CN111973271A - Preoperative ablation region simulation method and device for tumor thermal ablation - Google Patents

Abstract

A preoperative ablation region simulation method and device for tumor thermal ablation are provided, the method comprises the following steps: (1) selecting data suitable for training, and marking preoperative and postoperative images by a doctor to obtain preoperative livers, postoperative ablation regions and needle tracks; (2) registering the postoperative image to the preoperative image, adding rigid limitation in the registration process to ensure that the ablation region and the needle track region do not generate elastic deformation, and projecting the needle track information and the ablation region onto the preoperative image by utilizing a deformation field obtained by registration; (3) generating an ideal thermal field distribution map according to the projected needle track direction and position information and the corresponding power time adopted in the operation; (4) and establishing a convolutional neural network, cutting the ideal thermal field distribution map and the preoperative image according to a preoperative liver mask, combining cutting results as network input, learning by taking the deformed ablation region as a gold standard to obtain a regression model, and realizing the prediction of the ablation region in a test stage.

Description

Preoperative ablation area simulation method and device for tumor thermal ablation

technical field

The invention relates to the technical field of medical image processing, in particular to a preoperative ablation area simulation method for tumor thermal ablation, and a preoperative ablation area simulation device for tumor thermal ablation, which are mainly suitable for ablation operations based on deep learning guidance.

Background technique

During the microwave ablation procedure, the doctor inserts the ablation needle percutaneously into the position of the tissue lesion, and the anterior pole of the ablation needle emits microwaves, and the temperature of the surrounding lesion tissue rises after being irradiated by the microwave. The protein is inactivated, creating a region of ablation, thereby enabling the treatment of the tumor. In order to provide doctors with the best ablation method corresponding to different patients and realize individualized conformal ablation, it is necessary to design the position and direction of the needle in the preoperative planning stage according to the tumor and the positions of important tissue and blood vessels provided by the preoperative image. On this basis, the simulation generates the extent of the ablation area. By comparing the position and size relationship between the ablation area and the tumor, the position and direction of the needle insertion can be adjusted, so as to finally determine the best ablation plan suitable for the patient, and guide the actual ablation during the subsequent operation, so as to improve the overall operation efficiency. Therefore, in preoperative planning, according to the position and direction of the ablation needle, the prediction of the size and shape of the ablation area is the basis for designing the optimal ablation plan.

The cause of the ablation area is complex, and its size and shape are not only related to the power time of the ablation needle, but also affected by complex factors such as the heat sink effect of the surrounding blood vessels and the characteristics of the tissue and tumor tissue. Currently, the used ablation area generation models are divided into ellipsoid models and tissue-based Specific Absorption Rate (SAR) distribution models. For the first model, the ablation area is approximated as an ellipsoid, the direction of the long axis of the ellipsoid is the direction of needle insertion, and the size of the short axis of the long axis can be user-defined or set as an empirical value. The SAR distribution model reflects the increased temperature of the tissue after absorbing microwaves. By calculating the thermal damage through the Arrhenius equation, the range of the ablation area generated by the ablation needle can be determined. The SAR distribution model and the Arrhenius thermal damage equation include parameters such as tissue and blood density and specific heat, tissue thermal conductivity, and blood perfusion rate. Therefore, compared with the ellipsoid model, the SAR distribution model contains important factors that affect the generation of ablation regions. However, the SAR model needs accurate blood vessel segmentation results when calculating the vascular heat sink effect. At the same time, because the parameters related to different patient tissues and blood are different, and the real values of these parameters are difficult to obtain, the ablation area solved by the SAR distribution model is different from the The actual ablation effect is still quite different. This seriously reduces the clinical significance of preoperative planning for subsequent real operations.

SUMMARY OF THE INVENTION

In order to overcome the defects of the prior art, the technical problem to be solved by the present invention is to provide a preoperative ablation area simulation method for tumor thermal ablation, which not only avoids the acquisition of a large number of complex parameters and the segmentation of related tissues such as blood vessels, but also The predicted ablation area is closer to the real ablation effect, so that preoperative planning has clinical guiding significance for subsequent real operations.

