Physical thermal ablation energy regulation and control method based on intelligent feedback

Abstract

The invention belongs to the technical field of physical thermal ablation, and relates to a physical thermal ablation energy regulation and control method based on intelligent feedback, and the method comprises the steps: collecting point-shaped temperature data, impedance data and ultrasonic data, eliminating the asynchronization and heterogeneity problems through space-time alignment, and combining with focus region segmentation, thereby achieving the energy regulation and control of the physical thermal ablation. Discrete data and continuous images are integrated into a voxelization data cube containing space coordinates, physical attributes and semantic tags, then the temperature change rate and the impedance change rate are calculated based on the voxelization data cube, and then a morphological prediction network is fused with multi-modal features to output a voxel-level ablation probability graph. The method comprises the following steps: realizing prospective prediction of an ablation range, ensuring ablation accuracy by morphological deviation through a three-layer parallel regulation and control strategy, preventing and controlling local overheating by temperature gradient, avoiding tissue carbonization by an impedance change rate, and cooperatively deciding through multi-dimensional indexes to solve the contradiction that single regulation and control cannot give consideration to accurate coverage of a focus and protection of normal tissues; and misjudgment caused by abnormity of single data is avoided.

Description

Technical Field

[0001] This application belongs to the field of physical thermal ablation technology, and more specifically, relates to a physical thermal ablation energy regulation method based on intelligent feedback. Background Technology

[0002] Physical thermal ablation is an important minimally invasive technique for the clinical treatment of solid tumors. Its core principle is to release heat energy through the ablation needle, raising the temperature of the lesion tissue to above 60°C, causing coagulative necrosis of tumor cells, while minimizing damage to surrounding normal tissues. With the increasing demand for minimally invasive treatments, clinicians have raised their requirements for the precision and safety of thermal ablation. It is necessary to ensure that the ablation area completely covers the lesion, strictly control the ablation boundary, and avoid safety risks such as tissue overheating and carbonization and needle tract burns during the operation.

[0003] Currently, the main methods for controlling ablation equipment are to integrate a single-point temperature sensor on the ablation needle to monitor the needle tip temperature in real time, reducing power when the temperature exceeds a preset threshold and increasing power when it falls below the threshold; or by monitoring changes in tissue impedance, pausing ablation when the impedance suddenly increases. However, since single-point temperature only reflects the local thermal state of the needle tip, it cannot cover the temperature distribution of the entire lesion and surrounding tissue. Secondly, the acquisition frequency and timestamp of impedance monitoring are inconsistent with those of ultrasound images, making it difficult to establish a precise correlation between impedance changes, temperature distribution, and ablation morphology, resulting in one-sided control methods. Summary of the Invention

[0004] This invention provides a physical thermal ablation energy regulation method based on intelligent feedback, which aims to solve the technical problem that the current method of using single data leads to a one-sided regulation basis and insufficient regulation accuracy.

[0005] A method for regulating physical thermal ablation energy based on intelligent feedback includes the following steps: S1. Collect point temperature data, impedance data, and ultrasound data of the tissue around the needle tip. Perform spatiotemporal alignment on the acquired data, and then segment the lesion area of ​​the spatiotemporally aligned data to obtain a voxelized data cube. S2. Using voxelized data cubes, calculate the rate of temperature change and the rate of impedance change respectively; then, based on the morphological prediction network model, analyze the rate of temperature change, the rate of impedance change, and the voxelized data cubes, and output a voxel-level ablation probability map. S3. Compare the ablation probability map with the preset target shape, calculate the shape deviation, and simultaneously detect the temperature gradient and impedance change rate. Based on this, construct a three-layer parallel control strategy to control the thermal ablation device.

[0006] This invention collects point-like temperature data, impedance data, and ultrasound data. By eliminating the asynchronous and heterogeneous problems of multi-source data through spatiotemporal alignment, and combining this with lesion region segmentation, discrete data and continuous images are integrated into a voxelized data cube containing spatial coordinates, physical attributes, and semantic labels, providing comprehensive and accurate multidimensional data for subsequent regulation. Secondly, based on the voxelized data cube, the temperature change rate and impedance change rate are calculated. Then, a morphological prediction network is used to fuse multimodal features and output a voxel-level ablation probability map, enabling a forward-looking prediction of the ablation range. Furthermore, a three-layer parallel regulation strategy is employed to ensure ablation accuracy through morphological deviation, control local overheating through temperature gradient, and avoid tissue carbonization through impedance change rate. Through multi-dimensional index collaborative decision-making, the contradiction between precise lesion coverage and protection of normal tissue that cannot be achieved by single regulation is resolved, while also avoiding misjudgment caused by single data anomalies.

[0007] Preferably, the spatiotemporal alignment includes the following steps: Based on the ultrasonic data frame rate, an FPGA high-precision clock module is used as the master clock to provide a synchronous start signal for all sensors. For each frame of ultrasonic data, a time window with a predetermined time before and after each frame is taken, and the temperature and impedance data within the window are interpolated by Gaussian weighted average to obtain data synchronized with the ultrasonic frame. Using the coordinate system of the ultrasound data as a reference, the spatial coordinates of the ultrasound voxels are calculated based on the origin coordinates and voxel resolution. Then, the real-time coordinates of the temperature sensor and impedance electrode pair are obtained through electromagnetic navigation, and the index positions of the temperature sensor and impedance electrode pair in the ultrasound voxel coordinate system are calculated in reverse.

