A subcutaneous fat thermal perspective method based on physical mechanism and multi-data double driving

Abstract

The application discloses a subcutaneous fat thermal perspective method based on physical mechanism and multiple data double driving, and comprises the following steps: inputting surface temperature data of a biological tissue to be measured, external heat source action parameters, boundary condition parameters and tissue structure parameters into a prediction model, wherein the model comprises a basic temperature field prediction module and a temperature field correction module, the basic temperature field prediction result is corrected through a temperature field correction amount, and subcutaneous fat tissue temperature distribution under corresponding conditions is output; in a training stage, the basic temperature field prediction module trains a temperature field sample generated by numerical solution of a biological heat conduction control equation based on parameters calibrated through real experiment data, and learns a mapping relationship between different input conditions and a subcutaneous fat basic temperature field; and the temperature field correction module takes a temperature deviation of a subcutaneous local temperature measured value and the basic temperature field prediction result at a corresponding temperature measuring position as supervision information, learns a mapping relationship between different input conditions and the temperature field correction amount, and is optimized under multiple constraints.

Description

Technical Field

[0001] This invention belongs to the field of non-invasive biomedical detection technology, specifically involving a subcutaneous fat thermal imaging method based on physical mechanisms and multi-data dual-drive. Background Technology

[0002] In thermotherapy or energy therapy applied to biological tissues by external heat sources, the temperature distribution within the deep tissues is a crucial indicator for evaluating the effectiveness of the thermal effect and tissue safety. In treatment scenarios where subcutaneous adipose tissue is the primary target area, its temperature not only affects whether the target area achieves the expected thermal effect but also influences the thermal safety of the superficial tissue and surrounding normal tissue. However, the subcutaneous fat layer is located within the body, making it difficult to directly obtain its temperature using conventional non-contact methods. Existing direct measurement methods typically rely on thermocouples, needle-type temperature probes, and implanted miniature temperature sensors. While these methods can acquire localized internal temperatures, they suffer from high invasiveness, limited measurement points, and difficulty in achieving large-scale continuous monitoring. These methods are particularly unsuitable for non-surgical scenarios such as cosmetic fat reduction and non-invasive thermotherapy. Existing non-invasive methods mostly rely on surface infrared thermography, microwave thermometry, ultrasonic thermometry, and magnetic resonance thermometry. Among these, magnetic resonance thermometry suffers from complex equipment structures, high usage and maintenance costs, and severely limited applicable scenarios. Infrared thermography and microwave thermometry, on the other hand, have inherent defects such as insufficient tissue penetration, poor spatial resolution, and poor real-time temperature measurement, making it difficult to stably and accurately reflect the true three-dimensional temperature distribution of the subcutaneous fat layer. In addition to the above methods, existing technologies have also proposed internal temperature inversion methods based on tissue heat transfer models and neural network prediction methods that integrate physical mechanisms. The former usually uses surface temperature as a boundary condition and solves the internal temperature field of the tissue through numerical iteration. Its prediction accuracy is easily affected by the setting deviations of tissue thermophysical parameters, optical parameters, boundary heat transfer parameters, and tissue structural parameters, and it suffers from large computational load and limited parameter adaptability under complex dynamic conditions. The latter can use numerical simulation samples to learn the mapping relationship between external observation information and internal temperature. However, there are usually distribution differences between numerical simulation samples and real tissue thermal responses. Models trained solely on simulation samples are prone to inheriting parameter biases and model biases from forward numerical models. On the other hand, real-world experiments typically only obtain localized temperatures at a few locations within the subcutaneous fat layer using one or more temperature sensors, making it difficult to directly obtain a continuous two-dimensional or three-dimensional complete temperature field. Existing methods generally use locally measured temperatures to validate numerical models or directly mix them with numerical samples for model training. However, they have not yet fully addressed how to use sparse local measured information to spatially correct the simulation-reality deviation of the complete temperature field, or how to limit the unreasonable amplification, oscillation, or distortion that occurs when this local correction propagates to unmeasured regions. Summary of the Invention

[0003] To address the problems in existing technologies, such as the difficulty in non-invasively obtaining subcutaneous fat temperature, the high trauma and limited measurement range of invasive measurements, the sensitivity of pure physical modeling methods to parameter settings and boundary conditions, and the lack of physical constraints and insufficient generalization ability of pure data-driven methods, the purpose of this invention is to provide a subcutaneous fat thermal imaging method based on physical mechanisms and multi-data dual-drive, so as to achieve non-invasive prediction, temperature field reconstruction and thermal imaging visualization of subcutaneous fat temperature information.

[0004] In view of this, the present invention proposes a subcutaneous fat thermal imaging method based on physical mechanisms and multi-data dual-drive, comprising: The surface temperature data of the biological tissue to be tested, the parameters of the external heat source, the boundary condition parameters and the tissue structure parameters are input into the pre-trained prediction model. The prediction model includes a basic temperature field prediction module and a temperature field correction module. The temperature field correction amount output by the temperature field correction module is used to correct the basic temperature field prediction result output by the basic temperature field prediction module, and outputs the temperature distribution of subcutaneous adipose tissue under the corresponding conditions. During the model training phase, the basic temperature field prediction module trains on temperature field samples generated by numerically solving the biological heat conduction control equation based on parameters calibrated by real experimental data. This is used to learn the mapping relationship between different input conditions and the basic temperature field of subcutaneous fat. During the model training phase, the temperature field correction module uses the temperature deviation between the actual measured subcutaneous local temperature in real experiments and the predicted basic temperature field at the corresponding temperature measurement location as supervision information to learn the mapping relationship between different input conditions and the temperature field correction amount, and optimizes it under the constraints of tissue heat transfer physical consistency, correction amount amplitude, and spatial smoothness.

[0005] As an improvement to the above method, the surface temperature data includes at least one of the following: surface temperature distribution at a single moment acquired by an infrared thermal imager, surface temperature sequence within a time window, processed and extracted surface temperature statistical feature information, and other information that can characterize the thermal process.

[0006] As an improvement to the above method, the external heat source action parameters are used to characterize the action intensity, action time, action location and action trajectory of the external heat source. The boundary condition parameters are used to characterize the surface heat transfer conditions, cooling conditions, and initial temperature conditions. The tissue structure parameters are used to characterize the tissue layer structure information and the thickness information of each layer.

[0007] As an improvement to the above method, the basic temperature field prediction module includes: an input feature encoding unit, a feature fusion unit, and a decoding unit; the temperature field correction module includes: a correction feature extraction unit and a residual output unit.

[0008] As an improvement to the above method, the method further includes a training step for the prediction model, comprising: Acquire real experimental data, solve data based on the biological heat conduction control equation, and construct a sample set including experimental data and solution data; The data in the sample set is preprocessed and converted into a unified data representation format; Joint optimization objective for constructing a prediction model based on multi-source data constraints; The prediction model is trained based on the joint optimization objective to obtain a prediction model that meets the training requirements.

