Training methods, systems, application methods and systems of dual-energy CT imaging models

Abstract

This invention relates to a training method, system, application method, and system for a dual-energy CT imaging model, belonging to the fields of neural networks and medical imaging. The method includes: constructing a dual-energy CT imaging model; acquiring low-energy CT images and corresponding high-energy CT images, as well as a synthesized weight map; pre-training a weight map synthesis sub-model using the low-energy CT images as input and the synthesized weight map as the target; inputting the low-energy CT images into the dual-energy CT imaging model and training the weights of the dual-energy CT imaging sub-model using the high-energy CT images as the target, thereby obtaining a trained dual-energy CT imaging sub-model. This invention introduces enhanced attention gating into the dual-energy CT imaging model, which improves the model's attention to regions with significant differences between the high- and low-energy CT images in the synthesized weight map, thus enhancing the imaging accuracy of the dual-energy CT imaging model.

Description

Technical Field

[0001] This invention relates to the fields of neural networks and medical imaging, and in particular to a training method, system, application method, and system for a dual-energy CT imaging model. Background Technology

[0002] Computed tomography (CT) is one of the most widely used medical imaging modalities in clinical practice and is an indispensable part of disease diagnosis and image-guided radiotherapy.

[0003] Current research indicates that it is feasible to achieve dual-energy CT imaging using deep learning based on single-energy CT images. The earliest method for achieving dual-energy CT imaging was based on a cascaded neural network. Although this method can obtain dual-energy CT images from single-energy CT images, the imaging model has low accuracy, resulting in inaccurate dual-energy CT images. Therefore, improving the accuracy of dual-energy CT imaging is a pressing issue that needs to be addressed. Summary of the Invention

[0004] The purpose of this invention is to provide a training method, system, application method, and system for a dual-energy CT imaging model, which can effectively improve the accuracy of dual-energy CT imaging and obtain more accurate and reliable dual-energy CT images.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] On the one hand, this invention proposes a training method for a dual-energy CT imaging model, the training method comprising:

[0007] A dual-energy CT imaging model is constructed, comprising a weighted map synthesis sub-model and a dual-energy CT imaging sub-model. The weighted map synthesis sub-model includes a first encoder and a first decoder. The dual-energy CT imaging sub-model includes a second encoder, a second decoder, and an enhanced attention gating. The first decoder is connected to the second decoder through the enhanced attention gating.

[0008] Acquire low-energy CT images and corresponding high-energy CT images, as well as a composite weight map, wherein the composite weight map is obtained by synthesizing the low-energy CT images and the high-energy CT images;

[0009] Using the low-energy CT image as input and the synthesized weight map as the target, the weight map synthesis sub-model is pre-trained to obtain a trained weight map synthesis sub-model.

[0010] The low-energy CT images are input into the first encoder and the second encoder of the dual-energy CT imaging model, respectively. The weights of the dual-energy CT imaging sub-model are trained using the high-energy CT images as the target, and a trained dual-energy CT imaging sub-model is obtained.

[0011] Optionally, the weighted graph synthesis sub-model and / or the dual-energy CT imaging sub-model employs a U-Net network.

[0012] Optionally, a composite weight map is obtained by synthesizing the low-energy CT image and the high-energy CT image, specifically including:

[0013] The difference between the low-energy CT image and the high-energy CT image is obtained to obtain a high-low energy difference image.

[0014] The high-low energy difference image is subjected to hard thresholding and Gaussian filtering in sequence to obtain the noise-reduced high-low energy difference image.

[0015] The high- and low-energy difference image after noise reduction is standardized to obtain the synthetic weight map.

[0016] Optionally, the step of inputting the low-energy CT image into the first encoder and the second encoder of the dual-energy CT imaging model, respectively, and training the weights of the dual-energy CT imaging sub-model using the high-energy CT image as the target, to obtain a trained dual-energy CT imaging sub-model, specifically includes:

[0017] The low-energy CT image is input into the first encoder and the second encoder;

[0018] The outputs of the second encoder and the first decoder are simultaneously input into the enhanced attention gating;

[0019] The output of the enhanced attention gating is input into the second decoder;

[0020] The loss function is calculated based on the output of the second decoder and the high-energy CT image, and the weights of the dual-energy CT imaging sub-model are adjusted based on the loss function to obtain the trained dual-energy CT imaging sub-model.

