Fitting method and device for ablation field of electrode needle
The ablation field in the composite pulse electric field is fitted through the network structure model, and the problem of low precision in ablation field calculation in the prior art is solved, thereby achieving high-precision prediction of cell necrosis region, providing an effective reference for clinical treatment.
Patent Information
- Application Number
- CN202310967427.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-02
- Publication Date
- 2025-02-18
AI Technical Summary
When existing composite pulse electric field treats tumors, the lack of accurate ablation field calculations leads to low precision in surgical design, which may damage normal tissue or underablation of target areas, affecting the treatment effect.
By obtaining the position information of the target electrode needle insertion and surrounding tissue characteristics, fit the ablation field using a network structural model (such as the U-Net network architecture) to predict the size and shape of the cell necrotic region.
High-precision and rapid calculation ablation field fitting is achieved, providing more accurate prediction of cell necrosis region, and providing an effective reference for clinical treatment.
Smart Images
Figure CN119454225A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of composite pulse electric field ablation of tumors, and in particular to a fitting method and device for an ablation field of an electrode needle. Background Art
[0002] As a type of local ablation, the composite pulsed electric field (Nano Knife) can also be called irreversible electroporation. The working principle of the Nano Knife is that when the electric field strength increases to the millivolt level, the nanoscale micropores in the cell membrane will not automatically close and disappear after the electric field is removed, and the cell membrane will be irreversibly damaged, and the cells will die without the use of chemotherapy drugs. At the same time, the Nano Knife is a non-temperature-dependent local tumor ablation technology that will not cause protein denaturation, nor will it cause thermal damage to heat-sensitive tissues such as adjacent blood vessels, bile ducts, diaphragms, and gastrointestinal tracts. It can achieve ablation treatment of special parts that cannot be treated by thermal ablation.
[0003] At present, there are clinical trials to carry out research related to nanoknife, mainly focusing on clinical treatment and experimental research on liver cancer, pancreatic cancer, kidney tumors, bone tumors, intestinal tumors, etc. In the process of composite pulsed electric field treatment of tumors, how to accurately calculate the ablation field in real time is one of the prerequisites for reverse design of surgery.
[0004] Although compound pulsed electric field therapy has a good ablation effect on tumors, it is rarely used clinically at present. One of the important reasons is that doctors lack accurate calculation of the ablation field before surgery. When using low-precision ablation field estimation methods to design surgeries, doctors can only rely on experience to insert ablation needles into the patient's body, which may damage other normal tissues and may also cause under-ablation of the tumor target area, resulting in unsatisfactory actual ablation range.
[0005] Existing ablation field fitting methods are mainly based on traditional mathematical models, such as the ellipsoidal cylinder model and the Gaussian function model. However, these models simply describe the morphology of the ablation field and cannot accurately predict the distribution of the ablation field under different electrical parameters. In the actual ablation process, the ablation field model may change due to factors such as tissue structure, density, blood flow, and organ movement. Therefore, an accurate method is needed to fit the ablation field distribution in the composite pulse electric field. Summary of the invention
[0006] The purpose of the embodiment of the present invention is to provide a method and device for fitting the ablation field of an electrode needle. The fitting method can predict the ablation field morphology under the action of the target electrode needle based on factors such as the insertion position information of the target electrode needle and the tissue characteristics around the insertion position, thereby predicting the size and shape of the cell necrosis area under the action of the composite pulse electric field, providing a reference for clinical treatment. The method has the advantages of high accuracy and fast calculation speed.
[0007] In order to achieve the above-mentioned purpose, an embodiment of the present invention provides a fitting method for the ablation field of an electrode needle, the fitting method comprising: obtaining input sample data of a target electrode needle, wherein the input sample data is a plurality of three-channel images within a target area including a spatial area where the target electrode needle is located, and each of the plurality of three-channel images is formed by splicing together the following three cross-sectional images of the same size: a CT image, a cross-sectional image of a coarse model of the ablation field of the target electrode needle in the cross-section where the CT image is located, and a needle track mask image in the cross-section where the CT image is located; and inputting the input sample data of the target electrode needle into a network structure model to obtain the fitting ablation field of the target electrode needle.
[0008] Optionally, when there are two target electrode needles, the rough model of the ablation field of the target electrode needles is an elliptical cylinder with the positions of the two electrode needles as foci, and the height of the elliptical cylinder is the exposed length of the two electrode needles.
[0009] Optionally, the length of the minor axis of the elliptical cross section of the elliptical cylinder is equal to the distance between the two electrode needles.
