A computer tomography reconstruction method and related apparatus

Abstract

The application discloses a computer tomography reconstruction method and a related device, relates to the technical field of image reconstruction, and comprises the following steps: acquiring original projection domain data; performing denoising processing on the original projection domain data to obtain denoised projection domain data; inputting the denoised projection domain data into a computer tomography reconstruction model to obtain computer tomography in an image domain output by the computer tomography reconstruction model; and performing enhancement on the computer tomography in the image domain output by the computer tomography reconstruction model to obtain final computer tomography. The application improves the efficiency and effect of computer tomography reconstruction based on denoising of projection domain data, a computer tomography reconstruction model and enhancement of a model result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image reconstruction, and in particular, to a computer tomography reconstruction method and related device. BACKGROUND

[0002] CT (Computed Tomography) is a medical imaging technology, in which SPECT (Single-Photon Emission Computed Tomography) and PET (Positron Emission Tomography) are two CT technologies in nuclear medicine, and are collectively referred to as ECT (Emission Computed Tomography) because they are both imaging of gamma rays emitted from the human body. Taking SPECT as an example, SPECT devices are a type of commonly used clinical imaging device in the field of nuclear medicine, which can be used for functional evaluation of organs such as thyroid, kidney, heart, brain, and lung, and diagnosis of related diseases, and also plays a significant role in the clinical diagnosis and staging of various malignant tumors. Compared with traditional gamma cameras, SPECT devices not only reflect the function and morphology of organs and tissues, but also stereoscopically observe the changes of organs and diseases, and can display the above information in different directions.

[0003] Traditional computer tomography reconstruction techniques include an iterative reconstruction-based method and a traditional filter back-projection reconstruction-based method. However, the reconstruction efficiency and effect of the traditional computer tomography reconstruction techniques need to be improved.

[0004] Therefore, how to provide a computer tomography reconstruction method to improve the efficiency and effect of computer tomography reconstruction has become a technical problem to be solved by those skilled in the art. SUMMARY

[0005] In view of the above problems, the present application provides a computer tomography reconstruction method and related device to achieve the purpose of improving the efficiency and effect of computer tomography reconstruction. The specific scheme is as follows:

[0006] The first aspect of the present application provides a computer tomography reconstruction method, comprising:

[0007] obtaining original projection domain data;

[0008] performing denoising processing on the original projection domain data to obtain denoising-processed projection domain data;

[0009] input the denoised projection domain data into a computer tomography reconstruction model to obtain a computer tomography in an image domain output by the computer tomography reconstruction model; the computer tomography reconstruction model is trained based on projection domain training data and computer tomography training data in the image domain;

[0010] enhance the computer tomography in the image domain output by the computer tomography reconstruction model to obtain a final computer tomography.

[0011] In a possible implementation, the denoising processing of the original projection domain data to obtain the denoised projection domain data comprises:

[0012] input the original projection domain data into a trained denoising model to obtain denoised projection domain data output by the denoising model; the denoising model is trained based on noisy projection domain data as training samples, denoised projection domain data as sample labels, and projection domain data output by the model approaching the denoised projection domain data as a training target.

[0013] In a possible implementation, the computer tomography reconstruction model is trained in the following manner:

[0014] determine training data, the training data comprising computer tomography training data in an image domain and projection domain training data corresponding to the computer tomography training data in the image domain;

[0015] obtain a pre-constructed generator and discriminator;

[0016] based on the projection domain training data and the computer tomography training data in the image domain, perform multiple iteration training on the generator and the discriminator to obtain a trained generator and a trained discriminator; the trained generator is the computer tomography reconstruction model.

[0017] In a possible implementation, in each iteration training process, based on the projection domain training data and the computer tomography training data in the image domain, the generator and the discriminator are trained, comprising:

[0018] use the projection domain training data as training samples, use the computer tomography training data in the image domain as sample labels, and only train the generator;

[0019] use the computer tomography training data in the image domain as training samples, use the projection domain training data as sample labels, and only train the discriminator;

[0020] Fix the parameters of the discriminator, take the projection domain training data as the training sample and the sample label, and retrain the generator.

[0021] In a possible implementation, the loss function for training the generator and the discriminator based on the projection domain training data and the computed tomography training data in the image domain in each iteration training process includes:

[0022] The generator loss, the discriminator loss, the supervision loss of the corresponding network layer, and the cascade network loss;

[0023] The generator loss is used to represent the similarity between the output of the generator and the computed tomography training data in the image domain when only the generator is trained with the projection domain training data as the training sample and the computed tomography training data in the image domain as the sample label.

