Ultralow-dose CT (computed tomography) imaging method and application thereof in puncture path guidance

Abstract

The invention discloses an ultra-low dose CT imaging method and application of the ultra-low dose CT imaging method in puncture path guidance, the low dose CT imaging method obtains an artifact-free global image through three steps, and the global image retains key information of a region of interest ROI of an internal scanning image and recovers detail features of a non-region of interest ROI. The scanning view in the operation process is reduced through the internal scanning technology, the radiation dose can be greatly reduced, meanwhile, the convolutional neural network is introduced to extract prior information of primary full scanning to repair a subsequent internal scanning image to be processed, cup-shaped artifacts and metal artifacts are effectively inhibited, the image quality is improved, and the image quality is improved. And detail features of a non-ROI (Region of Interest) in the prior image are recovered, and clearer and more accurate positioning basis is provided for an operator. The radiation risk is reduced, and meanwhile the accuracy of clinical puncture operation is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to an ultra-low-dose CT imaging method and its application in puncture path guidance. Background Art

[0002] Computed tomography (CT) is a non-invasive imaging technology that can obtain high-resolution tomographic structural images of the inside of an object in a non-invasive manner. At present, during the puncture needle puncture process, CT imaging is required to guide the needle to accurately puncture the target lesion. This CT imaging method is called CT-guided puncture method. The main advantage of the CT-guided puncture method is that it can visualize the surgical process without obstruction, so that accurate puncture is located in the target area for imaging. The CT-guided puncture method requires multiple CT scans to accurately locate the puncture needle. The CT-guided puncture method requires repeated CT imaging, so the patient needs to be exposed to high-dose X-rays multiple times, which increases the radiation dose received. In addition, the presence of metal needles will cause severe metal artifacts in the CT image, making the target lesion and adjacent tissue structures unclear, affecting the judgment of the lesion location. Therefore, it is necessary to try to reduce the radiation dose during CT scanning and minimize artifacts to improve positioning accuracy, puncture efficiency and surgical quality.

[0003] At present, some studies have adopted low-dose CT scanning technology, such as reducing the X-ray tube current, to minimize the radiation dose received by patients during CT-guided puncture. In order to improve the reconstruction quality of low-dose CT data, a variety of correction and reconstruction algorithms have been proposed, including reconstruction algorithms based on image filtering, iterative reconstruction, and deep learning. Among them, deep learning technology can not only effectively eliminate the noise in the reconstructed image, but also effectively suppress the metal artifacts introduced by the puncture needle. However, in order to ensure that the measurement data has a sufficient signal-to-noise ratio, the above methods still have limitations in reducing radiation dose.

[0004] Therefore, in view of the shortcomings of the existing technology, it is necessary to provide an ultra-low-dose CT imaging method and its application in path guidance to solve the shortcomings of the existing technology. Summary of the invention

[0005] The first object of the present invention is to avoid the shortcomings of the prior art and provide an ultra-low-dose CT imaging method. The ultra-low-dose CT imaging method is based on the internal imaging technology, and the ultra-low-dose CT imaging method can significantly reduce the radiation dose, while obtaining an artifact-free global image that retains the key information of the region of interest ROI and restores the detailed features of the non-region of interest ROI.

[0006] The above-mentioned purpose of the present invention is achieved by the following technical measures:

[0007] Provided is an ultra-low-dose CT imaging method, which is carried out by the following steps:

[0008] S1. Obtain a CT scan image before the operation, and use the CT scan image as a prior image. The CT scan image is a global CT image obtained by a full-dose CT scan;

[0009] S2. Obtain a to-be-processed internal scan image of the region of interest during the operation. The to-be-processed internal scan image is a local CT image obtained by using an internal scan protocol and scanning the region of interest outside the human body;

[0010] S3. Input the prior image in S1 and the to-be-processed internal scan image in S2 into a trained artifact correction model for feature extraction and image reconstruction to obtain an artifact-free global image. The artifact-free global image is an artifact-free global image that retains the key information of the region of interest (ROI) of the internal scan image and restores the detailed features of the non-region of interest (ROI) of the prior image.

