulse generator, an ultrasonic generator and a phototherapy device, which are used to perform positive stimulation treatment on the nerve segment according to the analysis results. The treatment unit 104 integrates a variety of physical treatment methods. The electromagnetic pulse generator can generate a pulsed magnetic field of a specific frequency and intensity for nerve activation; the ultrasonic generator can perform deep tissue stimulation to promote nerve repair; the phototherapy device achieves the purpose of regulating nerve function through light of a specific wavelength. These treatment methods can be used alone or in combination as needed.

[0069] The control unit 105 is connected to the detection unit 101, the signal processing unit 102, the intelligent analysis unit 103 and the treatment unit 104, and is used to control and adjust the treatment parameters of the treatment unit 104. Preferably, the control unit 105 monitors the treatment effect in real time through a feedback control algorithm, and dynamically adjusts the treatment parameters accordingly.

[0070] According to the artificial intelligence-based neural function detection and balance system 100 provided by the present invention, the integration of neural function detection and treatment is realized, breaking through the limitations of traditional single detection methods. Through multimodal signal acquisition and artificial intelligence analysis, the accuracy and objectivity of neural function evaluation are significantly improved. At the same time, the intelligent treatment function of the system can realize the formulation of precise and personalized treatment plans, greatly improving the diagnosis and treatment effects of nervous system diseases.

[0071] In one embodiment of the present invention, the multimodal sensor array comprises:

[0072] A surface electrode array and / or a needle electrode array for collecting bioelectric signals of the human nervous system, with a sampling frequency of 1 kHz - 10 kHz. Exemplarily, the surface electrode array is composed of multiple circular electrode pieces with a diameter of 2 - 5 mm, supported by a flexible substrate, and can closely fit the skin surface. The needle electrode array is composed of micro-needle electrodes and can achieve signal acquisition at different depths. Both of these electrode arrays use differential amplification processing, combined with a high sampling frequency of 1 kHz - 10 kHz, and can accurately capture rapidly changing bioelectric signals such as neuron action potentials and local field potentials.

[0073] A pressure sensor array for detecting nerve segment pressure signals, with a detection accuracy of 0.01 N - 1 N. Exemplarily, the pressure sensor array uses highly sensitive MEMS piezoelectric sensors and can achieve pressure distribution detection with high spatial resolution, and can accurately reflect the pressure changes in the tissues around the nerve.

[0074] An infrared temperature sensor array for detecting temperature signals at the nerve conduction site, with a detection accuracy of 0.1 °C. Exemplarily, the infrared temperature sensor array uses high-precision thermopile sensors and is equipped with a temperature compensation circuit to ensure stable measurement accuracy at different ambient temperatures and can detect the temperature changes at the nerve conduction site in real time.

[0075] The multimodal sensor array is arranged in the dense nerve segment areas around the spine, wrist, and ankle.

[0076] That is to say, in terms of the layout of the sensors, according to the characteristics of human nerve anatomy, these sensors are arranged in three key areas: the area around the spine focuses on covering important nerve segments such as C1 - C6 and L2 - L5, and adopts a "pin" - shaped layout to ensure comprehensive signal acquisition; the wrist part is arranged for important nerve trunks such as the median nerve and ulnar nerve, and adopts a circular array layout; the ankle part focuses on detecting the tibial nerve and peroneal nerve and adopts a "U" - shaped layout.

[0077] Through the collaborative work of each sensor, the multimodal sensor array realizes the synchronous acquisition of bioelectric, pressure, and temperature signals of the nervous system. Its layout design ensures the accuracy and reliability of data acquisition, and also improves the detection comfort of patients, creating favorable conditions for long - term detection.

[0078] Refer to Figure 2 As shown, in some embodiments of the present invention, the intelligent analysis unit 103 includes a deep learning model 1031, an expert system 1032, a nerve function evaluation module 1033, and a treatment plan generator 1034.

[0079] The deep learning model 1031 is used to analyze the characteristic parameters of the neural signal, and the characteristic parameters include at least frequency, amplitude and phase. The deep learning model 1031 adopts a multi-layer convolutional neural network (CNN) structure to process multimodal neural signal data. The model includes a feature extraction layer, a time-frequency analysis layer and a classification layer. The feature extraction layer can extract the frequency characteristics (including energy distribution of frequency bands such as α, β, and γ), amplitude characteristics (including peak value, root mean square value, etc.), and phase characteristics (including phase synchronization, phase locking value, etc.) of the neural signal through multi-scale convolution operations. The model adopts an attention mechanism, which can automatically identify and focus on abnormal signal fragments and improve the accuracy of feature extraction.