[0010] Optionally, the needle channel mask image is constructed using the needle channel position information of the target electrode needle, wherein the grayscale of the background of the needle channel mask image is a first grayscale, the grayscale of the exposed electrode portion on the needle channel connected to the positive pulse pin is a second grayscale, and the grayscale of the exposed electrode portion on the needle channel connected to the negative pulse pin is a third grayscale.
[0011] Optionally, the network structure model is trained in the following manner: obtaining input sample data and measured ablation field data of a sample electrode needle, wherein the input sample data is a plurality of three-channel images within a set area of the insertion position of the sample electrode needle, and each of the plurality of three-channel images is formed by stitching together the following three cross-sectional images of the same size: a CT image, a cross-sectional image of a rough model of the ablation field of the sample electrode needle in the cross-section where the CT image is located, and a needle track mask image in the cross-section where the CT image is located; and using the input sample data and the measured ablation field data to train the network structure model so that the loss function between the measured ablation field data and the output sample data is minimized or less than a set value, wherein the output sample data is obtained by processing the input sample data through the network structure model.
[0012] Optionally, the network structure model is a U-Net network architecture, which includes multiple downsampling layers and multiple upsampling layers, and the multiple downsampling layers have the same number of layers as the multiple upsampling layers, wherein the last layer of the multiple downsampling layers is connected to the first layer of the multiple upsampling layers using a channel self-attention module, and the downsampling layers except the last layer of the multiple downsampling layers are connected to the upsampling layers except the first layer of the multiple upsampling layers through a jump connection module.
[0013] Optionally, the down-sampling layers except the last layer in the multiple down-sampling layers are connected to the up-sampling layers except the first layer in the multiple up-sampling layers by using a channel attention module.
[0014] Optionally, the measured ablation field data is acquired based on the conductivity distribution data in a set area, wherein, when the sample electrode needles are two electrode needles, the set area is a cube, the length and width of the cube are first set multiples of the distance between the two electrode needles, and the height of the cube is a second set multiple of the exposed length of the two electrode needles.
[0015] Optionally, the fitting method further includes: calculating a loss function between the measured ablation field data and the output sample data in the set area, wherein the loss function is expressed by the following formula:
[0016]
[0017] Among them, Loss(w) is the loss function, w is the network parameter to be optimized, N is the number of pixels in the output image, n is the current coordinate, P(x n ) is the measured ablation field data value at the current coordinate, is the output sample data value of the network structure model at the current coordinate.
[0018] Optionally, Dropout regularization is used to correct the network structure model.
[0019] On the other hand, the present invention provides a fitting device for the ablation field of an electrode needle, the fitting device comprising: a sample data acquisition module, used to acquire input sample data and output sample data, wherein the input sample data is a plurality of three-channel images within a target area including the spatial area where the target electrode needle is located, and each of the plurality of three-channel images is formed by splicing the following three cross-sectional images of the same size: a CT image, a cross-sectional image of a rough model of the ablation field of the target electrode needle in the cross-section where the CT image is located, and a needle track mask image in the cross-section where the CT image is located; and an ablation field fitting module, used to fit the ablation field of the target electrode needle according to the constructed network structure model.
[0020] Through the fitting method of the above technical solution, the present invention can predict the ablation field morphology and electric field strength distribution under the action of the target electrode needle based on factors such as the target electrode needle insertion position information and the tissue characteristics around the insertion position, thereby predicting the size and shape of the cell necrosis area under the action of the composite pulse electric field, and providing a reference for clinical treatment. Compared with traditional empirical or semi-empirical prediction methods, the present invention can more accurately predict the electric field strength distribution of the target electrode needle in the tissue based on the network structure model, while taking into account the physical characteristics of the tissue, so as to better simulate the conduction and conversion process of the electric field in the tissue. This method has the advantages of high prediction accuracy and fast calculation speed.
[0021] At the same time, the technical solution provided by the present invention can also make personalized treatment more effective. The present invention can select appropriate electrode needles and insertion parameters according to different types of tumors and tissue structures, and perform personalized prediction and adjustment, thereby improving the treatment effect. Therefore, compared with traditional fixed parameter treatment methods, the present invention can better meet the needs of individualized treatment and improve the treatment effect and survival rate.
