CN-121648482-B - AI-driven transcranial optical biological regulation intelligent control method, equipment and storage medium

CN121648482BCN 121648482 BCN121648482 BCN 121648482BCN-121648482-B

Abstract

The present disclosure relates to AI-driven transcranial optical bioregulation intelligent control methods, apparatus, and storage media. The method comprises the steps of obtaining multiple groups of resting state data of a user of the transcranial optical biological regulating device and stimulation state data of controllable external stimulation, extracting time sequence features by an AI chip through a space-time convolution network, carrying out interactive modeling on the time sequence features of different modes through a transducer structure of a multi-head attention mechanism, dynamically calculating weights of various fused feature vectors through a self-attention mechanism, mapping to obtain physiological feature patterns, storing the physiological feature patterns as a personalized baseline model, calculating optimal working parameters through a deep reinforcement learning algorithm, and dynamically regulating the optimal working parameters according to the personalized baseline model and the deep reinforcement learning algorithm after the optimal working parameters are executed. The method and the device realize personalized and real-time adjustment of the stimulation parameters, do not need group average threshold information, and have high accuracy and good effect of adjustment control.

Inventors

YI MING
WAN YOU
WANG JIAXIN
WANG TING
CONG XIN
XING YUANWEI
Jiang Jiankuo

Assignees

北京大学
北大医学部(泰州)医药健康产业创新中心

Dates

Publication Date: 20260508
Application Date: 20260205

Claims (6)