[0009] As an improvement to the above method, the real experimental data includes at least: surface temperature data, external heat source parameters, boundary condition parameters, and tissue structure parameters; and temperature sensors are deployed at the subcutaneous target depth of the experimental sample to collect measured data on the change of local subcutaneous temperature with laser treatment time as a label.

[0010] As an improvement to the above method, the data obtained based on the biological heat conduction control equation specifically includes: Establish a layered tissue calculation model that includes at least the epidermis, dermis and subcutaneous fat layer, and set corresponding parameter information for each layer. The parameter information includes at least layer thickness parameters, thermophysical parameters and parameters related to the effect of external heat sources. Establish the temperature field governing equations; for the isolated tissue scenario, the governing equations adopt a simplified form that excludes blood perfusion terms and tissue metabolic heat terms, satisfying the following equation:

[0011] For the living tissue scenario, based on tissue characteristics and prediction accuracy requirements, blood perfusion rate, blood temperature, and tissue metabolic heat terms are further introduced into the control equations; satisfying the following equation:

[0012] in, This indicates heat conduction due to temperature difference. Indicates time, Indicates tissue temperature, Indicates tissue density, Indicates the specific heat capacity of the tissue. Indicates the thermal conductivity of the tissue. This represents the heat source term generated within the tissue by an external heat source. Indicates blood perfusion rate, This indicates the specific heat capacity of blood. Indicates blood temperature. This indicates the pyrogen content of tissue metabolism. Indicates tissue state, used to distinguish between isolated tissue and living tissue; Temperature field samples are generated through numerical solutions.

[0013] As an improvement to the above method, the multi-source data constraints include: basic temperature field consistency constraints. Consistency constraints of measured local temperatures Physical consistency constraints and correction constraints They respectively satisfy the following formulas:

[0014]

[0015] Where M represents the total number of samples in the basic temperature field. Indicates the first Each basic temperature field sample includes the corresponding operating condition input information; Indicates the first The spatial location corresponding to each basic temperature field sample. This indicates that the basic temperature field prediction module is in the first... The predicted base temperature values ​​output at the corresponding operating conditions and spatial locations of each base temperature field sample. This represents the numerical temperature value obtained at the corresponding working condition and spatial location after the biological heat conduction control equation, after the parameters have been calibrated using real experimental data, is numerically solved. This represents the total number of constrained samples for local temperature measurements; Indicates the first The working condition input information corresponding to each measured constraint sample. This indicates the thermocouple temperature measurement position of the nth measured constraint sample; This represents the measured subcutaneous local temperature value of the nth measured constrained sample at the corresponding temperature measurement location. This indicates the final temperature field prediction result.

[0016]

[0017] in, This represents the total number of physical constraint sampling points. Indicates the first The operating condition input information corresponding to each physical constraint sampling point; They represent the first The spatial location corresponding to each physical constraint sampling point; Indicates the first The final temperature prediction value output by each physical constraint sampling point at the corresponding working condition and corresponding spatial location; , and These represent the tissue density, specific heat capacity, and thermal conductivity, respectively. Indicates the external heat source in the th The heat source term formed at each physical constraint sampling point; Indicates the first The tissue state-related equivalent heat exchange term at each physical constraint sampling point; These are the smoothness constraint weighting coefficients; This indicates the temperature correction amount output at the corresponding operating condition and spatial location. Joint optimization objective for constructing a prediction model based on multi-source data constraints for:

[0018] in, , , and This represents the weight coefficients corresponding to each constraint term. These coefficients are preset based on the training stage, prediction error, physical residual, and model convergence status, and are adjusted in stages or dynamically to achieve a balance between data-driven learning and physical mechanism constraints.

[0019] As an improvement to the above method, the prediction model is trained by using a phased optimization method, a joint optimization method, or a combination of both.

[0020] As an improvement to the above method, the temperature distribution of subcutaneous adipose tissue under the corresponding conditions is: single-point subcutaneous temperature value, temperature sequence at a certain depth, two-dimensional temperature distribution of subcutaneous adipose layer, three-dimensional temperature distribution of subcutaneous adipose layer, or dynamic temperature evolution result. The output of the prediction model also includes: uncertainty information or error range of the prediction results, used to assess the reliability of the prediction results.

[0021] Compared with the prior art, the advantages of the present invention are: 1. In the application stage, this invention can achieve non-invasive prediction of the subcutaneous fat temperature field based on surface temperature response information and related working parameters.

[0022] 2. This invention integrates real experimental data with sample data obtained by numerical solution based on tissue heat transfer mechanism to construct a multi-source data system for predicting subcutaneous fat temperature field, which makes up for the shortcomings of single data sources in terms of internal temperature information acquisition and operating condition coverage.

[0023] 3. This invention embeds the physical mechanism of tissue heat transfer into the subcutaneous fat temperature prediction model, enabling the model to meet physical constraints while performing data-driven learning, thereby improving the physical consistency, stability and interpretability of the prediction results.

[0024] 4. This invention can output local temperature values ​​of the subcutaneous fat layer, two-dimensional or three-dimensional temperature distribution, temperature profile and thermal imaging visualization results. It can be used not only for internal temperature prediction, but also for thermal process assessment, thermal safety analysis and auxiliary formulation of thermotherapy plans.

[0025] 5. This invention is not limited to a single external heat source scenario, and has good method versatility and scalability, and can be applied to the prediction of internal temperature field during thermal processes in various biological tissues. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating a subcutaneous fat thermal imaging method driven by both physical mechanisms and multiple data sources. Figure 2 This is a schematic diagram of multi-source data acquisition according to the present invention; Figure 3 The block diagram of the two-step segmented prediction model that integrates physical mechanisms in this invention. Detailed Implementation

[0027] This invention proposes a method for predicting the temperature field of subcutaneous fat. First, the parameters required for the biological heat conduction control equation are calibrated using the surface temperature and local subcutaneous temperature under the same thermal action condition. Then, the calibrated parameters are used to numerically solve the biological heat conduction control equation to generate complete temperature field samples under different conditions, which are used to train the basic temperature field prediction model. Furthermore, the simulation-measurement deviation at sparse subcutaneous temperature measurement locations in real experiments is used to perform residual correction on the basic temperature field, which is constrained by physical laws, correction amplitude, and spatial smoothness. Thus, in the practical application stage, the two-dimensional or three-dimensional temperature field of subcutaneous fat can be obtained only based on non-invasive surface temperature and related working condition parameters.

[0028] like Figure 1 As shown, the method includes the following steps: S1. Multi-source data acquisition and sample set construction Multi-source data for predicting subcutaneous fat temperature field was acquired, and a multi-source sample set was constructed. The multi-source data included at least two categories: one category consisted of data obtained from real experiments, and the other category consisted of sample data obtained through numerical solutions to the biological heat conduction governing equations. For example... Figure 2 As shown.