[0021] Optionally, when training the weighted graph synthesis sub-model and the dual-energy CT imaging sub-model, the mean square error function is used as the loss function, and the adaptive moment estimation algorithm is used to optimize the model network parameters.

[0022] On the other hand, the present invention also proposes an application method for a dual-energy CT imaging model, the application method comprising:

[0023] Acquire low-energy CT images of the target;

[0024] The target low-energy CT image is input into the dual-energy CT imaging model to obtain the predicted target high-energy CT image; the dual-energy CT imaging model is a dual-energy CT imaging model trained by the training method described above.

[0025] On the other hand, the present invention also proposes a training system for a dual-energy CT imaging model, the training system comprising:

[0026] A dual-energy CT imaging model construction module is used to construct a dual-energy CT imaging model. The dual-energy CT imaging model includes a weighted map synthesis sub-model and a dual-energy CT imaging sub-model. The weighted map synthesis sub-model includes a first encoder and a first decoder. The dual-energy CT imaging sub-model includes a second encoder, a second decoder, and an enhanced attention gating. The first decoder is connected to the second decoder through the enhanced attention gating.

[0027] The image acquisition module is used to acquire low-energy CT images and corresponding high-energy CT images, as well as a composite weight map, wherein the composite weight map is obtained by synthesizing the low-energy CT images and the high-energy CT images;

[0028] The pre-training module for the weighted graph synthesis sub-model is used to pre-train the weighted graph synthesis sub-model with the low-energy CT image as input and the synthesized weighted graph as the target, so as to obtain a trained weighted graph synthesis sub-model.

[0029] The dual-energy CT imaging sub-model training module is used to input the low-energy CT image into the first encoder and the second encoder of the dual-energy CT imaging model respectively, and to train the weights of the dual-energy CT imaging sub-model with the high-energy CT image as the target, so as to obtain the trained dual-energy CT imaging sub-model.

[0030] Optionally, the weighted graph synthesis sub-model and / or the dual-energy CT imaging sub-model employs a U-Net network.

[0031] Optionally, the training system further includes:

[0032] The difference unit is used to subtract the low-energy CT image and the high-energy CT image to obtain a high-low energy difference image.

[0033] The noise reduction unit is used to perform hard thresholding and Gaussian filtering on the high-low energy difference image in sequence to obtain the noise-reduced high-low energy difference image.

[0034] The standardization unit is used to standardize the denoised high-low energy difference image to obtain the synthetic weight map.

[0035] On the other hand, the present invention also proposes an application system for a dual-energy CT imaging model, the application system comprising:

[0036] The target image acquisition module is used to acquire low-energy CT images of the target.

[0037] The target image prediction module is used to input the target low-energy CT image into the dual-energy CT imaging model to obtain the predicted target high-energy CT image; the dual-energy CT imaging model is a dual-energy CT imaging model trained by the training method of the dual-energy CT imaging model described above.

[0038] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0039] This invention provides a training method, system, and application method for a dual-energy CT imaging model. The training method includes: constructing a dual-energy CT imaging model, wherein the dual-energy CT imaging model includes a weight map synthesis sub-model and a dual-energy CT imaging sub-model, the weight map synthesis sub-model including a first encoder and a first decoder; the dual-energy CT imaging sub-model including a second encoder, a second decoder, and an enhanced attention gating; the first decoder being connected to the second decoder through the enhanced attention gating; acquiring low-energy CT images and corresponding high-energy CT images, as well as a synthesized weight map, wherein the synthesized weight map is obtained by synthesizing the low-energy CT images and the high-energy CT images; pre-training the weight map synthesis sub-model using the low-energy CT images as input and the synthesized weight map as the target, to obtain a trained weight map synthesis sub-model; inputting the low-energy CT images into the first encoder and the second encoder of the dual-energy CT imaging model respectively, and training the weights of the dual-energy CT imaging sub-model using the high-energy CT images as the target, to obtain a trained dual-energy CT imaging sub-model, thereby obtaining a trained dual-energy CT imaging model.