The technical solution of the present invention is: this preoperative ablation area simulation method for tumor thermal ablation includes the following steps:

(1) Select the data suitable for training, and the doctor annotates the preoperative and postoperative images to obtain the preoperative liver, the postoperative ablation area and the needle track;

(2) Register the postoperative image to the preoperative image, and add rigid constraints during the registration process to prevent elastic deformation between the ablation area and the needle track area, and use the deformation field obtained by registration to project the needle track information and the ablation area. onto the preoperative image;

(3) Generate an ideal thermal field distribution map according to the projected needle track direction and position information and the corresponding power time used in the operation;

(4) Establish a convolutional neural network, crop the ideal heat field distribution map and the preoperative image according to the preoperative liver mask, combine the cropping results as the network input, and use the deformed ablation area as the gold standard for learning to obtain regression model, implemented in the testing phase

Prediction of the ablation area now.

The present invention utilizes retrospective ablation data, establishes a regression model through deep learning on the basis of real ablation results, and predicts the ablation area. Different from the ellipsoid and SAR distribution model prediction methods, the present invention takes the real ablation area obtained after surgery as the gold standard, takes the preoperative image and the ideal thermal field generated by the ablation needle as the input, and uses the convolutional neural network to establish a regression model. It not only avoids the acquisition of a large number of complex parameters and the segmentation of related tissues such as blood vessels, but also takes the real ablation area as the gold standard, and the predicted ablation area is closer to the real ablation effect, so that preoperative planning has clinical guiding significance for subsequent real operations.

Also provided is a preoperative ablation area simulation device for thermal tumor ablation, including:

The labeling module is configured to select data suitable for training. The doctor labels the preoperative and postoperative images to obtain the preoperative liver, the postoperative ablation area and the needle track;

The registration module is configured to register the postoperative image to the preoperative image. In the registration process, rigid constraints are added to prevent elastic deformation between the ablation area and the needle track area. Projecting the ablation area onto the preoperative image;

a heat field map generation module, which is configured to generate an ideal heat field distribution map according to the projected needle track direction and position information and the corresponding power time used in the operation;

The deep learning module is configured to build a convolutional neural network, crop the ideal heat field distribution map and the preoperative image according to the preoperative liver mask, merge the cropping results as the network input, and use the deformed ablation area as the gold standard. Learn, obtain a regression model, and achieve prediction of ablation regions in the testing phase.

Description of drawings

FIG. 1 is a flowchart of a preoperative ablation area simulation method for thermal tumor ablation according to the present invention.

Detailed ways

As shown in Figure 1, the preoperative ablation area simulation method for tumor thermal ablation includes the following steps:

(1) Select the data suitable for training, and the doctor annotates the preoperative and postoperative images to obtain the preoperative liver, the postoperative ablation area and the needle track;

(2) Register the postoperative image to the preoperative image, and add rigid constraints during the registration process to prevent elastic deformation between the ablation area and the needle track area, and use the deformation field obtained by registration to project the needle track information and the ablation area. onto the preoperative image;

(3) Generate an ideal thermal field distribution map according to the projected needle track direction and position information and the corresponding power time used in the operation;

(4) Establish a convolutional neural network, crop the ideal heat field distribution map and the preoperative image according to the preoperative liver mask, combine the cropping results as the network input, and use the deformed ablation area as the gold standard for learning to obtain regression A model that achieves prediction of ablation regions during the testing phase.

The present invention utilizes retrospective ablation data, establishes a regression model through deep learning on the basis of real ablation results, and predicts the ablation area. Different from the ellipsoid and SAR distribution model prediction methods, the present invention takes the real ablation area obtained after surgery as the gold standard, takes the preoperative image and the ideal thermal field generated by the ablation needle as the input, and uses the convolutional neural network to establish a regression model. It not only avoids the acquisition of a large number of complex parameters and the segmentation of related tissues such as blood vessels, but also takes the real ablation area as the gold standard, and the predicted ablation area is closer to the real ablation effect, so that preoperative planning has clinical guiding significance for subsequent real operations.

The ablation data for each patient includes preoperative and postoperative MR images, as well as the information of the ablation needle power and ablation time used during the operation. Preferably, in the step (1), in order to enable the network to learn the influence of the vascular heat sink effect, the images marked by the doctor on the preoperative and postoperative images and the images for subsequent registration and prediction training are all in the venous phase, and the venous phase Compared with other stages, the contrast between the internal liver blood vessels and the liver is stronger; the outlines of the liver and the ablation area are drawn; the needle track is marked by marking the needle point of the ablation needle and the ablation needle and ablation on the postoperative image. The world coordinate of the intersection point of the region. Since the ablation needle is nearly rigid, it is regarded as a ray passing through two marked points in the postoperative image, where the vertex of the ray is the needle tip point.

