Radio frequency ablation effect evaluation method and system based on edge prediction and storage medium

Abstract

The invention relates to the technical field of ablation evaluation, in particular to a radiofrequency ablation effect evaluation method and system based on edge prediction and a storage medium, and the method comprises the following steps: obtaining a real-time ablation image in a radiofrequency ablation process; performing projection fusion on the ablation edema zone of the real-time ablation image and the original tumor edge of the original tumor image by using a pre-established edema zone projection model to obtain a real-time ablation effect evaluation image; in the real-time ablation effect evaluation image, pixel coordinates between the ablation edema zone and the original tumor edge are matched, and a real-time ablation effect evaluation value is obtained. According to the method, the pre-established edema zone projection model is utilized to perform projection fusion on the ablation edema zone of the real-time ablation image and the original tumor edge of the original tumor image, the distance between the edema zone and the tumor edge is intuitively and accurately evaluated, and the association relationship between the real-time ablation operation parameter and the real-time ablation effect evaluation value is established, so that the accuracy of the evaluation result is improved. And the ablation evaluation efficiency is improved on the basis of accurate evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of ablation evaluation, and in particular to a method, system and storage medium for evaluating radiofrequency ablation effects based on edge prediction. Background Art

[0002] Radiofrequency ablation technology mainly relies on radiofrequency therapeutic devices with ablation and cutting functions, and the treatment mechanism is mainly thermal effect. When radiofrequency current flows through human tissue, the rapid change of electromagnetic field causes polar water molecules in the tissue to move at high speed, generating heat (i.e. endogenous thermal effect), causing the water inside and outside the cells to evaporate, dry, shrink and fall off, resulting in aseptic necrosis, thereby achieving the purpose of treatment.

[0003] Radiofrequency ablation will result in an edema zone after ablation. Currently, CT or MRI scans are used to obtain tissue images after ablation. It is difficult to intuitively assess the distance between the edema zone and the tumor edge, making it impossible to accurately assess in real time whether the ablation is complete. Summary of the invention

[0004] The purpose of the present invention is to provide a method, system and storage medium for evaluating the effect of radiofrequency ablation based on edge prediction, so as to solve the technical problem in the prior art that the distance between the edema zone and the tumor edge is intuitively evaluated, thereby failing to accurately evaluate in real time whether the ablation is complete.

[0005] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:

[0006] A method for evaluating the effect of radiofrequency ablation based on edge prediction comprises the following steps:

[0007] Acquire real-time ablation images during radiofrequency ablation;

[0008] Using a pre-established edema zone projection model, the ablation edema zone of the real-time ablation image and the original tumor edge of the original tumor image are projected and fused to obtain a real-time ablation effect evaluation image;

[0009] In the real-time ablation effect evaluation image, the pixel coordinates between the ablation edema zone and the edge of the original tumor are matched to obtain a real-time ablation effect evaluation value;

[0010] The real-time ablation operation parameters corresponding to the real-time ablation image are associated with the real-time ablation effect evaluation value to obtain a real-time ablation effect evaluation model that predicts the real-time ablation effect evaluation value based on the real-time ablation operation parameters.

[0011] As a preferred embodiment of the present invention, the method for constructing the edema zone projection model comprises:

[0012] Acquire multiple ablation images with ablated edema zones as sample images, and acquire original tumor images corresponding to the sample images;

[0013] Using a CNN neural network to extract features from a sample image and an original tumor image corresponding to the sample image, to obtain a feature map of the sample image and a feature map of the original tumor image;

[0014] Bidirectionally mapping the feature coordinates in the feature map of each sample image with the feature coordinates in the feature map of the original tumor image one by one, to construct a bidirectional mapping relationship between the features of the sample image and the original tumor image;

[0015] The feature bidirectional mapping relationship is:

[0016] R(XY)=FY(i,j) T FX(i l ,i l );

[0017] R(YX)=FX(i,j) T FX(i l ,i l );

[0018] Where R(XY) is the feature mapping relationship from the sample image to the original tumor image, R(YX) is the feature mapping relationship from the original tumor image to the sample image, FY(i,j) is the feature map vector of the original tumor image at the (i,j) coordinate, and FX(i l ,i l ) is the sample image in (i l ,i l ) coordinates, FX(i,j) is the feature map vector of the sample image at the (i,j) coordinates, FY(i l ,i l ) is the original tumor image at (i l ,i l ) coordinates;

[0019] Obtaining a feature bidirectional matching graph of the sample image and the original tumor image through the feature bidirectional mapping relationship;

[0020] The feature bidirectional matching graph is:

[0021]