[0022] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:
[0024] Figure 1 is a schematic flow chart of a method for fitting an ablation field of an electrode needle provided according to an embodiment of the present invention;
[0025] Figure 2 2. It is a schematic diagram of a U-Net network architecture of a network structure model provided according to an embodiment of the present invention;
[0026] Figure 3a is a schematic diagram of the principle of a channel attention module provided according to an embodiment of the present invention;
[0027] Figure 3b is a schematic diagram of the principle of a channel self-attention module provided according to an embodiment of the present invention;
[0028] Figure 4 is a schematic flow chart of a training method for a network structure model provided according to an embodiment of the present invention;
[0029] Figure 5 It is a schematic diagram of a fitting device for an ablation field of an electrode needle provided according to an embodiment of the present invention. DETAILED DESCRIPTION
[0030] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.
[0031] The present invention provides a method for fitting the ablation field of an electrode needle, such as Figure 1 As shown, the fitting method may include steps S110 - S120 .
[0032] Step S110, obtaining input sample data of the target electrode needle.
[0033] The input sample data may be a plurality of three-channel images in a target region including a spatial region where a target electrode needle is located. Each of the plurality of three-channel images may be formed by splicing the following three cross-sectional images of the same size: a CT image, a cross-sectional image of a coarse model of an ablation field of a target electrode needle in a cross-sectional plane where the CT image is located, and a needle track mask image in a cross-sectional plane where the CT image is located.
[0034] In one embodiment, when the target electrode needles are two electrode needles, the exposed parts of the needle tips of the two electrode needles are the two electrodes for discharge, and the remaining parts can be provided with an insulator to block. When the two electrode needles are inserted into the human body in parallel, the initial electric field distribution (i.e., the rough model of the ablation field) can be set as an elliptical cylinder with the positions of the two electrode needles as the focus. Among them, the height of the elliptical cylinder can be the exposed length of the two electrode needles, and the electric field at the end points of the exposed electrode needles is perpendicular to the plane of the electrode without depressions or protrusions. Preferably, the short axis length of the elliptical cross section of the elliptical cylinder can be equal to the set multiple of the distance between the two electrode needles. The set multiple can be between 1-1.5, preferably between 1-1.2. For example, it can take a value of 1, in which case the short axis length of the elliptical cross section of the elliptical cylinder is equal to the distance between the two electrode needles.
[0035] In one embodiment, a needle track mask image can be constructed using the needle track position information of the target electrode needle, wherein the grayscale of the background of the needle track mask image can be a first grayscale, the grayscale of the exposed electrode portion on the needle track connected to the positive pulse pin can be a second grayscale, and the grayscale of the exposed electrode portion on the needle track connected to the negative pulse pin can be a third grayscale. The first grayscale is different from the second grayscale and the first grayscale is different from the third grayscale, in order to distinguish the background information and the needle track information in the needle track mask image. The second grayscale and the third grayscale can be the same or different. When the second grayscale and the third grayscale are different, the positions of the positive and negative electrode needles can be effectively distinguished. For example, the first grayscale can take a value of 0, the second grayscale can take a value of 1, and the third grayscale can take a value of 2.
[0036] Step S120 , inputting the input sample data of the target electrode needle into the network structure model to obtain the fitted ablation field of the target electrode needle.
[0037] In one embodiment, the network structure model may be a U-Net network architecture, for example, a 3D-UNet network architecture. The U-Net network architecture may include multiple downsampling layers, multiple upsampling layers, and skip connections, that is, the image-to-image conversion task may be achieved through a fully convolutional network in the form of downsampling-upsampling-skip connections. Among them, the downsampling layer may be understood as an encoder, which can gradually reduce the spatial dimension of the pooling layer, and the upsampling layer may be understood as a decoder, which can gradually repair the details and spatial dimensions of the object. There is usually a shortcut connection (i.e., a skip connection) between the encoder and the decoder, which can help the decoder better repair the details of the target.
[0038] Specifically, Figure 2 As shown in the figure, the downsampling layer in the first half of U-Net is a contraction path, which is used for feature extraction. This contraction path follows the typical architecture of the convolutional network, and the number of feature channels can be doubled after each downsampling. The upsampling layer in the second half is an expansion path. Each step in this expansion path first uses deconvolution. Each time deconvolution is used, the number of feature channels is halved and the size of the feature map is doubled. After deconvolution, the result of deconvolution is concatenated with the feature map of the corresponding step in the contraction path. In addition, the number of downsampling layers is the same as that of upsampling layers, and the pixels of each downsampling layer are the same as those of the upsampling layer. And each downsampling layer is connected to the upsampling layer of the corresponding pixel through a jump connection module. Therefore, the output image of the network structure model is an accurate ablation field prediction map with the same size as the input image.