[0024] The discriminator loss is used to represent the similarity between the output of the discriminator and the projection domain training data when only the discriminator is trained with the computed tomography training data in the image domain as the training sample and the projection domain training data as the sample label.

[0025] The cascade network loss is used to represent the similarity between the output of the discriminator and the projection domain training data when the generator is retrained with the parameters of the discriminator fixed and the projection domain training data as the training sample and the sample label.

[0026] The supervision loss of the corresponding network layer is used to represent the feature map similarity of the corresponding layer structure in the generator and the discriminator in each iteration training process.

[0027] In a possible implementation, the enhancement of the computed tomography in the image domain output by the computed tomography reconstruction model includes:

[0028] Inputting the computed tomography in the image domain output by the computed tomography reconstruction model into the trained enhancement model to obtain the enhanced computed tomography output by the enhancement model as the final computed tomography; the enhancement model is trained with the unenhanced computed tomography as the training sample, the enhanced computed tomography as the sample label, and the model output computed tomography approaching the enhanced computed tomography as the training target.

[0029] The second aspect of the present application provides a computed tomography reconstruction device, including:

[0030] An acquisition unit is configured to acquire original projection domain data;

[0031] a denoising processing unit, configured to perform denoising processing on the original projection domain data to obtain denoised projection domain data;

[0032] a reconstruction unit, configured to input the denoised projection domain data into a computer tomography reconstruction model to obtain a computer tomography in an image domain output by the computer tomography reconstruction model;

[0033] an enhancement unit, configured to enhance the computer tomography in the image domain output by the computer tomography reconstruction model to obtain a final computer tomography.

[0034] In a possible implementation, the denoising processing unit is specifically configured to:

[0035] input the original projection domain data into a trained denoising model to obtain denoised projection domain data output by the denoising model; the denoising model is trained by taking noisy projection domain data as training samples, taking denoised projection domain data as sample labels, and taking the projection domain data output by the model approaching the denoised projection domain data as a training target.

[0036] In a possible implementation, the apparatus further includes a computer tomography reconstruction model training unit, which includes:

[0037] a training data determination unit, configured to determine training data, the training data including computer tomography training data in an image domain and projection domain training data corresponding to the computer tomography training data in the image domain;

[0038] a generator and discriminator acquisition unit, configured to acquire a pre-constructed generator and discriminator;

[0039] a training unit, configured to perform multiple iteration training on the generator and the discriminator based on the projection domain training data and the computer tomography training data in the image domain to obtain a trained generator and a trained discriminator; the trained generator is the computer tomography reconstruction model.

[0040] In a possible implementation, in each iteration training process, the training unit is specifically configured to:

[0041] take the projection domain training data as training samples, take the computer tomography training data in the image domain as sample labels, and train only the generator;

[0042] take the computer tomography training data in the image domain as training samples, take the projection domain training data as sample labels, and train only the discriminator;

[0043] retraining the generator with the projection domain training data as training samples and sample labels.

[0044] In a possible implementation, the loss function in each iteration of the training process comprises:

[0045] a generator loss, a discriminator loss, a supervision loss of a corresponding network layer, and a cascade network loss.

[0046] The generator loss is used to represent the similarity between the output of the generator and the computed tomography training data of the image domain when only the generator is trained with the computed tomography training data of the image domain as sample labels and the projection domain training data as training samples.

[0047] The discriminator loss is used to represent the similarity between the output of the discriminator and the projection domain training data when only the discriminator is trained with the projection domain training data as sample labels and the computed tomography training data of the image domain as training samples.

[0048] The cascade network loss is used to represent the similarity between the output of the discriminator and the projection domain training data when the generator is retrained with the projection domain training data as training samples and sample labels and the parameters of the discriminator are fixed.

[0049] The supervision loss of the corresponding network layer is used to represent the similarity between the feature maps of the corresponding layer structures of the generator and the discriminator in each iteration of the training process.

[0050] In a possible implementation, the enhancement unit comprises:

[0051] The image domain computed tomography input output by the computed tomography reconstruction model is input into the trained enhancement model to obtain enhanced computed tomography output by the enhancement model as the final computed tomography. The enhancement model is trained with unenhanced computed tomography as training samples, enhanced computed tomography as sample labels, and the model output approaching the enhanced computed tomography as a training target.

[0052] The third aspect of the present application provides a computer program product comprising computer readable instructions that, when executed on an electronic device, cause the electronic device to implement the computed tomography reconstruction method of the first aspect or any implementation manner of the first aspect.

[0053] The fourth aspect of the present application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0054] The memory is configured to store a computer program.