[0029] The data obtained from the actual experiments include at least measured surface temperature data, measured subcutaneous temperature data, external heat source parameters, tissue structure parameters, and boundary condition parameters. The measured surface temperature data characterizes the temperature response of biological tissue surfaces under the influence of an external heat source; the measured subcutaneous temperature data characterizes the local temperature response of subcutaneous adipose tissue. Specifically, an infrared thermal imager is used to collect the dynamic temperature distribution of the laser-irradiated area and its surrounding tissue surface, obtaining a temperature sequence of surface temperature changes over time; simultaneously, temperature sensors are deployed at the target subcutaneous depth of the experimental sample to collect measured data on the change of local subcutaneous temperature with laser exposure time. The external heat source parameters characterize information such as the intensity, duration, location, and trajectory of the external heat source; the tissue structure parameters characterize the layered structure and thickness of each layer; and the boundary condition parameters characterize information such as surface heat transfer conditions, cooling conditions, and initial temperature conditions.

[0030] The sample data obtained by numerically solving the biological heat conduction control equation includes at least surface temperature data and internal temperature field data under different operating conditions. This data is used to supplement the temperature field information that is difficult to obtain under real experimental conditions and to expand the operating condition coverage of the sample under different tissue parameters, heat source parameters and boundary conditions.

[0031] Furthermore, the parameters related to the numerical solution of the biological heat conduction control equation can be calibrated or corrected based on real experimental data, making the temperature field obtained by the numerical solution closer to the temperature response characteristics under real experimental conditions. Specifically, firstly, based on the tissue type and structural characteristics of the experimental samples, the initial parameters required for the numerical solution of the biological heat conduction control equation are set. These initial parameters include, but are not limited to, tissue thermal conductivity, density, specific heat capacity, blood perfusion rate, metabolic heat production parameters, laser absorption coefficient, scattering coefficient, and surface convective heat transfer coefficient. Then, under the same laser power, irradiation time, spot size, and ambient temperature conditions, the tissue surface temperature and subcutaneous target depth temperature obtained by the numerical solution of the biological heat conduction control equation are compared with the surface temperature data collected by the infrared thermal imager and the subcutaneous measured temperature data collected by the temperature sensor, respectively.

[0032] Based on the temperature error between the two, the relevant parameters for numerically solving the biological heat conduction control equation are iteratively corrected. The temperature error may include one or more of the following: peak temperature error, heating rate error, steady-state temperature error, and temperature change error during the cooling phase. By minimizing the deviation between the numerically calculated temperature and the experimentally measured temperature, the corrected model parameters are determined, thereby enabling the numerically solved temperature field to more accurately reflect the temperature response process of subcutaneous adipose tissue and its adjacent tissues under real laser irradiation conditions.

[0033] Furthermore, the aforementioned real experimental data and numerical solution sample data are organized to construct a multi-source sample set for subsequent processing.

[0034] The preferred external heat source is a laser heat source, but it is not limited to lasers; other external heat sources that can cause tissue temperature rise can also be used.

[0035] S2. Multi-source data preprocessing and unified representation The multi-source sample set constructed in step S1 is preprocessed. Data cleaning, outlier handling, unit unification, time alignment, and spatial coordinate unification are performed on the real experimental data and numerical solution sample data. Surface temperature data, subcutaneous local temperature data, external heat source parameters, boundary condition parameters, and tissue structure parameters are converted into a unified data representation form for subsequent construction, optimization, and prediction of the subcutaneous fat temperature field prediction model.

[0036] In some implementations, the numerical solution sample data can be perturbed to reduce the distribution difference between the simulated samples and the real experimental data.

[0037] S3. Construct a subcutaneous fat temperature field prediction model that integrates the physical mechanism of tissue heat transfer with multi-source data constraints. Based on the unified data representation obtained in step S2, a subcutaneous fat temperature field prediction model is constructed that integrates the physical mechanism of tissue heat transfer with multi-source data constraints. This prediction model is used to output the predicted subcutaneous fat temperature field under corresponding working conditions, provided that surface temperature response information, external heat source parameters, boundary condition parameters, and tissue structure parameters are known.

[0038] In some implementations, the input variables of the model are denoted as:

[0039] in, The surface temperature response information includes one or more of the following: surface temperature distribution at a single moment acquired by an infrared thermal imager, surface temperature sequence within a time window, processed and extracted surface temperature statistical feature information, or other information that can characterize the thermal process. Indicates the parameters of the external heat source; Indicates boundary condition parameters; Represent organizational structure parameters; let If the spatial location variable is represented, then the predicted result of the subcutaneous fat temperature field can be expressed as:

[0040] in, Indicates position Predicted subcutaneous fat temperature at the site This represents a temperature field prediction model. Indicates model parameters.

[0041] In some embodiments, the temperature field prediction model includes a basic temperature field prediction module and a temperature field correction module. The temperature field correction amount output by the temperature field correction module corrects the basic temperature field prediction result output by the basic temperature field prediction module. Let the basic temperature field prediction result be:

[0042] in, This indicates the predicted results of the basic temperature field. This represents the basic temperature field prediction model. This represents its model parameters.

[0043] Furthermore, the local temperature constraint information obtained from real experiments can be represented as multiple measured constraint samples. Each measured constraint sample includes at least the input condition information under the corresponding operating condition, the thermocouple temperature measurement location, and the measured subcutaneous local temperature value at that location. The temperature measurement locations of different measured constraint samples can be the same or different. Based on the local temperature constraint information, the basic temperature field prediction results are collaboratively corrected to obtain the temperature field correction amount:

[0044] in, This represents the temperature field correction amount. This represents the temperature field correction model. Indicates its model parameters, This represents the local temperature constraint information obtained from actual experiments.

[0045] The final predicted result of the subcutaneous fat temperature field can be expressed as:

[0046] like Figure 3 The diagram shows a two-step segmented prediction model that integrates physical mechanisms.

[0047] Furthermore, the prediction model is based on the biological heat conduction control equation during both sample construction and optimization. The numerical solution sample data, corrected by real experimental data, is used to characterize the variation law of the subcutaneous fat basal temperature field under different working conditions; the surface temperature data and measured subcutaneous local temperature data obtained from real experiments are used to constrain and correct the overall temperature field; physical consistency constraints and correction amount constraints are further introduced during the model optimization process to improve the physical rationality and spatial stability of the prediction results.

[0048] In some implementations, the biological heat conduction control equation can be constructed in a simplified, extended, or modified form. For ex vivo tissue scenarios, the control equation can be in a simplified form that does not include blood perfusion terms and tissue metabolic heat terms, or the relevant effects can be incorporated into the equivalent heat transfer term; for in vivo tissue scenarios, blood perfusion rate, blood temperature, and tissue metabolic heat terms can be further introduced into the control equation, depending on tissue characteristics and prediction accuracy requirements.

[0049] Furthermore, the optimization objectives of the prediction model include at least the following: basic temperature field consistency constraint, local temperature measurement consistency constraint, physical consistency constraint, and correction constraint. Specifically, the basic temperature field consistency constraint is used to ensure consistency between the predicted basic temperature field and the numerically solved sample temperature field after parameter correction; the local temperature measurement consistency constraint is used to ensure consistency between the final temperature field prediction result and the measured subcutaneous local temperature at the actual experimental temperature measurement location; the physical consistency constraint is used to ensure that the model output satisfies the tissue heat transfer control equation, boundary conditions, initial conditions, and prior parameter laws; and the correction constraint is used to limit the amplitude and spatial fluctuation of the temperature field correction, thereby improving the stability of the correction results and reducing the risk of the predicted results in unmeasured areas deviating from a reasonable physical distribution.