[0022] Wherein, g(XY) is the feature matching map from the sample image to the original tumor image, g(YX) is the feature matching map from the original tumor image to the sample image, R(XY) is the feature mapping relationship from the sample image to the original tumor image, R(YX) is the feature mapping relationship from the original tumor image to the sample image, n is the first dimension of the feature mapping relationship, and γ is a constant term;

[0023] The feature bidirectional matching map is used as the input of the projection module, and the projection image of the ablation edema zone in the sample image on the original tumor image and the projection image of the tumor edge in the original tumor image on the sample image are used as the output of the projection module respectively;

[0024] Acquire a first projection parameter by which the projection module projects the ablation edema zone in the sample image onto the original tumor image, and a second projection parameter by which the projection module projects the tumor edge in the original tumor image onto the sample image;

[0025] Iteratively optimizing the projection module and the CNN neural network through projection loss to obtain an optimal first projection parameter and an optimal second projection parameter;

[0026] The projection loss is:

[0027]

[0028] Where L(θ1,θ2) is the projection loss, GXY k (θ1,g(XY)) is the projection image of the ablation edema zone in the kth sample image on the kth original tumor image, GX is the sample image, GYX k (θ2,g(YX)) is the projection image of the tumor edge in the kth original tumor image on the kth sample image, GY is the original tumor image, θ1 is the first projection parameter, θ2 is the second projection parameter, m is the total number of sample images, and All are L2 norm forms;

[0029] The optimal first projection parameter and the optimal second projection parameter are combined to obtain the optimal projection parameter used by the projection regression module to optimally project and fuse the ablation edema zone of the sample image and the original tumor edge of the original tumor image;

[0030] The optimal projection parameters are:

[0031] θ max =M(θ1,(θ2) -1 );

[0032] In the formula, θ max is the optimal projection parameter, θ1 is the first projection parameter, θ2 is the second projection parameter, M(θ1,(θ2) -1 ) is θ1 and (θ2) -1 The arithmetic mean of

[0033] The input of the projection module corresponding to the optimal projection parameter is updated to a feature matching map in the direction from the sample image to the original tumor image, and the output of the projection module corresponding to the optimal projection parameter is updated to a projection image of the ablated edema zone in the sample image on the original tumor image;

[0034] Connecting the input of the updated projection module and the output of the CNN neural network through a feature mapping relationship in the direction from the sample image to the original tumor image to obtain the edema zone projection model;

[0035] The edema zone projection model is:

[0036] GXY=(CNN(GX),CNN(GY))→R(XY)→E(θ max );

[0037] Where GXY is the projection image of the ablated edema zone in the sample image on the original tumor image, GX is the sample image, GY is the original tumor image, CNN(GX) is the feature map of the sample image, CNN(GY) is the feature map of the original tumor image, R(XY) is the feature mapping relationship from the sample image to the original tumor image, E(θ max ) is the projection module corresponding to the optimal projection parameter, θ max is the optimal projection parameter.

[0038] As a preferred solution of the present invention, the method for acquiring the real-time ablation effect evaluation image includes:

[0039] Inputting the real-time ablation image and the original tumor image into the edema band projection model, and the edema band projection model outputs a projection image of the ablated edema band in the real-time ablation image on the original tumor image;

[0040] The projection image on the original tumor image is used as a real-time ablation effect evaluation image.

[0041] As a preferred solution of the present invention, the method for extracting the real-time ablation effect evaluation value includes:

[0042] In the real-time ablation effect evaluation image, edge detection is used to extract the ablation edema zone, and edge detection is used to extract the original tumor edge;

[0043] Matching the pixel coordinates of the ablation edema zone with the original tumor edge to obtain the real-time ablation effect evaluation value;

[0044] The real-time ablation effect evaluation value is:

[0045]

[0046] Where H is the real-time ablation effect evaluation value, P(D1) is the pixel coordinate of the ablated edema zone, and P(D2) is the pixel coordinate of the edge of the original tumor. is the L2 norm.

[0047] As a preferred solution of the present invention, the real-time ablation operation parameters include: a real-time ablation angle and a real-time ablation temperature of the ablation needle.

[0048] As a preferred solution of the present invention, the method for constructing the real-time ablation effect evaluation model includes:

[0049] Using the real-time ablation angle and the real-time ablation temperature of the ablation needle as inputs of the BP neural network, and using the real-time ablation effect evaluation value as output of the BP neural network;

[0050] By using the training of BP neural network, a real-time ablation effect evaluation model is constructed to characterize the mapping relationship between the real-time ablation angle and real-time ablation temperature of the ablation needle and the real-time ablation effect evaluation value;

[0051] The real-time ablation effect evaluation model is:

[0052] H = BP (S, term);

[0053] Where H is the real-time ablation effect evaluation value, BP is the neural network, S is the real-time ablation angle, and tem is the real-time ablation temperature.