An intelligent control method for AI-driven transcranial optical biological regulation, which is characterized by comprising the following steps: step 1, acquiring multiple groups of resting state data of a user of the transcranial optical biological regulation device, and acquiring stimulation state data of controllable external stimulation of the user of the transcranial optical biological regulation device; Step 2, the AI chip carries out preprocessing on the state data obtained in the step 1 to obtain preprocessed modal data, extracts time sequence features of the preprocessed modal data by utilizing a space-time convolution network, carries out interactive modeling on time sequence features of different modes by adopting a Transformer structure of a multi-head attention mechanism to obtain fused feature vectors, dynamically calculates weights of the fused feature vectors by a self-attention mechanism, and maps the fused feature vectors to a multidimensional physiological state space to obtain a physiological feature map; step 3, calculating the optimal working parameters of the transcranial optical biological regulating device used by the user by the AI chip according to the physiological characteristic map and through a deep reinforcement learning algorithm; step 4, executing optimal working parameters by the transcranial optical biological regulation device, monitoring physiological signals of a user of the transcranial optical biological regulation device in real time, analyzing the degree of deviation of the physiological signals from the normal range by using a deep reinforcement learning algorithm according to a personalized baseline model by an AI chip, and dynamically adjusting the optimal working parameters according to the degree; The state space of the deep reinforcement learning algorithm is composed of fused feature vectors, the action space is a set of working parameters, a reward function is expressed in a weighted form, and the reward function considers the symptom relieving speed of a user of the transcranial optical biological regulating device, the comfort of the user of the transcranial optical biological regulating device, the power consumption of the transcranial optical biological regulating device and the temperature control of the transcranial optical biological regulating device; The stimulation state data of the controllable external stimulus comprises an electroencephalogram signal under the stimulation of a cognitive task, a heart rate variability signal under the stimulation of the cognitive task, a skin surface potential variation signal under the stimulation of the cognitive task, an electroencephalogram signal under the stimulation of emotion, a heart rate variability signal under the stimulation of emotion, a skin surface potential variation signal under the stimulation of emotion, an electroencephalogram signal under the stimulation of emotion, a heart rate variability signal under the stimulation of sense organ and a skin surface potential variation signal under the stimulation of sense organ; The AI chip analyzes the degree of deviation of the physiological signal from the normal range of the physiological signal according to the personalized baseline model by utilizing a deep reinforcement learning algorithm and dynamically adjusts the optimal working parameter according to the degree, wherein the AI chip carries out similarity matching on the physiological signal and the personalized baseline model to judge the degree of deviation of the current state from the normal range, and the AI chip adjusts the optimal working parameter of the transcranial optical biological adjusting device according to the degree of deviation of the current state from the normal range by utilizing the deep reinforcement learning algorithm, so that the updated optimal working parameter can be re-executed in the step 4.
2. The AI-driven transcranial optical biological regulation intelligent control method according to claim 1, wherein the step 2 is implemented by an electroencephalogram classification model based on an identity embedding and cascading transducer architecture, and the electroencephalogram classification model based on the identity embedding and cascading transducer architecture comprises a shallow convolution feature coding module, a cascading transducer module for extracting coarse and fine granularity features and an identity embedding module; The shallow convolution feature coding module is used for extracting short-time dependence and local space dynamic features of each preprocessed modal data by using a space-time convolution network to obtain time sequence features, and the shallow convolution feature coding module is defined as follows: , Wherein, the Representing the output value of the shallow convolutional feature coding block, Representing the input of a convolutional feature encoder, The convolution function is represented as a function of the convolution, A batch normalization function is represented and, Representing an exponential linear element activation function, Representing the maximum pooling function, Representing a rearrangement function which, Representing a position code that is learnable; The cascade transducer module for extracting the coarse and fine granularity characteristics is used for interactively modeling the time sequence characteristics of different modes by adopting a transducer structure of a multi-head attention mechanism to obtain fused characteristic vectors, and each cascade transducer module for extracting the coarse and fine granularity characteristics of the cascade transducer module for extracting the coarse and fine granularity characteristics adopts the multi-head self-attention mechanism, so that the requirements are satisfied: , Wherein, the Representing a multi-headed attention mechanism function, The query vector is represented as a result of which, The key vector is represented as a vector of keys, A vector of values is represented and, The number of attention heads representing the multi-head attention mechanism, An index representing the attention header is shown, , Representative attention head Is used for the output matrix of the (c), An output matrix representing an attention header with index 0, An output matrix representing an attention header with index 1, Representing the index as Is provided with an output matrix of the attention head of (a), Represent the first A query matrix of the individual attention headers, Represent the first The key matrix of the individual attention header, Represent the first A matrix of values for the individual attention heads, Representing an attention function; the coarse-granularity embedding of the cascaded fransformer module for coarse-granularity feature extraction is defined as follows: , Wherein, the Represents the coarse granularity characteristics output by the transducer module for extracting the coarse granularity characteristics, Representing a forward-computing neural network, Representing a layer normalization function; fine-grained embedding of cascaded Transformer modules for coarse-fine grained feature extraction The definition is as follows: , The output of the cascaded fransformer module for coarse and fine granularity feature extraction is defined as: , Wherein, the Representing the final output of the cascaded fransformer module for coarse and fine granularity feature extraction, Represents the fine granularity characteristics output by the transducer module for extracting the coarse granularity characteristics, Representing a random inactivation function; the identity embedding module is used for embedding Mapping to an identity space, searching personalized embedded vectors of users, and fusing the personalized embedded vectors and the personalized embedded vectors And (3) performing individual brain state evaluation, wherein the following conditions are satisfied: , Wherein, the Representing the presentation to be The result of the mapping to the identity space, Representing the full-join transform function, The representation of the embedded function is made, Representation of The function ‌ is activated.
