Intelligent cooperative control method and system for multi-mode phototherapy equipment

Abstract

The invention provides an intelligent cooperative control method and system for a multi-mode phototherapy device, and the method comprises the steps: obtaining physiological parameters of a plurality of users to form a data set, and constructing a spectrum feedback model in combination with historical phototherapy data and environment data; on the basis of the model and historical phototherapy target data, a multi-mode spectrum parameter model is generated through a preset neural network, and cooperative control over multi-light-source equipment is achieved; physiological parameter changes of a target user are monitored in real time, and monitoring data are sent to an edge computing node; the edge node dynamically adjusts a spectral parameter combination scheme according to the monitoring data and a safety threshold, and generates a treatment effect evaluation index; and establishing a user spectrum feedback feature library based on the evaluation indexes, and continuously optimizing a spectrum feedback model through incremental learning. According to the invention, intelligent cooperative control of the multi-mode phototherapy equipment can be realized, the energy utilization efficiency of the phototherapy equipment is improved, the spectrum feedback difference influence among different users is reduced, and the accuracy and safety of the phototherapy effect are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control, and particularly relates to an intelligent collaborative control method and system of a multi-modal light therapy device. BACKGROUND

[0002] The light therapy device is a therapeutic instrument taking light as an energy source, which mainly treats the treatment area through light emitted by a light therapy head. The existing light therapy device is not intelligent enough in control, has poor light therapy effect and has safety hazards. SUMMARY

[0003] The present application is based on the above problems, and proposes an intelligent collaborative control method and system of a multi-modal light therapy device. Through the present application, the intelligent collaborative control of the multi-modal light therapy device can be realized, the energy utilization efficiency of the light therapy device is improved, the influence of the spectral feedback difference between different users is reduced, and the accuracy and safety of the light therapy effect are enhanced.

[0004] Therefore, one aspect of the present application proposes an intelligent collaborative control method of a multi-modal light therapy device, comprising:

[0005] Obtaining physiological parameters of a plurality of first users to obtain a physiological data set;

[0006] Obtaining historical light therapy data of the first user and historical environmental data corresponding to the historical light therapy data;

[0007] Based on the physiological data set, the historical light therapy data and the historical environmental data, a spectral feedback model is constructed;

[0008] According to the spectral feedback model and historical light therapy target data extracted from the historical light therapy data, a multi-modal spectral parameter model is generated in combination with a preset first neural network;

[0009] According to the multi-modal spectral parameter model, a plurality of different light sources in the multi-modal light therapy device are collaboratively controlled;

[0010] Real-time monitoring of changes in physiological parameters of a target user obtains monitoring data, and the monitoring data is sent to an edge computing node;

[0011] The edge computing node dynamically adjusts the spectral parameter combination scheme in the multi-modal spectral parameter model according to the monitoring data and a preset safety threshold through a reinforcement learning algorithm, and simultaneously generates a treatment effect evaluation index;

[0012] Based on the treatment effect evaluation index, a user spectral feedback feature library is established, and the spectral feedback model is continuously optimized through an incremental learning method.

[0013] Optionally, the step of constructing a spectral feedback model based on the physiological data set, the historical phototherapy data, and the historical environmental data includes:

[0014] The physiological data set is layered according to tissue levels to form layer parameter sets for the epidermis, dermis, and subcutaneous tissue layers;

[0015] Applying wavelet transform to perform feature extraction on the layer parameter set of each layer, the historical phototherapy data and the historical environment data to identify the absorption and scattering characteristics of each layer of tissue for different wavelength spectra under different environments and different phototherapy control parameters;

[0016] Based on the photon transmission Monte Carlo simulation technology, a spectral feedback prediction model for each layer of tissue is established;

[0017] The spectral feedback prediction models of each layer are integrated through a bidirectional long short-term memory network, and a coupling feedback mechanism reflecting the inter-layer spectral energy transfer effect is constructed.

[0018] Optionally, the first neural network includes:

[0019] A feature fusion layer based on graph neural network to process the correlation between physiological data of different dimensions;

[0020] Multiple parallel spectral parameter optimization sub-networks, each sub-network corresponds to the optimization calculation of a specific spectral parameter;

[0021] An attention-based parameter combination layer to generate the final multimodal spectral parameter model;

[0022] An uncertainty quantification module is used to evaluate the reliability of the generated multimodal spectral parameter models.

[0023] Optionally, the step of integrating the spectral feedback prediction models of each layer through a bidirectional long short-term memory network and constructing a coupling feedback mechanism that reflects the inter-layer spectral energy transfer effect includes:

[0024] A single-layer feedback sub-model was constructed, including: for the epidermis, a spectral absorption-scattering parameter mapping based on the improved Kuz-Mink model was established, focusing on characterizing the absorption characteristics of melanin to light of different wavelengths; for the dermis, an optical interaction model of collagen and elastic fibers based on the finite-difference time-domain method was constructed, focusing on characterizing the response of vascular microcirculation to spectral energy; for the subcutaneous tissue layer, a deep tissue light diffusion model based on the radiation transfer equation was established, focusing on characterizing the scattering and absorption characteristics of fat and connective tissue to light; a Bayesian optimization algorithm was applied to each sub-model, and model parameters were adjusted based on historical data to minimize the root mean square error between the model prediction value and the measured value;

[0025] A sub-model coupling feedback mechanism is constructed, including: establishing an inter-layer spectral energy transfer matrix to describe the energy attenuation law of different wavelengths of light during the transmission process from the surface layer to the deep layer; constructing an inter-layer heat conduction model to represent the diffusion process of light energy converted into heat energy among the layers of tissues; designing a tissue physiological response conversion function to map the local temperature change to the dynamic change of the physiological parameters; integrating the above three components to form a closed-loop feedback path, so that the response change of the deep tissue can inversely affect the optical properties of the surface tissue;

[0026] Model integration and verification, including: using a hierarchical recurrent neural network structure to integrate each sub-model as a network node into a unified spectral feedback model; introducing a time dimension to construct a time series prediction mechanism based on a gated recurrent unit, so that the model can predict the dynamic changes of the tissue response during the illumination process; using an independent test data set to evaluate the prediction accuracy of the integrated model through cross-validation method; adjusting the model integration parameters based on the verification results to optimize the weight distribution among the sub-models.

[0027] Optionally, the step of generating a multi-modal spectral parameter model based on the spectral feedback model and the historical phototherapy target data extracted from the historical phototherapy data, in combination with a preset first neural network, comprises:

[0028] Extracting feedback feature vectors of the epidermis layer, the dermis layer, and the subcutaneous tissue layer from the spectral feedback model, each feedback feature vector containing spectral absorption coefficients, scattering coefficients, heat conduction coefficients, and physiological response sensitivity parameters;

[0029] Standardizing each layer feedback feature vector to eliminate the influence of different physical dimensions and numerical ranges;

[0030] Constructing an inter-layer coupling relationship matrix to represent the mutual influence strength between different layers, generating a 144-dimensional coupling feature descriptor;

[0031] Performing dimensionality reduction processing on the coupling feature descriptor based on principal component analysis to generate a 64-dimensional compact feature representation;

[0032] Constructing a first neural network and performing wavelength selection and energy distribution calculation, time series parameter optimization calculation, and spatial distribution pattern generation based on the first neural network;

[0033] Integrating to generate a multi-modal spectral parameter model and verifying it.

[0034] Optionally, the step of cooperatively controlling a plurality of different light sources in the multi-modal phototherapy device based on the multi-modal spectral parameter model comprises:

[0035] Resolving a spectral parameter combination scheme from the multi-modal spectral parameter model;

[0036] A first near field communication network is constructed between the main controller of the light therapy device and the micro control units of the light sources for cooperative control;

[0037] The micro control units are initialized and parameter configured;

[0038] The main controller generates a global timing schedule table based on the spectral parameter combination scheme, and decomposes the global timing schedule table into local scheduling instructions for each micro control unit;

[0039] The local scheduling instructions are sent to the corresponding micro control units in time sequence in batches through the first near field communication network;

[0040] After receiving the scheduling instructions, each micro control unit stores them in a local instruction queue and sorts them according to the time stamp;

[0041] A distributed clock synchronization mechanism is established, and all micro control units take the clock of the main controller as the reference to maintain time synchronization through the network time protocol;

[0042] An instruction preloading mechanism is designed to load the instructions to be executed into an execution buffer;

[0043] Each micro control unit performs real-time cooperation and monitoring according to the local scheduling instructions;

[0044] A cooperative effect evaluation model is established to perform cooperative optimization and adaptive adjustment.