[0011] Preferably, the training process of the above-mentioned artifact correction model is as follows:

[0012] A1. Obtain a CT puncture image during a previous operation process, and segment the puncture needle mask in the CT puncture image to correspondingly obtain a puncture needle mask library composed of a plurality of puncture needle masks;

[0013] A2. Obtain multiple groups of training CT images. The training CT images do not contain puncture needles, and each level of training CT image is a multi-sequence image. Randomly select one image in each group of training CT images as a prior image, and in the same group of training CT images, select one of the images adjacent to the prior image as a simulation image. Finally, randomly select a puncture needle mask from the puncture needle mask library obtained in A1 to simulate the simulation image to obtain a paired data pair. The paired data pair is composed of an artifact-containing local image X baseline and the paired artifact-free global image X target and the artifact-free global image X target is used as a reference image. The artifact-containing local image X baseline is an internal scan image with a puncture needle and artifacts, and the artifact-free global image X target is a global scan image with a puncture needle and no artifacts;

[0014] A3. Input the prior image X prior and the artifact-containing local image X baseline in A2 into the artifact correction model to correct the cupping artifact and the metal artifact, and obtain a target artifact-removed image

[0015] A4. Use the loss function Loss to process the target artifact-removed image in A3 and the artifact-free global image X in A2 target constrained by mean squared error loss and structural similarity loss to obtain a loss value;

[0016] A5. According to the loss value in A4, use the gradient descent method to optimize the parameters of the artifact correction model , then determine whether the training end condition is reached. If not, return to A3; if yes, enter A6;

[0017] A6. Define the current artifact correction model as the trained artifact correction model.

[0018] Preferably, the above A1 is specifically to pre-obtain multi-sequence CT puncture images of different objects at different times during the puncture needle process. The CT puncture images of the same object include an initial image X ref without a puncture needle and multiple subsequent images X punct during the operation process, where the initial image X ref does not contain a puncture needle, and the subsequent image X punct contains a puncture needle. Then, according to the initial image X ref , segment the puncture needle masks in all subsequent images X punct correspondingly, and obtain a puncture needle mask library composed of multiple puncture needle masks.

[0019] In the above A1, perform threshold segmentation on the residual image between the initial image X ref and the subsequent image X punct , and then perform binary processing on the puncture needle to obtain the puncture needle mask.

[0020] Preferably, the above A3 is specifically to input the prior image X prior of A2 and the local image X baseline with artifacts into the artifact correction model The encoder of the artifact correction model extracts the prior feature information in the prior image X prior , and then the decoder of the artifact correction model repairs the missing part of the local image X baseline with artifacts, and finally obtains the target artifact-removed image

[0021] The loss function Loss is represented by Equation (1),

[0022]

[0023] where is the mean square error loss, is the structural similarity loss, and α is a hyperparameter that balances the mean square error loss and the structural similarity loss component, is the loss value.

[0024] The target artifact-free image is represented by Equation (2),

[0025]

[0026] wherein, are the learnable parameters in the artifact correction model

[0027] Preferably, the above training end condition is reaching the maximum number of training rounds.

[0028] Preferably, the above artifact correction model is a convolutional neural network.

[0029] In A2, paired data pairs are obtained by simulating the simulation image through the XCIST simulation toolkit.

[0030] The second object of the present invention is to avoid the deficiencies of the prior art and provide an application of an ultra-low dose CT imaging method in puncture path guidance. When the ultra-low dose CT imaging method is applied in puncture path guidance, the radiation dose can be greatly reduced, and at the same time, an artifact-free global image that retains the key information of the region of interest (ROI) and restores the detailed features of the non-region of interest (ROI) can be obtained. Through this artifact-free global image, the puncture needle can be accurately guided into the path, improving the accuracy of the puncture operation.