[0054] As a preferred solution of the present invention, the projection module consists of a convolutional layer and a fully connected layer.

[0055] As a preferred solution of the present invention, the loss function of the real-time ablation effect evaluation model is the mean square error between the output value of the real-time ablation effect evaluation model and the real-time ablation effect evaluation value obtained according to the real-time ablation effect evaluation image.

[0056] As a preferred embodiment of the present invention, the present invention provides a radiofrequency ablation effect evaluation system based on edge prediction, which is applied to a radiofrequency ablation effect evaluation method based on edge prediction. The system includes:

[0057] An image acquisition unit, used for acquiring real-time ablation images during radiofrequency ablation;

[0058] An image processing unit, configured to project and fuse the ablation edema band of the real-time ablation image and the original tumor edge of the original tumor image using a pre-established edema band projection model, so as to obtain a real-time ablation effect evaluation image;

[0059] An image ablation evaluation unit, used to match the pixel coordinates between the ablation edema zone and the edge of the original tumor in the real-time ablation effect evaluation image to obtain a real-time ablation effect evaluation value;

[0060] The parameter ablation evaluation unit is used to predict the real-time ablation effect evaluation value according to the real-time ablation operation parameters using the real-time ablation effect evaluation model.

[0061] As a preferred embodiment of the present invention, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, a method for evaluating the effect of radiofrequency ablation based on edge prediction is implemented.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] The present invention utilizes a pre-established edema band projection model to perform projection fusion on the ablation edema band of the real-time ablation image and the original tumor edge of the original tumor image, thereby obtaining a real-time ablation effect evaluation image, intuitively and accurately evaluating the distance between the edema band and the tumor edge, and establishing a correlation between real-time ablation operation parameters and real-time ablation effect evaluation values, thereby improving the efficiency of ablation evaluation on the basis of accurate evaluation through the correlation. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0065] Figure 1 A flow chart of a method for evaluating radiofrequency ablation effect based on edge prediction provided by an embodiment of the present invention;

[0066] Figure 2 A block diagram of a radiofrequency ablation effect evaluation system based on edge prediction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] like Figure 1 As shown, the present invention provides a method for evaluating the effect of radiofrequency ablation based on edge prediction, comprising the following steps:

[0069] Acquire real-time ablation images during radiofrequency ablation;

[0070] Using the pre-established edema zone projection model, the ablation edema zone of the real-time ablation image and the original tumor edge of the original tumor image are projected and fused to obtain a real-time ablation effect evaluation image;

[0071] In the real-time ablation effect evaluation image, the pixel coordinates between the ablation edema zone and the edge of the original tumor are matched to obtain a real-time ablation effect evaluation value;

[0072] The real-time ablation operation parameters corresponding to the real-time ablation image are associated with the real-time ablation effect evaluation value to obtain a real-time ablation effect evaluation model that predicts the real-time ablation effect evaluation value based on the real-time ablation operation parameters.

[0073] In order to ensure the accuracy of ablation effect evaluation, the present invention establishes an edema band projection model, and fuses the ablation edema band projection of the real-time ablation image during radiofrequency ablation onto the original tumor image, so that the ablation edema band and the edge of the original tumor are intuitively displayed on the projection image of the real-time ablation image, so that the coordinate distance between the ablation edema band and the edge of the original tumor can be accurately calculated. The coordinate distance shows the overlap of the ablation edema band and the edge of the original tumor, that is, the smaller the coordinate distance, the more the ablation edema band and the edge of the original tumor overlap, and the better the tumor ablation effect. Therefore, the present invention can accurately determine the ablation effect through image projection fusion. In other words, by projecting the real-time ablation image on the original tumor image, the ablation edema band and the edge of the original tumor are placed on the same image layer, and whether the ablation is complete can be accurately and intuitively determined.

[0074] When establishing the edema band projection model, the present invention obtains a feature matching graph from the sample image to the original tumor image and a feature matching graph from the original tumor image to the sample image through the feature mapping relationship in the direction from the sample image to the original tumor image and the feature mapping relationship in the direction from the original tumor image to the sample image to ensure the projection accuracy of the edema band projection model, thereby realizing feature matching in two directions: from the sample image to the original tumor image and from the original tumor image to the sample image, so as to enable bidirectional optimization of the projection module in the edema band projection model.