[0045] Optionally, the edge computing node dynamically adjusts the spectral parameter combination scheme in the multi-modal spectral parameter model according to the monitoring data and the preset safety threshold through a reinforcement learning algorithm, and simultaneously generates a treatment effect evaluation index, including:

[0046] The monitoring data is preprocessed;

[0047] The safety threshold is dynamically evaluated and updated in combination with the preprocessed monitoring data;

[0048] Reinforcement learning environment modeling and state definition are performed;

[0049] A deep Q network algorithm is constructed;

[0050] Parameter adjustment decisions are generated and executed based on the constructed reinforcement learning model and Q network algorithm;

[0051] A multi-dimensional effect evaluation system is established, and a time decay weight function is designed;

[0052] A personal baseline comparison mechanism is established to compare the current effect with the user's historical best effect and generate an improvement index;

[0053] An integrated score algorithm is constructed to integrate the indicators of each dimension into a comprehensive score of 0-100 points, and a detailed effect analysis report is generated.

[0054] Optionally, based on the treatment effect evaluation index, the step of establishing a user spectrum feedback feature library and continuously optimizing the spectrum feedback model through incremental learning includes:

[0055] Based on the treatment effect evaluation index, extract the corresponding user spectrum feedback features, including the sensitivity coefficient of the user to different wavelengths of light spectrum, the spectrum energy absorption rate, and the spectrum penetration depth parameter;

[0056] Correlate and label the spectrum feedback features with the corresponding treatment effect evaluation indexes to form user spectrum feedback feature samples;

[0057] Store the user spectrum feedback feature samples according to user age, gender, skin color type, and treatment site to obtain an initial user spectrum feedback feature library;

[0058] Establish an indexing mechanism in the initial user spectrum feedback feature library, which includes a first-level index based on user physiological characteristics and a second-level index based on treatment effect types, and assign a unique identifier to each user spectrum feedback feature sample and record its creation time, update time, and usage frequency to obtain a user spectrum feedback feature library;

[0059] Extract new user spectrum response feature samples from the user spectrum feedback feature library as input data for incremental learning;

[0060] Perform data cleaning and normalization on the new samples to eliminate outliers and unify the data format;

[0061] Calculate the similarity between the preprocessed new samples and historical training samples to identify representative incremental learning samples;

[0062] Based on the incremental learning samples, update the parameters of the epidermis response sub-model, the dermis response sub-model, and the subcutaneous tissue response sub-model using an online learning algorithm;

[0063] During model updating, keep the core weight parameters of the original model and fine-tune the local parameters related to the new samples;

[0064] Evaluate the prediction accuracy of the updated model through cross-validation method, and when the accuracy improvement reaches the preset threshold, solidify and store the updated model parameters;

[0065] The updated multi-level spectral response model was used to retrospectively verify historical treatment cases, and the model's prediction accuracy and treatment effect compliance rate were calculated.

[0066] When the model performance index is lower than the preset standard, the model retraining mechanism is automatically triggered, and more diverse training samples are added to the user's spectral response feature library;

[0067] The model performance evaluation results are fed back to the first neural network to achieve closed-loop optimization of the overall control system.

[0068] Optionally, the method for constructing the inter-layer spectral energy transfer matrix includes:

[0069] The skin tissue is divided into m discrete layers, each layer no thicker than 0.2 mm;

[0070] For each layer, calculate the absorption coefficient μa and scattering coefficient μ of the tissue layer for each wavelength of light s ;

[0071] Based on Monte Carlo photon transport simulation, the probability distribution of photons transmitted from layer i to layer j is calculated;

[0072] An m×m dimensional transfer matrix T is constructed, where the element T(i, j) represents the proportion of light energy incident on the i-th layer transferred to the j-th layer.

[0073] Another aspect of the present invention provides an intelligent collaborative control system for a multimodal phototherapy device, which is used to perform an intelligent collaborative control method for a multimodal phototherapy device, comprising: a multimodal phototherapy device, an edge computing node, and a cloud server;

[0074] The cloud server is configured as follows:

[0075] Acquiring physiological parameters of a plurality of first users to obtain a physiological data set;

[0076] Acquiring historical phototherapy data of the first user and historical environmental data corresponding to the historical phototherapy data;

[0077] constructing a spectral feedback model based on the physiological data set, the historical light therapy data, and the historical environmental data;

[0078] generating a multimodal spectral parameter model based on the spectral feedback model and the historical phototherapy target data extracted from the historical phototherapy data in combination with a preset first neural network;

[0079] Coordinately controlling multiple different light sources in a multimodal phototherapy device according to the multimodal spectral parameter model;

[0080] Real-time monitoring of physiological parameter changes of target users to obtain monitoring data, and sending the monitoring data to the edge computing node;

[0081] Based on the treatment effect evaluation indicators, a user spectral feedback feature library is established, and the spectral feedback model is continuously optimized through incremental learning;

[0082] The edge computing node is configured to dynamically adjust the spectral parameter combination scheme in the multimodal spectral parameter model through a reinforcement learning algorithm according to the monitoring data and a preset safety threshold, and generate a treatment effect evaluation indicator.

[0083] Using the technical solution of the present invention, the intelligent collaborative control method of a multimodal phototherapy device includes: obtaining physiological parameters of multiple first users to obtain a physiological data set; obtaining historical phototherapy data of the first users and historical environmental data corresponding to the historical phototherapy data; constructing a spectral feedback model based on the physiological data set, the historical phototherapy data and the historical environmental data; generating a multimodal spectral parameter model based on the spectral feedback model and the historical phototherapy target data extracted from the historical phototherapy data in combination with a preset first neural network; collaboratively controlling multiple different light sources in the multimodal phototherapy device based on the multimodal spectral parameter model; real-time monitoring of changes in the physiological parameters of the target user to obtain monitoring data, and sending the monitoring data to an edge computing node; the edge computing node dynamically adjusts the spectral parameter combination scheme in the multimodal spectral parameter model through a reinforcement learning algorithm based on the monitoring data and a preset safety threshold, and simultaneously generates a treatment effect evaluation index; based on the treatment effect evaluation index, establishing a user spectral feedback feature library, and continuously optimizing the spectral feedback model through incremental learning. Through the solution of the present invention, intelligent collaborative control of multimodal phototherapy equipment can be achieved, the energy utilization efficiency of the phototherapy equipment is improved, the impact of spectral feedback differences between different users is reduced, and the accuracy and safety of the phototherapy effect are enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Fig. 1 This is a flow chart of an intelligent collaborative control method for a multimodal phototherapy device provided by one embodiment of the present invention;

[0085] Fig. 2 It is a schematic block diagram of an intelligent collaborative control system for a multimodal phototherapy device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0086] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0087] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description.

[0088] The terms "first", "second", and the like in the description and in the claims of the present application and in the above drawings refer to different objects, not to a particular order. In addition, the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product, or device.

[0089] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment can be included in at least one embodiment of the application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily mutually exclusive of other embodiments. It is explicitly and implicitly understood that the embodiments described herein can be combined with other embodiments.

[0090] The intelligent collaborative control method and system of a multi-modal light therapy device according to some embodiments of the present application will be described below with reference to Figs. 1-2 .

[0091] As shown in Fig. 1 , one embodiment of the present application provides an intelligent collaborative control method of a multi-modal light therapy device, comprising:

[0092] Obtaining physiological parameters of a plurality of first users to obtain a physiological data set;

[0093] The physiological data set includes skin temperature, skin spectral reflectance, skin water content, melanin concentration, hemoglobin oxygenation state, hyaluronic acid content, and microcirculation blood flow, etc.

[0094] Obtaining historical light therapy data of the first user and historical environmental data corresponding to the historical light therapy data;

[0095] The historical light therapy data includes pressure distribution data of the multi-modal light therapy device on the skin, light therapy control parameters (such as wavelength, power, pulse frequency, duty cycle, etc.) of each light source module, and other working parameters of the multi-modal light therapy device, etc.

[0096] Based on the physiological data set, the historical light therapy data, and the historical environmental data, a spectral feedback model is constructed;

[0097] The spectral feedback model includes an epidermis layer feedback sub-model, a dermis layer feedback sub-model, and a subcutaneous tissue feedback sub-model, and a coupling feedback mechanism is formed between the sub-models.

[0098] According to the spectral feedback model and historical phototherapy target data extracted from the historical phototherapy data, a multi-modal spectral parameter model is generated in combination with a preset first neural network.