[0031] The above object of the present invention is achieved by the following technical measures:

[0032] Provide an application of an ultra-low dose CT imaging method as described above in puncture path guidance.

[0033] ​Provided is a method for ultra-low dose CT imaging and the application of the ultra-low dose CT imaging method in puncture path guidance. The low dose CT imaging method is carried out by the following steps: S1. Obtain the CT scan image before the operation, and use the CT scan image as the prior image. The CT scan image is a global CT image obtained by full-dose CT scanning; S2. Obtain the to-be-processed internal scan image of the region of interest during the operation. The to-be-processed internal scan image is a local CT image obtained by using an internal scanning protocol to scan the region of interest outside the human body; S3. Input the prior image in S1 and the to-be-processed internal scan image in S2 into the trained artifact correction model for feature extraction and image reconstruction to obtain an artifact-free global image. The artifact-free global image is an artifact-free global image that retains the key information of the region of interest (ROI) of the internal scan image and restores the detailed features of the non-region of interest (ROI) of the prior image. Through the internal scanning technology, the present invention reduces the scanning field of view during the operation, can greatly reduce the radiation dose, and at the same time introduces a convolutional neural network to extract the prior information of the initial full scan to repair the subsequent to-be-processed internal scan image, effectively suppressing cupping artifacts and metal artifacts to improve the image quality, and restoring the detailed features of the non-region of interest (ROI) of the prior image to obtain a clearer and more accurate artifact-free global image. While reducing the radiation risk, the present invention improves the accuracy of clinical puncture operations, provides a new technical path for the development of CT-guided puncture technology, and has broad application prospects and popularization value. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The present invention will be further described with reference to the accompanying drawings, but the content in the drawings does not constitute any limitation to the present invention.

[0035] Figure 1 It is a schematic flowchart of a method for ultra-low dose CT imaging.

[0036] Figure 2 It is a schematic diagram of the CT scanning range, where Figure 2 (a) is a schematic diagram of conventional CT scanning, Figure 2 (b) is a schematic diagram of internal scanning.

[0037] Figure 3 It is a schematic diagram of the training process of the artifact correction model.

[0038] Figure 4 It is a diagram of the results before and after correction in Example 2. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The technical solution of the present invention will be further described in conjunction with the following embodiments.

[0040] Example 1

[0041] An ultra-low dose CT imaging method, such asFigure 1 , which is carried out by the following steps:

[0042] S1. Obtain the CT scan image before the operation, and use the CT scan image as the prior image. The CT scan image is a global CT image obtained by full-dose CT scanning;

[0043] S2. Obtain the to-be-processed internal scan image of the region of interest during the operation. The to-be-processed internal scan image is a local CT image obtained by using an internal scanning protocol and scanning the region of interest outside the human body;

[0044] S3. Input the prior image of S1 and the to-be-processed internal scan image of S2 into the trained artifact correction model for feature extraction and image reconstruction to obtain an artifact-free global image. The artifact-free global image is an artifact-free global image that retains the key information of the region of interest (ROI) of the internal scan image and restores the detailed features of the non-region of interest (ROI) of the prior image. Among them, the artifact correction model is a convolutional neural network.

[0045] It should be noted that the to-be-processed internal scan image of the present invention contains a puncture needle used for puncture. This puncture needle is used for operations such as puncturing in the body, collecting liquid or tissue samples, infusing or injecting drugs, etc. Since the current puncture needles are mainly made of materials such as stainless steel and titanium alloy, a large number of metal artifacts will appear in the to-be-processed internal scan image, resulting in unclear display of structures such as the target lesion and adjacent tissues. The internal scanning protocol of the present invention is a change in the exposure method. In actual application, the exposure parameters in the internal scanning protocol, whether to use a contrast agent, etc. also need to be selected according to clinical needs. However, the exposure parameters, whether to use a contrast agent, etc. are not within the scope of the present invention's research, so they will not be elaborated here.