[0055] The processor is configured to execute the computer program, so that the electronic device can implement the computer tomography reconstruction method of the first aspect or any implementation manner of the first aspect.

[0056] The fifth aspect of the present application provides a computer storage medium, the storage medium carries one or more computer programs, when the one or more computer programs are executed by an electronic device, the electronic device can implement the computer tomography reconstruction method of the first aspect or any implementation manner of the first aspect.

[0057] By the above technical solution, the computer tomography reconstruction method and related device provided by the present application comprises: obtaining original projection domain data; performing denoising processing on the original projection domain data to obtain denoised projection domain data; inputting the denoised projection domain data into a computer tomography reconstruction model to obtain a computer tomography in an image domain output by the computer tomography reconstruction model; and enhancing the computer tomography in the image domain output by the computer tomography reconstruction model to obtain a final computer tomography. The present application realizes the improvement of the efficiency and effect of computer tomography reconstruction based on the denoising of projection domain data, the computer tomography reconstruction model and the enhancement of model results. BRIEF DESCRIPTION OF DRAWINGS

[0058] The above and other features, advantages, and aspects of the present disclosure will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings in which:

[0059] Figure 1 A flowchart of a computer tomography reconstruction method provided by an embodiment of the present application;

[0060] Figure 2 A structure diagram of a denoising model example disclosed by an embodiment of the present application;

[0061] Figure 3 A structure diagram of a generator and discriminator example disclosed by an embodiment of the present application;

[0062] Figure 4 A structure diagram of an enhancement model example disclosed by an embodiment of the present application;

[0063] Figure 5A structural schematic diagram of a computer tomography reconstruction device provided by an embodiment of the present application is shown in FIG. 1.

[0064] Figure 6 A structural schematic diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 1. DETAILED DESCRIPTION

[0065] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application. The terms used in the embodiment part of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0066] The embodiments of the present application are described below in conjunction with the drawings. It is known to those of ordinary skill in the art that, as technology develops and new scenarios appear, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0067] The terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the terms used in this way can be interchanged under appropriate circumstances, and this is only a distinguishing way used in the description of the embodiments of the present application to describe the objects with the same attributes. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, so that the processes, methods, systems, products or devices containing a series of units do not have to be limited to those units, but can include other units not clearly listed or inherent to these processes, methods, products or devices.

[0068] CT (Computed Tomography, computed tomography) is a medical imaging technology, and SPECT (Single-Photon Emission Computed Tomography, single-photon emission computed tomography) and PET (Positron Emission Tomography, positron emission tomography) are two CT technologies in nuclear medicine, which are collectively referred to as ECT (Emission Computed Tomography, emission computed tomography) because they are both imaging of gamma rays emitted from the patient's body. Taking SPECT as an example, SPECT equipment is a commonly used clinical imaging device in the field of nuclear medicine, which can be used for functional evaluation and diagnosis of related diseases of thyroid, kidney, heart, brain, lung and other organs, and also plays a major role in the clinical diagnosis and staging of various malignant tumors. Compared with traditional gamma cameras, SPECT equipment not only reflects the function and morphology of organs and tissues, but also stereoscopically observes the changes of organs and diseases, and can display the above information in different directions.

[0069] Traditional computer tomography reconstruction techniques include an iterative reconstruction based method and a traditional filtered back projection reconstruction based method. However, the reconstruction efficiency and the reconstruction effect of the traditional computer tomography reconstruction techniques need to be improved.

[0070] To solve the above problems, the embodiments of the present application provide a computer tomography reconstruction method. The computer tomography reconstruction method of the embodiments of the present application will be described in detail below in combination with the drawings.

[0071] Reference Figure 1 , Figure 1 A flowchart of a computer tomography reconstruction method provided by the embodiments of the present application is shown in FIG. 1. The computer tomography reconstruction method provided by the embodiments of the present application can include steps S101-S104, which will be described in detail below. Figure 1

[0072] S101: Obtain original projection domain data;

[0073] In the present application, the original projection domain data is the projection domain data collected by a computer tomography device.

[0074] S102: Perform denoising processing on the original projection domain data to obtain denoised projection domain data;

[0075] S103: Input the denoised projection domain data into a computer tomography reconstruction model to obtain an image domain computer tomography output by the computer tomography reconstruction model; the computer tomography reconstruction model is trained based on projection domain training data and image domain computer tomography training data;

[0076] S104: Enhance the image domain computer tomography output by the computer tomography reconstruction model to obtain a final computer tomography.