[0075] When training the projection module in the edema band projection model, the present invention selects to optimize the projection module simultaneously in two directions, from the sample image to the original tumor image and from the original tumor image to the sample image, so that the projection module after training is able to fuse the ablated edema band and the edge of the original tumor to the same image level in the direction from the sample image to the original tumor image. At this time, the ablated edema band is used as the projection subject and the edge of the original tumor is used as the projection object, and the edge of the original tumor and the ablated edema band can be fused to the same image level in the direction from the original tumor image to the sample image. The two-way projection can exchange the projection subject and object, and the projection fusion in the two directions constrains each other to eliminate the projection error in a single direction, so as to obtain a more accurate projection fusion result, or a more accurate image registration result of the edge of the original tumor and the ablated edema band, which ultimately affects the accuracy of the ablation effect evaluation.

[0076] The present invention utilizes projection loss when mutually constraining the projection fusion in two directions, and minimizes the difference between the projection image of the ablation edema band in the sample image on the original tumor image and the sample image, to ensure that the ablation edema band has minimal content loss when projected to the image layer where the tumor edge is located, and maximizes the retention / restoration of the pixel features of the ablation edema band, ultimately ensuring the calculation accuracy of the real-time ablation effect evaluation value.

[0077] Similarly, the difference between the projection image of the original tumor edge in the original tumor image on the sample image and the original tumor image is minimized, ensuring that the original tumor edge has minimal content loss when projected to the image layer where the ablation edema zone is located, maximizing the retention / restoration of the pixel features of the tumor edge, and ultimately ensuring the calculation accuracy of the real-time ablation effect evaluation value.

[0078] In summary, the present invention optimizes the projection module simultaneously in two directions, from the sample image to the original tumor image and from the original tumor image to the sample image, through projection loss constraints, to obtain the optimal module parameter settings (optimal first projection parameters and optimal second projection parameters) of the projection module in the two directions, from the sample image to the original tumor image and from the original tumor image to the sample image, and combines the optimal first projection parameters and the optimal second projection parameters by taking the arithmetic mean to obtain the optimal projection parameters of the projection module, thereby ensuring that the projection module can eliminate the projection error in a single direction.

[0079] Furthermore, the present invention establishes a correlation between real-time ablation operation parameters and real-time ablation effect evaluation values ​​on the basis of intuitive and accurate evaluation of ablation effects through real-time ablation effect evaluation images, and encapsulates the correlation into a real-time ablation effect evaluation model, which can directly predict the real-time ablation effect evaluation value based on the real-time ablation operation parameters. Since the training data of the real-time ablation effect evaluation model comes from the real-time ablation effect evaluation image, the real-time ablation effect evaluation model can inherit the accuracy of the image ablation evaluation, and at the same time, there is no need to perform the image data processing process of the above-mentioned image ablation evaluation, thereby achieving the technical effect of improving the efficiency of ablation evaluation on the basis of accurate evaluation and meeting the timeliness requirements of real-time ablation evaluation.

[0080] Moreover, the present invention can perform real-time evaluation of ablation effects during the ablation process. The ablation evaluation can be performed at any time during the ablation process, and the ablation operation can be adjusted according to the evaluation results. It is not necessary to perform the ablation evaluation only after the ablation is completed. It has strong real-time performance and is more practical.

[0081] In order to ensure the accuracy of ablation effect evaluation, the present invention establishes an edema band projection model, and the ablation edema band projection of the real-time ablation image during radiofrequency ablation is fused onto the original tumor image, so that the ablation edema band and the edge of the original tumor are intuitively displayed on the projection image of the real-time ablation image, thereby accurately calculating the coordinate distance between the ablation edema band and the edge of the original tumor. The coordinate distance shows the overlap between the ablation edema band and the edge of the original tumor, as follows:

[0082] The method for constructing the edema zone projection model includes:

[0083] Acquire multiple ablation images with ablated edema bands as sample images, and acquire original tumor images corresponding to the sample images;

[0084] The CNN neural network is used to extract features from the sample image and the original tumor image corresponding to the sample image to obtain a feature map of the sample image and a feature map of the original tumor image;

[0085] Bidirectionally mapping the feature coordinates in the feature map of each sample image with the feature coordinates in the feature map of the original tumor image one by one, and constructing a bidirectional mapping relationship between the features of the sample image and the original tumor image;

[0086] The feature bidirectional mapping relationship is:

[0087] R(XY)=FY(i,j) T FX(i l ,i l );

[0088] R(YX)=FX(i,j) T FX(i l ,il );

[0089] Where R(XY) is the feature mapping relationship from the sample image to the original tumor image, R(YX) is the feature mapping relationship from the original tumor image to the sample image, FY(i,j) is the feature map vector of the original tumor image at the (i,j) coordinate, and FX(i l ,i l ) is the sample image in (i l ,i l ) coordinates, FX(i,j) is the feature map vector of the sample image at the (i,j) coordinates, FY(i l ,i l ) is the original tumor image at (i l ,i l ) coordinates;

[0090] The feature bidirectional matching graph of the sample image and the original tumor image is obtained through the feature bidirectional mapping relationship;