[0046] Since the precise positioning of the puncture needle usually only requires a relatively small scanning field of view, while traditional CT scanning in the prior art usually uses full-field scanning, resulting in non-target tissues outside the region of interest being exposed to unnecessary radiation doses, thereby increasing the potential radiation risk to patients. The present invention obtains a prior image by full-dose CT scanning before the puncture operation, and then only locally scans the region of interest during the operation. Thus, non-target tissues outside the region of interest are less exposed to radiation doses. The advantage of obtaining a prior image by full-dose CT scanning before the operation in the present invention is that it can provide a global image with high signal-to-noise ratio, which is convenient for subsequent positioning of the puncture position and lays a foundation for the subsequent internal scanning and reconstruction of the present invention. The advantage of obtaining the to-be-processed internal scan image by internal scanning in the present invention is that it only focuses on the region of interest and reduces the scanning field of view. Moreover, in actual use, multiple internal scans are often required during the puncture operation. Therefore, the present invention can minimize the range of objects exposed to X-rays. The ranges of conventional CT scanning and internal scanning are compared with Figure 2As shown, the prior image is obtained by Figure 2 the scanning method of (a), and the internal scanning image to be processed is obtained by Figure 2 the scanning method of (b). However, directly reconstructing the truncated data obtained by internal scanning will introduce sharp white cup-shaped artifacts in the reconstructed image. Therefore, the artifact correction model of the present invention can correct cup-shaped artifacts and metal artifacts, and at the same time repair the detailed features of the non-region of interest (ROI) in the prior image. The finally obtained artifact-free global image is the global image that removes cup-shaped artifacts and metal artifacts and restores the detailed features of the non-region of interest (ROI) in the prior image. The function of the artifact correction model of the present invention is to consider that during the continuous CT scanning process, the scanning position is consistent and the change of tissue structure is extremely small. The prior image reconstructed from the global CT image obtained by full-dose CT scanning and the subsequent internal scanning reconstructed image are simultaneously input into the artifact correction model to predict the artifact-free global image. Moreover, the internal scanning of the present invention is a local scanning technology and does not need to invade the body. And the internal scanning imaging technology is a conventional technology of computer tomography technology. Those skilled in the art should know its principle and method and will not be elaborated here one by one.

[0047] As Figure 3 shown, the training process of the artifact correction model is as follows:

[0048] A1. Obtain the CT puncture image in the prior operation process, and segment the puncture needle mask in the CT puncture image to correspondingly obtain a puncture needle mask library composed of multiple puncture needle masks;

[0049] A2. Obtain multiple groups of training CT images. The training CT images do not contain puncture needles, and each level of training CT image is a multi-sequence image. Randomly select one image as the prior image in each group of training CT images, and select one of the images adjacent to the prior image in the same group of training CT images as the simulation image. Finally, randomly select a puncture needle mask from the puncture needle mask library obtained in A1 to simulate the simulation image to obtain a paired data pair. The paired data pair is composed of the local image X with artifacts baseline and the paired artifact-free global image X target . And the artifact-free global image X target is used as the reference image; where the local image X with artifacts baseline is the internal scanning image with puncture needles and artifacts, and the artifact-free global image X target is the global scanning image with puncture needles and no artifacts; specifically, the paired data pair is obtained by simulating the simulation image through the XCIST simulation toolbox;

[0050] A3. Input the prior image X prior and the local image X with artifacts baseline in A2 into the artifact correction model Thereby, correct the cupping artifacts and metal artifacts to obtain the target artifact-free image

[0051] A4. Through the loss function Loss, for the target artifact-free image in A3 and the artifact-free global image X in A2 target Use the mean square error loss and the structural similarity loss for constraint to obtain the loss value;

[0052] A5. According to the loss value in A4, use the gradient descent method to optimize the parameters of the artifact correction model Then determine whether the training end condition is reached. If not, return to A3; if yes, enter A6; where the training end condition is reaching the maximum number of training rounds;

[0053] A6. Define the current artifact correction model as the trained artifact correction model.