[0099] The multi-modal spectral parameter model includes spectral energy distribution of multiple different wavelengths, pulse frequency, irradiation time, and spatial distribution mode. It can be understood that the phototherapy system does not use a single type of light, but rather uses at least three different wavelengths of light, and for each wavelength of light, the system precisely controls the following four key parameters: spectral energy distribution, which refers to the energy intensity distribution in each wavelength range, for example, near-infrared light (800 nm) can be configured as 20 mW / cm 2 , visible light (600 nm) is configured as 15 mW / cm 2 , and ultraviolet light (360 nm) is configured as 10 mW / cm 2 . This distribution is precisely designed based on the different penetration depths and mechanisms of different wavelengths on the various layers of skin tissue; pulse frequency, which refers to the rhythm pattern of light source emission rather than continuous irradiation, for example, some wavelengths can use a pulse frequency of 10 Hz (flicker 10 times per second), and other wavelengths can use 50 Hz. The pulse mode can avoid tissue overheating, while at certain specific frequencies, it can enhance biological response; irradiation time, which refers to the working duration of each wavelength light source, for example, near-infrared light can need to be continuously irradiated for 15 minutes, while ultraviolet light can only need 3 minutes, which is determined based on the mechanism and safety considerations of different wavelengths of light; spatial distribution mode, which refers to the irradiation distribution pattern of the light source in the treatment area, for example, some wavelengths can be concentrated in the center of the treatment area, while other wavelengths can be uniformly distributed or in a specific geometric pattern, and this spatial distribution can be achieved through controllable diffraction gratings and optical lens systems to ensure that tissues at different depths and areas can receive the optimal dose of light. This multi-parameter combination method realizes the synergistic effect of multi-wavelength light, which can act on tissues at different depths at the same time; by precisely controlling each parameter, the problem of single light source being difficult to balance penetration depth and safety is avoided; personalized light programs are realized, providing customized treatment for different users' skin conditions; energy utilization efficiency is improved, unnecessary light is reduced, and the risk of thermal effect and potential harm is reduced.

[0100] According to the multi-modal spectral parameter model, multiple different light sources in the multi-modal phototherapy device are cooperatively controlled.

[0101] Each light source in the multi-modal light therapy device is configured with an independent control unit, and the control units share information and work cooperatively through a communication network.

[0102] Real-time monitoring of the physiological parameter changes of the target user obtains monitoring data, and the monitoring data is sent to an edge computing node;

[0103] The edge computing node dynamically adjusts the spectral parameter combination scheme in the multi-modal spectral parameter model according to the monitoring data and a preset safety threshold through a reinforcement learning algorithm, and simultaneously generates a treatment effect evaluation index;

[0104] Based on the treatment effect evaluation index, a user spectral feedback feature library is established, and the spectral feedback model is continuously optimized through an incremental learning method.

[0105] The technical scheme of this embodiment can realize intelligent collaborative control of the multi-modal light therapy device, improve the energy utilization efficiency of the light therapy device, reduce the influence of spectral feedback differences between different users, and enhance the accuracy and safety of the light therapy effect.

[0106] In some possible embodiments of the present application, the step of constructing the spectral feedback model based on the physiological data set, the historical light therapy data and the historical environmental data comprises:

[0107] The physiological data set is hierarchically processed according to the organization level to form layer parameter sets of the epidermis layer, the dermis layer and the subcutaneous tissue layer;

[0108] Wavelet transform is applied to each layer parameter set, the historical light therapy data and the historical environmental data to extract features and identify the absorption and scattering characteristics of different wavelengths of light spectrum for each layer of tissue under different environments and different light therapy control parameters;

[0109] Based on the photon transport Monte Carlo simulation technology, a spectral feedback prediction model of each layer of tissue is established;

[0110] A coupling feedback mechanism reflecting the interlayer spectral energy transfer effect is constructed by integrating the spectral feedback prediction models of each layer (including the epidermis layer feedback submodel, the dermis layer feedback submodel and the subcutaneous tissue feedback submodel, etc.) through a bidirectional long short-term memory network (Bi-LSTM).

[0111] In this embodiment, the prediction accuracy of the model for multi-level spectral feedback of the tissue is improved, providing a reliable basis for subsequent calculation of the spectral parameter combination scheme.

[0112] In some possible embodiments of the present application, the first neural network comprises:

[0113] A feature fusion layer based on graph neural network to process the correlation between physiological data of different dimensions;

[0114] Multiple parallel spectral parameter optimization sub-networks, each sub-network corresponds to the optimization calculation of a specific spectral parameter;

[0115] An attention-based parameter combination layer to generate the final multimodal spectral parameter model;

[0116] An uncertainty quantification module is used to evaluate the reliability of the generated multimodal spectral parameter models.

[0117] The solution of this embodiment improves the personalized accuracy of the spectral parameter combination solution, and at the same time ensures the safety and reliability of the parameter solution through uncertainty quantification.

[0118] In some possible implementations of the present invention, the step of integrating the spectral feedback prediction models of each layer through a bidirectional long short-term memory network and constructing a coupled feedback mechanism that reflects the inter-layer spectral energy transfer effect includes:

[0119] The steps for constructing a single-layer feedback sub-model include: for the epidermis, establishing a spectral absorption-scattering parameter mapping based on the modified Kubelka-Munk Model, focusing on characterizing the absorption characteristics of melanin to light of different wavelengths; for the dermis, constructing an optical interaction model of collagen and elastic fibers based on the finite-difference time-domain method (FDTD), focusing on characterizing the response of vascular microcirculation to spectral energy; for the subcutaneous tissue layer, establishing a deep tissue light diffusion model based on the radiative transfer equation (RTE), focusing on characterizing the scattering and absorption characteristics of fat and connective tissue to light; applying the Bayesian optimization algorithm to each sub-model, adjusting the model parameters based on historical data, and minimizing the root mean square error between the model prediction value and the measured value; through the above steps, a dedicated spectral feedback prediction model (i.e., spectral feedback sub-model) for different anatomical layers of the skin is constructed, achieving an accurate description of the specific optical behavior of each layer of tissue.

[0120] The steps for constructing the sub-model coupling feedback mechanism include: establishing an interlayer spectral energy transfer matrix to describe the energy attenuation law during the transmission of light of different wavelengths from the surface to the deep layer; constructing an interlayer heat conduction model to characterize the diffusion process between tissue layers after light energy is converted into heat energy; designing a tissue physiological response conversion function to map local temperature changes into dynamic changes in physiological parameters such as blood flow and oxygen content; integrating the above three components to form a closed-loop feedback path, so that the response changes of deep tissue can reversely affect the optical properties of surface tissue; through the above steps, dynamic coupling between different layers of the skin is achieved, overcoming the limitation of traditional single-layer models that ignore interlayer interactions.

[0121] The model integration and verification steps include: using a hierarchical recurrent neural network (RNN) structure to integrate each sub-model as a network node into a unified spectral feedback model; introducing the time dimension and constructing a timing prediction mechanism based on gated recurrent units (GRUs) to enable the model to predict the dynamic changes in tissue response during illumination; using an independent test dataset through cross-validation to evaluate the predictive accuracy of the integrated model; adjusting the model integration parameters based on the verification results and optimizing the weight distribution between sub-models. Through these steps, a complete spectral feedback model with spatiotemporal dynamic prediction capabilities was formed, achieving a comprehensive characterization of the multi-level tissue response during phototherapy.

[0122] The adaptive optimization and personalized adjustment steps include: designing a differential sensitivity function to quantify the deviation between the model prediction value and the actual monitored value; using an incremental learning algorithm to dynamically update the model parameters based on the real-time feedback data monitored; constructing individual user feature vectors and incorporating user-specific factors (such as age, gender, skin color, etc.) into the model adjustment process; establishing model interpretability indicators and generating parameter sensitivity analysis reports, so that the system can identify and optimize key influencing factors; through the above steps, the model's continuous self-optimization and personalized adaptation are achieved, ensuring the model's universality and pertinence among different users.

[0123] In some possible implementations of the present invention, the interlayer heat conduction model is implemented using a three-dimensional finite element method, specifically including:

[0124] Establish a three-dimensional skin tissue geometric model, including the anatomical structure of the epidermis, dermis and subcutaneous tissue;

[0125] Set the thermophysical parameters of each layer of tissue, including thermal conductivity, specific heat capacity and density;

[0126] Convert light energy absorption into volume heat source, the intensity of the heat source is proportional to the local light energy absorption rate;

[0127] Solve the biological heat conduction equation (Pennes equation) and calculate the distribution of temperature field over time and space;

[0128] Considering the heat dissipation effect of blood flow, blood perfusion rate is introduced as a temperature-dependent variable;

[0129] This embodiment achieves accurate simulation of the photothermal conversion process and provides dynamic distribution data of the temperature field for predicting the physiological response of tissues.