[0008] Preferably, obtaining the voxelized data cube includes the following steps: Denoising and contrast enhancement are performed on the spatiotemporally aligned ultrasound images. The enhanced images are then input into a deep learning model, which outputs lesion area masks, blood vessel masks, and neural masks. The spatiotemporally aligned data is combined with the lesion region mask, blood vessel mask, and neural mask output by the model to form a voxelized data cube; The voxelized data cube uses the ultrasound voxel spatial index and ultrasound frame index as dimensions. Each voxel contains spatial coordinates, real-time temperature, tissue conductivity, ultrasound echo intensity, as well as lesion area mask, vascular mask, and neural mask. In the process of synthesizing the voxelized data cube, inverse distance weighted interpolation is used to fill the sensing blind spots for discrete point data of temperature and impedance.

[0009] Preferably, the calculation of the temperature change rate includes the following steps: The temperature difference between each voxel and the surrounding voxels is calculated using ultrasound imaging data. Then, the thermal diffusivity corresponding to the current ablation organ is retrieved from the tissue type database. The thermal diffusivity is multiplied by the temperature gradient to obtain the thermal diffusivity contribution value of the corresponding voxel. By combining the spatial gradient of ultrasound intensity, the square of the current during ablation, and tissue impedance, and then dividing by the voxel volume, the energy deposition per unit volume of tissue is obtained; then, based on tissue density and tissue specific heat capacity, the energy deposition is divided by the product of density and tissue specific heat capacity to obtain the energy deposition temperature rise. The blood flow velocity in the current area is monitored by color Doppler ultrasound and converted into a blood perfusion coefficient. The difference between the current temperature of the corresponding voxel and the body's baseline temperature is then calculated. The perfusion coefficient is multiplied by the temperature difference to obtain the blood flow heat dissipation and cooling effect. The temperature change rate is obtained by adding the heat diffusion contribution value to the energy deposition temperature rise and subtracting the blood flow heat dissipation temperature drop.

[0010] Preferably, the calculation of the impedance change rate includes the following steps: Calculate the difference between the current impedance and the maximum impedance, and then, in conjunction with the temperature and temperature gradient, calculate the rate of change of impedance over time.

[0011] Preferably, the morphology prediction network model includes: Input layer: Input data includes voxelized data cubes, temperature change rate, and impedance change rate; Spatiotemporal feature extraction layer: Extracts local spatial and temporal correlation features from input features through 3D convolution operations; The generated correlation features are classified into impedance-related feature groups and temperature-related feature groups to obtain impedance feature groups and temperature feature groups. The obtained impedance feature groups and temperature feature groups are then converted into data formats suitable for LSTM network processing. The converted impedance characteristic set and temperature characteristic set are input into two independent channels of the dual-channel LSTM, including the impedance channel and the temperature channel; The impedance channel is used to receive data from the impedance feature set and to learn the rate of change of impedance-related features over time. The temperature channel is used to receive data of temperature feature groups and to learn the rate of change of temperature-related features over time. Then, based on the input gate, forget gate, and output gate of LSTM, the characteristic signals of the impedance channel and the temperature channel are received simultaneously. The dynamic relationship between the cell state and the hidden state of the two channels is automatically correlated, and then the coupling feature is output. Then, a transposed convolution operation is used to restore the coupled features to three-dimensional spatial resolution, generating an ablation probability map.

[0012] Preferably, the three-layer parallel control strategy includes PID power control based on morphological deviation, fuzzy rule control based on temperature gradient, and emergency circuit breaker control based on impedance change. When the absolute value of the ablation morphology deviation exceeds a set threshold, the PID power control of the morphology deviation is activated. The temperature gradient-based fuzzy rule control is activated when the maximum temperature gradient within the tissue exceeds a safety threshold. The emergency circuit breaker control based on impedance mutation executes multi-level responses based on predefined rules to achieve safety protection.

[0013] Preferably, the PID power control includes the following steps: The power adjustment contribution value of the proportional element is calculated based on the ablation morphology deviation and proportional gain. Based on historical time-series data of ablation morphology deviation, the power adjustment contribution value of the integral stage is calculated. Based on the rate of change of ablation morphology deviation and the differential gain, the power adjustment contribution value of the differential element is calculated. The power adjustment amount is obtained by adding the power adjustment contributions of the proportional, integral, and derivative components. The calculation cycle of the PID power control is synchronized with the calculation cycle of the ultrasonic frame rate and the morphological prediction network model.

[0014] Preferably, the fuzzy rule control based on temperature gradient includes the following steps: The membership function is used to determine the category of temperature gradient and impedance change rate; then, for each combination of temperature gradient category and impedance change rate category, the corresponding pulse parameters are retrieved from the rule base. Then, the weight of each combination parameter is calculated based on the product of the membership degree of the temperature gradient and the rate of change of impedance; then, the pulse parameters of all combinations are weighted and averaged to obtain the pulse parameters of the fuzzy rule control output, and the thermal ablation device is controlled based on the output pulse parameters.