[0054] Specifically, A1 of the present invention is to pre-obtain multi-sequence CT puncture images of different objects at different times during the puncture needle process. The CT puncture images of the same object include the initial image X without puncture ref and multiple subsequent images X during the operation process punct , where the initial image X ref does not contain a puncture needle, and the subsequent image X punct contains a puncture needle. Then, according to the initial image X ref correspondingly segment the puncture needle masks in all subsequent images X punct to obtain a puncture needle mask library composed of multiple puncture needle masks. Specifically, perform threshold segmentation on the residual image between the initial image X ref and the subsequent image X punct , and then perform binarization processing on the puncture needle to obtain the puncture needle mask.

[0055] It should be noted that the training of the artifact correction model of the present invention is carried out under the condition of less training CT images and CT puncture images. Based on the training CT images with multi-sequence images, randomly select one image as the prior image, and then select the image adjacent to the prior image sequence as the simulation image. The advantage of this is that a large amount of training data can be obtained. Taking the training CT images with 30 sequence images as an example, these 30 training CT images can generate at most 15 pairs of simulation images and prior images. In each pair of simulation images and prior images, then randomly select a puncture needle mask from the puncture needle mask library to simulate the simulation image to obtain the paired data pair. Among them, the local image X with artifacts baselineThe simulation process is to insert the simulation image into the puncture needle mask in the simulation software, perform internal scanning and reconstruction, and the reconstruction result contains internal scanning artifacts and metal artifacts; the artifact-free global image X target The simulation process is to first perform full scanning and reconstruction on the simulation image in the simulation software, and then insert the puncture needle mask into the reconstruction result, and the obtained reconstruction result does not contain internal scanning artifacts and metal artifacts.

[0056] It should also be noted that although the prior image and the simulation image of the present invention are CT images of different layers, the anatomical structures of adjacent sequential images still have similarities and can be used to guide image restoration. The purpose of the present invention not using the same image as the prior image and the simulation image respectively is that since the network needs to input the prior image and the artifact-containing image simultaneously, due to the inertia of the network, when using the same image, it may not learn and directly output the prior image, and the purpose of removing artifacts cannot be achieved.

[0057] In A3, specifically, the prior image X of A2 prior and the locally artifact-containing image X baseline are input into the artifact correction model The artifact correction model The encoder of extracts the prior feature information in the prior image X prior and then repairs the missing part of the locally artifact-containing image X through the decoder of the artifact correction model baseline to finally obtain the target artifact-removed image

[0058] The loss function Loss is represented by Equation (1),

[0059]

[0060] wherein, is the mean square error loss, is the structural similarity loss, and α is a hyperparameter that balances the mean square error loss and the structural similarity loss component, is the loss value.

[0061] The target artifact-removed image is represented by Equation (2),

[0062]

[0063] wherein, are the learnable parameters in the artifact correction model .

[0064] This ultra-low-dose CT imaging method reduces the scanning field of view during the operation through internal scanning technology, which can greatly reduce the radiation dose. At the same time, a convolutional neural network is introduced to extract the prior information of the initial full scan to repair the subsequent internal scan images to be processed, effectively suppressing cupping artifacts and metal artifacts to improve the image quality, restoring the detailed features of the non-region of interest (ROI) in the prior image, and reducing the radiation risk.

[0065] Example 2

[0066] An application of the ultra-low-dose CT imaging method of Example 1: Use the XCIST toolkit to simulate and reconstruct the chest and abdomen projections with / without metal implants to obtain the prior image X prior , and perform truncation processing on the projection data containing metal, respectively retaining the data of the middle 400, 300, and 200 detector units to simulate the internal scan data, and perform reconstruction to obtain the local image X with artifacts containing ring artifacts and metal artifacts baseline and the reference artifact-free global image X target .

[0067] In this embodiment, the metal shape is obtained by segmenting the puncture needles from 150 groups of multi-sequence images collected from Zhujiang Hospital. The metal puncture needle artifact simulation material in this example selects titanium, which is the most commonly used metal implant in clinical practice, and its density is 4.54 g / cm 3 .