[0130] In some possible implementations of the present invention, the step of generating a multimodal spectral parameter model based on the spectral feedback model and the historical phototherapy target data extracted from the historical phototherapy data in combination with a preset first neural network includes:

[0131] extracting feedback feature vectors of epidermis layer, dermis layer and subcutaneous tissue layer from the spectral feedback model, each feedback feature vector containing spectral absorption coefficient, scattering coefficient, thermal conductivity coefficient and physiological response sensitivity parameter;

[0132] normalizing the feedback feature vectors of each layer to eliminate the influence of different physical dimensions and numerical ranges;

[0133] constructing an interlayer coupling relationship matrix to represent the mutual influence strength between different layers, and generating a 144-dimensional coupling feature descriptor;

[0134] performing dimensionality reduction processing on the coupling feature descriptor based on principal component analysis to generate a 64-dimensional compact feature representation;

[0135] In this embodiment, the above steps realize efficient encoding and feature extraction of the spectral feedback model output information, and provide structured input data for the neural network.

[0136] constructing a first neural network and performing wavelength selection and energy distribution calculation, timing parameter optimization calculation, and spatial distribution pattern generation operations according to the first neural network;

[0137] constructing the first neural network, including: designing a multi-branch feature fusion layer containing three parallel convolutional neural network branches to process feature information (compact feature representation) of the epidermis layer, dermis layer and subcutaneous tissue layer respectively; constructing a cross-layer attention mechanism module to calculate the correlation weights between different layer features through self-attention and generate a 64x64 attention weight matrix; designing a parameter decoupling network structure containing four special subnetworks: wavelength selection network, energy distribution network, timing parameter network and spatial distribution network; introducing residual connection and batch normalization layer to prevent gradient disappearance and accelerate network convergence; designing a dynamic activation function selection mechanism to dynamically select ReLU, Swish or GELU activation function according to input features; extracting multiple treatment target features from historical phototherapy target data and constructing a multi-objective optimization constraint function according to the treatment target features; through the above steps, an adaptive neural network architecture with multi-branch and multi-level processing capability is constructed, which can simultaneously optimize multiple associated spectral parameters.

[0138] In some possible embodiments of the present application, the step of extracting multiple treatment target features from historical phototherapy target data and constructing a multi-objective optimization constraint function according to the treatment target features includes:

[0139] extracting historical phototherapy treatment records from a user spectral feedback feature library, the historical phototherapy treatment records including pre-treatment and post-treatment skin state comparison data, spectral parameter setting records and treatment effect evaluation results;

[0140] Data cleaning is performed on the historical phototherapy treatment records to remove invalid data with insufficient treatment time, incomplete parameter records, and missing effect evaluation;

[0141] The effective data is classified and labeled according to the treatment target types, including skin repair, pigment improvement, blood circulation promotion, and anti-aging;

[0142] Based on the obtained classification data, the skin repair treatment target features are extracted, including skin barrier function recovery index, epidermal cell regeneration rate, and inflammation subsidence degree;

[0143] The pigment improvement treatment target features are extracted, including melanin distribution uniformity, spot area reduction rate, and skin brightness improvement degree;

[0144] The blood circulation promotion treatment target features are extracted, including microvessel expansion degree, blood flow velocity growth rate, and tissue oxygenation degree improvement rate;

[0145] The anti-aging treatment target features are extracted, including collagen synthesis activation degree, skin elasticity recovery rate, and fine line depth reduction rate;

[0146] The above-mentioned various treatment target features are numerically quantified to establish a standardized feature evaluation system;

[0147] Pearson correlation coefficient analysis method is used to calculate the correlation strength between each treatment target feature;

[0148] The treatment target feature combinations with synergistic effect and the treatment target feature pairs with competitive relationship are identified;

[0149] Based on the correlation analysis results, a treatment target feature weight matrix is constructed, which reflects the contribution degree of different features to the overall treatment effect;

[0150] Based on the obtained treatment target feature weight matrix, a multi-objective optimization objective function group is established, including:

[0151] Treatment effect maximization objective function: max F1(x) = Σwa·Ta(x), where wa is the weight of the a-th treatment target feature, and Ta(x) is the corresponding feature value function;

[0152] Treatment time minimization objective function: min F2(x) = α·tt + β·Σttb, where α and β are time weight coefficients, tt is the single treatment time, and ttb is the b-th light source cooperative working time;

[0153] Energy consumption minimization objective function: min F3(x) = Σγ k ·P k• tt k where γ k is the energy consumption coefficient of the kth light source, P k is the power, tt k is the actual working time of the kth light source;

[0154] Establish safety constraints: set the maximum energy density threshold of each wavelength spectrum, the longest continuous irradiation time limit and the skin temperature rise limit;

[0155] Establish device performance constraints: determine the power adjustment range of each light source, the pulse frequency adjustable range and the geometric constraints of the spatial distribution mode;

[0156] Establish user comfort constraints: based on user historical feedback data, set the pain perception threshold, photosensitivity limit and treatment site adaptability constraints;

[0157] Convert the above constraints into a set of mathematical inequalities: gc(x)≤0 (c=1,2,...,q), where x is the spectrum parameter combination vector, and q represents the total number of constraints;

[0158] Use non-dominated sorting genetic algorithm (NSGA-II) as a multi-objective optimization solver, set the population size, crossover probability and mutation probability;

[0159] Establish a Pareto optimal solution set evaluation mechanism to maintain the distribution of the solution set through crowding distance calculation;

[0160] Develop solution set screening criteria: prefer solutions that perform well on the treatment effect objective function and meet all constraints;

[0161] When multiple solutions have similar treatment effects, choose the solution with the smallest energy consumption or the shortest treatment time as the final parameter combination scheme;

[0162] Establish an adaptive update mechanism for constraint functions, and when new treatment case data accumulates to a preset threshold, automatically trigger constraint function reconstruction;

[0163] Based on the latest treatment effect feedback, dynamically adjust the treatment target feature weight and constraint boundary parameters;

[0164] Update the multi-objective optimization model parameters through online learning to ensure that the constraint functions always reflect the latest treatment needs and safety standards.

[0165] In this embodiment, through multi-objective collaborative optimization, the problem of deterioration of other indicators caused by single-objective optimization is avoided, the treatment effect is improved, and the balance optimization of effect, time and energy consumption is realized; feature extraction based on historical data makes the treatment scheme more in line with actual needs, improves the treatment target achievement rate, and reduces the number of invalid treatments; multi-level constraint conditions ensure the safety of the treatment process, reduce the incidence of adverse reactions, and significantly improve user satisfaction; energy consumption optimization reduces the operating cost of the equipment, and treatment time optimization improves the efficiency of the equipment and user experience; the dynamic updating mechanism enables the system to adapt to different user groups and emerging treatment needs, and the system performance improves continuously after long-term use, with good scalability; this multi-objective optimization constraint function construction method based on historical data driving has significant innovation in the field of intelligent medical device control, especially the mathematical modeling and collaborative optimization of multi-dimensional targets such as treatment effect, safety and efficiency, which provides a complete technical solution for the intelligent control of phototherapy equipment.

[0166] The wavelength selection and energy distribution calculation step includes: the wavelength selection network selects the optimal 3-5 wavelength combination from the pre-defined spectrum library based on the input tissue response characteristics, and the wavelength library contains 12 candidate wavelengths in the range of 280-1000nm; for each selected wavelength, the energy distribution network calculates its optimal energy density, and the output range is 5-50mW / cm 2 The energy distribution calculation considers the absorption threshold and safety limit of each layer of tissue; a multi-objective optimization constraint function is executed to simultaneously satisfy the dual objectives of maximizing treatment effect and minimizing thermal damage risk; a soft gating mechanism is used to normalize the energy distribution of different wavelengths to ensure that the total energy does not exceed the safety threshold of 100mW / cm 2 ; Through the above steps, intelligent wavelength selection and energy distribution optimization based on tissue response characteristics are realized, ensuring the rationality and safety of the spectrum parameters.

[0167] The timing parameter optimization calculation step includes: the timing parameter network calculates the optimal pulse frequency of each wavelength based on the tissue thermal conductivity characteristics, and the frequency range is 0.1-100Hz; according to the penetration depth and mechanism of action of different wavelengths in the tissue, the optimal irradiation time of each wavelength is calculated, and the time range is 30 seconds-30 minutes;

[0168] A timing coordination mechanism is designed to ensure that the timing parameters of multiple wavelengths work collaboratively and avoid mutual interference and energy waste; an adaptive timing adjustment algorithm is introduced to dynamically adjust the pulse frequency and irradiation time according to real-time monitoring of the tissue response; a timing safety check module is constructed to ensure that any combination of timing parameters will not cause tissue overheating or other adverse reactions; through the above steps, intelligent coordination and optimization of multi-wavelength light source timing parameters are realized, improving the safety and effectiveness of the phototherapy process.