[0080] Expert system 1032 compares the characteristic parameters of the detected neural signals with the standard database based on the neuroanatomy database. Expert system 1032 has built a comprehensive neuroanatomy database that contains a large amount of clinically verified standard data. The database is classified according to different age groups, genders and physical conditions, and a detailed reference range of neurological function indicators is established. The system uses a fuzzy logic reasoning engine to perform multi-dimensional comparisons of the neural signal characteristics detected in real time with the standard data in the database. Through the set weight coefficients and threshold conditions, the degree of deviation of the signal characteristics can be identified, and a preliminary abnormal judgment result can be given.

[0081] The neurological function assessment module 1033 is used to output the location, type and degree of abnormal nerve segments to obtain the assessment results. The neurological function assessment module 1033 uses a multi-layer decision tree algorithm to comprehensively analyze the output results of the deep learning model 1031 and the expert system 1032. This module can accurately locate the anatomical location of abnormal nerve segments and classify the abnormal types into specific categories such as conduction block, overexcitement, and signal attenuation. At the same time, according to the degree to which the signal deviates from the standard range, the degree of abnormality is quantified into three levels: mild (Level 1), moderate (Level 2), and severe (Level 3). The assessment results are output in the form of a standardized electronic report, which contains detailed data analysis charts and clinical recommendations.

[0082] The treatment plan generator 1034 automatically generates a personalized treatment plan based on the evaluation results and transmits it to the treatment unit 104. The treatment plan generator 1034 establishes a dynamically optimized treatment strategy library based on the machine learning algorithm. The module first selects a suitable treatment plan combination from the treatment strategy library based on the results of the neurological function evaluation. Then, the treatment parameters are finely adjusted in combination with the individual characteristics of the patient (such as age, medical history, treatment contraindications, etc.). For example, for conduction block type abnormalities, electromagnetic pulse therapy of appropriate frequency will be recommended; for overexcited type abnormalities, a phototherapy plan with a specific wavelength will be recommended. The treatment plan generator 1034 also has a self-learning function, which can continuously optimize the treatment plan based on the feedback of the treatment effect.

[0083] In this embodiment, the intelligent analysis unit 103 realizes accurate identification of neurological dysfunction and formulation of personalized treatment plans through the organic combination of deep learning technology and expert knowledge. The adaptive learning ability and multi-dimensional analysis method significantly improve the accuracy of diagnosis and the pertinence of treatment. At the same time, the standardized evaluation process and automated solution generation function greatly improve the efficiency of clinical diagnosis and treatment, and provide doctors with reliable auxiliary decision support.

[0084] In one embodiment of the present invention, the electromagnetic pulse generator is used to generate a therapeutic pulse with a frequency of 0.1-1000 Hz and a magnetic field strength of 0.1-2.0 T. The ultrasonic generator is used to generate a frequency of 0.5-3 MHz and a magnetic field strength of 0.1-2.0 W / cm 2 The phototherapy device is used to emit 780-1100nm near-infrared light and 3-100μm far-infrared light, with an energy density range of 1-100J / cm 2 .

[0085] The treatment unit 104 realizes the selective control of multi-mode physical therapy and a wide range of adjustable parameters through the combination of electromagnetic pulse, ultrasound and phototherapy technology, meets the treatment needs of different types of neurological dysfunctions, and ensures the safety and effectiveness of the treatment process.

[0086] In some embodiments of the present invention, the treatment unit 104 further includes a feedback module for real-time acquisition of changes in physiological parameters during the treatment process. The control unit 105 is further configured to adjust the treatment parameters according to the changes in the physiological parameters, wherein the physiological parameters include nerve conduction velocity, heart rate variability, skin resistance, and local body temperature.