[0077] In the present application, the final computer tomography can be applied to clinical analysis and processing based on computer tomography.

[0078] ​This embodiment provides a computed tomography (CT) reconstruction method, comprising: acquiring original projection domain data; denoising the original projection domain data to obtain denoised projection domain data; inputting the denoised projection domain data into a computed tomography reconstruction model to obtain a computed tomography image of the image domain output by the computed tomography reconstruction model; and enhancing the computed tomography image of the image domain output by the computed tomography reconstruction model to obtain the final computed tomography image. This application improves the efficiency and effectiveness of computed tomography reconstruction based on projection domain data denoising, a computed tomography reconstruction model, and the enhancement of the model results.

[0079] In one possible implementation, the step of denoising the original projection domain data to obtain denoised projection domain data includes:

[0080] The original projection domain data is input into the trained denoising model to obtain the denoised projection domain data output by the denoising model. The denoising model is trained using noisy projection domain data as training samples, denoised projection domain data as sample labels, and the training objective is for the projection domain data output by the model to approximate the denoised projection domain data.

[0081] In this application, a denoising model can be pre-constructed, which includes an encoding group and a decoding group, wherein the encoding group and the decoding group have a symmetrical structure, and the corresponding layer structures of the encoding group and the decoding group are skip-connected.

[0082] For ease of understanding, please refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of an example denoising model disclosed in an embodiment of this application, as shown below. Figure 2 As shown, the encoding group of this denoising model consists of two 32x (3x3) convolutional layers, two 64x (3x3) convolutional layers, two 128x (3x3) convolutional layers, two 256x (3x3) convolutional layers, two 512x (3x3) convolutional layers, and pooling layers. Figure 2 (Not shown in the image) The decoding group consists of a symmetrical structure, consisting of two 256x (3x3) convolutional layers, two 128x (3x3) convolutional layers, two 64x (3x3) convolutional layers, two 32x (3x3) convolutional layers, and an upsampling layer ( Figure 2 (not shown in the image), where corresponding layers of the encoding and decoding groups are fused, for example, two 32x(3x3) convolutional layers of the encoding and decoding groups are skipped.

[0083] After the de-noising model is constructed, the noisy projection domain data and the de-noised projection domain data corresponding to the noisy projection domain data can be determined. In a possible implementation, the image domain computer tomography generated in the clinic can be acquired first, the corresponding projection domain data is obtained by converting the image domain computer tomography through Radon inverse transform, noise (such as Poisson noise) is added to the projection domain data to generate noisy projection domain data, and the de-noised projection domain data corresponding to the noisy projection domain data is the projection domain data before the noise is added.

[0084] After the noisy projection domain data and the de-noised projection domain data corresponding to the noisy projection domain data are determined, the de-noising model can be trained with the noisy projection domain data as the training sample, the de-noised projection domain data as the sample label, and the projection domain data output by the model approaching the de-noised projection domain data as the training target, to obtain the trained de-noising model.

[0085] It should be noted that, in a possible implementation, when the de-noising model is trained with the noisy projection domain data as the training sample, the de-noised projection domain data as the sample label, and the projection domain data output by the model approaching the de-noised projection domain data as the training target, the loss function can be mean-square error (MSE), the optimizer can be Adam, and the evaluation index can be signal-to-noise ratio.

[0086] In a possible implementation, the training manner of the computer tomography reconstruction model includes the following steps:

[0087] S201: determining training data, the training data including image domain computer tomography training data and projection domain training data corresponding to the image domain computer tomography training data;

[0088] In the present application, the image domain computer tomography generated in the clinic can be acquired as the image domain computer tomography training data, and the corresponding projection domain data is obtained by converting the image domain computer tomography generated in the clinic through Radon inverse transform, which is the projection domain training data.

[0089] S202: acquiring a pre-constructed generator and discriminator;

[0090] In this application, both the generator and the discriminator can employ a U-Net network structure. The U-Net network structure used by the generator and the discriminator can be the same or different. In one possible implementation, the U-Net network structure used by the generator includes an encoder group and a decoder group, wherein the encoder and decoder groups are symmetrical, and corresponding layers of the encoder and decoder groups are skip-connected. The U-Net network structure used by the discriminator can also include an encoder group and a decoder group, wherein the encoder and decoder groups are symmetrical, and corresponding layers of the encoder and decoder groups are skip-connected.

[0091] For ease of understanding, please refer to Figure 3 , Figure 3 This is a schematic diagram illustrating the structure of an example generator and discriminator disclosed in an embodiment of this application.