[0169] Generating a spatial distribution pattern includes: generating a spatial irradiation pattern for each wavelength light source based on the geometric shape and tissue distribution characteristics of the treatment area using a spatial distribution network; dividing the treatment area into N×N sub-areas (N=8-16) using a grid-based spatial discretization method, and independently calculating the light intensity for each sub-area; introducing a spatial smoothing constraint to prevent excessive differences in light intensity between adjacent areas and ensure the continuity of the spatial distribution; designing a multi-wavelength spatial coordination algorithm to optimize the distribution combination of different wavelengths in space to achieve a synergistic effect in the spatial dimension; constructing a spatial distribution visualization module to generate a spatial heat map of light intensity to facilitate parameter verification and adjustment; through the above steps, intelligent distribution optimization of multi-wavelength light sources in the spatial dimension is achieved to ensure the uniformity and pertinence of light within the treatment area.

[0170] A multimodal spectral parameter model is integrated and generated, and verified, including: fusing the outputs of the four sub-networks of wavelength selection, energy distribution, timing parameters, and spatial distribution in the first neural network to form a complete multimodal spectral parameter model; designing a parameter consistency verification module to ensure the compatibility and coordination between parameters of different dimensions; verifying the expected effects of the parameter combination scheme through Monte Carlo simulation, and predicting tissue responses and possible risks; based on the simulation results, fine-tuning and optimizing the parameter combination to ensure the feasibility and safety of the scheme; generating a confidence assessment of the parameter combination scheme, and quantifying the reliability of the scheme and the expected range of effects; through the above steps, a verified complete multimodal spectral parameter model is formed, providing reliable guidance for subsequent equipment control.

[0171] In some possible implementations of the present invention, the step of generating a multimodal spectral parameter model based on the spectral feedback model and the historical phototherapy target data extracted from the historical phototherapy data in combination with a preset first neural network further includes:

[0172] The adaptive learning and optimization steps include: building an online learning mechanism to continuously adjust the neural network parameters based on actual treatment effect feedback; designing an experience replay buffer to store historical parameter combination schemes and their effect evaluations for network training; introducing a meta-learning algorithm to enable the network to quickly adapt to the individual characteristics of new users; building an uncertainty quantification module to evaluate the prediction confidence of parameter combination schemes; and establishing a model version management mechanism to support incremental updates and rollback operations of network parameters.

[0173] This embodiment achieves continuous adaptive optimization of the neural network, enabling the system to continuously improve the accuracy of parameter calculation as usage experience accumulates.

[0174] In some possible implementations of the present invention, the step of collaboratively controlling multiple different light sources in a multimodal phototherapy device according to the multimodal spectral parameter model includes:

[0175] Resolving the spectral parameter combination scheme from the multi-modal spectral parameter model includes: the main controller of the phototherapy device receives the multi-modal spectral parameter model and resolves the multi-modal spectral parameter combination scheme, which includes the spectral energy distribution matrix E(λ, t, x, y) of M wavelengths, the pulse frequency vector F(λ), the irradiation time vector T(λ), and the spatial distribution mode matrix S(λ, x, y); based on the light source type and wavelength range, the parameter combination scheme is decomposed into K sub-task packages (K≥3), each corresponding to the control parameters of a specific light source; a unique task identifier and timestamp are assigned to each sub-task package to ensure the integrity and synchronization of parameter transmission; through CRC check and digital signature mechanism, the sub-task package is encapsulated to prevent data damage and tampering during transmission; according to the processing capacity and cache capacity of each light source micro control unit, the data volume and transmission priority of the sub-task package are dynamically adjusted; through the above steps, the intelligent decomposition and preprocessing of the complex spectral parameter scheme are realized, laying a data foundation for subsequent distributed control.

[0176] Building a first near field communication network for collaborative control between the main controller of the phototherapy device and the micro control units of each light source, including: establishing a star topology near field communication network based on IEEE 802.15.4 protocol, the main controller as the coordinator and each light source micro control unit as the terminal node; setting network communication parameters: carrier frequency 2.4GHz, data transmission rate 250kbps, communication distance limited within 2 meters, ensuring low delay and high reliability; assigning each micro control unit a 16-bit network address and an 8-bit logical channel, supporting simultaneous control of up to 255 light sources; establishing a time synchronization mechanism, through the main controller regularly broadcasting time stamps, all micro control units maintain a time synchronization accuracy of ±1ms; configuring network security strategy, including AES-128 encryption, key rotation and identity authentication, to prevent unauthorized devices from accessing; setting network redundancy and fault recovery mechanism, when the main communication path fails, automatically switch to the backup communication channel; through the above steps, a high-reliability, low-delay near field communication network infrastructure is established, providing stable communication guarantee for multi-light source collaborative control.

[0177] The microcontroller units are initialized and parameter configured, including: each microcontroller unit automatically executes a hardware self-test program after startup to detect the working status of the light source drive circuit, temperature sensor, and light intensity detector; the microcontroller unit sends a device information report to the main controller, including status parameters such as light source type, wavelength range, maximum output power, and current temperature; the main controller issues a personalized control parameter configuration file for each microcontroller unit based on the device information report; the microcontroller unit receives and parses the control parameters, including PWM duty cycle, current limit value, temperature protection threshold, fault detection rules, etc.; a parameter validity check is performed to ensure that all configuration parameters are within the safe operating range. If any abnormal parameters are found, the microcontroller unit is reported to the main controller and reconfiguration is requested; after completing the parameter configuration, the microcontroller unit enters the ready state and sends a ready confirmation signal to the main controller; through the above steps, it is ensured that each light source microcontroller unit is correctly configured, creating conditions for the precise execution of collaborative control.

[0178] The collaborative control timing scheduling steps include: the main controller generates a global timing scheduling table based on the spectral parameter combination scheme, which defines the start, modulation, and stop times of each light source on the time axis; decomposes the global timing scheduling table into local scheduling instructions for each microcontroller unit, each instruction contains information such as absolute timestamp, action type, parameter value, etc.; through the first near-field communication network, the local scheduling instructions are sent to the corresponding microcontroller unit in batches in chronological order to ensure the timing accuracy of instruction transmission; after each microcontroller unit receives the scheduling instruction, it stores it in a local instruction queue and sorts it according to the timestamp; establishes a distributed clock synchronization mechanism, and all microcontroller units use the clock of the main controller as the benchmark to maintain time synchronization through the Network Time Protocol (NTP); designs an instruction preloading mechanism to load the instructions to be executed into the execution buffer 500ms in advance to reduce the delay of real-time scheduling; through the above steps, precise timing coordination of multiple light sources is achieved to ensure that each light source can work synchronously according to the predetermined plan.

[0179] The micro control units cooperatively perform and monitor the steps in real time, including: each micro control unit accurately controls the switching, dimming, frequency adjustment, etc. of the light source according to the local scheduling instructions, and the execution accuracy reaches the microsecond level; during the operation of the light source, the micro control unit continuously monitors the output power, working temperature, current and voltage, etc. of the light source; the monitoring data is uploaded to the main controller in real time, the upload frequency is 10Hz, and the data contains time stamp, device ID, parameter value, state flag, etc. information; the main controller analyzes the collected monitoring data in real time to detect whether there are parameter deviations, equipment failures or cooperative disorders, etc. abnormal conditions; when an abnormality is detected, the main controller immediately calculates the compensation strategy and sends parameter adjustment instructions to the related micro control units through the near field network; after receiving the adjustment instructions, the micro control unit smoothly adjusts the light source parameters without interrupting the current operation, realizing seamless real-time correction; through the above steps, the real-time monitoring and dynamic adjustment of the cooperative working state of multiple light sources are realized, ensuring the stability and accuracy of the light therapy process.

[0180] The cooperative optimization and adaptive adjustment steps include: establishing a cooperative effect evaluation model, based on the actual output parameters of each light source and user physiological feedback, real-time calculation of the comprehensive effect of cooperative control; designing a distributed decision algorithm, each micro control unit can make preliminary parameter optimization decisions based on local information; information sharing between micro control units is realized through the near field network, each unit broadcasts its state information and optimization suggestions to neighboring nodes; the main controller collects the optimization suggestions of all micro control units, forms a global optimization strategy through a voting mechanism and weight distribution algorithm; the global optimization strategy is converted into specific parameter adjustment instructions and distributed to each micro control unit for execution; a cooperative learning mechanism is established, the system can learn from the results of each cooperative control and continuously optimize the cooperative strategy and parameter configuration; through the above steps, intelligent cooperative optimization based on real-time feedback is realized, enabling the multi-light source system to adapt to different use scenarios and user needs autonomously.