[0087] Preferably, the control unit 105 adopts an intelligent adaptive control algorithm, establishes a dynamic mapping model between treatment parameters and physiological indicators, and adopts a fuzzy PID control strategy to achieve smooth adjustment of treatment intensity. At the same time, a safety threshold is set to effectively prevent excessive parameter adjustment. Exemplarily, the control unit 105 can record the parameter change trend during the treatment process, establish a personalized treatment parameter database, and continuously optimize the treatment plan recommendation algorithm. For example, when an improvement in nerve conduction velocity is detected, the treatment intensity will be gradually reduced to maintain the best treatment effect. At the same time, the state of autonomic nervous function is evaluated according to the heart rate variability index, and the treatment frequency is adjusted accordingly; the energy output is adjusted by the change in skin resistance to avoid excessive stimulation of local tissues; the treatment time is controlled according to the change in local body temperature to ensure the safety of treatment.

[0088] This embodiment achieves precise regulation and personalized optimization of the treatment process by introducing a multi-parameter real-time feedback mechanism and intelligent adaptive control. It can timely detect and respond to changes in the body's state, dynamically adjust treatment parameters, and significantly improve the safety and effectiveness of treatment. Through data accumulation and intelligent learning, the treatment plan is continuously optimized, greatly improving the accuracy of neurological function treatment and the predictability of treatment effects.

[0089] Reference Figure 3 As shown, in some embodiments of the present invention, the signal processing unit 102 includes an adaptive filter 1021 , a wavelet transform processing module 1022 , an independent component analysis module 1023 and an artifact identification and elimination module 1024 .

[0090] The adaptive filter 1021 adopts the minimum mean square error criterion to perform noise reduction and interference suppression on the neural signal; and automatically adjusts the filter coefficient and bandwidth parameters in real time according to the frequency characteristics, signal-to-noise ratio and interference type of different types of neural signals. The filter can effectively suppress various types of noise such as power frequency interference and baseline drift. The filter adopts an adaptive weight update algorithm to dynamically adjust the filter coefficient by calculating the error between the signal and the expected output in real time. According to the characteristics of different neural signals, the optimal filter bandwidth is automatically selected. For example, a wider passband is selected for fast action potential signals, while narrowband filtering is used for slow wave signals. The filter can also include an automatic gain control circuit, which can automatically adjust the amplification factor according to the input signal strength to ensure that the signal always remains within the optimal dynamic range.

[0091] The wavelet transform processing module 1022 is used to extract the frequency and amplitude characteristics of the neural signal through multi-scale decomposition using a discrete wavelet transform algorithm. Exemplarily, the wavelet transform processing module 1022 uses a biomedically optimized wavelet basis function to accurately capture the transient characteristics of the neural signal. In addition, a multi-level decomposition strategy is used to decompose the neural signal into detail coefficients and approximate coefficients of different frequency bands. Through threshold processing and reconstruction algorithms, the characteristic components in the signal can be effectively extracted while suppressing non-characteristic components.

[0092] The independent component analysis module 1023 is used to separate the multi-channel aliased neural signals into several independent components based on the fast ICA algorithm and the statistical independence between the signal sources to extract the target neural signal. In other words, the independent component analysis module 1023 realizes signal source separation based on the fast ICA algorithm. In operation, the multi-channel original signal is first preprocessed and whitened to reduce the correlation between the data. Then, the mixing matrix of the signal is estimated through an iterative optimization algorithm, and the independent components are separated using the statistical independence criterion.

[0093] The artifact recognition and elimination module 1024 is used to identify and eliminate the features of ECG, EMG, and motion artifact interference signals using pattern recognition methods. Exemplarily, the artifact recognition and elimination module 1024 establishes a complete artifact feature library, which contains typical features of common interference sources such as ECG, EMG, EO, and motion artifacts. A multi-feature fusion strategy is adopted to integrate time domain, frequency domain, and morphological features for artifact recognition. Classifiers trained by deep learning algorithms can accurately identify various types of artifacts. For the identified artifacts, selective filtering and signal reconstruction methods can be used to eliminate them, retaining the effective neural signal components to the greatest extent.

[0094] Through multi-module collaborative signal processing, high-quality acquisition and processing of neural signals are achieved. Adaptive filtering ensures the purity of the signal, wavelet transform provides accurate time-frequency feature extraction capabilities, independent component analysis achieves effective separation of complex signals, and artifact identification and elimination ensures the authenticity of the signal. This comprehensive processing significantly improves the signal-to-noise ratio and feature extraction effect of neural signals, provides a high-quality data foundation for subsequent intelligent analysis, and greatly improves the accuracy and reliability of neural function detection.