[0015] Preferably, the emergency circuit breaker control for impedance abrupt changes includes a three-level response strategy; Level 1 response: When the impedance change rate exceeds a threshold, reduce the current output power to a predetermined percentage of the current power. Level 2 response: When the real-time impedance value exceeds a predetermined multiple of the baseline impedance before ablation, immediately switch to safe pulse mode; Level 3 Response: If the ablation morphology deviation exceeds the maximum permissible deviation and the maximum temperature gradient exceeds the predetermined threshold, an emergency stop signal is triggered, cutting off the energy output of the thermal ablation device and triggering an alarm.

[0016] The beneficial effects of this invention include: This invention collects point-like temperature data, impedance data, and ultrasound data. By eliminating the asynchronous and heterogeneous problems of multi-source data through spatiotemporal alignment, and combining this with lesion region segmentation, discrete data and continuous images are integrated into a voxelized data cube containing spatial coordinates, physical attributes, and semantic labels, providing comprehensive and accurate multidimensional data for subsequent regulation. Secondly, based on the voxelized data cube, the temperature change rate and impedance change rate are calculated. Then, a morphological prediction network is used to fuse multimodal features and output a voxel-level ablation probability map, enabling a forward-looking prediction of the ablation range. Furthermore, a three-layer parallel regulation strategy is employed to ensure ablation accuracy through morphological deviation, control local overheating through temperature gradient, and avoid tissue carbonization through impedance change rate. Through multi-dimensional index collaborative decision-making, the contradiction between precise lesion coverage and protection of normal tissue that cannot be achieved by single regulation is resolved, while also avoiding misjudgment caused by single data anomalies. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is an overall step diagram provided for an embodiment of the present invention.

[0019] Figure 2 A simplified diagram of step S2 provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the technical problems, technical solutions, and beneficial effects to be solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.

[0021] See Figure 1 As shown, a method for regulating physical thermal ablation energy based on intelligent feedback includes the following steps: S1. Collect point temperature data, impedance data, and ultrasound data of the tissue around the needle tip, perform spatiotemporal alignment on the acquired data, and segment the lesion area after spatiotemporal alignment to obtain a voxelized data cube; wherein, in this invention, the voxel is the core unit of three-dimensional spatial discretization, the smallest volume unit of medical imaging data, and a spatial container that carries physical properties. In this embodiment, a distributed fiber optic temperature sensor is used. 3-5 sensors are implanted at 0.5mm intervals along the ablation needle axis, or 4 peripheral sensors are implanted 5mm around the lesion to form a spatially distributed temperature measurement network. The temperature sensor has an accuracy of 1kHz to ensure the capture of rapid temperature changes. The raw data is transmitted to the signal conditioning module in the form of a voltage signal and is converted into a temperature value via analog-to-digital conversion. Its impedance data is obtained by forming a circuit between the ablation electrode and the reference electrode on the body surface, and the tissue impedance is measured using a high-frequency alternating current; the sampling rate of the impedance data is 500Hz.

[0022] Ultrasound data is acquired using an intraoperative real-time ultrasound system, supporting 2D dynamic scanning or 3D volumetric imaging, and simultaneously acquiring tissue morphology and blood flow information; a frame rate of 30fps is set (one frame is generated every 33 milliseconds), with a single frame image resolution of 512×512 pixels. During 3D imaging, multiple 2D images are superimposed to generate 1×1×1 mm³ voxel three-dimensional data, while simultaneously recording the probe spatial position at the moment of image acquisition (acquired through an electromagnetic positioning sensor).

[0023] For the k-th frame of ultrasound, take the time window , This represents the timestamp of the k-th ultrasound frame; Indicates the defined time interval; , The ultrasound frame rate is represented; based on the selected time window, the temperature and impedance data within the window are weighted and interpolated to obtain data synchronized with the ultrasound frame. For the temperature sensor window containing Each temperature point is used, among which Indicates the temperature sampling rate; Time window width; Number of temperature sampling points within the time window; calculated using Gaussian weighted average. Temperature value at time: ; ; ; In the formula: This represents the temperature value of the i-th temperature sensor at the k-th ultrasonic frame after time alignment. This represents the set of indices of temperature sampling points within the time window corresponding to the k-th ultrasound frame; This represents the Gaussian weight of the m-th temperature sampling point; This represents the original temperature value of the i-th temperature sensor at the m-th sampling point; This represents the timestamp of the m-th temperature sampling point; similarly, the impedance value of the j-th pair of impedance electrodes in the k-th ultrasonic frame is calculated after time alignment.

[0024] Using the coordinate system of ultrasound data Based on this, define the spatial coordinates of the ultrasound voxel: ; In the formula This indicates the index of the ultrasound voxel in the coordinate system, corresponding to the voxel position in the X, Y, and Z axes, respectively; This represents the spatial coordinates of the u-th voxel along the X-axis. This represents the spatial coordinates of the v-th voxel along the Y-axis. This represents the spatial coordinates of the w-th voxel along the z-axis. Indicates ultrasound data in The coordinates of the origin in the coordinate system are the spatial positions corresponding to the voxel index (0,0,0); The voxel resolution along the X-axis is 0.3 mm. The voxel resolution along the Y-axis is 0.3 mm. The voxel resolution along the Z-axis is 0.5 mm. The real-time coordinates of the temperature sensor are obtained through an electromagnetic navigation system. ; Substitute the spatial coordinates of the ultrasonic voxels to calculate the index of the temperature sensor in the ultrasonic voxel coordinate system: ; In the formula: This represents the index of the i-th temperature sensor in the ultrasonic voxel coordinate system; Indicates rounding down; Indicates the coordinates of the origin of the ultrasound body data; The spatial mapping of the impedance electrode pairs can be referenced from the spatial mapping of the temperature sensor, as the mapping logic principles of the two are the same.