[0091] The feature bidirectional matching graph is:

[0092]

[0093] Wherein, g(XY) is the feature matching map from the sample image to the original tumor image, g(YX) is the feature matching map from the original tumor image to the sample image, R(XY) is the feature mapping relationship from the sample image to the original tumor image, R(YX) is the feature mapping relationship from the original tumor image to the sample image, n is the first dimension of the feature mapping relationship, γ is a constant term, γ = 1 / 10000;

[0094] When establishing the edema band projection model, the present invention obtains a feature matching graph from the sample image to the original tumor image and a feature matching graph from the original tumor image to the sample image through the feature mapping relationship in the direction from the sample image to the original tumor image and the feature mapping relationship in the direction from the original tumor image to the sample image to ensure the projection accuracy of the edema band projection model, thereby realizing feature matching in two directions: from the sample image to the original tumor image and from the original tumor image to the sample image, so as to enable bidirectional optimization of the projection module in the edema band projection model.

[0095] The feature bidirectional matching map is used as the input of the projection module, and the projection image of the ablation edema zone in the sample image on the original tumor image and the projection image of the tumor edge in the original tumor image on the sample image are used as the output of the projection module respectively;

[0096] Acquire a first projection parameter by which the projection module projects the ablation edema zone in the sample image onto the original tumor image, and a second projection parameter by which the projection module projects the tumor edge in the original tumor image onto the sample image;

[0097] The projection module and CNN neural network are iteratively optimized through projection loss to obtain the optimal first projection parameter and the optimal second projection parameter;

[0098] When training the projection module in the edema band projection model, the present invention selects to optimize the projection module simultaneously in two directions, from the sample image to the original tumor image and from the original tumor image to the sample image, so that the projection module after training is able to fuse the ablated edema band and the edge of the original tumor to the same image level in the direction from the sample image to the original tumor image. At this time, the ablated edema band is used as the projection subject and the edge of the original tumor is used as the projection object, and the edge of the original tumor and the ablated edema band can be fused to the same image level in the direction from the original tumor image to the sample image. The two-way projection can exchange the projection subject and object, and the projection fusion in the two directions constrains each other to eliminate the projection error in a single direction, so as to obtain a more accurate projection fusion result, or a more accurate image registration result of the edge of the original tumor and the ablated edema band, which ultimately affects the accuracy of the ablation effect evaluation.

[0099] The projection loss is:

[0100]

[0101] Where L(θ1,θ2) is the projection loss, GXY k (θ1,g(XY)) is the projection image of the ablation edema zone in the kth sample image on the kth original tumor image, GX is the sample image, GYX k (θ2,g(YX)) is the projection image of the tumor edge in the kth original tumor image on the kth sample image, GY is the original tumor image, θ1 is the first projection parameter, θ2 is the second projection parameter, m is the total number of sample images, ||GXY k (θ1,g(XY))-GX|| 2 2 and || GYX k (θ2,g(YX))-GY|| 2 2 are all L2 norm forms;

[0102] The present invention utilizes projection loss when mutually constraining the projection fusion in two directions, and minimizes the difference between the projection image of the ablation edema band in the sample image on the original tumor image and the sample image, to ensure that the ablation edema band has minimal content loss when projected to the image layer where the tumor edge is located, and maximizes the retention / restoration of the pixel features of the ablation edema band, ultimately ensuring the calculation accuracy of the real-time ablation effect evaluation value.

[0103] Similarly, the difference between the projection image of the original tumor edge in the original tumor image on the sample image and the original tumor image is minimized, ensuring that the original tumor edge has minimal content loss when projected to the image layer where the ablation edema zone is located, maximizing the retention / restoration of the pixel features of the tumor edge, and ultimately ensuring the calculation accuracy of the real-time ablation effect evaluation value.

[0104] The optimal first projection parameter and the optimal second projection parameter are combined to obtain the optimal projection parameter used by the projection regression module to optimally project and fuse the ablation edema zone of the sample image and the original tumor edge of the original tumor image;

[0105] The optimal projection parameters are:

[0106] θ max =M(θ1,(θ2) -1 );

[0107] In the formula, θ max is the optimal projection parameter, θ1 is the first projection parameter, θ2 is the second projection parameter, M(θ1,(θ2) -1 ) is θ1 and (θ2) -1 The arithmetic mean of

[0108] The present invention optimizes the projection module simultaneously in two directions, from the sample image to the original tumor image and from the original tumor image to the sample image, through projection loss constraints, to obtain the optimal module parameter settings (optimal first projection parameters and optimal second projection parameters) of the projection module in the two directions, from the sample image to the original tumor image and from the original tumor image to the sample image, and combines the optimal first projection parameters and the optimal second projection parameters by taking the arithmetic mean value to obtain the optimal projection parameters of the projection module, thereby ensuring that the projection module can eliminate the projection error in a single direction.