[0092] like Figure 3 As shown, the generator's encoding group consists of 3 layers of 32x (3x3) convolutional layers, 3 layers of 64x (3x3) convolutional layers, 3 layers of 128x (3x3) convolutional layers, 3 layers of 256x (3x3) convolutional layers, 3 layers of 512x (3x3) convolutional layers, and pooling layers. Figure 3 (Not shown in the image) The generator's decoding group consists of a symmetrical structure, consisting of 3 layers of 256x (3x3) convolutional layers, 3 layers of 128x (3x3) convolutional layers, 3 layers of 64x (3x3) convolutional layers, 3 layers of 32x (3x3) convolutional layers, and an upsampling layer ( Figure 3 (not shown in the image), where the corresponding layers of the generator's encoding group and decoding group are channel-fused, for example, the generator's encoding group and decoding group's three 32x(3x3) convolutional layers are skip-connected.

[0093] The discriminator's encoding group consists of two 32x (3x3) convolutional layers, two 64x (3x3) convolutional layers, two 128x (3x3) convolutional layers, two 256x (3x3) convolutional layers, two 512x (3x3) convolutional layers, and pooling layers. Figure 3 (Not shown in the image) The discriminator's decoding group consists of a symmetrical structure, consisting of two 256x (3x3) convolutional layers, two 128x (3x3) convolutional layers, two 64x (3x3) convolutional layers, two 32x (3x3) convolutional layers, and an upsampling layer ( Figure 3 (not shown in the image), where the corresponding layers of the discriminator's encoding group and decoding group are channel-fused, for example, the two 32x(3x3) convolutional layers of the discriminator's encoding group and decoding group are skipped.

[0094] It should be noted that the generator is mainly used for computer tomography reconstruction of the projection domain data into the image domain, and the discriminator is mainly used for converting the image domain computer tomography generated by the generator into projection domain data.

[0095] S203: Based on the projection domain training data and the image domain computer tomography training data, the generator and the discriminator are trained for multiple iterations to obtain a trained generator and a trained discriminator; the trained generator is the computer tomography reconstruction model.

[0096] Specifically, in each iteration training process, the process of training the generator and the discriminator based on the projection domain training data and the image domain computer tomography training data includes the following steps:

[0097] S301: Taking the projection domain training data as the training sample and the image domain computer tomography training data as the sample label, only the generator is trained;

[0098] When the projection domain training data is taken as the training sample and the image domain computer tomography training data is taken as the sample label, only the generator is trained, the projection domain training data can be input into the generator to obtain the image domain computer tomography output by the generator, and the image domain computer tomography output by the generator is close to the image domain computer tomography training data as the training target, that is, the generator loss is minimized, and only the generator is trained.

[0099] S302: Taking the image domain computer tomography training data as the training sample and the projection domain training data as the sample label, only the discriminator is trained;

[0100] When the image domain computer tomography training data is taken as the training sample and the projection domain training data is taken as the sample label, only the discriminator is trained, the image domain computer tomography training data can be input into the discriminator to obtain the projection domain data output by the discriminator, and the projection domain data output by the discriminator is close to the projection domain training data as the training target, that is, the discriminator loss is minimized, and only the discriminator is trained.

[0101] S303: Fixing the parameters of the discriminator, taking the projection domain training data as the training sample and the sample label, and retraining the generator.

[0102] In the case of fixing the parameters of the discriminator, taking the projection domain training data as the training sample and the sample label, and retraining the generator, the projection domain training data can be input into the generator to obtain the computed tomography of the image domain output by the generator; and the computed tomography of the image domain output by the generator can be input into the discriminator to obtain the projection domain data output by the discriminator; and the training goal is that the projection domain data output by the discriminator approaches the projection domain training data, that is, the cascade network loss is minimized, and the generator is retrained.

[0103] In a possible implementation, in each iteration training process, the loss function for training the generator and the discriminator based on the projection domain training data and the computed tomography of the image domain training data includes:

[0104] a generator loss, a discriminator loss, a supervision loss of a corresponding network layer, and a cascade network loss;

[0105] The generator loss is used to represent the similarity between the output of the generator and the computed tomography of the image domain training data when only the generator is trained with the projection domain training data as the training sample and the computed tomography of the image domain training data as the sample label.

[0106] The discriminator loss is used to represent the similarity between the output of the discriminator and the projection domain training data when only the discriminator is trained with the computed tomography of the image domain training data as the training sample and the projection domain training data as the sample label.

[0107] The cascade network loss is used to represent the similarity between the output of the discriminator and the projection domain training data when the generator is retrained with the parameters of the discriminator fixed, the projection domain training data as the training sample and the sample label.