[0025] Based on the above steps, the temperature data, impedance data and ultrasound data were spatiotemporally aligned, and then the spatiotemporally aligned ultrasound images were denoised and contrast enhanced. The denoising and contrast enhancement used wavelet threshold denoising to suppress speckle noise, and combined with CLAHE contrast enhancement to improve the clarity of lesion boundaries. The B-mode acoustic intensity, CHFI blood flow velocity, and elastic imaging stiffness values ​​in the denoised and enhanced ultrasound images are stitched together as a 3-channel input to obtain 3D ultrasound body data. The 3D U-Net++ segmentation model is used. The preprocessed 3D ultrasound data is input and three binary masks are output: target demask (i.e. lesion area mask), where 1 represents the lesion area and 0 represents normal tissue; blood vessel mask, where 1 represents blood vessels with a diameter greater than 2 mm and 0 represents non-blood vessel areas; and nerve mask, where 1 represents nerve bundles and 0 represents non-nerve areas. The spatiotemporally aligned data is combined with the lesion region mask, blood vessel mask, and neural mask output by the model to form a voxelized data cube. Since temperature and impedance are discrete point data, spatial interpolation is needed to extend them to the entire ultrasound volume space to fill the perception blind spots. Inverse distance weighted interpolation is used to calculate the temperature of any voxel using the time-aligned temperature points. ; In the formula: This represents the target ultrasound voxel at the k-th ultrasound frame. Temperature value after difference; Index representing the target ultrasound voxel; This indicates the total number of temperature sensors involved in the interpolation. This represents the time-aligned temperature value of the i-th temperature sensor at the k-th ultrasonic frame. Indicates the target ultrasound voxel Mapping voxels to the i-th temperature sensor Spatial distance; p represents the distance weighting index; Similarly, the impedance of any voxel is calculated based on the impedance points after time alignment, as shown in the temperature calculation formula for any voxel above; after performing inverse distance weighted interpolation on temperature and impedance, a voxelized data cube is constructed, in which the voxelized data cube has the ultrasound voxel spatial index and ultrasound frame index as dimensions, and each voxel contains spatial coordinates, real-time temperature, tissue conductivity, ultrasound echo intensity, and lesion area mask, vascular mask and neural mask. S2. Using voxelized data cubes, calculate the rate of temperature change and the rate of impedance change respectively; then, based on the morphological prediction network model, analyze the rate of temperature change, the rate of impedance change, and the voxelized data cubes, and output a voxel-level ablation probability map. The calculation of the temperature change rate includes the following steps: The temperature difference between each voxel and the surrounding voxels is calculated using ultrasound imaging data. Then, the thermal diffusivity corresponding to the current ablation organ is retrieved from the tissue type database. The thermal diffusivity is multiplied by the temperature gradient to obtain the thermal diffusivity contribution value of the corresponding voxel. By combining the spatial gradient of ultrasound intensity, the square of the current during ablation, and tissue impedance, and then dividing by the voxel volume, the energy deposition per unit volume of tissue is obtained; then, based on tissue density and tissue specific heat capacity, the energy deposition is divided by the product of density and tissue specific heat capacity to obtain the energy deposition temperature rise. The blood flow velocity in the current area is monitored by color Doppler ultrasound and converted into a blood perfusion coefficient. The difference between the current temperature of the corresponding voxel and the body's baseline temperature is then calculated. The perfusion coefficient is multiplied by the temperature difference to obtain the blood flow heat dissipation and cooling effect. The temperature change rate is obtained by adding the energy deposition temperature rise rate to the obtained thermal diffusion contribution value and subtracting the blood flow heat dissipation temperature drop rate; the specific expression for the temperature change rate is as follows: ; In the formula: This indicates the rate of change of tissue temperature over time. t represents the real-time temperature of the tissue at the voxel level; t represents time. Indicates the thermal diffusivity of the tissue; Represents the Laplace operator for the temperature field; Represents the energy deposition function; Indicates the blood perfusion coefficient; This indicates the body's basal body temperature; Indicates the blood perfusion coefficient; Indicates the contribution value of thermal diffusion; Indicates the temperature rise during energy deposition; Indicates the degree of heat dissipation and temperature reduction through blood flow; In this embodiment, the temperature change rate is calculated from three key physical processes: heat diffusion, energy deposition, and blood flow heat dissipation. This comprehensively considers various influencing factors of tissue heat transfer, making the calculation of the temperature change rate more closely match the actual thermal dynamic changes of the tissue during the thermal ablation process. This provides an accurate thermal state basis for subsequent energy regulation and helps to avoid over- or under-ablation caused by temperature calculation deviations.

