Self-adjusting laser power control method and system based on temperature prediction

Abstract

The application discloses a kind of self-adjusting laser power control method and system based on temperature prediction, belong to medical equipment intelligent control technical field.Through constructing a forward-looking adaptive control closed loop: system first real-time continuous acquisition and constructs the temperature sequence of target area;Using pre-trained time series prediction model, based on current and historical temperature and power data, predict the temperature change trend in future short term;When it is predicted that temperature will deviate from the preset safe and effective treatment window, calculate and output smooth laser power regulation instruction in advance, drive actuator to actively intervene.The application can dynamically compensate biological tissue thermal inertia and environmental disturbance, maintain the temperature of treatment area in the preset target range with high precision and high stability, significantly improve the safety and controllability of treatment, and provide a key intelligent control solution for the clinical transformation of photothermal therapy.

Description

Technical Field

[0001] This invention discloses a self-adjusting laser power control method and system based on temperature prediction, belonging to the field of intelligent control technology for medical devices. Background Technology

[0002] In the field of photothermal therapy, achieving precise and stable temperature control of the target area is a core challenge in ensuring the safety and effectiveness of treatment. Current mainstream laboratory-level temperature control methods heavily rely on the operator's personal experience. Specifically, operators need to intermittently observe the readings of equipment such as infrared thermal imagers and manually adjust the laser's output power or irradiation position based on this information. This observation-response mode is essentially a delayed and discrete manual intervention process.

[0003] Because of the inertia inherent in the thermodynamic processes of biological tissues, and because temperature monitoring and power regulation are isolated processes, the aforementioned methods have inherent control flaws. By the time the operator observes that the temperature is too high, harmful heat accumulation may have already occurred; conversely, by the time insufficient temperature is detected, the treatment window has already been wasted. This control method cannot make forward-looking judgments about future temperature trends, nor can it achieve continuous and smooth power fine-tuning, resulting in a coarse control process and poor repeatability of results, severely hindering the standardized implementation and clinical translation of complex photothermal therapy protocols.

[0004] While some laser therapy devices on the market possess temperature feedback capabilities, their control logic is mostly based on simple threshold-triggered switching. For example, the system only shuts down or reduces power when the monitored temperature reaches a set upper limit. This still fails to solve the temperature overshoot problem caused by thermal inertia and easily leads to repeated temperature oscillations. Therefore, there is an urgent need for an intelligent control method that can actively predict temperature change trends and make feedforward, adaptive, and continuous adjustments accordingly to overcome the current technical bottleneck of achieving precise temperature control in the complex environment of a living organism in photothermal therapy. Summary of the Invention

[0005] To achieve the above objectives, this application provides the following technical solution: According to a first aspect of the present invention, the present invention claims a method for self-adjusting laser power control based on temperature prediction, the method comprising the following steps: S1. Real-time acquisition of continuous temperature monitoring data of the target area irradiated by the laser, forming the current temperature sequence; S2. Based on the current temperature sequence and current laser power parameters, the predicted temperature value of the target area in the future within a predetermined time period is calculated by the prediction model; S3. Compare the predicted temperature value with the preset target temperature range to determine whether the predicted temperature value will deviate from the target temperature range. S4. If the judgment result of step S3 is yes, then based on the deviation trend and degree of the predicted temperature value relative to the target temperature range, and the current laser power parameters, calculate the adjusted laser power parameters required to make the predicted temperature value return to and stabilize within the target temperature range. S5. Based on the adjusted laser power parameters calculated in step S4, generate a corresponding laser power control command and send the control command to the laser emitting device to drive the laser emitting device to adjust its output power to the level corresponding to the adjusted laser power parameters; S6. After completing the power adjustment in step S5, return to step S1 and continue to monitor, predict and adjust the temperature in the next cycle, thereby achieving dynamic closed-loop stable control of the temperature in the target area.

[0006] Furthermore, the real-time acquisition of continuous temperature monitoring data of the target area irradiated by the laser in step S1 is achieved by a non-contact infrared thermal imaging temperature measurement device. The non-contact infrared thermal imaging temperature measurement device scans the target area at a first sampling frequency not lower than a preset one to obtain an infrared thermal image containing the surface temperature distribution of the target area. The pixel temperature values ​​corresponding to the preset region of interest are extracted from each frame of the infrared thermal image, and spatial averaging is performed to obtain a single-point temperature value representing the temperature of the target region at that moment. These values ​​are then arranged in chronological order to form the current temperature sequence. The region of interest is automatically defined in the initial stage by an image recognition algorithm based on the morphological characteristics of the target region, or it can be manually specified by the operator.

[0007] Furthermore, the prediction model in step S2 is a dynamic prediction model based on time series analysis. Its input is the current temperature sequence containing the temperature values ​​of the most recent N historical moments and the laser power parameter sequence of the corresponding time period. The output is the predicted temperature values ​​for the next M consecutive time points, where N and M are both positive integers. The prediction model is built and run through the following steps: S2.1. Combine the temperature values ​​[T(t-N+1), T(t-N+2), …, T(t)] of the most recent N times in the current temperature sequence with the corresponding laser power parameters [P(t-N+1), P(t-N+2), …, P(t)] to form a multi-dimensional input vector; the laser power parameters are parameters characterizing the laser output energy, including laser power value or duty cycle; S2.2 Input the multidimensional input vector constructed in step S2.1 into the pre-established prediction model; the prediction model performs nonlinear transformation and mapping on the multidimensional input vector through the weight parameters stored in its internal structure to capture the dynamic change pattern of the temperature sequence under the influence of laser power. S2.3 The prediction model calculates and outputs a sequence of predicted temperature values ​​[T'(t+1), T'(t+2), …, T'(t+M)] for the next M times starting from time t+1, based on the historical dynamics reflected by the multidimensional input vector; wherein, the predetermined time period corresponds to the time length from time t+1 to time t+M. S2.4 During operation, the prediction model does not rely on prior knowledge of the thermophysical parameters of the biological tissue in the target area, but characterizes the thermal response characteristics of the target area under laser irradiation through the learned relationship implicit in the weight parameters.

[0008] Furthermore, the specific process of calculating the adjusted laser power parameters required to regress and stabilize the predicted temperature value within the target temperature range in step S4 includes: S4.1 From the M predicted temperature values ​​obtained in step S2, identify the first predicted temperature value T'_dev that exceeds the upper limit of the target temperature range or is lower than its lower limit, and its corresponding future time t_dev; calculate the temperature deviation ΔT = |T'_dev - T_threshold| between T'_dev and the nearest upper limit threshold T_high or lower limit threshold T_low, where T_threshold is T_high or T_low; S4.2 Based on the recent trend of the current temperature sequence and the change of the laser power parameters during the same period, estimate the rate of temperature change or steady-state temperature change in the target area caused by a unit change in laser power under the current thermal environment, and define it as the current power-temperature influence coefficient K; the value of K is estimated by analyzing the response slope and / or steady-state offset of the current temperature sequence when a known change in laser power occurs within the most recent time window; S4.3. Based on the temperature deviation ΔT obtained in step S4.1, the current power-temperature influence coefficient K obtained in step S4.2, and the prediction time interval Δt from the current time to time t_dev, calculate the required laser power adjustment ΔP; the calculation of ΔP aims to generate an expected temperature change to offset the temperature deviation ΔT; the calculation takes into account the dynamic process of temperature change, so that after applying ΔP, the predicted temperature trajectory returns to the target temperature range at or before time t_dev; S4.4 Determination of Adjusted Parameters: The current laser power parameter P_current is combined with the calculated power adjustment amount ΔP to obtain the adjusted laser power parameter P_adjusted, wherein the combination method is P_adjusted = P_current + ΔP; at the same time, a boundary check is performed on P_adjusted to ensure that it does not exceed the minimum and maximum safe power range allowed by the laser emitting device; if it exceeds, it is clamped to the corresponding boundary value; S4.5 Smoothing: To avoid temperature oscillations caused by power step changes, a smoothing constraint is further applied to the calculated P_adjusted. The smoothing constraint requires that the rate of change between the adjusted power parameter and the previous power parameter does not exceed the preset maximum allowable rate of change. If it exceeds the maximum allowable rate of change, the adjustment amount is limited according to the maximum allowable rate of change, and the smoothed adjusted laser power parameter is recalculated.