[0095] Reference Figure 4 As shown, in one embodiment of the present invention, the deep learning model 1031 includes a temporal feature extraction network 1031a, a spatial feature extraction network 1031b, an attention mechanism module 1031c and a model adaptive update module 1031d.

[0096] The time series feature extraction network 1031a is used to model the time series of neural signals using a long short-term memory network structure, extract amplitude, frequency, and phase features, and analyze their time dependence. That is, the time series feature extraction network 1031a adopts a long short-term memory network (LSTM) structure to specifically process the time series features of neural signals. The network contains multiple layers of LSTM units, each of which has a three-gate structure of an input gate, a forget gate, and an output gate, which can effectively capture long-term and short-term time series dependencies. First, the input neural signal is divided into time windows, and then the features on each time scale are extracted through the LSTM layer. In the feature extraction process, the focus is on the amplitude change trend of the signal, the evolution of the frequency components, and the change in the phase relationship.

[0097] The spatial feature extraction network 1031b is used to use a graph convolutional neural network (GCN) to construct a spatial topology map with nerve segments as nodes, and analyze the spatial correlation and signal conduction characteristics between multiple nerve segments. Specifically, in the process of constructing the spatial topology map, each nerve segment is used as a node of the graph, and the anatomical and functional connection relationship between the segments is used as an edge, and the weight of the edge is dynamically adjusted according to the signal correlation. GCN aggregates and updates the node features through graph convolution operations, and can make full use of the spatial position relationship and signal conduction characteristics between nodes. The network adopts a multi-layer graph convolution structure to extract high-order spatial features layer by layer, and finally obtains a feature representation that reflects the complex interactions between nerve segments.

[0098] The attention mechanism module 1031c includes two sub-modules, channel attention and spatial attention, which are used to highlight the neural signal features at key time points and key spatial positions by learning the importance weights of signal features. Specifically, the channel attention sub-module learns the importance weights of different feature channels through the squeeze-and-excitation structure (i.e., SSE structure) to strengthen the extraction of key frequency components in the signal. The spatial attention sub-module uses a self-attention mechanism to calculate the correlation scores between spatial positions and highlight the positional features of abnormal neural segments. The outputs of the two sub-modules are integrated through a feature fusion network to form a comprehensive attention feature map, guiding the model to focus on spatiotemporal features with diagnostic value.

[0099] The model adaptive update module 1031d is used to optimize network parameters and improve the generalization and robustness of the model based on the online learning algorithm, using the treatment effect as a supervisory signal. Specifically, this module manages an experience replay buffer to store recent treatment cases and their effect evaluation results. By minimizing the error between the predicted results and the actual treatment effects, the network weights are updated by back propagation. At the same time, the module also includes a regularization mechanism and an early stopping strategy to prevent the model from overfitting and maintain good generalization ability. In order to cope with the individual differences of different patients, this module can adopt a meta-learning framework that can quickly adapt to new patient data distributions.

[0100] The above-mentioned deep learning model 1031 realizes multi-dimensional intelligent analysis of neural signals by combining four modules: temporal feature extraction, spatial feature extraction, attention mechanism and adaptive update. The model can not only accurately capture the temporal and spatial characteristics of neural signals, but also has the ability of continuous learning and adaptive optimization, which significantly improves the accuracy, reliability and adaptability of neural function detection.

[0101] Reference Figure 5 As shown, in one embodiment of the present invention, the treatment plan generator 1034 adopts a reinforcement learning algorithm, including a state space construction module 1034a, an action space construction module 1034b, a reward function design module 1034c and a policy network 1034d.

[0102] The state space construction module 1034a is used to parameterize the neural function state as a multidimensional feature vector, and the neural function state includes characteristic parameters of neural signals, physiological parameters and clinical indicators. Neural signal characteristic parameters such as frequency characteristics, amplitude characteristics, phase characteristics, etc., real-time physiological parameters such as nerve conduction velocity, heart rate variability, skin resistance, local body temperature, etc., clinical indicators such as symptom scores, functional assessment scales, etc. Preferably, the module uses feature normalization and dimensionality reduction techniques to unify parameters of different dimensions to a standard scale, and extracts the most representative feature combination through principal component analysis. In addition, the module also introduces a time series feature encoder to capture the dynamic change trend of state parameters and provide a complete state representation for the reinforcement learning algorithm.