[0026] The calculation of the rate of change of impedance includes the following steps: Calculate the difference between the current impedance and the maximum impedance, and then, in conjunction with the temperature and temperature gradient, calculate the rate of change of impedance over time. The specific expression is as follows: ; In the formula: This represents the rate of change of tissue impedance over time. Indicates real-time tissue impedance at the voxel level; This represents the constant of the rate of change of impedance; Indicates the activation energy of tissue denaturation; This represents the temperature sensitivity index. This represents the maximum impedance value when the tissue is fully carbonized. Indicates the magnitude of the temperature gradient; The physical power calculated using the above-mentioned rate of temperature change and rate of impedance change is obtained using the following solution method: The finite volume method is used to divide the continuous multimodal data cube into independent voxel units of 1 cubic millimeter, and each voxel is calculated independently. The calculation period is set to 33 milliseconds, which is fully synchronized with the ultrasound imaging frame rate, ensuring that each frame of ultrasound image is generated corresponds to one physical model calculation, avoiding data time misalignment. The boundary conditions and constraints used in the solution process include: Tissue surface voxels: Set the air-tissue exchange coefficient (e.g., fixed at 25 W / m²K), and calculate the additional heat dissipation of the surface voxels based on the difference between air temperature and tissue surface temperature to avoid overestimating the surface temperature. Vascular region voxels: After identifying the location of blood vessels using ultrasound, the temperature of the voxels within the vessels is fixed to the blood temperature. Simultaneously, the heat exchange with the blood vessels is additionally considered when calculating the voxels around the vessels. The calculation formula incorporates the heat exchange between the peripheral voxels and the blood vessels. This heat exchange amount is then subtracted from the original formula's result to simulate the actual heat dissipation effect of the blood vessels. The expression for the heat exchange between the peripheral voxels and the blood vessels is as follows: ; In the formula: This represents the convective heat transfer coefficient of the blood vessel wall; Indicates the voxel-blood vessel contact area; Indicates the real-time temperature of the voxel; Indicates blood temperature; Indicates voxel volume; Indicates tissue density; This indicates the specific heat capacity of the tissue.

[0027] See Figure 2 As shown, the morphology prediction network model includes: Input layer: Input data includes voxelized data cubes, temperature change rate, and impedance change rate. Before inputting the data into the input layer, the data is first filtered, and the specific steps are as follows: From the voxelized data cube, a target region of 32×32×32 voxels is extracted. The purpose of this step is to focus on the core region to be ablated and avoid inefficiency caused by calculating irrelevant regions. Seven key features were extracted from the voxelized data cube and physical model results as network input channels, including current tissue temperature, temperature gradient amplitude, current tissue impedance, impedance change rate, ultrasound intensity, temperature change rate, and binary mask of the target ablation area. The extracted data is then input into the input layer. Spatiotemporal feature extraction layer: Extracts local spatial and temporal correlation features from the input features through 3D convolution operations; for example: A 5×5×5 3D convolution kernel (covering the current voxel and the range of the two surrounding voxels) is used to perform convolution calculations on the input 32×32×32 voxels to capture the temperature, impedance, and acoustic intensity correlations of adjacent voxel components. The ReLU activation function is then applied to the convolution result to filter out invalid negative features and enhance the expressive power of useful features, ultimately generating a 64-channel feature map (each channel corresponds to a type of local correlation feature). The generated correlation features are classified into impedance-related feature groups and temperature-related feature groups to obtain impedance feature groups and temperature feature groups. These impedance and temperature feature groups are then converted into a data format suitable for LSTM network processing, for example: Based on the feature learning logic of 3D convolution, the features are divided into two categories according to whether they are related to impedance or temperature from the 64-channel 3D feature map. Impedance-related characteristic group : Among the 64 channels, the characteristic channels that primarily capture impedance and rate of change in relation to surrounding voxels were selected, including: The difference between the impedance value of a voxel and the impedance values ​​of adjacent voxels; The relationship between the rate of change of impedance and the surrounding voxel temperature and sound intensity; The correlation between resistance characteristics and the target ablation mask; (e.g., whether the region with rapid impedance change is within the target ablation zone). Temperature-related feature groups : Among the 64 channels selected, the feature channels that primarily capture temperature and its derived features in relation to surrounding voxels were identified. Their semantic features include: The temperature of a voxel, its temperature gradient, and the difference in temperature gradient between a voxel and its neighboring voxels; The correlation between the rate of temperature change predicted by the physical model and heat dissipation from the surrounding voxel blood flow; The correlation between temperature characteristics and vascular regions (e.g., whether areas with slow temperature rise are close to blood vessels). It should be noted that the classification is based on the loss feedback during model training, automatically learning the channels corresponding to impedance dynamic prediction and temperature dynamic prediction, without the need to manually specify the channel numbers. The input requirements for LSTM will be and The dimensions are adjusted separately. However, how to adjust the data to conform to the input format of the LSTM model is a conventional technique in this field, so it will not be described in detail in this embodiment.