[0040] This invention not only enables dual-energy CT imaging based on low-energy CT images, but also introduces enhanced attention gating into the dual-energy CT imaging model. By applying an attention mechanism to dual-energy CT imaging technology, and with enhanced attention gating between the first decoder of the weight map synthesis sub-model and the second decoder of the dual-energy CT imaging sub-model, the synthesized weight map is input into the enhanced attention gating of the dual-energy CT imaging sub-model. This enhanced attention gating improves the dual-energy CT imaging model's attention to areas with significant differences between high-energy and low-energy CT images in the synthesized weight map, thereby improving the imaging accuracy of the dual-energy CT imaging model in these important areas and obtaining more accurate and reliable dual-energy CT images. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. The following drawings are not intentionally drawn to scale to actual size; their focus is on illustrating the main points of the present invention.

[0042] Figure 1 A flowchart of the training method for the dual-energy CT imaging model provided in Embodiment 1 of the present invention;

[0043] Figure 2 This is an overall structural diagram of the dual-energy CT imaging model provided in Embodiment 1 of the present invention;

[0044] Figure 3 This is a flowchart of the weight graph synthesis module provided in Embodiment 1 of the present invention;

[0045] Figure 4 This is a diagram illustrating the operation mechanism of the enhanced attention gating provided in Embodiment 1 of the present invention;

[0046] Figure 5 A flowchart of the application method of the dual-energy CT imaging model provided in Embodiment 2 of the present invention;

[0047] Figure 6 This is a schematic diagram of the structure of the training system for the dual-energy CT imaging model provided in Embodiment 3 of the present invention;

[0048] Figure 7 This is a schematic diagram of the application system of the dual-energy CT imaging model provided in Embodiment 4 of the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] As indicated in this invention and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0051] While this invention makes various references to certain modules in systems according to embodiments of the invention, any number of different modules can be used and run on user terminals and / or servers. The modules are merely illustrative, and different aspects of the systems and methods may use different modules.

[0052] This invention uses flowcharts to illustrate the operations performed by the system according to embodiments of the invention. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously, as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0053] Computed tomography (CT) is one of the most widely used medical imaging modalities in clinical practice, and it is an indispensable part of disease diagnosis and image-guided radiotherapy. However, due to the limitations imposed by the attenuation characteristics of X-rays in human tissues, different tissues may correspond to the same CT value in single-energy CT images. Therefore, clinically used single-energy CT cannot effectively distinguish different tissues or quantitatively characterize tissue components.

[0054] The emergence of dual-energy CT has filled this gap. It utilizes two sets of X-rays with different energy spectra to obtain two sets of attenuation information. Image reconstruction using this attenuation information yields both high-energy and low-energy CT images. Furthermore, material decomposition algorithms are used to obtain the desired base material images, enabling the differentiation or quantitative characterization of tissue components. Therefore, dual-energy CT has a wide range of applications, such as virtual monochrome imaging, bone removal imaging, blood perfusion quantification, kidney cyst diagnosis, and real-time online monitoring of proton therapy, greatly expanding the application scope of CT imaging. However, compared to traditional single-energy CT, advanced clinical dual-energy CT technology is more complex and costly. Currently, the dual-energy CT market is largely monopolized by medical device giants in developed countries (such as Siemens, GE, and Philips). Therefore, developing new dual-energy CT imaging methods and equipment is beneficial for rapidly promoting the independent development of high-end medical devices and improving the level of self-sufficiency in this field.