[0103] The action space construction module 1034b is used to discretize the treatment parameters into a finite set of actions, including parameter combinations of three treatment modes: electromagnetic pulse, ultrasound and phototherapy. Exemplarily, this module discretizes the parameters of the three physical therapy modes: the frequency and intensity of electromagnetic pulse therapy are quantized into several gears (such as frequency is divided into three gears: low, medium and high, and intensity is divided into three gears: weak, medium and strong); the frequency and sound intensity of ultrasound therapy are also divided into grades; the wavelength selection and energy density of phototherapy are also mapped to the discrete parameter space. Through this parameter discretization design, a limited but sufficiently fine action space is formed, which not only ensures the controllability of the treatment parameters, but also significantly reduces the complexity of the decision space.

[0104] The reward function design module 1034c is used to integrate the three evaluation indicators of treatment effect, safety and comfort, and construct a quantitative reward calculation model. For example, the treatment effect can be evaluated based on the degree of improvement of neurological function, including changes in objective indicators and improvements in subjective symptoms; safety can monitor the fluctuation range of physiological parameters to ensure that it does not exceed the safety threshold; comfort is based on the treatment experience score reported by the patient. The reward calculation adopts a weighted summation method, and the weight coefficient is determined by expert experience and data analysis. The module also introduces a time decay factor to give a higher weight to recent treatment effects, and designs a phased reward mechanism to encourage continuous improvement during the treatment process.

[0105] The policy network 1034d is used to use a deep deterministic policy gradient algorithm to output the optimal treatment parameter combination through alternating optimization of the value network and the policy network 1034d. Specifically, the policy network 1034d uses a deep deterministic policy gradient (DDPG) algorithm to achieve the learning of the optimal treatment strategy. The network architecture includes two parts: the Actor network (policy network) and the Critic network (value network): the Actor network is responsible for directly outputting continuous treatment parameters according to the current state, using a deep neural network structure, and improving the accuracy of parameter prediction through multi-layer perceptrons and residual connections; the Critic network evaluates the action value and helps the policy network 1034d optimize parameter selection. The training process of the policy network 1034d adopts an experience replay mechanism to store and reuse historical treatment experience, and introduces a target network to stabilize the training process. In order to deal with the exploration-utilization trade-off of treatment decisions, the network uses a noise injection mechanism to maintain a large exploration space in the early stages of training, and gradually reduces the degree of exploration as the training proceeds.

[0106] This treatment plan generator 1034 based on reinforcement learning realizes intelligent control of treatment parameters through accurate construction of state space, reasonable discretization of action space, multi-dimensional evaluation of reward function and deep learning optimization of strategy network 1034d. It can autonomously learn the optimal treatment strategy, dynamically adjust the parameter combination according to the individual characteristics of the patient and treatment feedback, and maximize the treatment effect while ensuring the safety and comfort of treatment.

[0107] Reference Figure 6 As shown, in one embodiment of the present invention, it also includes a neural pathway imaging unit 106, and the neural pathway imaging unit 106 includes an electrical impedance tomography module 1061, a neural function area marking module 1062, a treatment process visualization module 1063 and an anatomical structure alignment module 1064.

[0108] The electrical impedance tomography module 1061 is used to use multi-frequency multi-channel electrical impedance measurement technology to reconstruct the conductivity distribution of nerve tissue by solving inverse problems and construct a three-dimensional model of the nerve conduction pathway. Specifically, by arranging multiple groups of excitation electrodes and measurement electrodes on the surface of the human body, weak current signals of different frequencies (typical range 1kHz-1MHz) are injected. The module adopts a time division multiplexing scheme to switch different excitation-measurement electrode pair combinations in turn to obtain a complete transimpedance measurement data set. During the data acquisition process, synchronous detection is used to improve the signal-to-noise ratio, and effective suppression of common-mode interference is achieved through reference electrodes. In the reconstruction algorithm, the Newton iteration method can be used to solve the nonlinear inverse problem, and the regularization constraint is combined to ensure the stability of the solution. Through the multi-frequency data fusion strategy, the dispersion characteristics of the tissue at different frequencies are extracted to achieve effective distinction between nerve tissue and surrounding structures. Finally, a three-dimensional conductivity distribution map with high spatial resolution is output to clearly show the anatomical direction of the nerve conduction pathway.