[0028] The converted impedance characteristic set and temperature characteristic set are input into two independent channels of the dual-channel LSTM, including the impedance channel and the temperature channel; The impedance channel is used to receive data from the impedance feature set and to learn the rate of change of impedance-related features over time. The temperature channel is used to receive data of temperature feature groups and to learn the rate of change of temperature-related features over time. Then, based on the input gate, forget gate, and output gate of LSTM, the characteristic signals of the impedance channel and temperature channel are received simultaneously. The dynamic relationship between the cell state and the hidden state of the two channels is automatically correlated, and then the coupling characteristics of 32 units are output. Then, a transposed convolution operation is used to restore the coupled features to three-dimensional spatial resolution, generating an ablation probability map. Specifically: Using a 1×1×1 transposed convolution kernel, the 32-unit coupled features output by the LSTM are progressively upsampled to restore the resolution of 32×32×32 voxels from the low-resolution feature map; Apply the Sigmoid activation function to the result after transposed convolution to transform the feature values ​​into probability values ​​between 0 and 1, and obtain a voxel-level ablation probability map. Voxels with an ablation probability greater than 0.5 in the probability map are identified as ablation zones, and voxels with a probability less than or equal to 0.5 are identified as unablated zones, thus converting the probability map into a clear three-dimensional ablation range prediction.

[0029] Since thermal ablation requires precise control of the ablation morphology to cover the lesion and protect the surrounding tissue, predicting the ablation morphology in advance helps to adjust the energy input in a timely manner. Multimodal data (voxelated data, temperature change rate, impedance change rate) contains rich tissue state and thermal dynamic information. By constructing a network model, the spatiotemporal correlation of the data components can be fully explored to achieve accurate prediction of the ablation morphology, providing a target reference for subsequent energy regulation based on morphology deviation.

[0030] S3. Compare the ablation probability map with the preset target shape, calculate the shape deviation, and simultaneously detect the temperature gradient and impedance change rate. Based on this, construct a three-layer parallel control strategy to control the thermal ablation device.

[0031] The morphological deviation is calculated as follows: ; In the formula: Indicates morphological deviation; Indicates the predicted ablation volume; Indicates the target ablation volume; Indicates the weighting parameter; Indicates the target number of sampling points; This represents the target surface normal vector at the i-th sampling point; This represents the three-dimensional coordinates of the i-th sampling point on the predicted ablation morphology; This represents the three-dimensional coordinates of the i-th sampling point on the target ablation morphology; The three-layer parallel control strategy includes PID power control based on morphological deviation, fuzzy rule control based on temperature gradient, and emergency circuit breaker control based on impedance change. PID power control based on ablation morphology deviation: When the absolute value of the ablation morphology deviation exceeds a set threshold, PID power control based on the morphology deviation is activated; the PID power control includes the following steps: The power adjustment contribution value of the proportional element is calculated based on the ablation morphology deviation and proportional gain. Based on historical time-series data of ablation morphology deviation, the power adjustment contribution value of the integral stage is calculated. Based on the rate of change of ablation morphology deviation and the differential gain, the power adjustment contribution value of the differential element is calculated. The power adjustment contribution values ​​of the proportional, integral, and derivative components are added together to obtain the power adjustment amount, as shown in the following expression: ; In the formula: Indicates the amount of power adjustment; Indicates proportional gain; This indicates a deviation in the ablation pattern; Indicates integral gain; Represents the integral of morphological deviation; Represents differential gain; Indicates the rate of change of morphological deviation; The calculation cycle of the PID power control is synchronized with the calculation cycle of the ultrasonic frame rate and the morphology prediction network model, i.e., it is calculated once every 33ms. The power adjustment range is ±40W; New power , This indicates the power output before adjustment; The new power must meet the constraints of minimum safe power and maximum safe power. If the new power is less than the minimum safe power, it will be output at the minimum safe power; if the new power is greater than the maximum safe power, it will be output at the maximum safe power.

[0032] Fuzzy rule control based on temperature gradient: The fuzzy rule control based on temperature gradient is activated when the maximum temperature gradient within the tissue exceeds a safety threshold; the control steps are as follows: The membership function is used to determine the category of temperature gradient and impedance change rate; for example, three fuzzy sets, low, medium and high, are defined for temperature gradient. Low fuzzy set: When the temperature gradient is in the range of 0-5℃ / mm, it completely belongs to the low fuzzy set and the membership degree is 1; in the range of 5-10℃ / mm, the membership degree decreases linearly to 0 as the temperature gradient increases; when the temperature gradient is greater than or equal to 10℃ / mm, it does not belong to the low fuzzy set at all and the membership degree is 0. For fuzzy sets: when the temperature gradient is less than 5℃ / mm or greater than or equal to 20℃ / mm, the membership degree is 0; in the range of 5-15℃ / mm, the membership degree increases linearly to 1 as the temperature gradient increases; in the range of 1-20℃ / mm, the membership degree remains at 1; in the range of 20-25℃ / mm, the membership degree decreases linearly to 0 as the temperature gradient increases. Highly fuzzy sets: When the temperature gradient is less than 15℃ / mm, the membership degree is 0; in the range of 15-30℃ / mm, the membership degree increases linearly to 1 as the temperature gradient increases; when it is greater than or equal to 30℃ / mm, it is completely a member (membership degree is 1).

[0033] The definition rules for the fuzzy set of impedance change rate are as follows: Slow fuzzy set: The membership degree is 1 when the impedance change rate is in the range of 0 - 20Ω / s; in the range of 20 - 60Ω / s, the membership degree decreases linearly to 0 as the impedance change rate increases; when it is greater than or equal to 60Ω / s, the membership degree is 0.