[0050] Basic temperature field consistency constraints for:

[0051] Where M represents the total number of samples in the basic temperature field. Indicates the first Each basic temperature field sample includes the corresponding operating condition input information; Indicates the first The spatial location corresponding to each basic temperature field sample. This indicates that the basic temperature field prediction module is in the first... The predicted base temperature values ​​output at the corresponding operating conditions and spatial locations of each base temperature field sample. This represents the numerical temperature value obtained at the corresponding operating conditions and spatial location after numerically solving the biological heat conduction control equation with parameters calibrated using real experimental data. Let the first The input variables corresponding to each measured constraint sample are: Temperature measurement location is The measured subcutaneous local temperature at this location was [value missing]. Then the local temperature measurement consistency constraint It can be represented as:

[0052] in, This represents the total number of constrained samples for local temperature measurements; This represents the working condition input information corresponding to the nth measured constraint sample; This indicates the thermocouple temperature measurement position of the nth measured constraint sample; This represents the measured subcutaneous local temperature value of the nth measured constrained sample at the corresponding temperature measurement location.

[0053] The correction constraint may include at least a correction magnitude constraint and a smoothness constraint to limit the unreasonable amplification of the overall temperature field by local measured information and reduce the risk of significant distortion in the prediction results of unmeasured areas. In some embodiments, the correction constraint... It can be represented as:

[0054] Physical consistency constraints for:

[0055] in, This represents the total number of physical constraint sampling points. Indicates the first The operating condition input information corresponding to each physical constraint sampling point; They represent the first The spatial location corresponding to each physical constraint sampling point; , and These represent the tissue density, specific heat capacity, and thermal conductivity, respectively. Indicates the first The final temperature prediction value output by each physical constraint sampling point at the corresponding working condition and corresponding spatial location; Indicates the external heat source in the th The heat source term formed at each physical constraint sampling point; Indicates the first The tissue state-related equivalent heat exchange term at each physical constraint sampling point; These are the smoothness constraint weighting coefficients; This indicates the temperature correction amount output at the corresponding operating condition and spatial location.

[0056] In some embodiments, the prediction model may further include a reverse consistency constraint on the surface temperature response or a continuity constraint on the layered structure interface. The reverse consistency constraint on the surface temperature response is used to ensure consistency between the surface temperature response derived or mapped from the predicted internal temperature location and the actual surface temperature response information; the continuity constraint on the layered structure interface is used to ensure that the temperature distribution or heat transfer at the interface of adjacent structure layers meets the continuity requirement.

[0057] In some implementations, the prediction model may be implemented using a physical information neural network, a neural operator model, DeepONet, a residual network, or other model structures that can achieve the fusion of physical constraints and multi-source data, but is not limited to a specific model form.

[0058] By constructing the model as described above, the prediction model can learn the variation law of the subcutaneous fat base temperature field under different working conditions by using numerical solution sample data after parameter correction, and can also use local temperature information from real experiments to collaboratively correct the overall temperature field. Under the combined effect of base temperature field consistency constraints, local temperature measurement consistency constraints, physical consistency constraints, and correction amount constraints, the risk of the predicted temperature results in unmeasured areas deviating from the reasonable physical distribution is reduced, thereby achieving rapid, accurate, and physically consistent prediction of the subcutaneous fat temperature field.

[0059] S4. Joint solution and optimization based on physical constraints and data-driven approaches Based on the subcutaneous fat temperature field prediction model constructed in step S3, the prediction model is jointly solved and optimized using numerical solution sample data corrected by real experimental data, as well as surface temperature data and measured subcutaneous local temperature data obtained from real experiments.

[0060] In some implementations, the joint solution and optimization process includes: constructing a basic temperature field sample set, constructing a local temperature measured constraint sample set, establishing a joint optimization objective that includes basic temperature field consistency constraints, local temperature measured consistency constraints, physical consistency constraints, and correction constraints, and iteratively updating the model parameters based on the joint optimization objective to obtain an optimized subcutaneous fat temperature field prediction model.

[0061] Furthermore, the joint optimization objective can be expressed as:

[0062] in, This represents the basic temperature field consistency constraint term. This represents a constraint term indicating the consistency of local temperature measurements. Represents physical consistency constraints. Indicates the correction constraint term; to This represents the weight coefficient corresponding to each constraint term.

[0063] In some implementations, the model parameters can be solved using a phased optimization approach, a joint optimization approach, or a combination thereof. Preferably, the basic temperature field prediction model can first be pre-trained based on the corrected numerical solution sample data to learn the distribution patterns of the subcutaneous fat basic temperature field under different working conditions; then, combined with local temperature constraint information and physical consistency constraints obtained from real experiments, the overall prediction model can be further optimized to improve the consistency between the final prediction results and the actual tissue heat transfer process.

[0064] Through the above-mentioned joint solution and optimization process, the prediction model can effectively integrate numerical solution sample information and real experimental temperature measurement information under the constraints of physical mechanisms, thereby improving the accuracy, stability and physical rationality of the subcutaneous fat temperature field prediction results.

[0065] S5. Achieve subcutaneous fat temperature field prediction The surface temperature data of the biological tissue to be tested, the parameters of the external heat source, the boundary condition parameters, and the tissue structure parameters are input into the prediction model after joint optimization in step S4. The model outputs the temperature distribution of subcutaneous adipose tissue under the corresponding conditions. The predicted temperature of subcutaneous adipose tissue includes at least one or more of the following: local temperature value at a certain depth, two-dimensional temperature distribution of subcutaneous adipose layer, three-dimensional temperature distribution of subcutaneous adipose layer, and temperature profile along the depth direction.

[0066] Furthermore, the prediction process can be performed in real time or non-real time, supporting both steady-state temperature field prediction based on single-time-time input data and dynamic temperature field evolution prediction based on time-series input data.

[0067] Furthermore, in some embodiments, the prediction model may also output confidence information, uncertainty estimation information, or error range information corresponding to the temperature field prediction results, so as to provide a reference for subsequent thermotherapy evaluation.

[0068] S6, Temperature Field Characterization and Thermal Perspective Visualization Output The subcutaneous fat temperature field predicted by S5 is visualized to form a thermal imaging result that can be intuitively identified and analyzed.

[0069] The thermal visualization results include at least one or more of the following: a pseudo-color temperature field cloud map, using different colors to represent the temperature levels of different regions of the fat layer; an isotherm map, used to indicate the spatial distribution of specific temperature thresholds; a temperature profile curve, used to display the temperature change trend along a specified path; a heat dose distribution map, used to calculate the equivalent heat dose based on temperature and time history, reflecting the thermal effect; three-dimensional temperature field rendering, used to display the temperature distribution in three-dimensional space; and a dynamic temperature evolution video, used to display the temperature field's change process over time.