[0009] Furthermore, the generation of the corresponding laser power control command in step S5 includes converting the adjusted laser power parameters into a control signal format recognizable by the laser emitting device; The control signal format is an analog voltage signal, a pulse width modulation signal, or a digital communication protocol message; After receiving the control command, the laser emitting device, under the action of its internal power control circuit, adjusts its actual output power from the current value to the value specified by the adjusted laser power parameter.

[0010] Furthermore, the method also includes an initialization and parameter configuration step, which is performed before the start of the loop execution of S1, including: S01. Set target temperature range: Receive user input or load the upper limit threshold T_high and lower limit threshold T_low of the target temperature range from a pre-stored treatment plan; S02. Configure control parameters: Set the length of the predetermined future time period, the sampling frequency of temperature monitoring, the maximum allowable rate of change of power adjustment, and the safe operating upper and lower limits of laser power; S03. Start-up and calibration: Control the laser emitting device to output initial safe power and start the temperature monitoring device to verify the validity and stability of the temperature monitoring data and confirm that the target area is correctly identified and locked.

[0011] Furthermore, the method also includes a security monitoring and intervention step, which is executed concurrently with steps S1 to S6, including: S7.1 Real-time safety verification: In each control cycle, in addition to adjusting based on prediction, the latest measured temperature value in the current temperature sequence is directly monitored. If the measured temperature value exceeds the preset absolute safety temperature limit, the highest priority emergency stop command is immediately generated to force the laser power to be reduced to zero or the lowest safety level and to issue an alarm. S7.2 Predictive Failure Detection: Continuously evaluate the predictive performance of the prediction model described in step S2, and determine whether the prediction has failed by comparing the recent predicted temperature value with the subsequent actual monitored temperature value; if a failure is predicted, automatically switch to the proportional-integral-derivative backup control mode based on the measured temperature deviation, and record the failure event for subsequent analysis.

[0012] Furthermore, the method also includes an adaptive optimization step during operation, which is executed periodically or under triggered conditions during stable system operation, including: S81, Data Cache: Continuously stores the input data, prediction results, control commands, and subsequent actual temperature feedback data in a time-aligned manner in a circular buffer to form a historical operation dataset; S82. Performance evaluation: Periodically analyze the historical operating dataset and calculate the evaluation index of temperature control accuracy. The evaluation index includes the standard deviation of temperature fluctuation within the target temperature range, the number of overshoots, and the percentage of time the temperature is maintained within the target range. S83. Parameter Fine-tuning: Based on the performance evaluation results of step S82, if a downward trend in control accuracy is found or the optimal level is not reached, the internal parameters used in the control process are automatically fine-tuned. The internal parameters include, but are not limited to, the time window length used in step S4 to estimate the power-temperature influence coefficient K, and the gain coefficient used to calculate the power adjustment amount ΔP. The fine-tuning adopts an optimization method based on gradient descent or heuristic search to improve the stability and accuracy of temperature control in the next stage. S84. Model update trigger judgment: Based on long-term historical operating data, analyze the changes in the thermal response characteristics of the target area. If it is determined that the characteristics have drifted, an early warning is triggered, prompting that the prediction model in step S2 needs to be updated to adapt to the new thermal environment.

[0013] Furthermore, the method supports multimodal treatment protocols, wherein the preset target temperature range is not fixed and changes dynamically according to a predefined time-temperature curve; In step S3, the predicted temperature value is compared with the target temperature curve showing dynamic changes at the current time and the corresponding future time. In step S4, when calculating the adjusted laser power parameters, the goal is to ensure that the predicted temperature value tracks the dynamically changing target temperature curve. The time-temperature curve defines the temperature level and duration required to be maintained at different stages of the treatment process, in order to realize complex treatment strategies based on different temperatures.

[0014] According to a second aspect of the present invention, the present invention claims protection for a temperature-predictive-based self-adjusting laser power control system, comprising: One or more processors; A memory storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the temperature-predictive-based self-adjusting laser power control method.

[0015] This invention discloses a self-adjusting laser power control method based on temperature prediction, belonging to the field of intelligent control technology for medical devices. This method aims to solve the problems of reliance on human experience, control lag, and insufficient precision in traditional photothermal therapy. Its core lies in constructing a forward-looking adaptive control closed loop: the system first continuously collects and constructs the temperature sequence of the target area in real time; then, using a pre-trained time-series prediction model, based on current and historical temperature and power data, it predicts the temperature change trend in the near future; when the predicted temperature is about to deviate from the preset safe and effective treatment window, the system calculates in advance and outputs a smooth laser power adjustment command, driving the actuator to actively intervene. Through this combination of prediction-feedforward and real-time feedback, this invention can dynamically compensate for the thermal inertia of biological tissue and environmental interference, maintaining the temperature of the treatment area within the preset target range with high precision and stability. This method can be widely applied to various treatment scenarios requiring precise temperature control, such as tumor photothermal ablation and wound photothermal antibacterial treatment, significantly improving the safety and controllability of treatment, and providing a key intelligent control solution for the clinical translation of photothermal therapy. Attached Figure Description

[0016] Figure 1 The following is a flowchart of a self-adjusting laser power control method based on temperature prediction, which is claimed for protection in an embodiment of the present invention. Figure 2 A structural block diagram of a temperature-predictive-based self-adjusting laser power control system is claimed for an embodiment of the present invention. Figure 3 A flowchart for initialization and parameter configuration. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0018] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of those features. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications in the embodiments of this application, such as up, down, left, right, front, back, etc., are only used to explain the relative positional relationships and movements between components in a specific orientation as shown in the accompanying drawings. If the specific orientation changes, the directional indications will change accordingly. Furthermore, the terms "including" and "having," 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 may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0019] References to embodiments herein mean that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] According to the first embodiment of the present invention, referring to Figure 1 This invention claims protection for a self-adjusting laser power control method based on temperature prediction, the method comprising the following steps: S1. Real-time acquisition of continuous temperature monitoring data of the target area irradiated by the laser, forming the current temperature sequence; S2. Based on the current temperature sequence and current laser power parameters, the predicted temperature value of the target area in the future within a predetermined time period is calculated by the prediction model; S3. Compare the predicted temperature value with the preset target temperature range to determine whether the predicted temperature value will deviate from the target temperature range. S4. If the judgment result of step S3 is yes, then based on the deviation trend and degree of the predicted temperature value relative to the target temperature range, and the current laser power parameters, calculate the adjusted laser power parameters required to make the predicted temperature value return to and stabilize within the target temperature range. S5. Based on the adjusted laser power parameters calculated in step S4, generate a corresponding laser power control command and send the control command to the laser emitting device to drive the laser emitting device to adjust its output power to the level corresponding to the adjusted laser power parameters; S6. After completing the power adjustment in step S5, return to step S1 and continue to monitor, predict and adjust the temperature in the next cycle, thereby achieving dynamic closed-loop stable control of the temperature in the target area.

[0021] In this embodiment, the method is executed by a control system and includes an initialization phase and a continuously operating closed-loop control phase; the closed-loop control phase specifically includes the following steps: The control system uses an infrared thermal imager integrated with the laser optical path to continuously scan and image the target area of ​​the living biological tissue irradiated by the laser beam at a constant high sampling frequency. For each frame of the acquired infrared thermal image, the control system first accurately segments the pixel set corresponding to the target area from the image according to a predefined or automatically identified contour boundary. Then, it calculates the arithmetic mean of the temperature values ​​of all pixels in the pixel set and uses this average value as the representative temperature measurement value of the target area at the current sampling time. The control system stores these continuously acquired representative temperature measurements in a first-in-first-out data buffer in chronological order to form and update a current temperature sequence reflecting the temperature changes over a recent period of time. At the same time, the control system synchronously reads the current laser output power setting value from the laser drive unit.