[0108] The supervision loss of the corresponding network layer is used to represent the similarity of the feature maps of the corresponding layer structure in the generator and the discriminator in each iteration training process.

[0109] It should be noted that in the U-Net network, there is a high similarity between the feature maps of the corresponding network layers, resulting in waste of computing resources. To solve this problem, the supervision loss of the corresponding network layer is added in each iteration training process in the present application, that is, the similarity of the feature maps of the corresponding layer structure in the generator and / or the discriminator is calculated, and if the similarity is greater than a preset threshold, one of the corresponding network layers can be distilled and optimized to avoid repeated calculation of similar feature maps and save computing resources. The Euclidean distance can be used to measure the similarity between the feature maps.

[0110] In addition, it should be noted that the generator loss, the discriminator loss, the supervised loss of the corresponding network layer, and the cascade network loss can all be mean-square error (MSE), the optimizer can be Adam, and the evaluation criterion can be accuracy.

[0111] In a possible implementation, the enhancement of the computed tomography in the image domain output by the computed tomography reconstruction model to obtain the final computed tomography includes:

[0112] inputting the computed tomography in the image domain output by the computed tomography reconstruction model into the trained enhancement model to obtain the enhanced computed tomography output by the enhancement model as the final computed tomography; the enhancement model is trained by taking the unenhanced computed tomography as a training sample, taking the enhanced computed tomography as a sample label, and taking the computed tomography output by the model to approach the enhanced computed tomography as a training target.

[0113] In the present application, the enhancement model can be constructed in advance, and the enhancement model can be composed of super residual network components, and each super residual network component is composed of convolution, batch normalization, and ReLU activation. For ease of understanding, refer to Figure 4 , Figure 4 The structure of an enhancement model example disclosed in an embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, the enhancement model is composed of N super residual network components. The value of N can be determined based on scene requirements. In a possible implementation, the value of N can be 16. Figure 4

[0114] After the enhancement model is constructed, the unenhanced computed tomography and the enhanced computed tomography can be determined. In an implementable manner, the enhanced computed tomography can be the image domain computed tomography generated clinically, and the unenhanced computed tomography can be the computed tomography generated by the computed tomography reconstruction model using the projection domain data corresponding to the image domain computed tomography generated clinically.

[0115] After the unenhanced computed tomography and the enhanced computed tomography are determined, the enhancement model can be trained by taking the unenhanced computed tomography as a training sample, taking the enhanced computed tomography as a sample label, and taking the computed tomography output by the model to approach the enhanced computed tomography as a training target.

[0116] ​It should be noted that, when training the enhancement model, the loss function can be a perception loss, the optimizer can be Adam, and the evaluation standard can be a peak signal-to-noise ratio, the training sample is unenhanced computer tomography imaging, the sample label is enhanced computer tomography imaging, and the training target is that the computer tomography imaging output by the model approaches the enhanced computer tomography imaging.

[0117] The above introduces a computer tomography reconstruction method provided by the embodiment of the application, and the following introduces a device for executing the computer tomography reconstruction method.

[0118] Please refer to Figure 5 , Figure 5 FIG. 1 is a structural schematic diagram of a computer tomography reconstruction device provided by an embodiment of the application. As shown in the figure, the computer tomography reconstruction device includes: Figure 5

[0119] The acquisition unit 11 is configured to acquire original projection domain data.

[0120] The denoising processing unit 12 is configured to perform denoising processing on the original projection domain data to obtain denoising-processed projection domain data.

[0121] The reconstruction unit 13 is configured to input the denoising-processed projection domain data into a computer tomography reconstruction model to obtain an image domain computer tomography output by the computer tomography reconstruction model.

[0122] The enhancement unit 14 is configured to enhance the image domain computer tomography output by the computer tomography reconstruction model to obtain a final computer tomography.

[0123] In a possible implementation, the denoising processing unit is specifically configured to:

[0124] input the original projection domain data into a trained denoising model to obtain denoising-processed projection domain data output by the denoising model; the denoising model is trained by taking noisy projection domain data as a training sample, taking denoising projection domain data as a sample label, and taking the projection domain data output by the model approaching the denoising projection domain data as a training target.

[0125] In a possible implementation, the device further includes a computer tomography reconstruction model training unit, and the computer tomography reconstruction model training unit includes:

[0126] A training data determination unit is configured to determine training data, and the training data includes image domain computer tomography training data and projection domain training data corresponding to the image domain computer tomography training data.

[0127] ​The generator and discriminator obtaining unit is configured to obtain a pre-constructed generator and discriminator.