[0070] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0071] Example This embodiment provides a subcutaneous fat thermal imaging method based on both physical mechanisms and multi-data dual-drive. Taking laser treatment of layered biological tissue with epidermis, dermis, and subcutaneous fat layers as an example, it illustrates a method for predicting subcutaneous fat temperature based on a combination of multi-source data acquisition and a physical mechanism embedded model. This method simultaneously acquires surface temperature response, local subcutaneous temperature reference information, laser treatment parameters, and tissue structure parameters during laser treatment. Combined with numerical solution sample data corrected from real experimental data, it constructs a model capable of predicting the subcutaneous fat temperature field from the surface thermal response, thereby outputting the subcutaneous fat layer temperature distribution and thermal imaging visualization results.

[0072] Specifically as follows: S1. Multi-source data acquisition and sample set construction In this embodiment, a biological tissue with a layered structure is selected as the research object, and the biological tissue includes at least the epidermis, dermis and subcutaneous fat layer.

[0073] 1. Acquisition of real experimental data This invention utilizes an external heat source to heat biological tissue and obtains real temperature response data under the influence of the external heat source. Preferably, the external heat source is a laser heat source, but the invention is not limited to this; it is also applicable to other external heat source scenarios that can cause tissue temperature rise.

[0074] Specifically, infrared thermal imaging equipment is used to collect temperature data on the surface of biological tissue, obtaining a data sequence of surface temperature changes over time. Temperature sensors are placed at predetermined locations in the subcutaneous fat layer. These sensors can be thermocouples, temperature probes, or other temperature measuring devices suitable for acquiring localized temperatures within the tissue, used to obtain measured values ​​of the subcutaneous local temperature. The predetermined locations can be one or more temperature measurement points at a predetermined depth from the tissue surface; different measurement points can be located at different depths or lateral positions within the fat layer. Simultaneously, external heat source parameters are recorded, including at least heat source power, duration of action, spot size, and location of action. Boundary condition parameters are also recorded, including at least ambient temperature, convective heat transfer conditions, and cooling method. Tissue structure parameters are also acquired simultaneously, including at least the thickness information of each tissue layer and information on its layered structure.

[0075] The aforementioned surface temperature data, subcutaneous local temperature data, external heat source parameters, boundary condition parameters, and tissue structure parameters together constitute the real experimental dataset.

[0076] In this embodiment, a set of real experimental samples includes at least the surface temperature time series under the same heating condition, the corresponding external heat source parameters, boundary condition parameters, tissue structure parameters, and the measured subcutaneous local temperature values ​​at one or more preset temperature measurement locations.

[0077] 2. Numerical Solution Sample Data Acquisition To expand the range of operating conditions covered by the samples required for model training, this embodiment further constructs a numerical solution process based on the biological heat conduction control equation to generate temperature field sample data under different operating conditions.

[0078] Specifically, a layered tissue calculation model is first established, which includes at least the epidermis, dermis, and subcutaneous fat layer. Based on this, corresponding parameter information is set for each tissue layer. This parameter information includes at least layer thickness parameters, thermophysical parameters, and parameters related to the effect of an external heat source. The thermophysical parameters may include density, specific heat capacity, and thermal conductivity; these parameters can be initially set based on publicly available literature, empirical databases, preset standard parameters, or existing experimental results.

[0079] Furthermore, considering that different individual states, tissue conditions, and actual operating conditions of the hyperthermia device may lead to deviations between parameters and reference values, at least some of the parameters can be calibrated, corrected, or recalibrated using experimentally measured temperature data. The calibrable parameters preferably include one or more of boundary heat transfer-related parameters, external heat source-related parameters, and equivalent heat transfer-related parameters. In this embodiment, iterative correction of the parameters to be determined is preferably performed by comparing the deviations between the numerically solved surface temperature response, local deep temperature response, and the corresponding results from actual experiments. The calibration method can be as follows: under given external heat source conditions and temperature measurement conditions, the calculated results are compared with the actually measured surface temperature, deep temperature, or temperature rise process, and the parameters to be determined are adjusted based on the comparison results to improve the consistency between the numerically solved results and the actual thermal response.

[0080] After parameter setting or calibration, a temperature field control model for the internal tissue is established based on the external heat source conditions, initial tissue temperature conditions, and boundary heat transfer conditions. This model describes the heat transfer and evolution process within the layered tissue under the influence of an external heat source. Preferably, the control model comprehensively considers the internal tissue conduction process, the heat transfer process between the tissue surface and the external environment, and the energy deposition and attenuation process of the external heat source in each tissue layer. For laser-based scenarios, the control model further includes heat source characterization related to laser power, duration of action, location of action, spot size, and tissue absorption characteristics to reflect the heat input characteristics of the laser in the epidermis, dermis, and subcutaneous fat layer.

[0081] Subsequently, the control model is solved using numerical methods to obtain temperature distribution results under different combinations of tissue parameters, external heat source parameters, and boundary conditions. The numerical method can be the finite element method, finite difference method, finite volume method, or other suitable numerical solution methods. Preferably, the solution process can be implemented through a multiphysics simulation platform, but is not limited to a specific software platform.

[0082] Finally, the surface temperature information, subcutaneous temperature information, temperature field distribution results and their corresponding operating parameters obtained from the numerical solution are uniformly organized to construct a numerical solution sample dataset for subsequent model training, verification or optimization.

[0083] 3. Construction of multi-source sample sets Furthermore, the real experimental dataset and the numerical solution sample dataset corrected by experiments are uniformly organized to construct a multi-source sample set for subsequent model training, constraint optimization and performance evaluation.

[0084] S2. Multi-source data preprocessing and unified representation The real experimental data and numerical solution sample data obtained in step S1 are preprocessed and uniformly represented. Specifically, this includes: removing abnormal or severely missing samples, and making necessary corrections to local abnormal data; standardizing the units, time, space, and numerical values ​​of temperature information, external heat source parameters, boundary condition parameters, and tissue structure parameters from different sources; using the processed surface temperature information, external heat source parameters, boundary condition parameters, and tissue structure parameters as model inputs, and using the corresponding subcutaneous fat layer temperature information as model output labels to construct a unified multi-source sample representation.

[0085] In some implementations, the numerical solution sample data can also be perturbed to reduce the distribution difference between the simulated samples and the real experimental samples.

[0086] S3. Construct a subcutaneous fat temperature field prediction model that integrates the physical mechanisms of tissue heat transfer. Based on the unified data representation obtained in step S2, a subcutaneous fat temperature field prediction model is constructed that integrates the physical mechanism of tissue heat transfer with multi-source data constraints. This prediction model is used to output the predicted subcutaneous fat temperature field under corresponding working conditions, provided that surface temperature response information, external heat source parameters, boundary condition parameters, and tissue structure parameters are known.

[0087] In this embodiment, the prediction model adopts a two-step segmented structure that combines the aforementioned basic temperature field prediction with local constraint correction. First, based on surface temperature response information, external heat source parameters, boundary condition parameters, and tissue structure parameters, the basic temperature field prediction result of subcutaneous fat is obtained. Then, combined with local temperature constraint information obtained from real experiments, the basic temperature field prediction result is collaboratively corrected to obtain the final subcutaneous fat temperature field prediction result.