[0181] The fault handling and system recovery steps include: establishing a multi-level fault detection mechanism: local fault detection at the micro control unit level, connection fault detection at the communication network level, and cooperative fault detection at the system level; when a single light source fault is detected, the system automatically redistributes the load of the light source to other normally working light sources, maintaining the overall light therapy effect; for communication faults, a backup communication channel and offline working mode are enabled to ensure that each micro control unit can continue to perform the preset control task; a fault isolation mechanism is established to prevent a single point failure from affecting the normal operation of the entire system; a system state snapshot and recovery function is designed to periodically save the system working state, and after fault recovery, the system can quickly recover to the normal working state; a fault log and diagnosis system is established to record all fault events and processing procedures, providing data support for subsequent system optimization; through the above steps, a perfect fault handling and system recovery mechanism is built, improving the reliability and robustness of the multi-light source cooperative control system.

[0182] In some possible embodiments of the present application, the edge computing node dynamically adjusts the spectral parameter combination scheme in the multi-modal spectral parameter model according to the monitoring data and the preset safety threshold through a reinforcement learning algorithm, and simultaneously generates a treatment effect evaluation index, including:

[0183] The monitoring data is preprocessed, including: the edge computing node receives multi-dimensional monitoring data from the distributed sensors in real time, including skin temperature change rate Delta T / Delta t, spectral reflectance change Delta R(lambda), microcirculation blood flow change Delta BF, skin moisture content change Delta MC, etc.; the original monitoring data is subjected to digital filtering processing, and a 5th order Butterworth low-pass filter is used to eliminate high-frequency noise; the multi-dimensional monitoring data is converted into a standardized physiological feedback vector R(t)=[r1(t), r2(t),..., r n (t)], where n>=8, and each component represents a specific physiological parameter; a sliding window mechanism is established to maintain a history data buffer with a length of 200 sampling points, which is used to calculate the change trend and statistical characteristics of the physiological parameters; abnormal data points caused by sensor failure or external interference are identified and removed through an outlier detection algorithm to ensure data quality; through the above steps, real-time acquisition and preprocessing of high-quality feedback signals are realized, providing a reliable data foundation for subsequent intelligent decision-making.

[0184] The safety threshold is dynamically evaluated and updated in combination with the preprocessed monitoring data, including: establishing a multi-level safety threshold system, including an absolute safety threshold T_abs (based on medical standards), a relative safety threshold T_rel (based on individual characteristics of the user), and a dynamic safety threshold T_dyn (based on real-time physiological state calculation); according to the age, gender, skin color, health status and other individual characteristics of the current user, the personalized relative safety threshold is calculated through the Bayesian inference algorithm; based on the statistical characteristics (mean, variance, rate of change) of the preprocessed monitoring data, the safety threshold range is dynamically adjusted using adaptive control theory; a safety margin calculation model is established, setting a 20% safety margin for each physiological parameter to ensure that the system works within a safe region; a threshold out-of-boundary warning mechanism is designed, which triggers a warning when the physiological parameter approaches 80% of the safety threshold and starts preventive parameter adjustment; a threshold learning and updating mechanism is established, which continuously optimizes the threshold setting based on historical safety data to improve the accuracy and adaptability of the threshold; through the above steps, a dynamically adaptive multi-level safety guarantee system is constructed to provide a reliable safety boundary for spectral parameter adjustment.

[0185] The reinforcement learning environment modeling and state definition are performed, including: defining a system state space S, including a current physiological response vector, a spectral parameter vector, a safety margin vector, a time factor, etc., with a state dimension of 32; designing an action space A, including spectral energy adjustment amount Δe(λ), pulse frequency adjustment amount Δf(λ), illumination time adjustment amount Δt(λ), and spatial distribution adjustment parameter Δs(λ), with each action dimension discretized into 21 levels (-10 to +10); constructing a reward function R(s, a), considering the improvement degree of treatment effect, the safety guarantee level, the energy efficiency, and the user comfort in four dimensions, with weights of 0.4, 0.3, 0.2, and 0.1 respectively; establishing a state transition probability model P(s'|s, a), based on the spectral feedback model and the user physiological characteristics to predict the state transition probability; designing a termination condition judgment mechanism, including target achievement termination, safety threshold triggering termination, and maximum time limit termination; through the above steps, a complete reinforcement learning environment model is established, providing a standardized problem framework for intelligent decision-making algorithms.

[0186] The deep Q network (DQN) algorithm is constructed, including: constructing a deep neural network structure, including an input layer (32 neurons), 3 hidden layers (128 neurons in each layer, using ReLU activation function), and an output layer (action value function Q value); designing a double network architecture, the main network is used for action selection, and the target network is used for Q value calculation, the target network is soft updated every 500 steps, and the update coefficient τ=0.01; an experience replay buffer is established, with a capacity of 10,000 experience samples, using a priority replay mechanism, and preferentially selecting samples with larger TD errors for training; an ε-greedy strategy is implemented for action selection, with an initial ε=0.9, linearly decaying to 0.1 with the training step number, balancing exploration and utilization; a learning rate adaptive adjustment mechanism is set, with an initial learning rate of 0.001, dynamically adjusted according to the convergence of the loss function; the NoisyNetworks technology is introduced, adding learnable noise to the network parameters to improve exploration efficiency; through the above steps, an efficient deep reinforcement learning algorithm is realized, which can find the optimal strategy in complex continuous control problems.

[0187] Based on the constructed reinforcement learning model and Q network algorithm, parameter adjustment decisions are generated and executed, including: based on the current state s_t, calculating the Q values of all possible actions through the trained Q network algorithm, and selecting the action a* with the maximum Q value; converting the selected action a* into specific spectral parameter adjustment instructions, including adjustment amplitude, adjustment direction, execution timing, etc.; verifying the safety of the parameter adjustment instructions to ensure that the adjusted parameter combination does not violate any safety threshold constraints; calculating the expected effect and risk assessment of parameter adjustment, generating an adjustment confidence score (0-100 points); when the confidence score is higher than the threshold (75 points), execute parameter adjustment; below the threshold, use a conservative strategy or request user confirmation; send the parameter adjustment instructions to the light source control system through a secure channel, and monitor the execution results; through the above steps, an intelligent parameter adjustment decision based on reinforcement learning is realized, ensuring the effectiveness and safety of the adjustment.

[0188] The treatment effect evaluation index is constructed, including: establishing a multi-dimensional effect evaluation system, including objective physiological indicators (accounting for 60%) and subjective feeling indicators (accounting for 40%); the objective physiological indicators include: 8 sub-indicators such as skin temperature improvement rate, blood flow improvement rate, tissue oxygen content improvement rate, spectral reflection characteristic change and the like; the subjective feeling indicators are collected through an intelligent inquiry system, including 5 dimensions such as comfort degree, pain relief degree, overall satisfaction and the like; a time decay weight function is designed, the recent effect weight is higher, the historical effect weight gradually decays, and the weight function is w (t) = e^ (-at), wherein a = 0.1; a personal baseline comparison mechanism is established, the current effect is compared with the historical best effect of the user to generate a relative improvement index; a comprehensive score algorithm is constructed, the indicators in each dimension are weighted and fused into a comprehensive score of 0-100 points, and a detailed effect analysis report is generated; through the above steps, a scientific and comprehensive personalized treatment effect evaluation system is established, which provides a quantitative basis for treatment scheme optimization.

[0189] Online learning and model updating are performed, including: establishing a continuous learning mechanism, adding the state-action-reward sequence of this time to the experience replay buffer after completing each treatment cycle; designing an incremental learning algorithm to avoid the problem of catastrophic forgetting, and protecting important network parameters through the elastic weight consolidation (EWC) technology; implementing a federated learning framework to collaboratively optimize the reinforcement learning model using multi-user data while protecting user privacy; establishing a model performance monitoring mechanism to track key indicators such as prediction accuracy, decision effect, and safety performance of the model; designing an adaptive learning rate adjustment strategy to dynamically adjust the learning parameters according to the changes in model performance; establishing a model version management and rollback mechanism to quickly restore to a stable version when the performance of the new model decreases; through the above steps, the reinforcement learning model is continuously optimized and adapted to the individual, and the system performance is continuously improved with experience.