[0128] The training unit is configured to perform multiple iteration training on the generator and the discriminator based on the projection domain training data and the computed tomography training data in the image domain, to obtain a trained generator and a trained discriminator; and the trained generator is the computed tomography reconstruction model.

[0129] In a possible implementation, during each iteration training process, the training unit is specifically configured to:

[0130] train only the generator by taking the projection domain training data as training samples and taking the computed tomography training data in the image domain as sample labels;

[0131] train only the discriminator by taking the computed tomography training data in the image domain as training samples and taking the projection domain training data as sample labels;

[0132] retrain the generator by fixing the parameters of the discriminator, taking the projection domain training data as training samples and sample labels.

[0133] In a possible implementation, during each iteration training process, the loss function comprises:

[0134] a generator loss, a discriminator loss, a supervision loss of a corresponding network layer, and a cascade network loss.

[0135] The generator loss is used to represent the similarity between the output of the generator and the computed tomography training data in the image domain when only the generator is trained by taking the projection domain training data as training samples and taking the computed tomography training data in the image domain as sample labels.

[0136] The discriminator loss is used to represent the similarity between the output of the discriminator and the projection domain training data when only the discriminator is trained by taking the computed tomography training data in the image domain as training samples and taking the projection domain training data as sample labels.

[0137] The cascade network loss is used to represent the similarity between the output of the discriminator and the projection domain training data when the generator is retrained by fixing the parameters of the discriminator, taking the projection domain training data as training samples and sample labels.

[0138] The supervision loss of the corresponding network layer is used to represent the feature map similarity of the corresponding layer structure in the generator and the discriminator during each iteration training process.

[0139] In a possible implementation, the enhancement unit comprises:

[0140] The computer tomography input of the image domain output by the computer tomography reconstruction model is trained by the trained enhancement model, and the enhanced computer tomography output by the enhancement model is obtained as the final computer tomography; the enhancement model is trained by taking the unenhanced computer tomography as a training sample, taking the enhanced computer tomography as a sample label, and taking the computer tomography output by the model approaching the enhanced computer tomography as a training target.

[0141] The embodiments of the present application also provide an electronic device. Referring to Figure 6 Fig. 1 shows a structural schematic diagram of an electronic device suitable for implementing the electronic device in the embodiments of the present application. The electronic device in the embodiments of the present application can include, but is not limited to, a fixed terminal such as a mobile phone, a notebook computer, a PDA (Personal Digital Assistant), a PAD (Tablet Personal Computer), a desktop computer, and the like. Figure 6 The electronic device shown is only an example, and should not bring any limitation to the functions and use range of the embodiments of the present application.

[0142] As shown in Figure 6 The electronic device can include a processing device (for example, a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 602 or loaded from a storage device 608 to a random access memory (RAM) 603. In the state that the electronic device is powered on, the RAM 603 also stores various programs and data required for the operation of the electronic device. The processing device 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0143] Generally, the following devices can be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a memory card, a hard disk, etc.; and a communication device 609. The communication device 609 can allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 The electronic device is shown with various devices, but it should be understood that all the shown devices are not required to be implemented or possessed. More or fewer devices can be alternatively implemented or possessed.

[0144] The embodiment of the present application further provides a computer program product comprising computer readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the computer tomography reconstruction methods provided by the embodiments of the present application.

[0145] The embodiment of the present application further provides a computer readable storage medium, which carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement any of the computer tomography reconstruction methods provided by the embodiments of the present application.

[0146] In addition, it should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments. In addition, the connection relationship between the modules in the apparatus embodiments provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines.

[0147] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can also be realized by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. Generally, functions completed by computer programs can be easily realized by corresponding hardware, and specific hardware structures for realizing the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better embodiment. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, U disk, mobile hard disk, ROM, RAM, magnetic disk or optical disk, etc., including a plurality of instructions to make a computer device (which can be a personal computer, training device, or network device, etc.) execute the methods described in various embodiments of the present application.

[0148] In the above embodiments, all or part can be realized by software, hardware, firmware or any combination thereof. When realized by software, it can be realized in the form of a computer program product in whole or in part.

[0149] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, training device or data center to another website, computer, training device or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be stored by the computer or a data storage device such as a training device, a data center, etc. integrated with one or more available media sets. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)), etc.