3. The AI-driven transcranial optical biological regulation intelligent control method according to claim 1, wherein training of the personalized baseline model is achieved through a cloud, preprocessing comprises band-pass filtering, denoising, artifact removal and standardization, the optimal working parameters comprise optimal combinations of various working parameters, and the working parameters comprise luminous area information, light intensity information, wavelength information and pulse frequency information.
AI-driven transcranial optical bioregulation intelligent control device, characterized in that it comprises: the biosensor module is used for acquiring multiple groups of resting state data of a user of the transcranial optical biological regulating device and acquiring stimulation state data of controllable external stimulation of the user of the transcranial optical biological regulating device; The AI chip is used for acquiring resting state data and stimulation state data of the biosensor module, preprocessing the state data to obtain preprocessed modal data, extracting time sequence features of the preprocessed modal data by utilizing a space-time convolution network, performing interactive modeling on the time sequence features of different modalities by adopting a transducer structure of a multi-head attention mechanism to obtain fused feature vectors, dynamically calculating weights of the fused feature vectors by a self-attention mechanism, mapping the fused feature vectors to a multidimensional physiological state space to obtain physiological feature patterns, and storing the physiological feature patterns as personalized baseline models; the AI chip is used for calculating the optimal working parameters of the transcranial optical biological adjustment device used by the user through a deep reinforcement learning algorithm according to the personalized baseline model; the optical modulation module is used for regulating and controlling the light waves of the transcranial optical biological regulation device according to the optimal working parameters; the power management module is used for supplying power to the working area of the transcranial optical biological regulation device according to the optimal working parameters; the biosensor module is also used for monitoring physiological signals of a user of the transcranial optical biological regulation device in real time when the transcranial optical biological regulation device executes the work of optimal working parameters; The AI chip is also used for obtaining the physiological signal monitored by the biosensor module in real time, analyzing the degree of deviation of the physiological signal from the normal range by using a deep reinforcement learning algorithm and dynamically adjusting the optimal working parameter according to the degree; The state space of the deep reinforcement learning algorithm is composed of fused feature vectors, the action space is a set of working parameters, a reward function is expressed in a weighted form, and the reward function considers the symptom relieving speed of a user of the transcranial optical biological regulating device, the comfort of the user of the transcranial optical biological regulating device, the power consumption of the transcranial optical biological regulating device and the temperature control of the transcranial optical biological regulating device; The stimulation state data of the controllable external stimulus comprises an electroencephalogram signal under the stimulation of a cognitive task, a heart rate variability signal under the stimulation of the cognitive task, a skin surface potential variation signal under the stimulation of the cognitive task, an electroencephalogram signal under the stimulation of emotion, a heart rate variability signal under the stimulation of emotion, a skin surface potential variation signal under the stimulation of emotion, an electroencephalogram signal under the stimulation of emotion, a heart rate variability signal under the stimulation of sense organ and a skin surface potential variation signal under the stimulation of sense organ; The AI chip is used for analyzing the degree of deviation of the physiological signal from the normal range of the physiological signal by utilizing a deep reinforcement learning algorithm according to the personalized baseline model and dynamically adjusting the optimal working parameter according to the degree, wherein the AI chip is particularly used for carrying out similarity matching on the physiological signal and the personalized baseline model to judge the degree of deviation of the current state from the normal range, the AI chip is used for adjusting the optimal working parameter of the transcranial optical biological regulating device according to the degree of deviation of the current state from the normal range by utilizing the deep reinforcement learning algorithm, and restarting the function of the biosensor module for monitoring the physiological signal of a user of the transcranial optical biological regulating device in real time when the transcranial optical biological regulating device executes the work of the optimal working parameter by utilizing the updated optimal working parameter.
5. The AI-driven transcranial optical biological regulation intelligent control device of claim 4 wherein the operating parameters include illumination location information, light intensity information, wavelength information, pulse frequency information, and the power management module supplies power to the operating region of the transcranial optical biological regulation apparatus based on the illumination location information, and the optical modulation module regulates the light waves of the transcranial optical biological regulation apparatus based on the light intensity information, the wavelength information, and the pulse frequency information.
6. The AI-driven transcranial optical bio-adjustment intelligent control device of claim 4, wherein training of the personalized baseline model is achieved through a cloud, the AI chip provides computing force through the AI chip or through the AI chip and the cloud, the biosensor module comprises an electroencephalogram sensor, a heart rate variability sensor and a skin electric response sensor, the optical modulation module comprises a light-giving system for regional luminescence, a temperature real-time monitoring system for real-time temperature monitoring of the light-giving system, and a temperature control system for performing constant temperature control on the light-giving system according to a temperature implementation monitoring result of the temperature real-time monitoring system, and the transcranial optical bio-adjustment device is a head-mounted transcranial optical bio-adjustment device.