Preferably, in the step (1), after the labeling is completed, the mask images of the preoperative liver L and the postoperative ablation area A, as well as the position and direction parameters of the ablation needle are obtained respectively.

There may be differences in the patient's posture in the preoperative and postoperative images. At the same time, the liver is an elastic tissue. Due to the influence of breathing, the liver undergoes non-rigid deformation in the preoperative and postoperative images. Therefore, we will use a two-step registration method to achieve preoperative Matching of postoperative images. Preferably, the step (2) includes the following sub-steps:

(2.1) Rigid registration of postoperative images to preoperative images is performed, and the differences in translation and rotation caused by the patient's pose are eliminated through rigid registration, and the deformation matrix is obtained after rigid registration;

(2.2) The non-rigid registration algorithm was used to solve the deformation field in the liver region to further reduce the difference between preoperative and postoperative images.

Preferably, in the step (2.2), in order to avoid the deformation of the ablation region and the needle track, the ablation region and the needle track are kept as rigid bodies during the registration process; for the ablation region, the rigid body is maintained during the registration process through a rigid restriction term ; For the ablation needle, since the marked points are all within the ablation area, the rigidity of the ablation area also makes the ablation needle remain rigid during the registration process.

Preferably, in the step (3), the thermal field generated by the ideal SAR model is a symmetrical distribution with the ablation needle as the axis, and the thermal field image is obtained according to the deformed ablation needle information and the ideal SAR distribution model.

Preferably, in the step (4), the U-Net network is used to predict the ablation area, the input data is the combined image of the preoperative MR image and the thermal field image, and the deformed mask image of the ablation area will be used as the gold standard , the preoperative venous phase images provide important information on the characteristics of blood vessels and liver tumors, while the thermal field images provide important information on the characteristics of ablation needles.

Preferably, in the step (4), the preoperative image and the thermal field image are cropped by using the marked liver mask, and the loss function loss is the overlap ratio between the predicted ablation area and the real ablation area. Through learning, the prediction model of the ablation area is finally obtained, and for the input test data, fast and accurate prediction of the ablation area can be achieved.

Those of ordinary skill in the art can understand that all or part of the steps in the method of the above-mentioned embodiments can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium, and the program can be stored in a computer-readable storage medium. During execution, it includes each step of the method in the above embodiment, and the storage medium may be: ROM/RAM, magnetic disk, optical disk, memory card, and the like. Therefore, corresponding to the method of the present invention, the present invention also includes a preoperative ablation area simulation device for tumor thermal ablation, which is usually expressed in the form of functional modules corresponding to each step of the method. The device includes:

The labeling module is configured to select data suitable for training. The doctor labels the preoperative and postoperative images to obtain the preoperative liver, the postoperative ablation area and the needle track;

The registration module is configured to register the postoperative image to the preoperative image. In the registration process, rigid constraints are added to prevent elastic deformation between the ablation area and the needle track area. Projecting the ablation area onto the preoperative image;

a heat field map generation module, which is configured to generate an ideal heat field distribution map according to the projected needle track direction and position information and the corresponding power time used in the operation;

The deep learning module is configured to build a convolutional neural network, crop the ideal heat field distribution map and the preoperative image according to the preoperative liver mask, merge the cropping results as the network input, and use the deformed ablation area as the gold standard. Learn, obtain a regression model, and achieve prediction of ablation regions in the testing phase.

The above are only preferred embodiments of the present invention, and do not limit the present invention in any form. Any simple modifications, equivalent changes and modifications made to the above embodiments according to the technical essence of the present invention still belong to the present invention The protection scope of the technical solution of the invention.