Claims

1. A computed tomography reconstruction method, characterized by, The method comprises the following steps: obtaining original projection domain data; performing denoising processing on the original projection domain data to obtain denoised projection domain data; inputting the denoised projection domain data into a computer tomography reconstruction model to obtain a computer tomography image output by the computer tomography reconstruction model; the computer tomography reconstruction model is a trained generator obtained by performing multiple iteration training on the generator and a discriminator based on projection domain training data and computer tomography training data in an image domain; in each iteration training process, a loss function for training the generator and the discriminator based on the projection domain training data and the computer tomography training data in the image domain comprises a generator loss, a discriminator loss, a supervision loss of a corresponding network layer and a cascade network loss; performing enhancement on the computer tomography image output by the computer tomography reconstruction model to obtain a final computer tomography image.

2. The method of claim 1, wherein, The denoising processing on the original projection domain data to obtain the denoised projection domain data comprises the following steps: inputting the original projection domain data into a trained denoising model to obtain denoised projection domain data output by the denoising model; the denoising model is trained by taking noisy projection domain data as training samples, taking denoised projection domain data as sample labels and taking the projection domain data output by the model approaching the denoised projection domain data as a training target.

3. The method of claim 1, wherein, The training method of the computer tomography reconstruction model is as follows: determining training data, which comprises computer tomography training data in an image domain and projection domain training data corresponding to the computer tomography training data in the image domain; obtaining a pre-constructed generator and a discriminator; performing multiple iteration training on the generator and the discriminator based on the projection domain training data and the computer tomography training data in the image domain to obtain a trained generator and a trained discriminator; the trained generator is the computer tomography reconstruction model.

4. The method of claim 3, wherein, in each iteration training process, training the generator and the discriminator based on the projection domain training data and the computer tomography training data in the image domain comprises the following steps: training only the generator by taking the projection domain training data as training samples and the computer tomography training data in the image domain as sample labels; training only the discriminator by taking the computer tomography training data in the image domain as training samples and the projection domain training data as sample labels; re-training the generator by fixing the parameters of the discriminator and taking the projection domain training data as training samples and sample labels.

5. The method of claim 4, wherein, the generator loss is used to represent the similarity between the output of the generator and the computer tomography training data in the image domain when only the generator is trained by taking the projection domain training data as training samples and the computer tomography training data in the image domain as sample labels. The discriminator loss is used to represent that when only the discriminator is trained with the computed tomography training data in the image domain as training samples and the projection domain training data as sample labels, the similarity between the output of the discriminator and the projection domain training data. The cascade network loss is used to represent that when the parameters of the discriminator are fixed and the generator is trained again with the projection domain training data as training samples and sample labels, the similarity between the output of the discriminator and the projection domain training data. The supervision loss of the corresponding network layer is used to represent the similarity of feature maps of corresponding layers in the generator and the discriminator in each iteration training process.

6. The method of claim 1, wherein, The computer tomography in the image domain output by the computer tomography reconstruction model is enhanced to obtain a final computer tomography, including: The computer tomography in the image domain output by the computer tomography reconstruction model is input into the trained enhancement model to obtain enhanced computer tomography output by the enhancement model as the final computer tomography; the enhancement model is trained with unenhanced computer tomography as training samples, enhanced computer tomography as sample labels, and model output computer tomography tending to the enhanced computer tomography as a training target.

7. A computed tomography reconstruction apparatus, characterized by Including: An acquisition unit is configured to acquire original projection domain data. A denoising processing unit is configured to perform denoising processing on the original projection domain data to obtain denoising processed projection domain data. A reconstruction unit is configured to input the denoising processed projection domain data into a computer tomography reconstruction model to obtain computer tomography in the image domain output by the computer tomography reconstruction model; the computer tomography reconstruction model is a trained generator obtained by multiple iteration training of a generator and a discriminator based on projection domain training data and computed tomography training data in the image domain. In each iteration training process, the loss function for training the generator and the discriminator based on the projection domain training data and the computed tomography training data in the image domain includes: generator loss, discriminator loss, supervision loss of corresponding network layers, and cascade network loss. An enhancement unit is configured to enhance the computer tomography in the image domain output by the computer tomography reconstruction model to obtain a final computer tomography.

8. A computer program product, characterised in that, The computer readable instructions, when executed on an electronic device, cause the electronic device to implement the computed tomography reconstruction method of any one of claims 1-6.

9. An electronic device, comprising: The memory is configured to store computer programs. The processor is configured to execute the computer programs to enable the electronic device to implement the computed tomography reconstruction method of any one of claims 1-6. ​ 10. A computer-readable storage medium, characterized in that, The storage medium has one or more computer programs carried thereon, and when the one or more computer programs are executed by the electronic device, the electronic device can implement the computed tomography reconstruction method according to any one of claims 1 to 6.

Images