[0109] The neural function area marking module 1062 is used for automatically locating and marking abnormal nerve segments based on the image segmentation algorithm of deep learning, and displaying the quantitative score of the abnormal degree. In other words, the neural function area marking module 1062 is based on the deep learning framework and adopts the U-Net network structure for image segmentation. The network maintains high-resolution spatial detail information through the encoder-decoder structure and jump connection. During the network training process, a large amount of annotated clinical imaging data is used to construct a training set, and data enhancement technology is introduced to improve the generalization ability of the model. This module can adopt a multi-scale feature fusion mechanism, which can simultaneously process abnormal features of different spatial scales. For the identified abnormal areas, a scoring algorithm based on deep features is adopted to quantify the abnormal degree from multiple dimensions such as signal intensity, morphological features, texture features, etc., and the scoring results are intuitively displayed through color coding.

[0110] The treatment process visualization module 1063 is used to use three-dimensional rendering technology to display the spatial distribution and scope of treatment energy in real time, and to display the dynamic changes of tissue response during treatment through pseudo-color. Specifically, the treatment process visualization module 1063 uses real-time three-dimensional rendering technology based on GPU acceleration to build a visualization platform. In the display of treatment energy distribution, a ray tracing algorithm is used to simulate the energy transfer path, and the spatial distribution of energy is displayed through volume rendering technology. The energy intensity is displayed in the form of a heat map, and different intensity areas are clearly distinguished by gradient color scales. In the visualization of tissue response, a dynamic update mechanism is adopted, based on the real-time collected physiological parameter data, and the dynamic response characteristics such as tissue temperature changes and blood flow changes are displayed through pseudo-color mapping technology.

[0111] The anatomical structure registration module 1064 is used to adopt a non-rigid registration algorithm to spatially align the three-dimensional model of the nerve conduction pathway obtained by the electrical impedance tomography module 1061 and the abnormal nerve segment information marked by the nerve function area marking module 1062 with the standard human anatomical atlas to achieve accurate anatomical positioning of the abnormal nerve function area.

[0112] In other words, the anatomical structure registration module 1064 realizes the accurate spatial correspondence between the electrical impedance image and the standard anatomical atlas. The module adopts a hierarchical non-rigid registration strategy, firstly establishes a rough global correspondence through feature point matching, and then uses the B-spline deformation model for local fine registration. In the registration process, anatomical knowledge constraints are introduced to ensure that the deformation results conform to the physiological and anatomical characteristics. The module also includes a registration quality assessment mechanism, which quantifies the registration accuracy through multiple indicators such as mutual information and overlap coefficient. The registration results can be mapped bidirectionally, which can project abnormal markers onto the standard atlas, and can also superimpose anatomical structure information onto the original scanned image, providing accurate anatomical positioning reference for clinical diagnosis and treatment planning.

[0113] The neural pathway imaging unit 106 of this embodiment realizes accurate positioning and dynamic monitoring of abnormal neural function through the organic combination of electrical impedance imaging, deep learning segmentation, three-dimensional visualization and anatomical structure registration. This unit not only provides non-invasive and real-time neural conduction pathway imaging capabilities, but also can intuitively display the energy distribution and tissue response during the treatment process, significantly improving the accuracy and visualization of neural function detection and treatment.

[0114] Reference Figure 7 As shown, the neural function detection balance method based on artificial intelligence provided in an embodiment of the present invention is applied to the system as described above, including:

[0115] The detection unit 101 collects neural signals of the human nervous system, wherein the neural signals include bioelectric signals, pressure signals and temperature signals.

[0116] The signal processing unit 102 filters, amplifies and digitizes the neural signals collected by the detection module.

[0117] The intelligent analysis unit 103 establishes a neural function evaluation model and analyzes abnormal neural signals to obtain analysis results.

[0118] The treatment unit 104 performs positive stimulation treatment on the nerve segment according to the analysis result, and the positive stimulation treatment includes electromagnetic pulse treatment, ultrasonic treatment, and infrared light treatment.