[0022] The control system invokes an internally pre-trained time-series prediction model, using the current temperature sequence obtained in step A and the laser power setpoint sequence recorded synchronously as inputs. The prediction model internally performs nonlinear feature extraction and spatiotemporal correlation analysis on the input multidimensional time-series data to simulate the complex thermodynamic process in the target area under a given laser power input, caused by the combined effects of photothermal conversion, biological tissue heat conduction, and blood perfusion. Based on this analysis, the prediction model outputs a predicted temperature sequence for the target area over a future time period, describing the estimated temperature at each predicted time point if the current laser power remains constant.

[0023] The control system compares the obtained sequence of predicted temperature values ​​with the target temperature range preset by the user or specified in the treatment plan. The comparison is forward-looking, that is, the system checks the temperature value at each predicted time point in the future to determine whether it will exceed the upper threshold of the target temperature range or fall below the lower threshold of the target temperature range. This determination is intended to identify the trend of an unfavorable deviation in temperature, rather than waiting for the temperature to actually deviate before responding.

[0024] If the judgment indicates that the predicted temperature will deviate from the target range at some future time, the control system immediately initiates the power adjustment calculation process. This calculation first assesses the severity of the predicted deviation, including calculating the deviation between the predicted temperature and the threshold, and estimating the remaining time window from the current time to the predicted time when the deviation occurs. Second, the control system dynamically estimates a power-temperature influence coefficient by combining the current laser power value and the actual impact of historical power changes on temperature. This coefficient characterizes the rate of temperature change in the target area caused by a unit change in laser power under the current specific tissue environment and irradiation conditions. Finally, based on the above deviation, time window, and response coefficient, the system uses preset control logic to calculate the direction and magnitude of laser power adjustment required to pull the predicted temperature trajectory back and stabilize it within the target range within the future time window, thereby determining a specific adjusted laser power parameter value. The calculation process includes a safety boundary check to ensure that the result is within the laser's safe operating range.

[0025] The control system generates a corresponding digital or analog control signal based on the calculated adjusted laser power parameter value. This control signal is sent to the laser's power drive unit via a data bus or dedicated control line. The power drive unit analyzes the signal and precisely adjusts the output of its internal power supply or modulator accordingly, so that the actual emission power of the laser smoothly and without abrupt transitions from the current value to the new set value. The execution speed of the power adjustment is limited to avoid impacting the laser and tissue.

[0026] After completing a power adjustment, the control system does not wait for a fixed cycle, but immediately returns to the initial step and begins the operation of the next control cycle. The system continuously collects the latest temperature data, updates the historical sequence, and performs prediction, judgment, calculation and adjustment again, thus forming a continuous feedforward-feedback composite control loop guided by future temperature prediction. This dynamically counteracts environmental disturbances and changes in organizational characteristics, maintaining the true temperature of the target area within a preset narrow window.

[0027] Furthermore, the real-time acquisition of continuous temperature monitoring data of the target area irradiated by the laser in step S1 is achieved by a non-contact infrared thermal imaging temperature measurement device. The non-contact infrared thermal imaging temperature measurement device scans the target area at a first sampling frequency not lower than a preset one to obtain an infrared thermal image containing the surface temperature distribution of the target area. The pixel temperature values ​​corresponding to the preset region of interest are extracted from each frame of the infrared thermal image, and spatial averaging is performed to obtain a single-point temperature value representing the temperature of the target region at that moment. These values ​​are then arranged in chronological order to form the current temperature sequence. The region of interest is automatically defined in the initial stage by an image recognition algorithm based on the morphological characteristics of the target region, or it can be manually specified by the operator.

[0028] In this embodiment, before the treatment begins, the control system guides the operator to perform spatial and temperature calibration on the infrared thermal imager. Spatial calibration ensures that the field of view of the thermal imager and the laser irradiation area are precisely overlapped in space. This is accomplished by simultaneously displaying the visible light camera image and the infrared thermal image, and by registering them using preset marker points. Temperature calibration is performed by placing a reference blackbody with known emissivity near the target area, and by measuring and correcting it to ensure the absolute accuracy of infrared temperature measurement.

[0029] The target area is not fixed during treatment; the control system allows the definition of two areas: a core treatment area and a safety monitoring area; the core treatment area is the main area for temperature control, and its outline can be manually drawn by the user on an infrared thermogram or automatically identified by an image processing algorithm based on the temperature rise area of ​​the tissue in the early stage of laser irradiation; the safety monitoring area is a ring-shaped or larger area surrounding the core treatment area, used to monitor heat diffusion; the system constructs temperature sequences for these two areas respectively.

[0030] The raw pixel temperature values ​​obtained from the infrared thermal imager may contain noise. Before calculating the average temperature of the area, the control system applies a spatial domain median filter to the pixel set to eliminate the influence of individual abnormal pixels. When constructing the time series, the system also performs a simple time domain consistency check. If the average temperature value calculated at a certain sampling time changes drastically compared to the value at the previous time, the data at that time is marked as suspicious, and a reasonable value is estimated from the valid data before and after using interpolation to replace it. At the same time, the abnormal event is recorded.

[0031] The first-in-first-out (FIFO) data buffer is a structured circular storage queue that stores not only each sampling timestamp and its corresponding average temperature value, but also the laser's operating status at that moment, including the power setpoint, the actual power output, and the temperature of the laser's internal heat dissipation module. This multi-dimensional data together constitutes the complete context information required by the prediction model. The length of the buffer is configurable and sufficient to cover a period of time from the current moment back to a sufficiently past period to meet the prediction model's requirement for the length of historical data.

[0032] Furthermore, the prediction model in step S2 is a dynamic prediction model based on time series analysis. Its input is the current temperature sequence containing the temperature values ​​of the most recent N historical moments and the laser power parameter sequence of the corresponding time period. The output is the predicted temperature values ​​for the next M consecutive time points, where N and M are both positive integers. Reference Figure 2 The prediction model is built and run through the following steps: S2.1. Combine the temperature values ​​[T(t-N+1), T(t-N+2), …, T(t)] of the most recent N times in the current temperature sequence with the corresponding laser power parameters [P(t-N+1), P(t-N+2), …, P(t)] to form a multi-dimensional input vector; the laser power parameters are parameters characterizing the laser output energy, including laser power value or duty cycle; S2.2 Input the multidimensional input vector constructed in step S2.1 into the pre-established prediction model; the prediction model performs nonlinear transformation and mapping on the multidimensional input vector through the weight parameters stored in its internal structure to capture the dynamic change pattern of the temperature sequence under the influence of laser power. S2.3 The prediction model calculates and outputs a sequence of predicted temperature values ​​[T'(t+1), T'(t+2), …, T'(t+M)] for the next M times starting from time t+1, based on the historical dynamics reflected by the multidimensional input vector; wherein, the predetermined time period corresponds to the time length from time t+1 to time t+M. S2.4 During operation, the prediction model does not rely on prior knowledge of the thermophysical parameters of the biological tissue in the target area, but characterizes the thermal response characteristics of the target area under laser irradiation through the learned relationship implicit in the weight parameters.

[0033] In this embodiment, before calling the prediction model, the control system performs formatted preprocessing on the acquired raw data. First, it extracts data from the structured data buffer that traces back a fixed time period from the current moment, the time period being equal to the historical observation window required by the model. The extracted data includes: the temperature time series of the core treatment area, the time series of the synchronized laser power setpoint, and the average temperature series of the optional safety monitoring area. Second, it normalizes these time series data, scaling the temperature values ​​to a relative scale based on the target temperature range and the power values ​​to a percentage scale of the laser's maximum rated power to eliminate the influence of dimensions and improve the numerical stability of the model. Finally, it strictly aligns the processed time series according to the time step and concatenates them into a multidimensional input tensor, ready to be fed into the prediction model.