[0034] For fuzzy sets: when the impedance change rate is less than 20Ω / s or greater than or equal to 140Ω / s, the membership degree is 0; in the 20-100Ω / s range, the membership degree increases linearly to 1 as the impedance change rate increases; in the 100-120Ω / s range, the membership degree remains at 1; in the 120-140Ω / s range, the membership degree decreases linearly to 0 as the impedance change rate increases.

[0035] Fast fuzzy set: When the impedance change rate is less than 100Ω / s, the membership degree is 0; in the range of 100-200Ω / s, the membership degree increases linearly to 1 as the impedance change rate increases; when it is greater than or equal to 200Ω / s, the membership degree is 1.

[0036] For each combination of temperature gradient category and impedance change rate category, the corresponding pulse parameters are retrieved from the pre-established rule base, where the rule base pre-establishes a mapping of pulse parameters corresponding to the impedance and temperature combination categories; Then, the weight of each combination parameter is calculated based on the product of the membership degree of the temperature gradient and the rate of change of impedance; then, the pulse parameters of all combinations are weighted and averaged to obtain the pulse parameters of the fuzzy rule control output, and the thermal ablation device is controlled based on the output pulse parameters.

[0037] Emergency circuit breaker control based on impedance change: Safety protection is achieved through multi-level responses executed according to predefined rules. These multi-level responses include: Level 1 response: When the impedance change rate exceeds a threshold (e.g., 100Ω / s), the current output power is reduced to a predetermined percentage of the current power, such as 30%; if the current power is 100W, it is reduced to 30W according to the Level 1 response. Level 2 response: When the real-time impedance value exceeds a predetermined multiple (2 times) of the baseline impedance before ablation, immediately switch to safe pulse mode; Level 3 response: If the ablation morphology deviation exceeds the maximum permissible deviation (e.g., 10 mm²) and the maximum temperature gradient exceeds the predetermined threshold (20 °C / mm), an emergency stop signal is triggered, cutting off the energy output of the thermal ablation device and triggering an alarm.

[0038] In this embodiment, a three-layer parallel control strategy is constructed to control different aspects of the problem, namely morphological deviation, temperature gradient, and impedance mutation, forming a multi-level and multi-dimensional control system. PID control of morphological deviation ensures accurate ablation morphology, fuzzy rule control of temperature gradient responds to abnormal heat distribution, and emergency fuse control of impedance mutation achieves safety protection. The synergy of the three can improve the accuracy, adaptability and safety of energy control.

[0039] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for regulating physical thermal ablation energy based on intelligent feedback, characterized in that, Includes the following steps: S1. Collect point temperature data, impedance data, and ultrasound data of the tissue around the needle tip. Perform spatiotemporal alignment on the acquired data, and then segment the lesion area of ​​the spatiotemporally aligned data to obtain a voxelized data cube. S2. Using voxelized data cubes, calculate the rate of temperature change and the rate of impedance change respectively; then, based on the morphological prediction network model, analyze the rate of temperature change, the rate of impedance change, and the voxelized data cubes, and output a voxel-level ablation probability map. S3. Compare the ablation probability map with the preset target shape, calculate the shape deviation, and simultaneously detect the temperature gradient and impedance change rate. Based on this, construct a three-layer parallel control strategy to control the thermal ablation device.

2. The method for regulating physical thermal ablation energy based on intelligent feedback according to claim 1, characterized in that, The spatiotemporal alignment includes the following steps: Based on the ultrasonic data frame rate, an FPGA high-precision clock module is used as the master clock to provide a synchronous start signal for all sensors. For each frame of ultrasonic data, a time window with a predetermined time before and after each frame is taken, and the temperature and impedance data within the window are interpolated by Gaussian weighted average to obtain data synchronized with the ultrasonic frame. Using the coordinate system of the ultrasound data as a reference, the spatial coordinates of the ultrasound voxels are calculated based on the origin coordinates and voxel resolution. Then, the real-time coordinates of the temperature sensor and impedance electrode pair are obtained through electromagnetic navigation, and the index positions of the temperature sensor and impedance electrode pair in the ultrasound voxel coordinate system are calculated in reverse.

3. The method for regulating physical thermal ablation energy based on intelligent feedback according to claim 1, characterized in that, The process of obtaining the voxelized data cube includes the following steps: Denoising and contrast enhancement are performed on the spatiotemporally aligned ultrasound images. The enhanced images are then input into a deep learning model, which outputs lesion area masks, blood vessel masks, and neural masks. The spatiotemporally aligned data is combined with the lesion region mask, blood vessel mask, and neural mask output by the model to form a voxelized data cube; The voxelized data cube uses the ultrasound voxel spatial index and ultrasound frame index as dimensions. Each voxel contains spatial coordinates, real-time temperature, tissue conductivity, ultrasound echo intensity, as well as lesion area mask, vascular mask, and neural mask. In the process of synthesizing the voxelized data cube, inverse distance weighted interpolation is used to fill the sensing blind spots for discrete point data of temperature and impedance.