[0055] Deep learning is an artificial intelligence technology based on deep neural networks, capable of learning representations from data. Supported by large datasets of medical images, deep learning is widely applied in medical image processing tasks, such as image segmentation, registration, and denoising, and also provides a new approach for dual-energy CT imaging. In recent years, research has shown that it is feasible to achieve dual-energy CT imaging using deep learning based on single-energy CT images. Among these, the cascaded neural network-based dual-energy CT imaging method was the earliest research to utilize deep learning for dual-energy CT imaging. This method trains a cascaded neural network using clinical dual-energy CT image data (each set of dual-energy CT images includes a low-energy CT image and its corresponding high-energy CT image). Inputting the low-energy CT image into the trained network model yields the corresponding high-energy CT image, thus achieving dual-energy CT imaging based on single-energy CT images. However, this method has a simple model and limited data, resulting in inaccurate high-energy CT images that cannot meet the demand for precise high-energy CT images.

[0056] Based on this, the purpose of this invention is to provide a training method, system, application method and system for a dual-energy CT imaging model, which can effectively improve the accuracy of dual-energy CT imaging and obtain more accurate and reliable dual-energy CT images.

[0057] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] Example 1

[0059] like Figure 1 As shown, this embodiment provides a training method for a dual-energy CT imaging model, the training method including:

[0060] Step S1: Construct a dual-energy CT imaging model.

[0061] In this embodiment, the dual-energy CT imaging model includes a weighted map synthesis sub-model and a dual-energy CT imaging sub-model. The weighted map synthesis sub-model includes a first encoder and a first decoder. The dual-energy CT imaging sub-model includes a second encoder, a second decoder, and an enhanced attention gating. The first decoder is connected to the second decoder through the enhanced attention gating.

[0062] like Figure 2As shown, the dual-energy CT imaging model in this embodiment consists of two sub-modules: a weighted map synthesis sub-model and a dual-energy CT imaging sub-model. The weighted map synthesis sub-model can be called a weighted map generator. The weighted map synthesis module in the weighted map generator is used to generate a synthesized weighted map based on low-energy CT images and high-energy CT images. The synthesized weighted map represents the degree of difference in CT values ​​between the two energies of the high-energy and low-energy CT images. The dual-energy CT imaging sub-model, i.e., the dual-energy CT imaging module, is mainly used to predict high-energy CT images based on low-energy CT images.

[0063] In this embodiment, the weighted graph synthesis sub-model and / or the dual-energy CT imaging sub-model can use a U-Net network or other similar neural networks.

[0064] In this embodiment, both the first decoder and the second decoder can be four-layer structures, with each layer in the first decoder and each layer in the second decoder connected in a one-to-one correspondence. There are also four enhanced attention gates. For each layer in the first decoder in the weighted graph synthesis sub-model, a corresponding enhanced attention gate is set in the dual-energy CT imaging sub-model, so that each layer in the first decoder is connected to the corresponding layer in the second decoder after passing through a corresponding enhanced attention gate. This allows the output of each layer in the first decoder in the weighted graph synthesis sub-model to first be input to the corresponding enhanced attention gate, and then reach the corresponding layer in the second decoder.

[0065] Step S2: Obtain low-energy CT images and corresponding high-energy CT images, as well as synthesized weight maps.

[0066] It should be noted that the synthesized weight map in this embodiment is obtained by synthesizing the low-energy CT image and the corresponding high-energy CT image through the weight map synthesis module, or it can be a synthesized weight map obtained by other means.

[0067] like Figure 3 As shown, this embodiment uses a weighted map synthesis module to synthesize low-energy CT images and high-energy CT images to obtain a synthesized weighted map, specifically including:

[0068] The difference between the low-energy CT image and the high-energy CT image is obtained to obtain a high-low energy difference image.

[0069] The high-low energy difference image is subjected to hard thresholding and Gaussian filtering in sequence to obtain the noise-reduced high-low energy difference image.

[0070] The high- and low-energy difference image after noise reduction is standardized to obtain the synthetic weight map.