[0034] The prediction model is a deep neural network model whose architecture includes multiple sequentially connected functional layers. The input multidimensional time series tensor first passes through several one-dimensional convolutional layers. The function of these convolutional layers is to automatically extract local temporal pattern features from the raw data, such as identifying short-term trends in temperature rise or fall, and the rhythm of power changes. The feature maps output by the convolutional layers are then fed into a long short-term memory network layer or a gated recurrent unit layer. This recurrent neural network layer is specifically designed to process sequential data. Its internal gating mechanism can selectively remember long-term important information, such as the basal warming rate of tissues, and forget short-term irrelevant fluctuations, such as measurement noise, thereby modeling the long-range temporal dependence of temperature changes. Finally, the output of the recurrent layer is integrated and mapped through one or more fully connected layers, converting the learned high-dimensional spatiotemporal features into specific numerical predictions of future temperatures.

[0035] The prediction model performs a forward propagation calculation, starting from the preprocessed input tensor and passing through nonlinear transformations at each layer, finally outputting a vector. The length of this vector is equal to the preset future prediction step size, and each element in the vector represents the predicted temperature value of the core treatment area at a specific time interval in the future (e.g., every 1 second). The control system decodes this output vector back into actual temperature units. At the same time, the model may also output an uncertainty estimate or confidence interval associated with each predicted value. This information is used to assist decision-making, and a more conservative power regulation strategy is adopted when the prediction uncertainty is high.

[0036] Although the prediction model is a general model pre-trained on a large amount of experimental data, the control system performs a brief model warm-up and micro-adaptation process in the initial stage after each treatment is initiated. During this process, the system irradiates the target area with a low fixed power for a short period of time to collect the initial thermal response data of the specific object and tissue. The system inputs this initial data into the prediction model without updating the model weights. Instead, it uses the model's internal state, such as the hidden state of the recursive network, to initialize a contextual memory that better matches the current treatment scenario. This allows the model to adapt more quickly to the individual-specific thermodynamic characteristics and improves the accuracy of subsequent predictions.

[0037] Furthermore, the specific process of calculating the adjusted laser power parameters required to regress and stabilize the predicted temperature value within the target temperature range in step S4 includes: S4.1 From the M predicted temperature values ​​obtained in step S2, identify the first predicted temperature value T'_dev that exceeds the upper limit of the target temperature range or is lower than its lower limit, and its corresponding future time t_dev; calculate the temperature deviation ΔT = |T'_dev - T_threshold| between T'_dev and the nearest upper limit threshold T_high or lower limit threshold T_low, where T_threshold is T_high or T_low; S4.2 Based on the recent trend of the current temperature sequence and the change of the laser power parameters during the same period, estimate the rate of temperature change or steady-state temperature change in the target area caused by a unit change in laser power under the current thermal environment, and define it as the current power-temperature influence coefficient K; the value of K is estimated by analyzing the response slope and / or steady-state offset of the current temperature sequence when a known change in laser power occurs within the most recent time window; S4.3. Based on the temperature deviation ΔT obtained in step S4.1, the current power-temperature influence coefficient K obtained in step S4.2, and the prediction time interval Δt from the current time to time t_dev, calculate the required laser power adjustment ΔP; the calculation of ΔP aims to generate an expected temperature change to offset the temperature deviation ΔT; the calculation takes into account the dynamic process of temperature change, so that after applying ΔP, the predicted temperature trajectory returns to the target temperature range at or before time t_dev; S4.4 Determination of Adjusted Parameters: The current laser power parameter P_current is combined with the calculated power adjustment amount ΔP to obtain the adjusted laser power parameter P_adjusted, wherein the combination method is P_adjusted = P_current + ΔP; at the same time, a boundary check is performed on P_adjusted to ensure that it does not exceed the minimum and maximum safe power range allowed by the laser emitting device; if it exceeds, it is clamped to the corresponding boundary value; S4.5 Smoothing: To avoid temperature oscillations that may be caused by power step changes, a smoothing constraint is further applied to the calculated P_adjusted. The smoothing constraint requires that the rate of change between the adjusted power parameter and the previous power parameter does not exceed the preset maximum allowable rate of change. If it exceeds the maximum allowable rate of change, the adjustment amount is limited according to the maximum allowable rate of change, and the smoothed adjusted laser power parameter is recalculated.

[0038] In this embodiment, when a deviation is predicted in the temperature prediction sequence, the control system first performs a comprehensive analysis of the deviation. This analysis includes: identifying the first predicted time point T_high_exceed that will exceed the upper threshold and its excess value ΔT_high, and the first predicted time point T_low_exceed that will fall below the lower threshold and its deficiency value ΔT_low; calculating the time differences ΔT_high and ΔT_low from the current moment to these two time points; assessing the urgency of the deviation, generally indicating a higher urgency as the time difference is smaller; furthermore, the system analyzes the curve shape of the entire predicted temperature sequence to determine whether it is a monotonic deviation or an oscillating deviation, which is crucial for selecting the adjustment strategy.

[0039] The control system maintains a dynamically updated table of response coefficients, which reflects the relationship between the rate of power change and the rate of temperature change under the current tissue state and laser parameters. The identification process utilizes historical data within a recent short time window: the system finds the moment of the most recent measurable active change in laser power from the buffer, analyzes the curve of temperature change in the target area over time from that moment, calculates its initial slope as the transient response coefficient, and observes the final steady-state temperature offset as the steady-state response coefficient. Combining the nonlinear response of the current laser power level and the ambient temperature, a comprehensive response coefficient K applicable to the current moment is obtained by interpolating a pre-stored response surface or running a simplified online parameter estimation algorithm; this coefficient K is the core parameter for subsequent calculations.

[0040] The control system aims to offset predicted deviations by calculating the required power adjustment ΔP. The calculation follows the basic physical relationship: Required temperature change = Response coefficient × Power adjustment × Effective duration, but does not directly use a simple formula. Specifically: For predicted overheating, the system calculates the need to apply a negative temperature change trend to offset the overheating amount ΔT_high within the future timeframe. Considering thermal inertia, the system requires this negative adjustment to take effect early and continue in effect; therefore, the calculated ΔP is a negative power reduction, its magnitude being proportional to ΔT_high, inversely proportional to ΔT_high, and related to the reciprocal of the response coefficient K. For predicted low temperatures, the calculation logic is similar but in the opposite direction. If both overheating and low temperature risks, such as oscillations, are predicted simultaneously, the system calculates a compromise power adjustment to smooth temperature fluctuations, ensuring that both peak and trough values ​​return to the target range.

[0041] The initially calculated ΔP needs to undergo multiple constraint processing. First, amplitude constraint: check whether P_current + ΔP is within the laser's allowed hard safety range of [P_min, P_max]. If it exceeds this range, it is truncated to the boundary value. Second, rate of change constraint: check whether the change of the calculated ΔP relative to the previously executed power adjustment exceeds the preset maximum allowable rate of change per unit time to prevent hardware damage or temperature oscillation caused by drastic power changes. If it exceeds this range, ΔP is limited according to the maximum allowable rate of change. Finally, smoothness constraint: in some applications that pursue extreme smoothness, the system will perform low-pass filtering on ΔP to filter out high-frequency power adjustment jitter that may be caused by measurement noise, ensuring that the power command curve output to the laser is smooth.

[0042] After the above calculations and constraints, a safe, feasible, and smooth adjusted laser power parameter value P_adjusted is finally determined. The system records this value, along with the decision basis for this adjustment, such as the triggered deviation type, the response coefficient used, and the applied constraints, in the log. At the same time, the system outputs this P_adjusted value as part of the historical data in the next control cycle for future prediction and coefficient identification.