[0088] In some implementations, the model input variables are assumed to be...

[0089] in, The surface temperature response information includes one or more of the following: surface temperature distribution at a single moment acquired by an infrared thermal imager, surface temperature sequence within a time window, processed and extracted surface temperature statistical feature information, or other information that can characterize the thermal process. Indicates the parameters of the external heat source; Indicates boundary condition parameters; Represent organizational structure parameters; let If the spatial location variable is represented, then the predicted result of the subcutaneous fat basal temperature field can be expressed as:

[0090] in, This indicates the predicted results of the basic temperature field. This represents the basic temperature field prediction model.

[0091] Furthermore, the local temperature constraint information obtained from real experiments can be represented as multiple measured constraint samples. Each measured constraint sample includes at least the input condition information under the corresponding operating condition, the temperature measurement location, and the measured subcutaneous local temperature value at that location. Based on the local temperature constraint information, the basic temperature field prediction results are collaboratively corrected to obtain the temperature field correction amount.

[0092] in, This represents the temperature field correction amount. This represents the temperature field correction model.

[0093] The final predicted subcutaneous fat temperature field can be expressed as:

[0094] By using the above two-step segmented prediction method, we can first use numerical solution samples corrected by real experimental data to learn the variation law of the subcutaneous fat basal temperature field under different working conditions, and then use the local temperature constraint information obtained from real experiments to correct the basal temperature field, thereby improving the adaptability of the final prediction results to real working conditions as well as the accuracy and stability of the prediction results.

[0095] In this embodiment, the tissue heat transfer physical mechanism is mainly used to construct numerical solution samples and constrain the physical rationality of the predicted temperature field. In the actual application stage, the biological heat conduction control equation is not directly solved numerically online; instead, the prediction results are output by the trained subcutaneous fat temperature field prediction model.

[0096] Ex vivo tissue and living tissue are characterized as different tissue states within the tissue heat transfer description framework. For the ex vivo tissue scenario, the governing equations are simplified to exclude blood perfusion and tissue metabolic heat terms, or the relevant effects are incorporated into the equivalent heat transfer term, satisfying the following equation:

[0097] For the living tissue scenario, based on tissue characteristics and prediction accuracy requirements, blood perfusion rate, blood temperature, and tissue metabolic heat terms are further introduced into the control equations; satisfying the following equation:

[0098] in, This indicates heat conduction due to temperature difference. Indicates time, Indicates tissue temperature, Indicates tissue density, Indicates the specific heat capacity of the tissue. Indicates the thermal conductivity of the tissue. This represents the heat source term generated within the tissue by an external heat source. Indicates blood perfusion rate, This indicates the specific heat capacity of blood. Indicates blood temperature. This indicates the pyrogen content of tissue metabolism. Indicates the state of the tissue and is used to distinguish between isolated tissue and living tissue.

[0099] During the numerical sample generation process, corresponding microstructure parameters are set according to the microstructure of the sample, and the surface temperature response and internal complete temperature field under the corresponding working conditions are obtained through numerical solution. The numerical solution results are used to construct the complete temperature field samples required for training the prediction model.

[0100] During the training of the prediction model, the physical residuals of the predicted temperature field can also be calculated based on the aforementioned unified tissue heat transfer relationship, and these physical residuals can be used as physical consistency constraints. It should be noted that the tissue heat transfer control relationship is mainly used for numerical sample generation and model training constraints, and does not mean that the subcutaneous fat temperature can be solved time-by-time using this control relationship in the actual application stage.

[0101] In this embodiment, the above parameters are respectively taken in the epidermal layer, dermal layer and subcutaneous fat layer regions to characterize the differences in thermal conductivity of the layered tissues; at the interface of adjacent tissue layers, the conditions of continuous temperature and continuous heat flow can be further satisfied to ensure the physical rationality of the temperature field distribution.

[0102] In some implementations, the external heat source item The external heat source term can be constructed based on the power of the external heat source, the duration of action, the area of ​​action, the location of action, the trajectory of action, and the absorption and attenuation characteristics of the tissue to heat energy. Preferably, in a laser action scenario, the external heat source term can be constructed based on the laser power, spot size, duration of action, incident location, and tissue photothermal response characteristics to characterize the deposition process of laser energy in the superficial tissue and subcutaneous fat region; for other external heat sources that can cause tissue temperature rise, the external heat source term can also be adjusted according to the energy deposition mode of the corresponding heat source.

[0103] Furthermore, at the tissue surface boundary, surface boundary constraints can be set according to ambient temperature, convective heat transfer conditions, and cooling conditions to characterize the heat transfer behavior between the tissue surface and the outside world. During the model optimization process, a reverse consistency constraint on the surface temperature response can also be introduced to ensure consistency between the surface temperature response derived, mapped, or projected from the predicted internal temperature field and the actual surface temperature response information.

[0104] In some implementations, the optimization objectives of the prediction model include at least the following: basic temperature field consistency constraint, local temperature measurement consistency constraint, physical consistency constraint, and correction constraint. Specifically, the basic temperature field consistency constraint is used to ensure consistency between the predicted basic temperature field and the numerically solved sample temperature field corrected from real experimental data; the local temperature measurement consistency constraint is used to ensure consistency between the final predicted temperature field and the measured subcutaneous local temperature at the actual experimental temperature measurement location; the physical consistency constraint is used to ensure that the model output meets the requirements of the tissue heat conduction control equation, boundary conditions, initial conditions, and the continuity of the layered tissue interface; and the correction constraint is used to limit the amplitude and spatial fluctuation of the temperature field correction to avoid unreasonable amplification of local temperature measurement information in unmeasured areas and reduce the risk of the prediction results deviating from a reasonable physical distribution.

[0105] In this embodiment, the prediction model preferably employs a physically constrained neural network structure capable of receiving operating parameters and surface thermal response information and outputting temperature values ​​corresponding to spatial locations. In some implementations, the prediction model may employ a physical information neural network, a neural operator model, DeepONet, a residual network, or other model structures capable of fusing physical constraints and multi-source data, but is not limited to a specific model form. Preferably, while using numerical solutions to learn the basic temperature distribution law from samples, the prediction model also incorporates local temperature information from real experiments to correct the overall temperature field, making the model output more closely resemble the thermal response process of real tissues while satisfying the physical laws of tissue heat transfer.