[0071] This invention uses dual-energy CT image data collected clinically in a tertiary hospital equipped with dual-energy CT imaging equipment to synthesize a target synthetic weight map. First, an absolute difference map is obtained by subtracting the low-energy CT image and the high-energy CT image. Then, image noise interference is eliminated through two steps: hard thresholding and Gaussian filtering. Finally, the denoised absolute difference map is standardized to obtain the target synthetic weight map. In this embodiment, the numerical range of the synthetic weight map is 0 to 1. The larger the value, the greater the difference between the high-energy and low-energy CT images at that pixel, i.e., the higher the attention required. Then, the synthetic weight map is used as the target, and the low-energy CT image is used as the input to pre-train the weight map synthesis sub-model. Since the weight map synthesis sub-model adopts the classic U-Net structure, consisting of a first encoder and a first decoder, after the low-energy CT image is input into the trained weight map synthesis sub-model, the four-layer structure of the first decoder obtains four synthetic weight maps of different sizes.

[0072] This invention synthesizes a weighted map from low-energy CT images and high-energy CT images. After preprocessing such as hard thresholding and Gaussian filtering, a denoised synthesized weighted map is obtained, which effectively eliminates the interference of CT image noise on the training effect during the training process and improves the model training accuracy and dual-energy CT imaging accuracy.

[0073] Step S3: Using the low-energy CT image as input and the synthesized weight map as the target, pre-train the weight map synthesis sub-model to obtain the trained weight map synthesis sub-model.

[0074] Step S4: Input the low-energy CT image into the first encoder and the second encoder of the dual-energy CT imaging model, respectively. Using the high-energy CT image as the target, train the weights of the dual-energy CT imaging sub-model to obtain a trained dual-energy CT imaging sub-model. Specifically, this includes:

[0075] Step S4.1: Input the low-energy CT image into the first encoder and the second encoder.

[0076] Step S4.2: Simultaneously input the output of the second encoder and the output of the first decoder into the enhanced attention gating.

[0077] In this embodiment, the enhanced attention gating mechanism operates as follows: Figure 4As shown, the enhanced attention gating receives synthetic weight maps of different sizes output by the first decoder and low-energy CT feature maps output by the second encoder. The synthetic weight maps are then processed by a linear activation ReLU function, a convolutional layer, and a sigmoid function, and then merged with the low-energy CT feature maps to obtain a weighted feature map. The weighted feature map represents a low-energy CT image with weighted differences in CT values ​​for high and low energies. This enables the dual-energy CT imaging model to predict high-energy CT images based on low-energy CT images. Furthermore, because the enhanced attention gating during training makes the dual-energy CT imaging model pay more attention to regions with significant differences between high and low-energy CT images, its predicted high-energy CT images are more accurate and reliable.

[0078] Step S4.3: Input the output of the enhanced attention gating into the second decoder.

[0079] Step S4.4: Calculate the loss function based on the output of the second decoder and the high-energy CT image, and adjust the weights of the dual-energy CT imaging sub-model based on the loss function to obtain the trained dual-energy CT imaging sub-model, thereby obtaining the trained dual-energy CT imaging model.

[0080] In this embodiment, when training the weighted graph synthesis sub-model and the dual-energy CT imaging sub-model, the mean square error function is used as the loss function, and the adaptive moment estimation algorithm is used to optimize the model network parameters.

[0081] Since the weighted graph synthesis sub-model has been pre-trained, when training the dual-energy CT imaging sub-model, high-energy and low-energy CT images are used as the target and input, respectively. During the training process, the low-energy CT image is processed by the weighted graph synthesis sub-model to obtain four different sizes of synthesized weighted graphs. These synthesized weighted graphs are input from the first decoder to the corresponding enhanced attention gating, guiding the training process of the dual-energy CT imaging sub-model.

[0082] Since the dual-energy CT imaging sub-model in this embodiment also uses the classic U-Net neural network, four enhanced attention gates control different layers of the second decoder of the dual-energy CT imaging sub-model. After the first decoder of the weight map synthesis sub-model outputs four synthesized weight maps of different sizes, these four synthesized weight maps are respectively input into the corresponding enhanced attention gates. Through the enhanced attention gates acting on different layers of the second decoder of the dual-energy CT imaging sub-model, the dual-energy CT imaging sub-model increases its attention to areas with significant differences in high-energy and low-energy CT images. At the same time, considering that the mean squared error (MSE) can be used to calculate the similarity between the high-energy CT image output by the network and the real high-energy CT image, this embodiment uses MSE as the loss function during training and uses an adaptive moment estimation algorithm to fine-tune the network parameters, ultimately enabling the trained dual-energy CT imaging model to generate high-energy CT images with clinical accuracy.