[0043] Furthermore, the generation of the corresponding laser power control command in step S5 includes converting the adjusted laser power parameters into a control signal format recognizable by the laser emitting device; The control signal format is an analog voltage signal, a pulse width modulation signal, or a digital communication protocol message; After receiving the control command, the laser emitting device, under the action of its internal power control circuit, gradually adjusts its actual output power from the current value to the value specified by the adjusted laser power parameter.

[0044] In this embodiment, the control command generated by the control system includes two parts: a target power value P_adjusted and transition path parameters to reach the target value. The transition path parameters specify the curve shape followed by the power change from the current value P_current to P_adjusted, such as a linear ramp rise / fall, an S-curve smooth transition, etc. These parameters are encoded into specific protocol messages that the laser driver unit can parse. The protocol messages typically include a start symbol, command code, target power data field, transition time or slope field, checksum, and end symbol to ensure reliable transmission.

[0045] The encoded command is sent to the laser driver unit via a physical link such as USB, Ethernet, or analog voltage line. After transmission, the control system starts a timer to wait for the confirmation signal from the driver unit. The confirmation signal should include the received target power value for comparison. If no confirmation is received within the timeout period, or if the confirmation value does not match the sent value, the control system will determine that the command transmission has failed, immediately resend the command, maintain the laser power at the last confirmed safe value, and report the communication failure to the upper-level system.

[0046] Upon receiving a valid command, the laser driver unit's internal microcontroller generates a corresponding pulse width modulation signal or analog voltage signal based on the target value and transition path to control the pump source of the laser diode or solid-state laser. The driver unit typically has a local closed-loop control loop to quickly stabilize the output power and resist power fluctuations and device temperature drift. Simultaneously, the driver unit monitors the actual output optical power of the laser, diode temperature, coolant temperature, and other statuses in real time through its internal sensors, and periodically reports this status information to the control system. The control system uses this feedback information to verify whether the power regulation is executed correctly and for system health monitoring.

[0047] During the power regulation process, if the control system receives a new emergency temperature over-limit signal, such as the measured temperature momentarily exceeding the absolute safety threshold, or receives a hardware alarm from the drive unit, such as overheating or overcurrent, the system will immediately generate and send a highest priority emergency stop or rapid power reduction command, regardless of whether the current regulation command has been completed. This command will override any ongoing gradual regulation commands, requiring the drive unit to reduce the power to zero or a preset safe standby power as quickly as possible to ensure the safety of the patient.

[0048] Furthermore, referring to Figure 3 The method further includes an initialization and parameter configuration step, which is performed before the start of the loop execution of S1, including: S01. Set target temperature range: Receive user input or load the upper limit threshold T_high and lower limit threshold T_low of the target temperature range from a pre-stored treatment plan; S02. Configure control parameters: Set the length of the predetermined future time period, the sampling frequency of temperature monitoring, the maximum allowable rate of change of power adjustment, and the safe operating upper and lower limits of laser power; S03. Start-up and calibration: Control the laser emitting device to output initial safe power and start the temperature monitoring device to verify the validity and stability of the temperature monitoring data and confirm that the target area is correctly identified and locked.

[0049] In this embodiment, after the control system is powered on, it first performs a communication self-test on all connected hardware modules to confirm that the infrared thermal imager, laser drive unit, safety interlock device, etc. are all online and responding normally. Then, it initializes each hardware module to a preset standby state, such as turning off the laser output, starting the thermal imager to cool to a stable operating temperature, and resetting all control parameters to default values.

[0050] The system guides users through the graphical user interface to complete the treatment parameter settings. Users need to input or select: treatment mode such as constant temperature maintenance, step temperature increase, core target temperature and allowable fluctuation range (i.e., target temperature window), total treatment time, and laser wavelength to select the corresponding tissue absorption coefficient preset file. In addition, users also need to set control parameters such as prediction step size, control cycle, maximum power change rate, and emergency safety temperature threshold.

[0051] After the laser begins irradiation with extremely low safety guide power, the control system initiates an automatic calibration and learning period lasting several tens of seconds. During this period, the system performs the following operations: 1. Guides the user to accurately mark the boundaries of the treatment area on the infrared image using software tools; 2. Controls the laser power to make several small, predefined changes, such as a slight increase and then a return to normal, while simultaneously recording temperature response data at high frequency; 3. Utilizes this initial response data to quickly estimate the approximate range of basic thermal parameters for the specific treatment site, such as effective thermal diffusivity and perfusion rate, and accordingly initializes or fine-tunes the internal state parameters related to tissue characteristics in the prediction model, as well as the offset of the response coefficient lookup table in use, enabling the system to adapt to the current treatment subject more quickly.

[0052] After the learning period is completed, the system summarizes all initialization results, including hardware status, parameter settings, calibration quality, and initial learning data fit, and presents them to the user in a clear and visual way. After the user confirms that all statuses are normal, they issue a treatment start command. The control system then officially switches from the initialization stage to the closed-loop control stage, and the laser power will begin to make prediction-based adaptive adjustments according to the preset target temperature.

[0053] Furthermore, the method also includes a security monitoring and intervention step, which is executed concurrently with steps S1 to S6, including: S7.1 Real-time safety verification: In each control cycle, in addition to adjusting based on prediction, the latest measured temperature value in the current temperature sequence is directly monitored. If the measured temperature value exceeds the preset absolute safety temperature limit, the highest priority emergency stop command is immediately generated to force the laser power to be reduced to zero or the lowest safety level and to issue an alarm. S7.2 Predictive Failure Detection: Continuously evaluate the predictive performance of the prediction model described in step S2, and determine whether the prediction has failed by comparing the recent predicted temperature value with the subsequent actual monitored temperature value; if a failure is predicted, automatically switch to the proportional-integral-derivative backup control mode based on the measured temperature deviation, and record the failure event for subsequent analysis.

[0054] In this embodiment, in addition to the averaged and filtered core temperature sequence used for control decisions, the system also processes the raw data stream from the infrared thermal imager in parallel; a separate safety sentinel thread is set up, which directly reads the hotspot temperature of the highest pixel in each frame of thermal image and the average temperature of the safety monitoring area; once the hotspot temperature is detected to exceed a preset hard safety shutdown threshold that is higher than the upper limit of the control target, or the temperature of the safety monitoring area exceeds its allowable diffusion threshold, the thread will immediately interrupt the main control loop and forcibly execute the emergency shutdown procedure, which has a higher priority than any prediction-based adjustment instructions.

[0055] The system continuously compares the predicted temperature value with the subsequently measured temperature value; it calculates the prediction error statistics for short periods, such as the past 30 seconds, including the mean absolute error and the standard deviation of the error; when the mean absolute error continues to exceed the preset error tolerance, or the standard deviation of the error increases significantly, it indicates that the performance of the prediction model may have degraded in the current environment; at this time, the system will trigger a warning of reduced prediction reliability, and can automatically or after user confirmation, smoothly switch the core logic of the controller from the prediction-based feedforward-feedback mode to a robust, enhanced proportional-integral-derivative feedback mode based purely on the deviation between the current measured temperature and the target value, so as to ensure that the basic control function is not interrupted.

[0056] The control system monitors the communication latency and data packet loss rate with the laser drive unit in real time, and monitors the frame rate stability of the thermal imager; it also monitors the health indicators reported by the drive unit, such as the laser diode temperature and cooling efficiency; if any indicator exceeds the normal range, the system will record the fault event and degrade the operation according to the preset strategy, such as reducing the maximum allowable power limit, extending the control cycle, or suspending treatment, to prevent control failure due to hardware failure.

[0057] The system records key data in detail for each control cycle, including timestamps, measured temperature, predicted temperature, current power, calculated adjusted power, response coefficient, and any triggered warnings or abnormal events. These logs are stored in an immutable manner for post-treatment effect analysis, accident tracing, and as an offline data source for optimizing prediction models and control parameters.