[0119] The control unit 105 controls and adjusts the treatment parameters of the treatment unit 104 .

[0120] According to the artificial intelligence-based neural function detection and balance system and method provided by the embodiment of the present invention, a comprehensive collection of bioelectric signals, pressure signals and temperature signals of the nervous system is realized through a multimodal sensor array, and high-quality neural signal features are effectively extracted with the help of advanced signal processing technologies such as adaptive filtering, wavelet transform and independent component analysis. The deep learning model 1031 is used to analyze the time series and spatial characteristics of neural signals, and the expert system 1032 is combined to realize the accurate positioning and evaluation of abnormal nerve segments. In terms of treatment, through the coordinated application of various treatment methods such as electromagnetic pulses, ultrasound and phototherapy, the reinforcement learning algorithm is used to automatically generate personalized treatment plans, and the intelligent optimization of treatment parameters is realized. At the same time, the system also has a real-time feedback adjustment function, which can dynamically adjust the treatment strategy according to the changes in physiological parameters during the treatment process. Through electrical impedance tomography technology, the system visualizes the three-dimensional model of the neural conduction pathway and realizes real-time monitoring and precise positioning of the treatment process. This integrated and intelligent detection and treatment closed-loop system significantly improves the diagnosis and treatment accuracy and treatment effect of neurological dysfunction.

[0121] In this article, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0122] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0123] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A neural function detection and balance system based on artificial intelligence, characterized in that: include: A detection unit, comprising a multimodal sensor array, for collecting neural signals of a human nervous system, wherein the neural signals include bioelectric signals, pressure signals and temperature signals; A signal processing unit, electrically connected to the detection unit, for filtering, amplifying and digitally processing the neural signal collected by the detection module; An intelligent analysis unit, connected to the signal processing unit, is used to establish a neural function evaluation model and analyze abnormal neural signals to obtain analysis results; a treatment unit, connected to the intelligent analysis unit by signal, the treatment unit comprising an electromagnetic pulse generator, an ultrasonic generator and a phototherapy device, for performing positive stimulation treatment on the nerve segment according to the analysis result; The control unit is connected to the detection unit, the signal processing unit, the intelligent analysis unit and the treatment unit by signal, and is used for controlling and adjusting the treatment parameters of the treatment unit.

2. The neural function detection and balance system based on artificial intelligence according to claim 1 is characterized in that: The multimodal sensor array comprises: Surface electrode array and / or needle electrode array for collecting bioelectric signals of the human nervous system, with a sampling frequency of 1kHz-10kHz; A pressure sensor array for detecting nerve segment pressure signals with a detection accuracy of 0.01N-1N; Infrared temperature sensor array for detecting temperature signals at nerve conduction sites, with a detection accuracy of 0.1°C; The multimodal sensor array is arranged in the nerve segment-dense areas around the spine, wrists, and ankles.

3. The neural function detection and balance system based on artificial intelligence according to claim 1 is characterized in that: The intelligent analysis unit comprises: A deep learning model for analyzing characteristic parameters of neural signals, wherein the characteristic parameters include at least frequency, amplitude and phase; The expert system, based on the neuroanatomical database, compares the characteristic parameters of the detected neural signals with the standard database; A neurological function assessment module is used to output the location of abnormal nerve segments, the type and degree of abnormality, and obtain assessment results; A treatment plan generator automatically generates a personalized treatment plan based on the evaluation results and transmits it to the treatment unit.

4. The neural function detection and balance system based on artificial intelligence according to claim 1 is characterized in that: The electromagnetic pulse generator is used to generate therapeutic pulses with a frequency of 0.1-1000 Hz and a magnetic field strength of 0.1-2.0 T; The ultrasonic generator is used to generate 0.5-3MHz frequency, 0.1-2.0W / cm 2 Ultrasonic sound intensity; The phototherapy device is used to emit 780-1100nm near-infrared light and 3-100μm far-infrared light, with an energy density range of 1-100J / cm 2 .

5. The neural function detection and balance system based on artificial intelligence according to claim 1 is characterized in that: The treatment unit also includes a feedback module for real-time acquisition of changes in physiological parameters during the treatment process; the control unit is also used to adjust the treatment parameters according to the changes in the physiological parameters, and the physiological parameters include nerve conduction velocity, heart rate variability, skin resistance, and local body temperature.