[0083] like Figure 2 As shown, this invention mainly includes a weight map generator, an enhanced attention gating system, and a dual-energy CT imaging module. The weight map generator is also known as a weight map synthesizer, which is the weight map synthesis sub-model in this invention. The enhanced attention gating system and the dual-energy CT imaging module together form the dual-energy CT imaging sub-model. The weight map synthesis sub-model and the dual-energy CT imaging sub-model together form a complete dual-energy CT imaging model. First, this embodiment utilizes clinically acquired dual-energy CT image data and obtains a synthesized weight map through the weight map synthesis module. The values ​​of the synthesized weight map represent the degree of difference in CT values ​​between high-energy and low-energy CT images. Then, using a low-energy CT image as input and the synthesized weight map as the target, the weight map synthesis sub-model, i.e., the weight map generator, is pre-trained. The pre-trained weight map synthesis sub-model can automatically generate a synthesized weight map from low-energy image features to characterize the differences in dual-energy CT images. At this point, a dual-energy CT imaging sub-model with enhanced attention gating was trained using clinical dual-energy CT image data. During training, low-energy CT images were input into the trained weighted image synthesis sub-model and the dual-energy CT imaging sub-model. The trained weighted image synthesis sub-model outputs a synthesized weighted image, which is then input into the enhanced attention gating via the first decoder. This allows the weighted image synthesis sub-model to guide the dual-energy CT imaging sub-model to focus on tissue regions with significant differences in CT values ​​between high-energy and low-energy CT images through enhanced attention gating, thus improving the accuracy of predicting high-energy CT images. Furthermore, the process of synthesizing the weighted image employs noise reduction methods such as hard thresholding and Gaussian filtering, thereby eliminating the interference of CT noise during model training and further improving the accuracy of the dual-energy CT imaging model in predicting high-energy CT images.

[0084] This invention enables dual-energy CT imaging using neural network technology based on low-energy CT images. This imaging method solves the problems of dual-energy CT being a high-end medical imaging device monopolized by foreign manufacturers, and its high price and difficulty in widespread adoption. Furthermore, this invention introduces enhanced attention gating into the dual-energy CT imaging model, applying an attention mechanism to dual-energy CT imaging technology. Because enhanced attention gating is set between the first decoder of the weight map synthesis sub-model and the second decoder of the dual-energy CT imaging sub-model, the synthesized weight map is input into the enhanced attention gating of the dual-energy CT imaging sub-model. This enhanced attention gating improves the dual-energy CT imaging model's attention to areas with significant differences between high- and low-energy CT images in the synthesized weight map, thereby improving the imaging accuracy of the dual-energy CT imaging model in these important areas. Simultaneously, it effectively supervises the enhanced attention gating using only low-energy CT image features, eliminating the interference of CT image noise during training and obtaining more accurate and reliable dual-energy CT images.

[0085] Example 2

[0086] like Figure 5 As shown, corresponding to the training method of the dual-energy CT imaging model in Example 1, this example proposes an application method for the dual-energy CT imaging model, which includes:

[0087] Step A1: Acquire the target low-energy CT image.

[0088] Step A2: Input the target low-energy CT image into the dual-energy CT imaging model to obtain the predicted target high-energy CT image; the dual-energy CT imaging model is a dual-energy CT imaging model trained by the training method of the dual-energy CT imaging model described in Example 1. The training process is the same as in Example 1, and will not be repeated here.