[0068] Specifically, this embodiment uses the Mayo publicly available dataset for simulation data. Among them, the size of the simulated CT image with metal artifacts is 512×512, the distance from the source to the detector (DSD) is set to 950 mm, the distance from the source to the rotation center (DSO) is set to 550 mm, the size of the detector unit is 1 mm×1 mm, the number of detector units is 900, the pixel size of the reconstructed image is 0.78125 mm×0.78125 mm, and the number of projections is 1000. A total of 14,772 pairs of simulated data are simulated, of which 12,556 pairs are used as the training set, and the rest are used as the test set. It should be noted that the data of the test set and the training set in this embodiment come from different objects.

[0069] Specifically, during the training of the artifact correction model , the prior image X obtained by simulation prior , and the local image X with artifacts baseline are input into the artifact correction model to achieve ring artifact and metal artifact correction, and obtain the target artifact-free image Among them, the artifact correction model in this embodiment is NAFnet, the maximum number of training epochs is 50, and the learning rate linearly decreases from 1×10 -4 to 1×10-6 The optimizer is the Adam optimizer, and the hyperparameter α in the loss function is set to 10.

[0070] This embodiment demonstrates Figure 4 the artifact correction effect of reconstructed images with different detector channel numbers. Among them Figure 4 the first three columns in the first row respectively show the local images with artifacts reconstructed with different detector channel numbers. After being processed by the artifact correction model of the present invention, the corrected images in the second row are obtained respectively. The fourth image in the first row is a reference image without artifacts.

[0071] As can be seen from Figure 4 , obvious white cup-shaped artifacts appear in the projection data truncation area of the uncorrected image. At the same time, due to the presence of the metal puncture needle, significant bright and dark artifacts appear around the metal in the image. After being processed by the artifact correction model of the present invention, when the detector channel number is 400, the reconstructed image is highly close to the reference image; even when the channel number is reduced to 200, a high-quality reconstructed image can still be obtained. Therefore, it shows that the artifact correction model of the present invention can effectively achieve the approximate recovery of data outside the FOV during internal scanning, suppress the interference of metal artifacts on the image quality, and avoid the influence of such artifacts on the puncture needle positioning accuracy. This result verifies the feasibility and effectiveness of the ultra-low-dose CT-guided puncture method based on internal scanning proposed by the present invention.

[0072] Example 3

[0073] An application of an ultra-low-dose CT imaging method in puncture path guidance. The artifact-free global image obtained by the ultra-low-dose CT imaging method of the present invention in Example 2 can effectively remove cup-shaped artifacts and metal artifacts. Therefore, using the ultra-low-dose CT imaging method of the present invention in puncture path guidance can greatly reduce, and at the same time obtain an accurate artifact-free global image, thus providing a clearer and more accurate positioning basis for the operator, and further improving the accuracy of clinical puncture operations.

[0074] Example 4

[0075] A puncture path guidance method for ultra-low-dose CT imaging. This method is the same as the CT imaging method in Example 1 or 2. This puncture path guidance method can effectively remove cup-shaped artifacts and metal artifacts. Therefore, using the ultra-low-dose CT imaging method of the present invention in puncture path guidance can greatly reduce, and at the same time obtain an accurate artifact-free global image, thus providing a clearer and more accurate positioning basis for the operator, and improving the accuracy of clinical puncture operations.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the protection scope of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. An ultra-low dose CT imaging method, characterized in that, It is carried out by the following steps: S1. Obtain the CT scan image before the operation, and use the CT scan image as the prior image. The CT scan image is a global CT image obtained by full-dose CT scanning; S2. Obtain the to-be-processed internal scan image of the region of interest during the operation. The to-be-processed internal scan image is a local CT image obtained by using an internal scan protocol and scanning the region of interest outside the human body; S3. Input the prior image in S1 and the to-be-processed internal scan image in S2 into the trained artifact correction model for feature extraction and image reconstruction to obtain an artifact-free global image. The artifact-free global image is an artifact-free global image that retains the key information of the region of interest (ROI) of the internal scan image and restores the detailed features of the non-region of interest (ROI) of the prior image.