[0058] Furthermore, the method also includes an adaptive optimization step during operation, which is executed periodically or under specific triggering conditions during stable system operation, including: S81, Data Cache: Continuously stores the input data, prediction results, control commands, and subsequent actual temperature feedback data in a time-aligned manner in a circular buffer to form a historical operation dataset; S82. Performance evaluation: Periodically analyze the historical operating dataset and calculate the evaluation index of temperature control accuracy. The evaluation index includes the standard deviation of temperature fluctuation within the target temperature range, the number of overshoots, and the percentage of time the temperature is maintained within the target range. S83. Parameter Fine-tuning: Based on the performance evaluation results of step S82, if a downward trend in control accuracy is found or the optimal level is not reached, the internal parameters used in the control process are automatically fine-tuned. The internal parameters include, but are not limited to, the time window length used in step S4 to estimate the power-temperature influence coefficient K, and the gain coefficient used to calculate the power adjustment amount ΔP. The fine-tuning adopts an optimization method based on gradient descent or heuristic search to improve the stability and accuracy of temperature control in the next stage. S84. Model update trigger judgment: Based on long-term historical operating data, analyze the changes in the thermal response characteristics of the target area. If it is determined that the characteristics have drifted, an early warning is triggered, indicating that the prediction model in step S2 may need to be updated to adapt to the new thermal environment.

[0059] In this embodiment, the system calculates and updates a set of performance indicators reflecting the current control quality in real time during operation. The main indicators include: temperature stability index, standard deviation of temperature values ​​within the target temperature window, cumulative percentage of time within the target window where the temperature is within a preset range, overshoot / undershoot count, number of times the temperature exceeds the window, and power regulation activity, root mean square value of power change per unit time. These indicators are calculated in a rolling time window manner, and global statistical values ​​are provided since the start of treatment.

[0060] Based on real-time performance metrics, the system has a built-in optimizer. For example, if the temperature stability index is found to have deteriorated recently, but the power regulation activity is high, it indicates that the system may be over-regulating and causing oscillations. The optimizer will attempt to automatically fine-tune the core parameters, such as slightly increasing the time constant when calculating the power adjustment to make the regulation smoother, or reducing the weight of the short-term data window used to calculate the response coefficient K to reduce sensitivity to noise. Conversely, if the stability is good but the response is slow, the system will fine-tune in the opposite direction. All fine-tuning is performed within the pre-set parameter safety boundaries and is implemented in a small step and gradual manner to avoid causing control abrupt changes.

[0061] For long-term treatment, the system can identify different treatment phases. For example, in the initial treatment phase, during the tissue warming phase, a more aggressive power increase strategy is allowed; in the steady-state maintenance phase, a regulation strategy that pursues ultimate stability is adopted; and in the cooling phase before the end of treatment, a gradual power decrease strategy is adopted. By analyzing the temperature trajectory, the duration of treatment, and the preset treatment protocol, the system can automatically or semi-automatically switch between different sets of control strategy parameters. Each set of parameters contains a set of control parameters optimized for that phase.

[0062] Model performance degradation warning and data collection labeling: Long-term operation modules analyze the long-term trend of prediction errors. If it is found that the prediction model's error is systematically high for a certain type of tissue site, such as the liver or breast as labeled by the user, or for a certain type of treatment parameter, such as a specific power density range, the system will anonymize all context data of that treatment and label it as a valuable sample for improvement, and prompt whether to upload it to the cloud or local analysis library after obtaining permission. This provides a data accumulation mechanism for future iterative training and updating of more powerful general prediction models, or for building dedicated prediction models for specific scenarios. This process does not involve real-time model training during this treatment.

[0063] Furthermore, the method supports multimodal treatment protocols, wherein the preset target temperature range is not fixed and changes dynamically according to a predefined time-temperature curve; In step S3, the predicted temperature value is compared with the target temperature curve showing dynamic changes at the current time and the corresponding future time. In step S4, when calculating the adjusted laser power parameters, the goal is to ensure that the predicted temperature value tracks the dynamically changing target temperature curve. The time-temperature curve defines the temperature level and duration required to be maintained at different stages of the treatment process, in order to achieve complex treatment strategies based on different temperatures, including low-temperature hyperthermia, high-temperature ablation, or segmented temperature treatment.

[0064] The method is encapsulated as a control software module with a standardized interface. The control software module receives temperature data streams from temperature monitoring devices through a first data interface, sends power control commands to laser emitting devices through a second control interface, and receives control parameters and treatment protocols from an upper-level control system or user interface through a third configuration interface. The control software module can be deployed and run on photothermal therapy devices on different hardware platforms, providing the photothermal therapy devices with universal, temperature-predictive-based adaptive laser power control capabilities.

[0065] In this embodiment, the user can not only set a constant target temperature range, but also define a target temperature curve that changes over time as a treatment protocol. This curve is displayed on the graphical interface with the time axis as the horizontal axis and the temperature axis as the vertical axis. Users can drag key points of time and temperature to draw a line or select preset function waveforms such as ramp-up, plateau, or step-down. The system discretizes this curve into a high-resolution time-temperature lookup table and stores it in the control system.

[0066] During control operation, the comparison object is no longer a fixed target range, but a dynamic target temperature value read from the lookup table that corresponds to the current time and the future predicted time. The power adjustment calculation not only compensates for the deviation from the current target temperature caused by disturbances, but also considers in advance the feedforward power input required to achieve the changing trend of the target temperature curve in the future short period of time, such as requiring a temperature increase of 2°C in the next 10 seconds. The system combines the feedback correction amount with the feedforward pre-compensation amount to calculate a comprehensive power adjustment command, so that the controlled temperature can smoothly and accurately track the preset dynamic temperature trajectory, realizing complex treatment procedures such as slowly heating to the ablation temperature and maintaining it, and then rapidly cooling to the safe temperature.

[0067] The system maintains a treatment curve execution state machine, recording the current stage of the curve (e.g., heating phase, plateau phase, cooling phase) and the percentage of progress completed. This state information can be used to synchronously control other auxiliary equipment. For example, when entering the high-temperature plateau phase, it can automatically activate an additional local cooling system to protect the epidermis; or when the treatment curve is completed, it can automatically shut down the laser and save all data. The system also allows for dynamic adjustments to the remaining treatment curve during treatment, based on real-time efficacy feedback and within safety constraints, achieving a degree of personalized responsive treatment.

[0068] According to a second embodiment of the present invention, the present invention claims protection for a temperature-predictive-based self-adjusting laser power control system, comprising: One or more processors; A memory storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the temperature-predictive-based self-adjusting laser power control method.

[0069] The following specific embodiments illustrate how the method of the present invention can achieve precise and stable photothermal ablation therapy using a live mouse model carrying a subcutaneous tumor.

[0070] Before the experiment began, the operator started the treatment system, which integrated the intelligent photothermal control engine software module. The system first performed a hardware self-test, confirming that the near-infrared laser wavelength, high-frame-rate infrared thermal imager, water cooling system, and safety interlock devices were all communicating normally and had been reset to standby mode. Subsequently, the operator configured parameters through the graphical user interface: selecting the tumor high-temperature ablation treatment mode, setting the target temperature range as the high-temperature treatment window, which was based on a central temperature with a small fluctuation above and below; setting the total treatment duration; and configuring the safe maximum and minimum power limits for the laser. The system automatically loaded the default prediction model file and control parameter set that matched the laser wavelength used and common tumor tissues. After configuration, the system entered a ready state.

[0071] The mouse was secured to the treatment table, exposing the tumor. The laser was output at a very low, safe guiding power, with the spot covering the entire tumor area. The infrared thermal imager began imaging, and the real-time image was displayed on the screen. Using software tools, the operator manually and precisely outlined the tumor area on the infrared thermal image; this area was defined by the system as the core treatment zone. The system automatically generated a ring-shaped area surrounding the core zone, defined as the safety monitoring zone. Subsequently, the system initiated a brief automatic learning period: controlling the laser power to make several small, preset-amplitude step changes based on the guiding power, while simultaneously recording the temperature response curve of the core treatment zone at a very high frequency. The system analyzed this initial response data, quickly estimating the preliminary thermal response characteristics of this specific tumor under the current illumination conditions, and used this information to initialize the internal state parameters of the prediction model, as well as the baseline value used in subsequent control to estimate the power-temperature influence coefficient. After the learning period, the system reported the calibration and learning results, which the operator confirmed were correct.