[0089] Example 3

[0090] like Figure 6 As shown, corresponding to the training method of the dual-energy CT imaging model in Example 1, this embodiment proposes a training system for a dual-energy CT imaging model. The training system adopts the training method of the dual-energy CT imaging model in Example 1, and the training method is exactly the same as that in Example 1. The training system includes:

[0091] The dual-energy CT imaging model construction module M1 is used to construct a dual-energy CT imaging model, which includes a weighted map synthesis sub-model and a dual-energy CT imaging sub-model. The weighted map synthesis sub-model includes a first encoder and a first decoder. The dual-energy CT imaging sub-model includes a second encoder, a second decoder, and an enhanced attention gating. The first decoder is connected to the second decoder through the enhanced attention gating.

[0092] In this embodiment, the weighted graph synthesis sub-model and / or the dual-energy CT imaging sub-model can use a U-Net network or other similar neural networks, which can be determined according to actual needs.

[0093] The image acquisition module M2 is used to acquire low-energy CT images and corresponding high-energy CT images, as well as a composite weight map, wherein the composite weight map is obtained by synthesizing the low-energy CT images and the high-energy CT images.

[0094] The weight map can be obtained through the weight map synthesis module in the weight map synthesis sub-model. The weight map synthesis module includes:

[0095] The difference unit is used to subtract the low-energy CT image from the high-energy CT image to obtain a high-low energy difference image.

[0096] The noise reduction unit is used to perform hard thresholding and Gaussian filtering on the high-low energy difference image in sequence to obtain the noise-reduced high-low energy difference image.

[0097] The standardization unit is used to standardize the denoised high-low energy difference image to obtain the synthetic weight map.

[0098] The weighted graph synthesis sub-model pre-training module M3 is used to pre-train the weighted graph synthesis sub-model with the low-energy CT image as input and the synthesized weighted graph as the target, so as to obtain the trained weighted graph synthesis sub-model.

[0099] The dual-energy CT imaging sub-model training module M4 is used to input the low-energy CT image into the first encoder and the second encoder of the dual-energy CT imaging model respectively, and to train the weights of the dual-energy CT imaging sub-model with the high-energy CT image as the target, so as to obtain the trained dual-energy CT imaging sub-model.

[0100] Example 4

[0101] like Figure 7 As shown, corresponding to the application method of the dual-energy CT imaging model in Example 2, this example proposes an application system for the dual-energy CT imaging model, the application system including:

[0102] The target image acquisition module N1 is used to acquire low-energy CT images of the target.

[0103] The target image prediction module N2 is used to input the target low-energy CT image into the dual-energy CT imaging model to obtain the predicted target high-energy CT image. The dual-energy CT imaging model is a dual-energy CT imaging model trained by the training method of the dual-energy CT imaging model described in Example 1. The training process is the same as in Example 1 and will not be repeated here.

[0104] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in a common dictionary shall be interpreted as having a meaning consistent with their meaning in the context of the relevant art, and not as having an idealized or highly formalized meaning, unless expressly defined herein.

[0105] The foregoing description is illustrative of the invention and should not be construed as limiting it. Although several exemplary embodiments of the invention have been described, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the invention. Therefore, all such modifications are intended to be included within the scope of the invention as defined in the claims. It should be understood that the foregoing description is illustrative of the invention and should not be construed as limiting it to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The invention is defined by the claims and their equivalents.

Claims

1. A method for training a dual-energy CT imaging model, characterized in that, The training method comprises: A dual-energy CT imaging model is constructed, the dual-energy CT imaging model comprising a weight map synthesis submodel and a dual-energy CT imaging submodel, the weight map synthesis submodel comprising a first encoder and a first decoder; the dual-energy CT imaging submodel comprising a second encoder, a second decoder and an enhanced attention gate; the first decoder is connected with the second decoder through the enhanced attention gate; A low-energy CT image and a corresponding high-energy CT image and a synthesis weight map are obtained, the synthesis weight map being obtained by synthesizing the low-energy CT image and the high-energy CT image; wherein the synthesis weight map obtained by synthesizing the low-energy CT image and the high-energy CT image specifically comprises: the low-energy CT image and the high-energy CT image are subtracted to obtain a high-low energy difference image; the high-low energy difference image is sequentially subjected to hard threshold processing and Gaussian filtering processing to obtain a denoised high-low energy difference image; the denoised high-low energy difference image is subjected to standardization processing to obtain the synthesis weight map; The low-energy CT image is taken as input, and the synthesis weight map is taken as target to pre-train the weight map synthesis submodel to obtain a trained weight map synthesis submodel; The low-energy CT image is input into the first encoder and the second encoder of the dual-energy CT imaging model respectively, and the high-energy CT image is taken as target to train the weight of the dual-energy CT imaging submodel to obtain a trained dual-energy CT imaging submodel.