[0039] Each cube block represents a multi-channel feature map, and arrows of different sizes and orientations represent different operations. The short right arrow represents a 3x3x3 three-dimensional convolution operation. For example, the size of the convolution kernel of the network structure model can be fixed to 3, the stride (Stride) is 1, and the activation function can use the ReLu function. The downward arrow represents a 2x2x2 maximum pooling operation, and the upward arrow represents a 2x2x2 deconvolution operation. For example, the transposed convolution kernel size of the upsampling layer can be set to 2 and the stride (Stride) is 1.
[0040] By setting the network structure model to a U-Net network architecture, the present invention can simultaneously have a contraction path for capturing contextual information and a symmetric expansion path for allowing precise positioning. Compared with FCN (fully convolutional network), the upsampling process of U-Net still has a large number of channels, which enables the network to propagate contextual information to a higher level of resolution.
[0041] It is worth noting that the network structure model of the present invention adds two channel attention mechanisms based on the above U-Net network architecture, so as to assist the network structure model to focus on the important information in the input image that is worth judging and reasoning, while ignoring irrelevant information. Among them, the last layer of multiple downsampling layers can be connected to the first layer of multiple upsampling layers using a channel self-attention module ( Figure 2 In the figure, the downsampling layers except the last layer can be connected to the upsampling layers except the first layer of the multiple upsampling layers through the skip connection module. In addition, the downsampling layers except the last layer of the multiple downsampling layers can also be connected to the upsampling layers except the first layer of the multiple upsampling layers using the channel attention module (as shown in the figure circle B).
[0042] Specifically, the calculation principles of the two channel attention modules can be as follows Figure 3a and 3b As shown, Figure 3a It means that the channel attention module can be used to assign different weights to each channel through global average pooling and Sigmoid function. Figure 3b It means that the channel self-attention module can strengthen the connection between channels through the Sigmoid function.
[0043] In one embodiment, the network structure model can be obtained by data training. Figure 4 As shown, the network structure model can be obtained through steps S121-S122.
[0044] Step S121, obtaining input sample data and measured ablation field data of a sample electrode needle.
[0045] Among them, the input sample data can be multiple three-channel images within the set area of the sample electrode needle insertion position, and each of the multiple three-channel images can be spliced by the following three cross-sectional images of the same size: a CT image, a cross-sectional image of the rough model of the ablation field of the sample electrode needle in the cross-section where the CT image is located, and a needle track mask image in the cross-section where the CT image is located.
[0046] In one embodiment, the measured ablation field data can be obtained based on the conductivity distribution data in the set area. Specifically, since the conductivity of the tissue in the breakdown area is different from the conductivity of the non-breakdown area (it can also be said that the conductivity of the tissue before and after the breakdown is different), the conductivity distribution data in the set area can be measured when the ablation is completed under various parameter conditions (such as: tissue type, electrode radius, inter-electrode distance, etc.), and then the electric field distribution can be reconstructed based on the conductivity in the set area of the spatial area where the sample electrode needle in the tissue is located.
[0047] For example, in the case where the sample electrode needles are two electrode needles, the set area can be a cube, the length and width of the cube can be a first set multiple of the distance between the two electrode needles, and the height of the cube can be a second set multiple of the exposed length of the two electrode needles. The first set multiple can be between 1-1.2. For example, the value can be 1, in which case the length and width of the cube are equal to the distance between the two electrode needles. The second set multiple can be between 1.2-1.5. For example, the value can be 1.5, in which case the height of the cube is 1.5 times the exposed length of the two electrode needles. In addition, the center of the cube can be the midpoint of the line connecting the center points of the exposed parts of the two electrode needles. When measuring the conductivity of the target area of the cube, the target area of the cube needs to be divided into sufficiently small cubes to establish an accurate perforation area shape to obtain the electric field distribution between the two needles during ablation.
[0048] Step S122, use the input sample data and the measured ablation field data to train the network structure model, that is, use the obtained electric field distribution under given conditions as training and testing samples for model training, so that the loss function between the measured ablation field data and the output sample data is minimized or less than the set value.
[0049] Specifically, the CT image of the sample electrode needle obtained in step S121, the cross-sectional image of the rough model of the ablation field of the sample electrode needle in the cross section where the CT image is located, and the needle track mask image in the cross section where the CT image is located (for example, setting the background grayscale to 0, the main needle to 1, and the auxiliary needle to 2) can be used as training samples. The input image can be a three-channel composite image spliced by the above three cross-sectional images of the same size. The output sample data is an image obtained by processing the input sample data through the network structure model. Then, the loss function is continuously optimized through the network structure model so that the loss function between the measured ablation field data and the output sample data is minimized or less than the set value. For example, the Adam optimizer can be used to optimize the network structure model so that the simulated electric field distribution under given conditions is close to the real electric field distribution. It is worth noting that overfitting should be prevented during the training process. For example, Dropout regularization can be used to correct the network structure model.