[0106] It should be noted that the unified tissue heat transfer control equation in this embodiment is a preferred form based on the Pennes biological heat conduction equation, used to describe the heat conduction within layered biological tissues, energy deposition from external heat sources, tissue state-related heat exchange, and heat transfer processes between the tissue surface and the external environment under the influence of external heat sources. In another embodiment, other biological heat transfer control equations can also be used to construct the biological heat conduction control equation, such as non-Fourier biological heat transfer equations considering heat propagation hysteresis effects, two-phase hysteresis biological heat transfer equations, or biological heat transfer models based on porous media theory. The biological heat transfer control equation used should at least be able to characterize the heat conduction process within layered tissues, the energy deposition process of external heat sources within tissues, and the heat transfer process between the tissue surface and the external environment; for living tissues, it should also be able to characterize the effects of blood perfusion, tissue metabolic heat production, or other physiological thermoregulatory factors on tissue temperature changes. The specific network structure of the prediction model can be set according to the input data format and temperature field output dimension, but should at least include a basic temperature field prediction process based on surface temperature response and related operating parameters, and a temperature field correction process supervised using real experimental local temperature data.

[0107] S4. Joint solution or optimization based on physical constraints and data-driven approaches. Based on the subcutaneous fat temperature field prediction model constructed in step S3, the prediction model is jointly solved and optimized using numerical solution sample data corrected by real experimental data, as well as surface temperature data and measured subcutaneous local temperature data obtained from real experiments.

[0108] In this embodiment, the joint solution and optimization process includes: constructing a basic temperature field sample set, which mainly consists of numerical solution sample data corrected by real experimental data; constructing a local temperature measurement constraint sample set, which mainly consists of surface temperature data, subcutaneous local temperature measurement data, and corresponding operating parameters obtained from real experiments; establishing a joint optimization objective that includes basic temperature field consistency constraints, local temperature measurement consistency constraints, physical consistency constraints, and correction constraints; and iteratively updating the model parameters based on the joint optimization objective to obtain the optimized subcutaneous fat temperature field prediction model.

[0109] In some implementations, the joint optimization objective includes at least a basic temperature field consistency constraint, a local temperature measurement consistency constraint, a physical consistency constraint, and a correction constraint, which can be expressed as:

[0110] in, This represents the consistency constraint term for the basic temperature field, which is used to constrain the consistency between the predicted results of the basic temperature field and the numerically solved sample temperature field after correction based on real experimental data. This represents a local temperature measurement consistency constraint term, used to constrain the consistency between the final temperature field prediction result and the actual experimental temperature measurement location and the measured value of the subcutaneous local temperature. This represents a physical consistency constraint term, used to constrain the model output to meet the tissue heat conduction control equation, boundary conditions, initial conditions, and layered tissue interface continuity requirements described in step S3. This represents a correction constraint term, used to limit the magnitude and spatial fluctuation of the temperature field correction. to This represents the weight coefficient corresponding to each constraint term.

[0111] Furthermore, the basic temperature field consistency constraint term can be used to constrain the consistency between the basic temperature field results output by the basic temperature field prediction module and the basic temperature field sample set; the physical consistency constraint term can be used to constrain the model output to satisfy the layered heat transfer mechanism, surface heat transfer conditions, and heat conduction law under the action of external heat sources; the correction amount constraint term can be used to suppress the unreasonable amplification caused by local measured information in unmeasured areas, and reduce the risk of obvious oscillations or local distortions in the temperature field correction results.

[0112] In this embodiment, the model parameters can be solved using a phased optimization approach, a joint optimization approach, or a combination of both. Preferably, the basic temperature field prediction module can be pre-trained based on a basic temperature field sample set to learn the distribution patterns of the subcutaneous fat basic temperature field under different working conditions. Then, surface temperature data and measured subcutaneous local temperature data obtained from real experiments are introduced to further optimize the local constraint correction module and the overall model, so that the final prediction result is closer to the real tissue thermal response process while maintaining physical rationality.

[0113] In another embodiment, the basic temperature field sample set and the local temperature measured constraint sample set can be input into the model together, and the basic temperature field learning, local temperature correction and physical consistency constraint optimization can be completed simultaneously through joint training.

[0114] Furthermore, the weight coefficients of each constraint term can be preset, adjusted in stages, or dynamically according to the training stage, prediction error, physical residual, and model convergence status to achieve a balance between data-driven learning and physical mechanism constraints. Preferably, in the early stage of model training, the weight of the basic temperature field consistency constraint term can be appropriately increased, so that the model learns the basic temperature field distribution law first; in the later stage of model training, the weights of the local temperature measurement consistency constraint term and the correction amount constraint term can be appropriately increased, so that the model completes the stable correction of the overall temperature field under the constraints of real experiments.

[0115] Through the above-mentioned joint solution and optimization process, the prediction model can not only learn the variation law of the subcutaneous fat base temperature field under different working conditions by using numerical solution sample data corrected by real experimental data, but also combine the surface temperature data and measured subcutaneous local temperature data obtained from real experiments to collaboratively correct the overall temperature field, thereby improving the accuracy, stability and physical rationality of the subcutaneous fat temperature field prediction results.

[0116] It should be noted that the above-described joint solution and optimization process is only a preferred implementation in this embodiment, and the present invention is not limited to the specific optimization process described above. In other embodiments, the composition, solution order, weight setting method, and parameter update method of the constraint terms can be adjusted or replaced according to the sample size, data source, heat source type, tissue structure characteristics, and prediction accuracy requirements. As long as the optimization of the subcutaneous fat temperature field prediction model based on physical constraints and multi-source data collaboration can be achieved, it falls within the protection scope of the present invention.

[0117] S5. Achieve subcutaneous fat temperature field prediction After the model training is completed, the surface temperature data of the biological tissue under test under the action of an external heat source, the parameters of the external heat source, the boundary condition parameters and the tissue structure parameters are input into the prediction model to obtain the predicted results of the temperature field of the subcutaneous fat layer.

[0118] The prediction results can be single-point subcutaneous temperature values, temperature sequences at a certain depth, two-dimensional temperature distribution of the subcutaneous fat layer, three-dimensional temperature distribution of the subcutaneous fat layer, or dynamic temperature evolution results.

[0119] In some implementations, the model may also output uncertainty information or error range of the prediction results to assess the reliability of the prediction results.

[0120] In some implementations, once new surface temperature observation data or new subcutaneous temperature reference data are obtained, the model output results can be corrected or the model parameters can be further optimized to improve the model's prediction accuracy in specific scenarios.

[0121] S6, Temperature Field Characterization and Thermal Perspective Visualization Output The subcutaneous fat temperature prediction results output by the model are spatially mapped and visualized to generate thermal visualization results.

[0122] In this embodiment, the predicted discrete temperature values ​​can be mapped to corresponding spatial locations, and a two-dimensional temperature distribution map can be formed using pseudo-color encoding. Alternatively, depth temperature profile maps can be generated based on prediction results at different depths. Furthermore, a three-dimensional temperature field map can be generated based on three-dimensional spatial coordinates. For continuous prediction results at different time points, dynamic temperature evolution display results can also be generated.