2. The training method of claim 1, wherein, The weight map synthesis submodel and / or the dual-energy CT imaging submodel adopts a U-Net network.

3. The training method of claim 1, wherein, The low-energy CT image is input into the first encoder and the second encoder of the dual-energy CT imaging model respectively, and the high-energy CT image is taken as target to train the weight of the dual-energy CT imaging submodel to obtain a trained dual-energy CT imaging submodel, specifically comprising: The low-energy CT image is input into the first encoder and the second encoder; The output of the second encoder and the output of the first decoder are simultaneously input into the enhanced attention gate; The output of the enhanced attention gate is input into the second decoder; A loss function is calculated according to the output of the second decoder and the high-energy CT image, and the weight of the dual-energy CT imaging submodel is adjusted based on the loss function to obtain the trained dual-energy CT imaging submodel.

4. The training method of claim 1, wherein, When training the weight map synthesis submodel and the dual-energy CT imaging submodel, a mean square error function is adopted as a loss function, and a self-adaptive moment estimation algorithm is adopted to optimize the network parameters of the model.

5. A method for applying a dual-energy CT imaging model, characterized in that, The application method comprises: A target low-energy CT image is obtained; The target low-energy CT image is input into a dual-energy CT imaging model to obtain a predicted target high-energy CT image; the dual-energy CT imaging model is a dual-energy CT imaging model trained by the training method according to any one of claims 1-4. 6.A training system of a dual-energy CT imaging model, characterized in that, The training system comprises: The dual-energy CT imaging model construction module is configured to construct a dual-energy CT imaging model, the dual-energy CT imaging model comprising a weight map synthesis sub-model and a dual-energy CT imaging sub-model, the weight map synthesis sub-model comprising a first encoder and a first decoder; the dual-energy CT imaging sub-model comprising a second encoder, a second decoder and an enhanced attention gate; and the first decoder being connected to the second decoder through the enhanced attention gate. The image acquisition module is configured to acquire a low-energy CT image and a corresponding high-energy CT image and a synthesis weight map, the synthesis weight map being obtained by synthesizing the low-energy CT image and the high-energy CT image; wherein the training system further comprises: a difference unit configured to obtain a high-low energy difference image by differencing the low-energy CT image and the high-energy CT image; a noise reduction unit configured to obtain a denoised high-low energy difference image by sequentially performing hard threshold processing and Gaussian filtering processing on the high-low energy difference image; and a standardization unit configured to obtain the synthesis weight map by performing standardization processing on the denoised high-low energy difference image. The weight map synthesis sub-model pre-training module is configured to pre-train the weight map synthesis sub-model by taking the low-energy CT image as input and taking the synthesis weight map as target, to obtain a trained weight map synthesis sub-model. The dual-energy CT imaging sub-model training module is configured to input the low-energy CT image into the first encoder and the second encoder of the dual-energy CT imaging model respectively, to train the weight of the dual-energy CT imaging sub-model by taking the high-energy CT image as target, to obtain a trained dual-energy CT imaging sub-model.

7. The training system of claim 6, wherein, The weight map synthesis sub-model and / or the dual-energy CT imaging sub-model adopts a U-Net network.

8. An application system of a dual-energy CT imaging model, characterized in that, The application system comprises: a target image acquisition module configured to acquire a target low-energy CT image; a target image prediction module configured to input the target low-energy CT image into a dual-energy CT imaging model to obtain a predicted target high-energy CT image; and the dual-energy CT imaging model being a dual-energy CT imaging model trained according to the training method of a dual-energy CT imaging model according to any one of claims 1-4.

Images