4. The method for regulating physical thermal ablation energy based on intelligent feedback according to claim 1, characterized in that, The calculation of the temperature change rate includes the following steps: The temperature difference between each voxel and the surrounding voxels is calculated using ultrasound imaging data. Then, the thermal diffusivity corresponding to the current ablation organ is retrieved from the tissue type database. The thermal diffusivity is multiplied by the temperature gradient to obtain the thermal diffusivity contribution value of the corresponding voxel. By combining the spatial gradient of ultrasound intensity, the square of the current during ablation, and tissue impedance, and then dividing by the voxel volume, the energy deposition per unit volume of tissue is obtained; then, based on tissue density and tissue specific heat capacity, the energy deposition is divided by the product of density and tissue specific heat capacity to obtain the energy deposition temperature rise. The blood flow velocity in the current area is monitored by color Doppler ultrasound and converted into a blood perfusion coefficient. The difference between the current temperature of the corresponding voxel and the body's baseline temperature is then calculated. The perfusion coefficient is multiplied by the temperature difference to obtain the blood flow heat dissipation and cooling effect. The temperature change rate is obtained by adding the heat diffusion contribution value to the energy deposition temperature rise and subtracting the blood flow heat dissipation temperature drop.

5. The method for regulating physical thermal ablation energy based on intelligent feedback according to claim 1, characterized in that, The calculation of the rate of change of impedance includes the following steps: Calculate the difference between the current impedance and the maximum impedance, and then, in conjunction with the temperature and temperature gradient, calculate the rate of change of impedance over time.

6. The method for regulating physical thermal ablation energy based on intelligent feedback according to claim 1, characterized in that, The morphology prediction network model includes: Input layer: Input data includes voxelized data cubes, temperature change rate, and impedance change rate; Spatiotemporal feature extraction layer: Extracts local spatial and temporal correlation features from input features through 3D convolution operations; The generated correlation features are classified into impedance-related feature groups and temperature-related feature groups to obtain impedance feature groups and temperature feature groups. The obtained impedance feature groups and temperature feature groups are then converted into data formats suitable for LSTM network processing. The converted impedance characteristic set and temperature characteristic set are input into two independent channels of the dual-channel LSTM, including the impedance channel and the temperature channel; The impedance channel is used to receive data from the impedance feature set and to learn the rate of change of impedance-related features over time. The temperature channel is used to receive data of temperature feature groups and to learn the rate of change of temperature-related features over time. Then, based on the input gate, forget gate, and output gate of LSTM, the characteristic signals of the impedance channel and the temperature channel are received simultaneously. The dynamic relationship between the cell state and the hidden state of the two channels is automatically correlated, and then the coupling feature is output. Then, a transposed convolution operation is used to restore the coupled features to three-dimensional spatial resolution, generating an ablation probability map.

7. The method for regulating physical thermal ablation energy based on intelligent feedback according to claim 1, characterized in that, The three-layer parallel control strategy includes PID power control based on morphological deviation, fuzzy rule control based on temperature gradient, and emergency circuit breaker control based on impedance change. When the absolute value of the ablation morphology deviation exceeds a set threshold, the PID power control of the morphology deviation is activated. The temperature gradient-based fuzzy rule control is activated when the maximum temperature gradient within the tissue exceeds a safety threshold. The emergency circuit breaker control based on impedance mutation executes multi-level responses based on predefined rules to achieve safety protection.

8. The method for regulating physical thermal ablation energy based on intelligent feedback according to claim 7, characterized in that, The PID power control includes the following steps: The power adjustment contribution value of the proportional element is calculated based on the ablation morphology deviation and proportional gain. Based on historical time-series data of ablation morphology deviation, the power adjustment contribution value of the integral stage is calculated. Based on the rate of change of ablation morphology deviation and the differential gain, the power adjustment contribution value of the differential element is calculated. The power adjustment amount is obtained by adding the power adjustment contributions of the proportional, integral, and derivative components. The calculation cycle of the PID power control is synchronized with the calculation cycle of the ultrasonic frame rate and the morphological prediction network model.

9. The method for regulating physical thermal ablation energy based on intelligent feedback according to claim 7, characterized in that, The temperature gradient-based fuzzy rule control includes the following steps: The membership function is used to determine the category of temperature gradient and impedance change rate; then, for each combination of temperature gradient category and impedance change rate category, the corresponding pulse parameters are retrieved from the rule base. Then, the weight of each combination parameter is calculated based on the product of the membership degree of the temperature gradient and the rate of change of impedance; then, the pulse parameters of all combinations are weighted and averaged to obtain the pulse parameters of the fuzzy rule control output, and the thermal ablation device is controlled based on the output pulse parameters.

10. The method for regulating physical thermal ablation energy based on intelligent feedback according to claim 7, characterized in that, The emergency circuit breaker control for impedance surges includes a three-level response strategy; Level 1 response: When the impedance change rate exceeds a threshold, reduce the current output power to a predetermined percentage of the current power. Level 2 response: When the real-time impedance value exceeds a predetermined multiple of the baseline impedance before ablation, immediately switch to safe pulse mode; Level 3 Response: If the ablation morphology deviation exceeds the maximum permissible deviation and the maximum temperature gradient exceeds the predetermined threshold, an emergency stop signal is triggered, cutting off the energy output of the thermal ablation device and triggering an alarm.

Images