2. The ultra-low dose CT imaging method according to claim 1, wherein The training process of the artifact correction model is as follows: A1. Obtain the CT puncture image during the prior operation process, and segment the puncture needle mask in the CT puncture image to correspondingly obtain a puncture needle mask library composed of a plurality of puncture needle masks; A2. Obtain multiple groups of training CT images. The training CT images do not contain puncture needles, and each level of training CT images is a multi-sequence image. Randomly select one image from each group of training CT images as the prior image, and in the same group of training CT images, select one of the images in the sequence adjacent to the prior image as the simulation image. Finally, randomly select a puncture needle mask from the puncture needle mask library obtained in A1 to simulate the simulation image, and obtain a paired data pair. The paired data pair consists of a local image X with artifacts baseline and the paired global image X without artifacts target . The global image X without artifacts target is used as the reference image. Among them, the local image X with artifacts baseline is the internal scan image with a puncture needle and artifacts, and the global image X without artifacts target is the global scan image with a puncture needle and without artifacts; A3. Input the prior image X of A2 prior and the local image X with artifacts baseline into the artifact correction model to correct the cupping artifact and metal artifact, and obtain the target artifact-free image A4. Use the loss function Loss to process the target de-artifacted image in A3 and the artifact-free global image X in A2 target constrain them using mean squared error loss and structural similarity loss to obtain a loss value; A5. According to the loss value of A4, use the gradient descent method to optimize the parameters of the artifact correction model , and then determine whether the training end condition is reached. If not, return to A3; if yes, enter A6; A6. Define the current artifact correction model as the trained artifact correction model.

3. The ultra-low dose CT imaging method according to claim 2, wherein: The A1 is specifically pre-obtained multi-sequence CT puncture images of different objects at different times during the puncture needle process. The CT puncture images of the same object include an initial image X without puncture ref and multiple subsequent images X during the operation process punct , where the initial image X ref does not contain a puncture needle, and the subsequent image X punct contains a puncture needle. Then, according to the initial image X ref , the puncture needle masks in all subsequent images X punct are correspondingly segmented, and a puncture needle mask library composed of multiple puncture needle masks is correspondingly obtained.

4. The ultra-low dose CT imaging method according to claim 2, wherein: In the A1, the initial image X ref and the subsequent image X punct The residual image between them is threshold segmented, and then the puncture needle is binarized to obtain the puncture needle mask.

5. The ultra-low dose CT imaging method according to claim 2, wherein: The A3 specifically refers to inputting the prior image X of A2 prior and the local image X with artifacts baseline into the artifact correction model The encoder of the artifact correction model extracts the prior feature information in the prior image X prior , and then the decoder of the artifact correction model repairs the missing part of the local image X with artifacts baseline to finally obtain the target artifact-free image 6. The ultra-low dose CT imaging method according to claim 2, wherein: The loss function Loss is represented by Equation (1), where, l MSE is the mean squared error loss, l SSIM is the structural similarity loss, α is a hyperparameter that balances the mean squared error loss and the structural similarity loss components, is the loss value.

7. The ultra-low dose CT imaging method according to claim 2, characterized in that: The target artifact-free image is represented by Equation (2), Among them, are the learnable parameters of the artifact correction model.

8. The ultra-low dose CT imaging method according to claim 2, wherein: The training end condition is to reach the maximum number of training epochs; The artifact correction model is a convolutional neural network.

9. The ultra-low dose CT imaging method according to claim 2, wherein: In A2, paired data pairs are obtained by simulating the simulation image through the XCIST simulation toolkit.

10. Application of the ultra-low dose CT imaging method according to any one of claims 1 to 9 in puncture path guidance.

Images