[0050] In one embodiment, the fitting method may further include: calculating a loss function between the measured ablation field data and the output sample data in a set area. For example, the loss function may be expressed as follows:
[0051]
[0052] Among them, Loss(w) is the loss function, w is the network parameter to be optimized, N is the number of pixels in the output image, n is the current coordinate, P(x n ) is the measured ablation field data value at the current coordinate, It is the output sample data value of the network structure model at the current coordinate.
[0053] Through the above steps, the present invention can predict the ablation field morphology and electric field strength distribution under the action of the target electrode needle based on factors such as the target electrode needle insertion position information and the tissue characteristics around the insertion position, thereby predicting the size and shape of the cell necrosis area under the action of the composite pulse electric field, and providing a reference for clinical treatment. Compared with traditional empirical or semi-empirical prediction methods, the fitting method of the present invention can more accurately predict the electric field strength distribution of the target electrode needle in the tissue based on the network structure model, while taking into account the physical properties of the tissue, so as to better simulate the conduction and conversion process of the electric field in the tissue. This method has the advantages of high prediction accuracy and fast calculation speed.
[0054] At the same time, the technical solution provided by the present invention can also make personalized treatment more effective. The present invention can select appropriate electrode needles and insertion parameters according to different types of tumors and tissue structures, and perform personalized prediction and adjustment, thereby improving the treatment effect. Therefore, compared with traditional fixed parameter treatment methods, the present invention can better meet the needs of individualized treatment and improve the treatment effect and survival rate.
[0055] On the other hand, the present invention also provides a fitting device for the ablation field of the electrode needle, such as Figure 5 As shown, the fitting device may include a sample data acquisition module 210 and an ablation field fitting module 220 .
[0056] The sample data acquisition module 210 can be used to acquire input sample data and output sample data. The input sample data is a plurality of three-channel images in a target area including the spatial area where the target electrode needle is located, and each of the plurality of three-channel images is composed of the following three cross-sectional images of the same size: a CT image, a cross-sectional image of the ablation field rough model of the target electrode needle in the cross section where the CT image is located, and a needle track mask image in the cross section where the CT image is located.
[0057] The ablation field fitting module 220 can be used to fit the ablation field of the target electrode needle according to the constructed network structure model.
[0058] Through the fitting device, the present invention can predict the ablation field morphology and electric field strength distribution under the action of the target electrode needle based on factors such as the target electrode needle insertion position information and the tissue characteristics around the insertion position, thereby predicting the size and shape of the cell necrosis area under the action of the composite pulse electric field, and providing a reference for clinical treatment. Compared with traditional empirical or semi-empirical prediction methods, the fitting device of the present invention can more accurately predict the electric field strength distribution of the target electrode needle in the tissue based on the network structure model, while taking into account the physical characteristics of the tissue, so as to better simulate the conduction and conversion process of the electric field in the tissue. This method has the advantages of high prediction accuracy and fast calculation speed.
[0059] At the same time, the technical solution provided by the present invention can also make personalized treatment more effective. The present invention can select appropriate electrode needles and insertion parameters according to different types of tumors and tissue structures, and perform personalized prediction and adjustment, thereby improving the treatment effect. Therefore, compared with traditional fixed parameter treatment methods, the present invention can better meet the needs of individualized treatment and improve the treatment effect and survival rate.
[0060] On the other hand, the present invention also provides a layout method of electrode needles, which may include: laying out corresponding electrode needles according to the above-mentioned fitting ablation field. That is, corresponding surgical design can be performed according to the fitting ablation field, thereby providing doctors with a more accurate fitting ablation field as a surgical reference, and performing personalized prediction and adjustment, thereby improving the treatment effect.
[0061] On the other hand, the present invention further provides a layout system for electrode needles, which can be used to layout corresponding electrode needles according to the fitting ablation field of the fitting device.
[0062] Through the above-mentioned electrode needle layout method or layout system, the present invention can select appropriate electrode needles and insertion parameters according to different types of tumors and tissue structures. Therefore, compared with traditional fixed parameter treatment methods, the present invention can better meet individualized treatment needs and improve treatment effects and survival rates.