[0109] The input of the projection module corresponding to the optimal projection parameter is updated to a feature matching map in the direction from the sample image to the original tumor image, and the output of the projection module corresponding to the optimal projection parameter is updated to a projection image of the ablated edema zone in the sample image on the original tumor image;

[0110] The input of the updated projection module and the output of the CNN neural network are connected through a feature mapping relationship from the sample image to the original tumor image to obtain an edema zone projection model;

[0111] The projection model of the edema zone is:

[0112] GXY=(CNN(GX),CNN(GY))→R(XY)→E(θ max );

[0113] Where GXY is the projection image of the ablation edema zone in the sample image on the original tumor image, GX is the sample image, GY is the original tumor image, CNN(GX) is the feature map of the sample image, CNN(GY) is the feature map of the original tumor image, R(XY) is the feature mapping relationship from the sample image to the original tumor image, E(θ max ) is the projection module corresponding to the optimal projection parameter, θ max is the optimal projection parameter.

[0114] The method for acquiring the real-time ablation effect evaluation image includes:

[0115] The real-time ablation image and the original tumor image are input into the edema band projection model, and the edema band projection model outputs a projection image of the ablated edema band in the real-time ablation image on the original tumor image;

[0116] The projection image on the original tumor image is used as a real-time ablation effect evaluation image.

[0117] The method for extracting the real-time ablation effect evaluation value includes:

[0118] In the real-time ablation effect evaluation image, edge detection is used to extract the ablation edema zone, and edge detection is used to extract the original tumor edge;

[0119] Match the pixel coordinates of the ablation edema zone with the original tumor edge to obtain a real-time ablation effect evaluation value;

[0120] The real-time ablation effect evaluation value is:

[0121]

[0122] Where H is the real-time ablation effect evaluation value, P(D1) is the pixel coordinate of the ablated edema zone, and P(D2) is the pixel coordinate of the edge of the original tumor. is the L2 norm.

[0123] The real-time ablation operation parameters include: the real-time ablation angle and the real-time ablation temperature of the ablation needle.

[0124] The present invention also establishes a correlation between real-time ablation operation parameters and real-time ablation effect evaluation values ​​based on an intuitive and accurate evaluation of ablation effects through real-time ablation effect evaluation images, and encapsulates the correlation into a real-time ablation effect evaluation model, which can directly predict the real-time ablation effect evaluation value based on the real-time ablation operation parameters. Since the training data of the real-time ablation effect evaluation model comes from the real-time ablation effect evaluation image, the real-time ablation effect evaluation model can inherit the accuracy of the image ablation evaluation, and at the same time, there is no need to perform the image data processing process of the above-mentioned image ablation evaluation, thereby achieving the technical effect of improving the efficiency of ablation evaluation on the basis of accurate evaluation, and meeting the timeliness requirements of real-time ablation evaluation, as follows:

[0125] The method for constructing a real-time ablation effect evaluation model includes:

[0126] The real-time ablation angle and the real-time ablation temperature of the ablation needle are used as the input of the BP neural network, and the real-time ablation effect evaluation value is used as the output of the BP neural network;

[0127] By using the training of BP neural network, a real-time ablation effect evaluation model is constructed to characterize the mapping relationship between the real-time ablation angle and real-time ablation temperature of the ablation needle and the real-time ablation effect evaluation value;

[0128] The real-time ablation effect evaluation model is:

[0129] H = BP (S, term);

[0130] Where H is the real-time ablation effect evaluation value, BP is the neural network, S is the real-time ablation angle, and tem is the real-time ablation temperature.

[0131] The projection module consists of a convolutional layer and a fully connected layer. Specifically, the parameter regression layer consists of a convolutional layer with a convolution kernel size of 7×7, a convolutional layer with a convolution kernel size of 5×5, a convolutional layer with a convolution kernel size of 3×3, and a fully connected layer. Each convolution layer is followed by a batch normalization and ReLU activation function.

[0132] The loss function of the real-time ablation effect evaluation model is the mean square error between the output value of the real-time ablation effect evaluation model and the real-time ablation effect evaluation value obtained according to the real-time ablation effect evaluation image.