[0190] In some possible embodiments of the present application, based on the treatment effect evaluation index, a user spectral feedback feature library is established, and the spectral feedback model is continuously optimized in an incremental learning manner, including:

[0191] Based on the treatment effect evaluation index, user spectral feedback features corresponding thereto are extracted, including user sensitivity coefficients to different wavelength spectra, spectral energy absorption rates, and spectral penetration depth parameters;

[0192] The spectral feedback features are associated with the corresponding treatment effect evaluation indicators to form user spectral feedback feature samples;

[0193] The user spectral feedback feature samples are classified and stored according to user age, gender, skin color type, and treatment site to obtain an initial user spectral feedback feature library;

[0194] An indexing mechanism is established in the initial user spectral feedback feature library, which includes a first-level index based on user physiological characteristics and a second-level index based on treatment effect types, and each user spectral feedback feature sample is assigned a unique identifier, and its creation time, update time and usage frequency are recorded, obtaining a user spectral feedback feature library;

[0195] Extracting the newly added user spectral response feature samples from the user spectral feedback feature library as input data for incremental learning;

[0196] Performing data cleaning and normalization processing on the new samples to eliminate outliers and unify the data format;

[0197] Calculate the similarity between the preprocessed new samples and historical training samples to identify representative incremental learning samples;

[0198] Based on the incremental learning samples, an online learning algorithm is used to update the parameters of the epidermis response sub-model, the dermis response sub-model and the subcutaneous tissue response sub-model;

[0199] During model updating, the core weight parameters of the original model are maintained, and the local parameters related to the new samples are fine-tuned;

[0200] Evaluate the prediction accuracy of the updated model through cross-validation method, when the accuracy improves to the preset threshold, the updated model parameters are solidified and stored;

[0201] Use the updated multi-level spectral response model to perform backtracking verification on historical treatment cases, calculate the model prediction accuracy and treatment effect compliance rate;

[0202] When the model performance indicators are lower than the preset standard, automatically trigger the model retraining mechanism, and supplement more diversified training samples to the user spectral response feature library;

[0203] The model performance evaluation results are fed back to the first neural network to realize closed-loop optimization of the overall control system.

[0204] The scheme of the embodiment realizes personalized spectral parameter configuration for thousands of people with thousands of faces through the establishment of a user spectral response feature library, which can improve the prediction accuracy of treatment effect; The incremental learning mechanism enables the system to continuously learn from new user data, avoiding the computational overhead of repeated training, and the model adaptability is enhanced with the extension of usage time, and the long-term use effect is better; The indexing mechanism of the feature library realizes fast retrieval, shortens the response time, avoids redundant data storage, and improves storage efficiency; Based on the dynamic adjustment of the safety threshold of a large amount of user data, the safety risk caused by individual differences is reduced; The abnormal detection mechanism can timely discover and handle inappropriate treatment parameters.

[0205] In some possible embodiments of the present application, the method for constructing the inter-layer spectral energy transfer matrix comprises:

[0206] The skin tissue is divided into m discrete layers (m≥10), and each layer has a thickness of no more than 0.2 mm;

[0207] For each layer, the absorption coefficient μa and the scattering coefficient μs of the tissue of the layer to light of each wavelength are calculated; s ;

[0208] Based on Monte Carlo photon transport simulation, the probability distribution of the transfer of photons from the i-th layer to the j-th layer is calculated;

[0209] An m×m-dimensional transfer matrix T is constructed, wherein element T(i, j) represents the proportion of the light energy incident on the i-th layer that is transferred to the j-th layer.

[0210] The embodiment of the present application realizes accurate quantitative description of the transfer process of light energy in the multi-layer structure of the skin.

[0211] Please refer to Fig. 2 Another embodiment of the present application provides an intelligent collaborative control system of a multi-modal light therapy device, which is used to execute an intelligent collaborative control method of a multi-modal light therapy device, and comprises a multi-modal light therapy device, an edge computing node and a cloud server.

[0212] The cloud server is configured to:

[0213] Obtain physiological parameters of a plurality of first users to obtain a physiological data set;

[0214] Obtain historical light therapy data of the first user and historical environmental data corresponding to the historical light therapy data;

[0215] Based on the physiological data set, the historical light therapy data and the historical environmental data, a spectral feedback model is constructed;

[0216] According to the spectral feedback model and historical light therapy target data extracted from the historical light therapy data, a multi-modal spectral parameter model is generated in combination with a preset first neural network;

[0217] According to the multi-modal spectral parameter model, a plurality of different light sources in the multi-modal light therapy device are collaboratively controlled;

[0218] Real-time monitoring of changes in physiological parameters of a target user obtains monitoring data, and the monitoring data is sent to the edge computing node;

[0219] Based on a treatment effect evaluation index, a user spectral feedback feature library is established, and the spectral feedback model is continuously optimized through an incremental learning method;

[0220] The edge computing node is configured to dynamically adjust a spectrum parameter combination scheme in the multi-modal spectrum parameter model by an reinforcement learning algorithm according to the monitoring data and a preset safety threshold, while generating a treatment effect evaluation index.

[0221] It should be understood that, Fig. 2 The block diagram of the intelligent collaborative control system of the multi-modal light therapy device shown is only schematic, and the number of each module shown does not limit the protection scope of the present application. The intelligent collaborative control system of the multi-modal light therapy device provided in the embodiment can be used to execute each embodiment scheme of the corresponding intelligent collaborative control method of the multi-modal light therapy device. For the specific implementation process, please refer to the description of each method embodiment, which is not described here.

[0222] It should be noted that, for the foregoing method embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited by the action sequence described, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily necessary for the present application.

[0223] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0224] In several embodiments provided in the present application, it should be understood that the disclosed device can be implemented by other ways. For example, the device embodiments described above are only schematic, and the division of the above units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical or other forms.

[0225] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0226] In addition, each of the function units in each of the embodiments of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software function unit.

[0227] When the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable memory. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a memory and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the above-mentioned method of each embodiment of the present application. The aforementioned memory includes: a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0228] Those of ordinary skill in the art can understand that all or part of the steps in the above-mentioned embodiments can be completed by programs instructing relevant hardware, and the programs can be stored in a computer readable memory, which can include a flash disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0229] The embodiments of the present application are described in detail above, and the principles and implementation manners of the present application are described by applying specific examples. The above description of the embodiments is only used to help understand the method of the present application and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present application, the specific implementation manner and application range can be changed, and the above description of the present application should not be understood as a limitation.

[0230] Although the present application is disclosed as above, the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements without departing from the spirit and scope of the present application, and can make various changes and modifications, including the combination of different functions and implementation steps, including the implementation of software and hardware, which are all within the protection scope of the present application.

Claims

1. An intelligent collaborative control method for a multimodal phototherapy device, characterized in that: include: Acquiring physiological parameters of a plurality of first users to obtain a physiological data set; Acquiring historical phototherapy data of the first user and historical environmental data corresponding to the historical phototherapy data; constructing a spectral feedback model based on the physiological data set, the historical light therapy data, and the historical environmental data; generating a multimodal spectral parameter model based on the spectral feedback model and the historical phototherapy target data extracted from the historical phototherapy data in combination with a preset first neural network; Coordinately controlling multiple different light sources in a multimodal phototherapy device according to the multimodal spectral parameter model; Real-time monitoring of physiological parameter changes of target users to obtain monitoring data, and sending the monitoring data to the edge computing node; The edge computing node dynamically adjusts the spectral parameter combination scheme in the multimodal spectral parameter model according to the monitoring data and a preset safety threshold through a reinforcement learning algorithm, and generates a treatment effect evaluation index at the same time; Based on the treatment effect evaluation index, a user spectral feedback feature library is established, and the spectral feedback model is continuously optimized through incremental learning.

2. The intelligent collaborative control method of a multimodal phototherapy device according to claim 1, characterized in that: The step of constructing a spectral feedback model based on the physiological data set, the historical phototherapy data, and the historical environmental data includes: The physiological data set is layered according to tissue levels to form layer parameter sets for the epidermis, dermis, and subcutaneous tissue layers; Applying wavelet transform to perform feature extraction on the layer parameter set of each layer, the historical phototherapy data and the historical environment data to identify the absorption and scattering characteristics of each layer of tissue for different wavelength spectra under different environments and different phototherapy control parameters; Based on the photon transmission Monte Carlo simulation technology, a spectral feedback prediction model for each layer of tissue is established; The spectral feedback prediction models of each layer are integrated through a bidirectional long short-term memory network, and a coupling feedback mechanism reflecting the inter-layer spectral energy transfer effect is constructed.

3. The intelligent collaborative control method for a multimodal phototherapy device according to claim 2, characterized in that: The first neural network comprises: A feature fusion layer based on graph neural network to process the correlation between physiological data of different dimensions; Multiple parallel spectral parameter optimization sub-networks, each sub-network corresponds to the optimization calculation of a specific spectral parameter; An attention-based parameter combination layer to generate the final multimodal spectral parameter model; An uncertainty quantification module is used to evaluate the reliability of the generated multimodal spectral parameter models.