Claims (9)

1. A preoperative ablation area simulation method for tumor thermal ablation, characterized in that: it comprises the following steps: (1) Select the data suitable for training, and the doctor annotates the preoperative and postoperative images to obtain the preoperative liver, the postoperative ablation area and the needle track; (2) Register the postoperative image to the preoperative image, and add rigid constraints during the registration process to prevent elastic deformation between the ablation area and the needle track area, and use the deformation field obtained by registration to project the needle track information and the ablation area. onto the preoperative image; (3) Generate an ideal thermal field distribution map according to the projected needle track direction and position information and the corresponding power time used in the operation; (4) Establish a convolutional neural network, crop the ideal heat field distribution map and the preoperative image according to the preoperative liver mask, combine the cropping results as the network input, and use the deformed ablation area as the gold standard for learning to obtain regression A model that achieves prediction of ablation regions during the testing phase. 2 . The preoperative ablation area simulation method for tumor thermal ablation according to claim 1 , wherein in the step (1), the doctor annotates the preoperative and postoperative images and performs subsequent registration and prediction training. 3 . The images are all in the venous phase, and the liver blood vessels in the venous phase have stronger contrast with the liver than other phases; the outlines of the liver and the ablation area are drawn when they are marked; the needle track is marked by marking the ablation on the postoperative image. The world coordinates of the needle point of the needle and the intersection of the ablation needle and the ablation area. Since the ablation needle is almost a rigid body, it is regarded as a ray passing through two marked points in the postoperative image, where the vertex of the ray is the needle point. 3. The preoperative ablation area simulation method for tumor thermal ablation according to claim 2, wherein in the step (1), after the labeling is completed, the preoperative liver L and the postoperative ablation area A masks are obtained respectively. The model image, as well as the position and direction parameters of the ablation needle. 4. The preoperative ablation area simulation method for tumor thermal ablation according to claim 3, wherein the step (2) comprises the following sub-steps: (2.1) Rigid registration of postoperative images to preoperative images is performed, and the differences in translation and rotation caused by the patient's pose are eliminated through rigid registration, and the deformation matrix is obtained after rigid registration; (2.2) The non-rigid registration algorithm was used to solve the deformation field in the liver region to further reduce the difference between preoperative and postoperative images. 5 . The preoperative ablation area simulation method for tumor thermal ablation according to claim 4 , wherein in the step (2.2), in order to avoid deformation of the ablation area and the needle track, the ablation area is maintained during the registration process. 6 . It is a rigid body with the needle track; for the ablation area, the rigid body is maintained by the rigid restriction item during the registration process; for the ablation needle, since the marked points are all within the ablation area, the rigid restriction of the ablation area also makes the ablation needle in the registration process. remains a rigid body. 6 . The preoperative ablation area simulation method for tumor thermal ablation according to claim 5 , wherein in the step (3), the thermal field generated by the ideal SAR model is a symmetrical distribution with the ablation needle as the axis, 6 . According to the deformed ablation needle information and the ideal SAR distribution model, the thermal image is obtained. 7. The preoperative ablation area simulation method for tumor thermal ablation according to claim 6, characterized in that: in the step (4), U-Net network is used to realize the prediction of the ablation area, and the input data is preoperative The combined image of the MR image and the thermal field image, the deformed ablation area mask image will be used as the gold standard, the preoperative venous phase image provides important blood vessel and liver tumor characteristic information, and the thermal field image provides important ablation needle characteristic information . 8 . The preoperative ablation area simulation method for tumor thermal ablation according to claim 7 , wherein in the step (4), the preoperative image and the thermal field image are trimmed by using the marked liver mask. 9 . , and the overlap rate between the predicted ablation area and the real ablation area is used as the loss function loss. 9. A preoperative ablation area simulation device for tumor thermal ablation, characterized in that: it includes: a labeling module, which is configured to select data suitable for training, and the doctor labels preoperative and postoperative images to obtain preoperative liver, surgical Post-ablation area and needle track; The registration module is configured to register the postoperative image to the preoperative image. In the registration process, rigid constraints are added to prevent elastic deformation between the ablation area and the needle track area. Projecting the ablation area onto the preoperative image; a heat field map generation module, which is configured to generate an ideal heat field distribution map according to the projected needle track direction and position information and the corresponding power time used in the operation; The deep learning module is configured to build a convolutional neural network, crop the ideal heat field distribution map and the preoperative image according to the preoperative liver mask, merge the cropping results as the network input, and use the deformed ablation area as the gold standard. Learn, obtain a regression model, and achieve prediction of ablation regions in the testing phase.

Images