[0063] It should also be noted that “fitting” in each embodiment of the present invention can be replaced by “prediction”.
[0064] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0065] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.
Claims
1. A method for fitting an ablation field of an electrode needle, characterized in that: The fitting method comprises: Acquire input sample data of a target electrode needle, wherein the input sample data is a plurality of three-channel images in a target area including a spatial area where the target electrode needle is located, and each of the plurality of three-channel images is formed by stitching together the following three cross-sectional images of the same size: a CT image, a cross-sectional image of a coarse model of an ablation field of the target electrode needle in the cross section where the CT image is located, and a needle track mask image in the cross section where the CT image is located; and The input sample data of the target electrode needle is input into the network structure model to obtain the fitting ablation field of the target electrode needle.
2. The fitting method according to claim 1, characterized in that: In the case that the target electrode needles are two electrode needles, the rough model of the ablation field of the target electrode needles is an elliptical cylinder with the positions of the two electrode needles as foci, and the height of the elliptical cylinder is the exposed length of the two electrode needles.
3. The fitting method according to claim 2, characterized in that: The length of the minor axis of the elliptical cross section of the elliptical cylinder is equal to the distance between the two electrode needles.
4. The fitting method according to claim 1, characterized in that: The needle track mask image is constructed using the needle track position information of the target electrode needle, wherein the grayscale of the background of the needle track mask image is the first grayscale, the grayscale of the exposed electrode part on the needle track connected to the positive pulse pin is the second grayscale, and the grayscale of the exposed electrode part on the needle track connected to the negative pulse pin is the third grayscale.
5. The fitting method according to claim 1, characterized in that: The network structure model is trained in the following way: Acquire input sample data and measured ablation field data of a sample electrode needle, wherein the input sample data is a plurality of three-channel images in a set area of the sample electrode needle insertion position, and each of the plurality of three-channel images is formed by stitching together the following three cross-sectional images of the same size: a CT image, a cross-sectional image of a coarse model of the ablation field of the sample electrode needle in the cross section where the CT image is located, and a needle track mask image in the cross section where the CT image is located; and The network structure model is trained using the input sample data and the measured ablation field data so that the loss function between the measured ablation field data and the output sample data is minimized or less than a set value, wherein the output sample data is obtained by processing the input sample data through the network structure model.
6. The fitting method according to claim 5, characterized in that: The network structure model is a U-Net network architecture, and the U-Net network architecture includes multiple downsampling layers and multiple upsampling layers, and the number of layers of the multiple downsampling layers is the same as that of the multiple upsampling layers. Among them, the last layer of the multiple downsampling layers is connected to the first layer of the multiple upsampling layers using a channel self-attention module, and the downsampling layers except the last layer of the multiple downsampling layers are connected to the upsampling layers except the first layer of the multiple upsampling layers through a jump connection module.
7. The fitting method according to claim 6, characterized in that: The down-sampling layers except the last layer of the multiple down-sampling layers are also connected to the up-sampling layers except the first layer of the multiple up-sampling layers using a channel attention module.
8. The fitting method according to claim 5, characterized in that: The measured ablation field data is acquired based on the conductivity distribution data in the set area, wherein, when the sample electrode needles are two electrode needles, the set area is a cube, the length and width of the cube are first set multiples of the distance between the two electrode needles, and the height of the cube is second set multiples of the exposed length of the two electrode needles.
9. The fitting method according to claim 8, characterized in that: The fitting method further includes: calculating a loss function between the measured ablation field data and the output sample data in the set area, wherein the loss function is expressed by the following formula: Among them, Loss(w) is the loss function, w is the network parameter to be optimized, N is the number of pixels in the output image, n is the current coordinate, P(x n ) is the measured ablation field data value at the current coordinate, is the output sample data value of the network structure model at the current coordinate.
10. The fitting method according to claim 5, characterized in that: Dropout regularization is used to modify the network structure model.
11. A fitting device for an ablation field of an electrode needle, characterized in that: The fitting device comprises: a sample data acquisition module, used to acquire input sample data and output sample data, wherein the input sample data is a plurality of three-channel images in a target area including a spatial area where the target electrode needle is located, and each of the plurality of three-channel images is formed by splicing the following three cross-sectional images of the same size: a CT image, a cross-sectional image of a coarse model of the ablation field of the target electrode needle in the cross section where the CT image is located, and a needle track mask image in the cross section where the CT image is located; and The ablation field fitting module is used to fit the ablation field of the target electrode needle according to the constructed network structure model.