4. The intelligent collaborative control method for a multimodal phototherapy device according to claim 3, characterized in that: The steps of integrating the spectral feedback prediction models of each layer through a bidirectional long short-term memory network and constructing a coupling feedback mechanism that reflects the inter-layer spectral energy transfer effect include: A single-layer feedback sub-model was constructed, including: for the epidermis, a spectral absorption-scattering parameter mapping based on the improved Kuz-Mink model was established, focusing on characterizing the absorption characteristics of melanin to light of different wavelengths; for the dermis, an optical interaction model of collagen and elastic fibers based on the finite-difference time-domain method was constructed, focusing on characterizing the response of vascular microcirculation to spectral energy; for the subcutaneous tissue layer, a deep tissue light diffusion model based on the radiation transfer equation was established, focusing on characterizing the scattering and absorption characteristics of fat and connective tissue to light; a Bayesian optimization algorithm was applied to each sub-model, and model parameters were adjusted based on historical data to minimize the root mean square error between the model prediction value and the measured value; The sub-model coupling feedback mechanism is constructed, including: establishing an interlayer spectral energy transfer matrix to describe the energy attenuation law of light of different wavelengths during transmission from the surface layer to the deep layer; constructing an interlayer heat conduction model to characterize the diffusion process between tissue layers after light energy is converted into heat energy; designing a tissue physiological response conversion function to map local temperature changes to dynamic changes in physiological parameters; integrating these three components to form a closed-loop feedback path, so that the response changes of deep tissue can reversely affect the optical properties of surface tissue; Model integration and validation include: using a hierarchical recursive neural network structure to integrate each sub-model as a network node into a unified spectral feedback model; introducing the time dimension and constructing a timing prediction mechanism based on gated recurrent units, so that the model can predict the dynamic changes of tissue response during illumination; using an independent test data set through cross-validation to evaluate the prediction accuracy of the integrated model; adjusting the model integration parameters based on the validation results and optimizing the weight distribution between sub-models.

5. The intelligent collaborative control method for a multimodal phototherapy device according to claim 4, characterized in that: The step of generating a multimodal spectral parameter model based on the spectral feedback model and the historical phototherapy target data extracted from the historical phototherapy data in combination with a preset first neural network includes: Extracting feedback feature vectors of the epidermis, dermis, and subcutaneous tissue layers from the spectral feedback model, each feedback feature vector contains spectral absorption coefficient, scattering coefficient, thermal conductivity coefficient, and physiological response sensitivity parameters; Normalize the feedback feature vectors of each layer to eliminate the influence of different physical dimensions and numerical ranges; Construct an inter-layer coupling relationship matrix to characterize the mutual influence strength between different layers and generate a 144-dimensional coupling feature descriptor; The coupled feature descriptor is reduced in dimension based on principal component analysis to generate a 64-dimensional compact feature representation. Constructing a first neural network, and performing wavelength selection and energy distribution calculation, timing parameter optimization calculation, and spatial distribution pattern generation based on the first neural network; A multimodal spectral parameter model was generated and verified.

6. The intelligent collaborative control method for a multimodal phototherapy device according to claim 5, characterized in that: The step of collaboratively controlling a plurality of different light sources in a multimodal phototherapy device according to the multimodal spectral parameter model comprises: Determining a spectrum parameter combination solution from the multimodal spectrum parameter model; Establishing a first near-field communication network for coordinated control between the main controller of the phototherapy device and the microcontroller units of each light source; Initialize and configure parameters of the microcontroller unit; The main controller generates a global timing schedule based on the spectral parameter combination scheme; decomposes the global timing schedule into local scheduling instructions for each microcontroller unit; Sending the local scheduling instructions to the corresponding micro control units in batches in chronological order through the first near field communication network; After receiving the scheduling instructions, each microcontroller unit stores them in the local instruction queue and sorts them according to the timestamp; Establish a distributed clock synchronization mechanism, all microcontroller units use the main controller's clock as the benchmark and maintain time synchronization through the network time protocol; Design an instruction preloading mechanism to load the instructions to be executed into the execution buffer; Each microcontroller unit performs real-time coordination and monitoring according to local scheduling instructions; Establish a collaborative effect evaluation model and perform collaborative optimization and adaptive adjustment.

7. The intelligent collaborative control method for a multimodal phototherapy device according to claim 6, characterized in that: The edge computing node dynamically adjusts the spectral parameter combination scheme in the multimodal spectral parameter model through a reinforcement learning algorithm based on the monitoring data and a preset safety threshold, and generates a treatment effect evaluation index at the same time, including the following steps: Preprocessing the monitoring data; Dynamically evaluate and update safety thresholds based on pre-processed monitoring data; Perform reinforcement learning environment modeling and state definition; Build a deep Q network algorithm; Generate and execute parameter adjustment decisions based on the constructed reinforcement learning model and Q network algorithm; Establish a multi-dimensional effect evaluation system and design a time-attenuation weight function; Establish a personal baseline comparison mechanism to compare the current performance with the user's best historical performance and generate an improvement index; Build a comprehensive scoring algorithm, weight the indicators of each dimension into a comprehensive score of 0-100 points, and generate a detailed effect analysis report.

8. The intelligent collaborative control method for a multimodal phototherapy device according to claim 7, characterized in that: The step of establishing a user spectral feedback feature library based on the treatment effect evaluation index and continuously optimizing the spectral feedback model through incremental learning includes: Based on the treatment effect evaluation index, extracting corresponding user spectral feedback features, the spectral feedback features including the user's sensitivity coefficient to spectra of different wavelengths, spectral energy absorption rate, and spectral penetration depth parameters; Associating and marking the spectral feedback features with corresponding treatment effect evaluation indicators to form a user spectral feedback feature sample; Classify and store the user spectral feedback feature samples according to user age, gender, skin color type, and treatment area to obtain an initial user spectral feedback feature library; Establishing an indexing mechanism in the initial user spectral feedback feature library, the indexing mechanism includes a primary index based on the user's physiological characteristics and a secondary index based on the treatment effect type, assigning a unique identifier to each user spectral feedback feature sample, and recording its creation time, update time, and usage frequency, to obtain a user spectral feedback feature library; Extracting new user spectral response feature samples from the user spectral feedback feature library as input data for incremental learning; Performing data cleaning and normalization on the newly added samples, removing outliers and unifying the data format; Calculate the similarity between the preprocessed new samples and historical training samples to identify representative incremental learning samples; Based on the incremental learning samples, an online learning algorithm is used to update the parameters of the epidermis response sub-model, the dermis response sub-model, and the subcutaneous tissue response sub-model; During the model update process, the core weight parameters of the original model are maintained, and the local parameters related to the newly added samples are fine-tuned; The prediction accuracy of the updated model is evaluated through cross-validation. When the accuracy improvement reaches a preset threshold, the updated model parameters are solidified and stored. The updated multi-level spectral response model was used to retrospectively verify historical treatment cases, and the model's prediction accuracy and treatment effect compliance rate were calculated. When the model performance index is lower than the preset standard, the model retraining mechanism is automatically triggered, and more diverse training samples are added to the user's spectral response feature library; The model performance evaluation results are fed back to the first neural network to achieve closed-loop optimization of the overall control system.

9. The intelligent collaborative control method for a multimodal phototherapy device according to claim 8, characterized in that: The method for constructing the inter-layer spectral energy transfer matrix includes: The skin tissue is divided into m discrete layers, each layer no thicker than 0.2 mm; For each layer, calculate the absorption coefficient μa and scattering coefficient μ of the tissue layer for each wavelength of light s ; Based on Monte Carlo photon transport simulation, the probability distribution of photons transmitted from layer i to layer j is calculated; An m×m dimensional transfer matrix T is constructed, where the element T(i, j) represents the proportion of light energy incident on the i-th layer transferred to the j-th layer.

10. An intelligent collaborative control system for a multimodal phototherapy device, configured to execute the intelligent collaborative control method for a multimodal phototherapy device according to any one of claims 1 to 9, characterized in that: include: Multimodal phototherapy equipment, edge computing nodes, and cloud servers; The cloud server is configured as follows: Acquiring physiological parameters of a plurality of first users to obtain a physiological data set; Acquiring historical phototherapy data of the first user and historical environmental data corresponding to the historical phototherapy data; constructing a spectral feedback model based on the physiological data set, the historical light therapy data, and the historical environmental data; generating a multimodal spectral parameter model based on the spectral feedback model and the historical phototherapy target data extracted from the historical phototherapy data in combination with a preset first neural network; Coordinately controlling multiple different light sources in a multimodal phototherapy device according to the multimodal spectral parameter model; Real-time monitoring of physiological parameter changes of target users to obtain monitoring data, and sending the monitoring data to the edge computing node; Based on the treatment effect evaluation indicators, a user spectral feedback feature library is established, and the spectral feedback model is continuously optimized through incremental learning; The edge computing node is configured to dynamically adjust the spectral parameter combination scheme in the multimodal spectral parameter model through a reinforcement learning algorithm according to the monitoring data and a preset safety threshold, and generate a treatment effect evaluation indicator.

Images