[0072] The operator issues the command to begin treatment. The system immediately enters the closed-loop control phase.

[0073] The infrared thermal imager continuously acquires thermal images at a constant high frame rate. For each frame of the thermal image, the system automatically identifies the pixel sets of the core treatment area and the safety monitoring area, and calculates their spatial average temperature values ​​separately. Before calculation, spatial median filtering is applied to the pixel values ​​to remove noise. The average temperature value, along with information such as the timestamp and the current laser power setting, is stored in real time in a circular data buffer, forming a continuously updated temperature and power sequence.

[0074] During each control cycle, the system extracts the core area temperature sequence and synchronized laser power sequence from the buffer within the most recent historical time window. After standardizing these data, they are input into the prediction model. This model uses a combination of deep neural network structures, such as one-dimensional convolutional layers and recurrent neural network layers, to perform spatiotemporal feature analysis on the historical data, simulating the thermodynamic processes of tumor tissue under the combined effects of photothermal conversion, heat conduction, and blood perfusion. The model outputs a predicted sequence of core treatment area temperatures for a future period, such as the next few tens of seconds. Simultaneously, the system utilizes information acquired during the learning period to make the model's initial hidden state more closely match the current tumor, improving the accuracy of the prediction in the initial stage.

[0075] The system compares the predicted future temperature sequence with a preset target range for high-temperature treatment. This comparison is forward-looking. For example, the system determines that if the current power remains constant, the predicted temperature will exceed the upper threshold of the treatment window at a specific future time.

[0076] Once a predicted overtemperature trend is detected, the system immediately initiates calculations. First, it quantifies the degree of deviation, the overtemperature amplitude, and the expected occurrence time. Next, the system dynamically estimates the power-temperature impact coefficient under the current environment: it analyzes the most recent observable power change in the buffer zone and its resulting actual temperature change rate, combining this with the current power level to obtain a comprehensive coefficient K through internal logic. Then, based on the overtemperature amplitude, time window, and coefficient K, it uses the feedforward compensation principle to calculate the laser power reduction ΔP required to allow the predicted temperature trajectory to smoothly fall back to the target range. During the calculation process, the system strictly ensures that the reduced power value remains within the laser's safe operating range and imposes smoothness constraints on the power change rate to prevent abrupt changes.

[0077] The system encodes the calculated adjusted power value and a smooth transition time parameter into a digital command message recognizable by the laser driver unit, and sends it via the communication link. Upon receiving the command, the laser driver unit's internal closed-loop control circuit drives the laser diode pump source, causing the actual output power to decrease from the current value along a smooth curve, such as an S-shaped curve, to the newly set value. The driver unit then feeds back the actual output power and diode temperature status to the control system.

[0078] After the power adjustment command is issued, the system immediately returns to step A and begins the next control cycle. This cycle repeats continuously, forming a predictive-regulatory closed loop.

[0079] The following modules run in parallel throughout the treatment process: An independent thread monitors the highest pixel temperature hotspot and the average temperature of the safety monitoring area in real time from the raw thermal image data. When the hotspot temperature spikes abnormally and exceeds an absolute safety threshold far above the treatment window, the thread immediately overrides its control, forcing the laser to shut down urgently and issuing an audible and visual alarm.

[0080] The system's backend calculates control performance indicators in real time, such as the standard deviation of temperature fluctuations within the treatment window and the power adjustment frequency. During the ten-minute isothermal treatment, the system detected a slight increase in temperature fluctuations, but with frequent power adjustments. The built-in optimizer automatically fine-tunes control parameters accordingly, such as slightly increasing the filter time constant in the power adjustment calculation to make the control actions smoother. After adjustment, the temperature fluctuations converged again, the power adjustment frequency decreased, and the system became more stable.

[0081] The system continuously compares short-term predicted values ​​with actual measured values. Throughout the treatment process, the prediction error remained at a low and stable level, indicating that the prediction model was well-fitted and did not trigger a switch in control mode.

[0082] This embodiment employs a simple isothermal maintenance protocol. However, the system fully supports more complex dynamic protocols. For example, if the operator presets a stepped temperature rise curve: first rapidly raising the temperature to a lower level and maintaining it for a period to promote blood flow, then gradually raising it to the ablation temperature and maintaining it long-term. In step C, the system compares the predicted temperature sequence with the dynamically changing target temperature curve; in step D, when calculating the power adjustment, it comprehensively considers the feedback adjustment required to maintain the current stage temperature, as well as the feedforward power compensation required to reach the next stage target, thereby achieving accurate tracking of the entire dynamic temperature curve.

[0083] The core of the system used in this experiment is the aforementioned intelligent photothermal control engine. This engine drives infrared thermal imagers and lasers from different manufacturers through a hardware abstraction layer. The upper-level experimental control software initiates treatment and acquires data by calling APIs provided by the engine, such as startTherapy() and getRealTimeMetrics(). All control parameters and prediction models are loaded through external configuration files, demonstrating a modular and configurable architecture.

[0084] In mouse tumor treatment experiments using the method of this invention, the system successfully maintained the tumor area temperature stably within a preset high-temperature treatment window throughout the treatment phase. Temperature curves recorded by infrared thermal imaging showed that temperature fluctuations in the core treatment area were strictly limited to a very narrow range throughout the entire treatment period, with no overshoot or undershoot due to response delay. The laser power curve exhibited smooth, adaptive adjustment characteristics; when tissue thermal response changes, such as blood flow fluctuations, the power could be adjusted in advance to counteract interference. Compared to the traditional manual control group, the temperature stability under the control of this invention was significantly improved, and no burns to surrounding healthy tissues due to overheating or insufficient treatment due to underheating occurred. Post-treatment pathological analysis showed that the tumor area underwent uniform and thorough coagulative necrosis with clear boundaries.

[0085] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0086] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0087] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A self-adjusting laser power control method based on temperature prediction, characterized in that, The method includes the following steps: S1. Real-time acquisition of continuous temperature monitoring data of the target area irradiated by the laser, forming the current temperature sequence; S2. Based on the current temperature sequence and current laser power parameters, the predicted temperature value of the target area in the future within a predetermined time period is calculated by the prediction model; S3. Compare the predicted temperature value with the preset target temperature range to determine whether the predicted temperature value will deviate from the target temperature range. S4. If the judgment result of step S3 is yes, then based on the deviation trend and degree of the predicted temperature value relative to the target temperature range, and the current laser power parameters, calculate the adjusted laser power parameters required to make the predicted temperature value return to and stabilize within the target temperature range. S5. Based on the adjusted laser power parameters calculated in step S4, generate a corresponding laser power control command and send the control command to the laser emitting device to drive the laser emitting device to adjust its output power to the level corresponding to the adjusted laser power parameters; S6. After completing the power adjustment in step S5, return to step S1 and continue to monitor, predict and adjust the temperature in the next cycle, thereby achieving dynamic closed-loop stable control of the temperature in the target area. The specific process of calculating the adjusted laser power parameters in step S4 includes: S4.1 From the M predicted temperature values ​​obtained in step S2, identify the first predicted temperature value T'_dev that exceeds the upper limit of the target temperature range or is lower than its lower limit, and its corresponding future time t_dev; calculate the temperature deviation ΔT = |T'_dev - T_threshold| between T'_dev and the nearest upper limit threshold T_high or lower limit threshold T_low, where T_threshold is T_high or T_low; S4.2 Based on the recent trend of the current temperature sequence and the change of the laser power parameters during the same period, estimate the rate of temperature change or steady-state temperature change in the target area caused by a unit change in laser power under the current thermal environment, and define it as the current power-temperature influence coefficient K; the value of K is estimated by analyzing the response slope and / or steady-state offset of the current temperature sequence when a known change in laser power occurs within the most recent time window; S4.