[0123] Furthermore, high-temperature areas, low-temperature areas, or thermal risk areas can be marked according to preset temperature thresholds to form thermal risk distribution maps or effective temperature rise area maps. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A subcutaneous fat thermal imaging method based on physical mechanisms and multi-data dual-drive, comprising: The surface temperature data of the biological tissue to be tested, the parameters of the external heat source, the boundary condition parameters and the tissue structure parameters are input into the pre-trained prediction model. The prediction model includes a basic temperature field prediction module and a temperature field correction module. The temperature field correction amount output by the temperature field correction module is used to correct the basic temperature field prediction result output by the basic temperature field prediction module, and outputs the temperature distribution of subcutaneous adipose tissue under the corresponding conditions. During the model training phase, the basic temperature field prediction module trains on temperature field samples generated by numerically solving the biological heat conduction control equation based on parameters calibrated by real experimental data. This is used to learn the mapping relationship between different input conditions and the basic temperature field of subcutaneous fat. During the model training phase, the temperature field correction module uses the temperature deviation between the actual measured subcutaneous local temperature in real experiments and the predicted basic temperature field at the corresponding temperature measurement location as supervision information to learn the mapping relationship between different input conditions and the temperature field correction amount, and optimizes it under the constraints of tissue heat transfer physical consistency, correction amount amplitude, and spatial smoothness.

2. The subcutaneous fat thermal imaging method based on physical mechanism and multi-data dual-drive as described in claim 1, characterized in that, The surface temperature data includes at least one of the following: surface temperature distribution at a single moment acquired by an infrared thermal imager, surface temperature sequence within a time window, processed and extracted surface temperature statistical feature information, and other information that can characterize the thermal process.

3. The subcutaneous fat thermal imaging method based on physical mechanism and multi-data dual-drive as described in claim 1, characterized in that, The external heat source action parameters are used to characterize the intensity, duration, location, and trajectory of the external heat source action. The boundary condition parameters are used to characterize the surface heat transfer conditions, cooling conditions, and initial temperature conditions. The tissue structure parameters are used to characterize the tissue layer structure information and the thickness information of each layer.

4. The subcutaneous fat thermal imaging method based on physical mechanism and multi-data dual-drive as described in claim 1, characterized in that, The basic temperature field prediction module includes an input feature encoding unit, a feature fusion unit, and a decoding unit; the temperature field correction module includes a correction feature extraction unit and a residual output unit.

5. The subcutaneous fat thermal imaging method based on physical mechanism and multi-data dual-drive as described in claim 4, characterized in that, The method also includes a training step for the prediction model, comprising: Acquire real experimental data, solve data based on the biological heat conduction control equation, and construct a sample set including experimental data and solution data; The data in the sample set is preprocessed and converted into a unified data representation format; Joint optimization objective for constructing a prediction model based on multi-source data constraints; The prediction model is trained based on the joint optimization objective to obtain a prediction model that meets the training requirements.

6. The subcutaneous fat thermal imaging method based on physical mechanism and multi-data dual-drive as described in claim 5, characterized in that, The actual experimental data includes at least: surface temperature data, external heat source parameters, boundary condition parameters, and tissue structure parameters; and measured data of local subcutaneous temperature changes with laser treatment time, collected by deploying temperature sensors at the target depth location under the skin of the experimental sample, as labels.

7. The subcutaneous fat thermal imaging method based on physical mechanism and multi-data dual-drive as described in claim 5, characterized in that, The data obtained from solving the biological heat conduction control equation specifically includes: Establish a layered tissue calculation model that includes at least the epidermis, dermis and subcutaneous fat layer, and set corresponding parameter information for each layer. The parameter information includes at least layer thickness parameters, thermophysical parameters and parameters related to the effect of external heat sources. Establish the temperature field governing equations; for the isolated tissue scenario, the governing equations adopt a simplified form that excludes blood perfusion terms and tissue metabolic heat terms, satisfying the following equation: For the living tissue scenario, based on tissue characteristics and prediction accuracy requirements, blood perfusion rate, blood temperature, and tissue metabolic heat terms are further introduced into the control equations; satisfying the following equation: in, This indicates heat conduction due to temperature difference. Indicates time, Indicates tissue temperature, Indicates tissue density, Indicates the specific heat capacity of the tissue. Indicates the thermal conductivity of the tissue. This represents the heat source term generated within the tissue by an external heat source. Indicates blood perfusion rate, This indicates the specific heat capacity of blood. Indicates blood temperature. This indicates the pyrogen content of tissue metabolism. Indicates tissue state, used to distinguish between isolated tissue and living tissue; Temperature field samples are generated through numerical solutions.

8. The subcutaneous fat thermal imaging method based on physical mechanism and multi-data dual-drive as described in claim 5, characterized in that, The multi-source data constraints include: basic temperature field consistency constraints. Consistency constraints of measured local temperatures Physical consistency constraints and correction constraints They respectively satisfy the following formulas: Where M represents the total number of samples in the basic temperature field. Indicates the first Each basic temperature field sample includes the corresponding operating condition input information; Indicates the first The spatial location corresponding to each basic temperature field sample. This indicates that the basic temperature field prediction module is in the first... The predicted base temperature values ​​output at the corresponding operating conditions and spatial locations of each base temperature field sample. This represents the numerical temperature value obtained at the corresponding working condition and spatial location after the biological heat conduction control equation, after the parameters have been calibrated using real experimental data, is numerically solved. This represents the total number of constrained samples for local temperature measurements; Indicates the first The working condition input information corresponding to each measured constraint sample. This indicates the thermocouple temperature measurement position of the nth measured constraint sample; This represents the measured subcutaneous local temperature value of the nth measured constrained sample at the corresponding temperature measurement location. This indicates the final temperature field prediction result; in, This represents the total number of physical constraint sampling points. Indicates the first The operating condition input information corresponding to each physical constraint sampling point; They represent the first The spatial location corresponding to each physical constraint sampling point; Indicates the first The final temperature prediction value output by each physical constraint sampling point at the corresponding working condition and corresponding spatial location; , and These represent the tissue density, specific heat capacity, and thermal conductivity, respectively. Indicates the external heat source in the th The heat source term formed at each physical constraint sampling point; Indicates the first The tissue state-related equivalent heat exchange term at each physical constraint sampling point; These are the smoothness constraint weighting coefficients; This indicates the temperature correction amount output at the corresponding operating condition and spatial location; Joint optimization objective for constructing a prediction model based on multi-source data constraints for: in, , , and This represents the weight coefficients corresponding to each constraint term. These coefficients are preset based on the training stage, prediction error, physical residual, and model convergence status, and are adjusted in stages or dynamically to achieve a balance between data-driven learning and physical mechanism constraints.

9. The subcutaneous fat thermal imaging method based on physical mechanism and multi-data dual-drive as described in claim 5, characterized in that, The prediction model is trained using a phased optimization approach, a joint optimization approach, or a combination of both.

10. The subcutaneous fat thermal imaging method based on physical mechanism and multi-data dual-drive as described in claim 1, characterized in that, The temperature distribution of subcutaneous adipose tissue under the corresponding conditions is: single-point subcutaneous temperature value, temperature sequence at a certain depth, two-dimensional temperature distribution of subcutaneous adipose layer, three-dimensional temperature distribution of subcutaneous adipose layer, or dynamic temperature evolution result. The output of the prediction model also includes: uncertainty information or error range of the prediction results, used to assess the reliability of the prediction results.

Images