6. The neural function detection and balance system based on artificial intelligence according to claim 1 is characterized in that: The signal processing unit comprises: An adaptive filter, wherein the adaptive filter adopts a minimum mean square error criterion to perform noise reduction and interference suppression processing on the neural signal; and automatically adjusts the filter coefficient and bandwidth parameter in real time according to the frequency characteristics, signal-to-noise ratio and interference type of different types of neural signals; A wavelet transform processing module, which is used to extract the frequency and amplitude characteristics of neural signals through multi-scale decomposition using a discrete wavelet transform algorithm; Independent component analysis module, which is used to separate multi-channel aliased neural signals into several independent components based on the fast ICA algorithm and utilize the statistical independence between signal sources to extract the target neural signals; The artifact recognition and elimination module is used to recognize and eliminate the characteristics of electrocardiogram, electromyography and motion artifact interference signals by using pattern recognition methods.

7. The neural function detection and balance system based on artificial intelligence according to claim 3 is characterized in that: The deep learning model includes: The time series feature extraction network is used to model the time series of neural signals using a long short-term memory network structure, extract amplitude, frequency, and phase features, and analyze their time dependence; A spatial feature extraction network is used to construct a spatial topology map using a graph convolutional neural network with nerve segments as nodes, and analyze the spatial correlation and signal conduction characteristics between multiple nerve segments; The attention mechanism module includes two sub-modules: channel attention and spatial attention. It is used to highlight the neural signal features at key time points and key spatial positions by learning the importance weights of signal features. The model adaptive update module is used to optimize network parameters based on the online learning algorithm and use the treatment effect as a supervisory signal to improve the generalization ability and robustness of the model.

8. The neural function detection and balance system based on artificial intelligence according to claim 3 is characterized in that: The treatment plan generator adopts a reinforcement learning algorithm, including: A state space construction module, used for parameterizing the neural function state into a multi-dimensional feature vector, wherein the neural function state includes characteristic parameters of neural signals, physiological parameters and clinical indicators; An action space construction module is used to discretize the treatment parameters into a finite set of actions, including the parameter combinations of three treatment modes: electromagnetic pulse, ultrasound and phototherapy; The reward function design module is used to comprehensively evaluate the three evaluation indicators of treatment effect, safety and comfort, and to build a quantitative reward calculation model; The policy network is used to output the optimal treatment parameter combination by adopting a deep deterministic policy gradient algorithm through alternating optimization of the value network and the policy network.

9. The neural function detection and balance system based on artificial intelligence according to claim 1 is characterized in that: Also included is a neural pathway imaging unit, the neural pathway imaging unit comprising: The electrical impedance tomography module is used to reconstruct the conductivity distribution of nerve tissue by using multi-frequency and multi-channel electrical impedance measurement technology and to construct a three-dimensional model of the nerve conduction pathway by solving inverse problems; The neural function area labeling module is used to automatically locate and label abnormal neural segments based on deep learning image segmentation algorithms, and display quantitative scores of abnormality levels; The treatment process visualization module is used to display the spatial distribution and scope of treatment energy in real time using 3D rendering technology, and to display the dynamic changes of tissue response during treatment using pseudo-color; The anatomical structure registration module is used to use a non-rigid registration algorithm to spatially align the three-dimensional model of the nerve conduction pathway obtained by the electrical impedance tomography module and the abnormal nerve segment information marked by the nerve function area marking module with the standard human anatomical atlas to achieve accurate anatomical positioning of the abnormal nerve function area.

10. A neural function detection balance method based on artificial intelligence, applied to the system according to any one of claims 1 to 9, characterized in that: include: The detection unit collects neural signals of the human nervous system, wherein the neural signals include bioelectric signals, pressure signals and temperature signals; The signal processing unit filters, amplifies and digitizes the neural signals collected by the detection module; The intelligent analysis unit establishes a neural function assessment model and analyzes abnormal neural signals to obtain analysis results; The treatment unit performs positive stimulation treatment on the nerve segment according to the analysis result, and the positive stimulation treatment includes electromagnetic pulse treatment, ultrasonic treatment, and infrared light treatment; The control unit controls and adjusts the treatment parameters of the treatment unit.

Images