3. Based on the temperature deviation ΔT obtained in step S4.1, the current power-temperature influence coefficient K obtained in step S4.2, and the prediction time interval Δt from the current time to time t_dev, calculate the required laser power adjustment ΔP; the calculation of ΔP aims to generate an expected temperature change to offset the ΔT deviation; the calculation takes into account the dynamic process of temperature change, so that after applying ΔP, the predicted temperature trajectory returns to the target temperature range at or before time t_dev; S4.4 Determination of Adjusted Parameters: The current laser power parameter P_current is combined with the calculated power adjustment amount ΔP to obtain the adjusted laser power parameter P_adjusted, wherein the combination method is P_adjusted = P_current + ΔP; at the same time, a boundary check is performed on P_adjusted to ensure that it does not exceed the minimum and maximum safe power range allowed by the laser emitting device; if it exceeds, it is clamped to the corresponding boundary value; S4.5 Smoothing: To avoid temperature oscillations caused by power step changes, a smoothing constraint is further applied to the calculated P_adjusted. The smoothing constraint requires that the rate of change between the adjusted power parameter and the previous power parameter does not exceed the preset maximum allowable rate of change. If it exceeds the maximum allowable rate of change, the adjustment amount is limited according to the maximum allowable rate of change, and the smoothed adjusted laser power parameter is recalculated.

2. The method according to claim 1, characterized in that, In step S1, the continuous temperature monitoring data of the target area irradiated by the laser is acquired in real time through a non-contact infrared thermal imaging temperature measurement device. The non-contact infrared thermal imaging temperature measurement device scans the target area at a first sampling frequency not lower than the preset frequency, and obtains an infrared thermal image containing the surface temperature distribution of the target area at the current moment. The pixel temperature values ​​corresponding to the preset region of interest are extracted from each frame of the infrared thermal image and spatially averaged to obtain a single-point temperature value representing the temperature of the target region at the current moment. These values ​​are then arranged in chronological order to form the current temperature sequence. The region of interest is automatically defined in the initial stage by an image recognition algorithm based on the morphological characteristics of the target region, or it can be manually specified by the operator.

3. The method according to claim 1, characterized in that, The prediction model described in step S2 is a dynamic prediction model based on time series analysis. Its input is the current temperature sequence containing the temperature values ​​of the most recent N historical moments and the laser power parameter sequence of the corresponding time period. The output is the predicted temperature values ​​for the next M consecutive time points, where N and M are both positive integers. The prediction model is built and run through the following steps: S2.

1. Combine the temperature values ​​[T(t-N+1), T(t-N+2), …, T(t)] of the most recent N times in the current temperature sequence with the corresponding laser power parameters [P(t-N+1), P(t-N+2), …, P(t)] to form a multi-dimensional input vector; the laser power parameters are parameters characterizing the laser output energy, including laser power value or duty cycle; S2.2 Input the multidimensional input vector constructed in step S2.1 into the pre-established prediction model; the prediction model performs nonlinear transformation and mapping on the multidimensional input vector through the weight parameters stored in its internal structure to capture the dynamic change pattern of the temperature sequence under the influence of laser power. S2.3 The prediction model calculates and outputs a sequence of predicted temperature values ​​[T'(t+1), T'(t+2), …, T'(t+M)] for the next M times starting from time t+1, based on the historical dynamics reflected by the multidimensional input vector; wherein, the predetermined time period corresponds to the time length from time t+1 to time t+M. S2.4 During operation, the prediction model does not rely on prior knowledge of the thermophysical parameters of the biological tissue in the target area, but characterizes the thermal response characteristics of the target area under laser irradiation through the learned relationship implicit in the weight parameters.

4. The method according to claim 1, characterized in that, The generation of laser power control instructions in step S5 includes converting the adjusted laser power parameters into a control signal format that can be recognized by the laser emitting device. The control signal format is an analog voltage signal, a pulse width modulation signal, or a digital communication protocol message; After receiving the control command, the laser emitting device, under the action of its internal power control circuit, gradually adjusts its actual output power from the current value to the value specified by the adjusted laser power parameter.

5. The method according to claim 1, characterized in that, The method further includes an initialization and parameter configuration step, which is performed before the start of the loop execution of S1, including: S01. Set target temperature range: Receive user input or load the upper limit threshold T_high and lower limit threshold T_low of the target temperature range from a pre-stored treatment plan; S02. Configure control parameters: Set the length of the predetermined future time period, the sampling frequency of temperature monitoring, the maximum allowable rate of change of power adjustment, and the safe operating upper and lower limits of laser power; S03. Start-up and calibration: Control the laser emitting device to output initial safe power and start the temperature monitoring device to verify the validity and stability of the temperature monitoring data and confirm that the target area is correctly identified and locked.

6. The method according to claim 1, characterized in that, The method further includes a security monitoring and intervention step, which is executed concurrently with steps S1 to S6, including: S7.1 Real-time safety verification: In each control cycle, in addition to adjusting based on prediction, the latest measured temperature value in the current temperature sequence is directly monitored. If the measured temperature value exceeds the preset absolute safety temperature limit, the highest priority emergency stop command is immediately generated to force the laser power to be reduced to zero or the lowest safety level and to issue an alarm. S7.2 Prediction Failure Detection: Continuously evaluate the prediction performance of the prediction model described in step S2, and determine whether the prediction has failed by comparing the degree of agreement between the recent predicted temperature value and the subsequent actual monitored temperature value; if prediction failure is detected, automatically switch to the proportional-integral-derivative backup control mode based on the measured temperature deviation, and record the failure event for subsequent analysis.

7. The method according to claim 1, characterized in that, The method also includes an adaptive optimization step during operation, which is executed periodically or under triggered conditions during stable system operation, including: S81, Data Cache: Continuously stores the input data, prediction results, control commands, and subsequent actual temperature feedback data in a time-aligned manner in a circular buffer to form a historical operation dataset; S82. Performance evaluation: Periodically analyze the historical operating dataset and calculate the evaluation index of temperature control accuracy. The evaluation index includes the standard deviation of temperature fluctuation within the target temperature range, the number of overshoots, and the percentage of time the temperature is maintained within the target range. S83. Parameter Fine-tuning: Based on the performance evaluation results of step S82, if a downward trend in control accuracy is found or the optimal level is not reached, the internal parameters used in the control process are automatically fine-tuned. The internal parameters include the time window length used in step S4 to estimate the power-temperature influence coefficient K, and the gain coefficient used to calculate the power adjustment amount ΔP. The fine-tuning adopts an optimization method based on gradient descent or heuristic search to improve the stability and accuracy of temperature control in the next stage. S84. Model update trigger judgment: Based on long-term historical operating data, analyze the changes in the thermal response characteristics of the target area. If it is determined that the characteristics have drifted, an early warning is triggered, prompting that the prediction model in step S2 needs to be updated to adapt to the new thermal environment.

8. The method according to claim 1, characterized in that, The method supports multimodal treatment protocols, wherein the preset target temperature range is not fixed and changes dynamically according to a predefined time-temperature curve; In step S3, the predicted temperature value is compared with the dynamically changing target temperature curve value at the current time and the corresponding future time. In step S4, when calculating the adjusted laser power parameters, the goal is to ensure that the predicted temperature value tracks the dynamically changing target temperature curve. The time-temperature curve defines the temperature level and duration required to be maintained at different stages of the treatment process, in order to realize complex treatment strategies based on different temperatures.

9. A self-adjusting laser power control system based on temperature prediction, characterized in that, include: One or more processors; A memory having stored one or more programs that, when executed by one or more processors, cause the one or more processors to implement a temperature-predictive-based self-adjusting laser power control method according to any one of claims 1 to 8.

Images