Description

AI-driven transcranial optical biological regulation intelligent control method, equipment and storage medium Technical Field The disclosure relates to the technical field of intelligent parameter regulation and control, in particular to an intelligent control method, equipment and a storage medium for AI-driven transcranial optical biological regulation. Background Photo-biological modulation (PBM, photobiomodulation ‌) is a non-invasive technique for modulating biological tissue function by irradiation with light of a specific wavelength (typically 620-1100 nm). The action mechanism is mainly based on the absorption of cytochrome c oxidase in mitochondria to red light and near infrared light, and further a series of biochemical cascade reactions such as the increase of the yield of adenosine triphosphate, the release of nitric oxide, the regulation of active oxygen and the like are initiated. By virtue of non-invasive, high safety, etc., PBM has been developed for the treatment of various clinical diseases, such as improving ischemic stroke symptoms, treating traumatic brain injury, and intervening in mental diseases, neurological diseases (e.g., alzheimer's disease, parkinson's disease), neurological diseases, and improving cognitive memory in healthy individuals. The brain function is typically enhanced by wearable tPBM (TRANSCRANIAL PHOTOBIOMODULATION, transcranial optical bioregulation) devices or the like. Chinese patent publication No. CN119546229a entitled "method and apparatus for closed-loop transcranial optical bio-modulation stimulation using cognitive test" discloses a transcranial optical bio-modulation based physiological and neural stimulation device capable of providing continuous wave or pulsed light source for stimulating specific brain region, and also evaluating tPBM intervention effect by using physiological signals such as electroencephalogram (Electroencephalography, EEG) or heart rate variability (HEART RATE Variability, HRV), or modifying tPBM parameters according to these signals, in which the technical scheme is directed to cognitive test task, and the transcranial optical bio-modulation is combined with Event-related potential (Event-Related Potentials, ERP) recorded in the cognitive test, and the neural feedback modulation function is realized by time-frequency analysis of ERP and monitoring of other physiological signals. The existing transcranial optical biological regulation control method depends on a group average threshold value, cannot meet individual requirements, cannot respond to individual differences of patients, and has low regulation control accuracy and poor effect. Disclosure of Invention Based on this, it is necessary to provide an AI-driven intelligent control method, apparatus and storage medium for transcranial optical bio-modulation in view of the above-described problems. In order to solve the problems, the present disclosure adopts the following technical scheme: in a first aspect, the present disclosure provides an AI-driven transcranial optical bioregulation intelligent control method comprising the steps of: step 1, acquiring multiple groups of resting state data of a user of the transcranial optical biological regulation device, and acquiring stimulation state data of controllable external stimulation of the user of the transcranial optical biological regulation device; Step 2, the AI chip carries out preprocessing on the state data obtained in the step 1 to obtain preprocessed modal data, extracts time sequence features of the preprocessed modal data by utilizing a space-time convolution network, carries out interactive modeling on time sequence features of different modes by adopting a Transformer structure of a multi-head attention mechanism to obtain fused feature vectors, dynamically calculates weights of the fused feature vectors by a self-attention mechanism, and maps the fused feature vectors to a multidimensional physiological state space to obtain a physiological feature map; step 3, calculating the optimal working parameters of the transcranial optical biological regulating device used by the user by the AI chip according to the physiological characteristic map and through a deep reinforcement learning algorithm; And 4, executing optimal working parameters by the transcranial optical biological adjusting device, monitoring physiological signals of a user of the transcranial optical biological adjusting device in real time, analyzing the degree of deviation of the physiological signals from the normal range by using a deep reinforcement learning algorithm according to a personalized baseline model by an AI chip, and dynamically adjusting the optimal working parameters according to the degree. In a second aspect, the present disclosure provides an AI-driven transcranial optical bioregulation intelligent control device comprising: the biosensor module is used for acquiring multiple groups of resting state data of a user of the transcranial op