[0133] like Figure 2 As shown, in a second aspect of the present invention, the present invention provides a radiofrequency ablation effect evaluation system based on edge prediction, which is applied to a radiofrequency ablation effect evaluation method based on edge prediction, and the system includes:

[0134] An image acquisition unit, used for acquiring real-time ablation images during radiofrequency ablation;

[0135] An image processing unit, used to project and fuse the ablation edema zone of the real-time ablation image and the original tumor edge of the original tumor image using a pre-established edema zone projection model, so as to obtain a real-time ablation effect evaluation image;

[0136] An image ablation evaluation unit, used to match the pixel coordinates between the ablation edema zone and the edge of the original tumor in the real-time ablation effect evaluation image to obtain a real-time ablation effect evaluation value;

[0137] The parameter ablation evaluation unit is used to predict the real-time ablation effect evaluation value according to the real-time ablation operation parameters using the real-time ablation effect evaluation model.

[0138] In a third aspect of the present invention, the present invention provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, a method for evaluating the effect of radiofrequency ablation based on edge prediction is implemented.

[0139] The present invention utilizes a pre-established edema band projection model to perform projection fusion on the ablation edema band of the real-time ablation image and the original tumor edge of the original tumor image, to obtain a real-time ablation effect evaluation image, to intuitively and accurately evaluate the distance between the edema band and the tumor edge, and to establish a correlation between the real-time ablation operation parameters and the real-time ablation effect evaluation values, to improve the efficiency of ablation evaluation on the basis of accurate evaluation through the correlation.

[0140] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. A method for evaluating the effect of radiofrequency ablation based on edge prediction, characterized in that: The following steps are involved: Acquire real-time ablation images during radiofrequency ablation; Using a pre-established edema zone projection model, the ablation edema zone of the real-time ablation image and the original tumor edge of the original tumor image are projected and fused to obtain a real-time ablation effect evaluation image; In the real-time ablation effect evaluation image, the pixel coordinates between the ablation edema zone and the edge of the original tumor are matched to obtain a real-time ablation effect evaluation value; The real-time ablation operation parameters corresponding to the real-time ablation image are associated with the real-time ablation effect evaluation value to obtain a real-time ablation effect evaluation model that predicts the real-time ablation effect evaluation value based on the real-time ablation operation parameters.

2. The method for evaluating the effect of radiofrequency ablation based on edge prediction according to claim 1, characterized in that: The method for constructing the edema zone projection model comprises: Acquire multiple ablation images with ablated edema zones as sample images, and acquire original tumor images corresponding to the sample images; Using a CNN neural network to extract features from a sample image and an original tumor image corresponding to the sample image, to obtain a feature map of the sample image and a feature map of the original tumor image; Bidirectionally mapping the feature coordinates in the feature map of each sample image with the feature coordinates in the feature map of the original tumor image one by one, to construct a bidirectional mapping relationship between the features of the sample image and the original tumor image; The feature bidirectional mapping relationship is: R(XY)=FY(i,j) T FX(i l ,i l ); R(YX)=FX(i,j) T FX(i l ,i l ); Where R(XY) is the feature mapping relationship from the sample image to the original tumor image, R(YX) is the feature mapping relationship from the original tumor image to the sample image, FY(i,j) is the feature map vector of the original tumor image at the (i,j) coordinate, and FX(i l ,i l ) is the sample image in (i l ,i l ) coordinates, FX(i,j) is the feature map vector of the sample image at the (i,j) coordinates, FY(i l ,i l ) is the original tumor image at (i l ,i l ) coordinates; Obtaining a feature bidirectional matching graph of the sample image and the original tumor image through the feature bidirectional mapping relationship; The feature bidirectional matching graph is: Wherein, g(XY) is the feature matching map from the sample image to the original tumor image, g(YX) is the feature matching map from the original tumor image to the sample image, R(XY) is the feature mapping relationship from the sample image to the original tumor image, R(YX) is the feature mapping relationship from the original tumor image to the sample image, n is the first dimension of the feature mapping relationship, and γ is a constant term; The feature bidirectional matching map is used as the input of the projection module, and the projection image of the ablation edema zone in the sample image on the original tumor image and the projection image of the tumor edge in the original tumor image on the sample image are used as the output of the projection module respectively; Acquire a first projection parameter by which the projection module projects the ablation edema zone in the sample image onto the original tumor image, and a second projection parameter by which the projection module projects the tumor edge in the original tumor image onto the sample image; Iteratively optimizing the projection module and the CNN neural network through projection loss to obtain an optimal first projection parameter and an optimal second projection parameter; The projection loss is: Where L(θ1,θ2) is the projection loss, GXY k (θ1,g(XY)) is the projection image of the ablation edema zone in the kth sample image on the kth original tumor image, GX is the sample image, GYX k (θ2,g(YX)) is the projection image of the tumor edge in the kth original tumor image on the kth sample image, GY is the original tumor image, θ1 is the first projection parameter, θ2 is the second projection parameter, m is the total number of sample images, and All are L2 norm forms; The optimal first projection parameter and the optimal second projection parameter are combined to obtain the optimal projection parameter used by the projection regression module to optimally project and fuse the ablation edema zone of the sample image and the original tumor edge of the original tumor image; The optimal projection parameters are: i max =M(θ1,(θ2) -1 ); In the formula, θ max is the optimal projection parameter, θ1 is the first projection parameter, θ2 is the second projection parameter, M(θ1,(θ2) -1 ) is θ1 and (θ2) -1 The arithmetic mean of The input of the projection module corresponding to the optimal projection parameter is updated to a feature matching map in the direction from the sample image to the original tumor image, and the output of the projection module corresponding to the optimal projection parameter is updated to a projection image of the ablated edema zone in the sample image on the original tumor image; The input of the updated projection module and the output of the CNN neural network are connected through a feature mapping relationship in the direction from the sample image to the original tumor image to obtain the edema zone projection model; The edema zone projection model is: GXY=(CNN(GX),CNN(GY))→R(XY)→E(θ max ); Where GXY is the projection image of the ablated edema zone in the sample image on the original tumor image, GX is the sample image, GY is the original tumor image, CNN(GX) is the feature map of the sample image, CNN(GY) is the feature map of the original tumor image, R(XY) is the feature mapping relationship from the sample image to the original tumor image, E(θ max ) is the projection module corresponding to the optimal projection parameter, θ max is the optimal projection parameter.

3. The method for evaluating the effect of radiofrequency ablation based on edge prediction according to claim 2, characterized in that: The method for acquiring the real-time ablation effect evaluation image includes: Inputting the real-time ablation image and the original tumor image into the edema band projection model, and the edema band projection model outputs a projection image of the ablated edema band in the real-time ablation image on the original tumor image; The projection image on the original tumor image is used as a real-time ablation effect evaluation image.

4. The method for evaluating the effect of radiofrequency ablation based on edge prediction according to claim 3, characterized in that: The method for extracting the real-time ablation effect evaluation value includes: In the real-time ablation effect evaluation image, edge detection is used to extract the ablation edema zone, and edge detection is used to extract the original tumor edge; Matching the pixel coordinates of the ablation edema zone with the original tumor edge to obtain the real-time ablation effect evaluation value; The real-time ablation effect evaluation value is: Where H is the real-time ablation effect evaluation value, P(D1) is the pixel coordinate of the ablated edema zone, and P(D2) is the pixel coordinate of the original tumor edge. is the L2 norm.

5. The method for evaluating radiofrequency ablation effect based on edge prediction according to claim 4, characterized in that: The real-time ablation operation parameters include: the real-time ablation angle and the real-time ablation temperature of the ablation needle.

6. The method for evaluating radiofrequency ablation effect based on edge prediction according to claim 5, characterized in that: The method for constructing the real-time ablation effect evaluation model includes: Using the real-time ablation angle and the real-time ablation temperature of the ablation needle as inputs of the BP neural network, and using the real-time ablation effect evaluation value as output of the BP neural network; By using the training of BP neural network, a real-time ablation effect evaluation model is constructed to characterize the mapping relationship between the real-time ablation angle and real-time ablation temperature of the ablation needle and the real-time ablation effect evaluation value; The real-time ablation effect evaluation model is: H = BP (S, term); Where H is the real-time ablation effect evaluation value, BP is the neural network, S is the real-time ablation angle, and tem is the real-time ablation temperature.

7. The method for evaluating radiofrequency ablation effect based on edge prediction according to claim 6, characterized in that: The projection module consists of a convolutional layer and a fully connected layer.

8. The method for evaluating radiofrequency ablation effect based on edge prediction according to claim 7, characterized in that: The loss function of the real-time ablation effect evaluation model is the mean square error between the output value of the real-time ablation effect evaluation model and the real-time ablation effect evaluation value obtained according to the real-time ablation effect evaluation image.

9. A radiofrequency ablation effect evaluation system based on edge prediction, characterized in that: A method for evaluating the effect of radiofrequency ablation based on edge prediction applied to any one of claims 1 to 8, the system comprising: An image acquisition unit, used for acquiring real-time ablation images during radiofrequency ablation; An image processing unit, configured to project and fuse the ablation edema band of the real-time ablation image and the original tumor edge of the original tumor image using a pre-established edema band projection model, so as to obtain a real-time ablation effect evaluation image; An image ablation evaluation unit, used to match the pixel coordinates between the ablation edema zone and the edge of the original tumor in the real-time ablation effect evaluation image to obtain a real-time ablation effect evaluation value; The parameter ablation evaluation unit is used to predict the real-time ablation effect evaluation value according to the real-time ablation operation parameters using the real-time ablation effect evaluation model.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the processor executes the computer-executable instructions, the method according to any one of claims 1 to 8 is implemented.

Images