Three-dimensional ct imaging method, device, electronic equipment and storage medium

Abstract

The application provides a three-dimensional CT imaging method, device, electronic equipment and storage medium. The method comprises the following steps: collecting two DR images with a vertical relationship for an object to be imaged, and obtaining three-dimensional coordinate information of the object to be imaged; inputting the two DR images with the vertical relationship and the three-dimensional coordinate information into a preset three-dimensional volume reconstruction model to obtain three-dimensional volume data of the object to be imaged; and slicing the three-dimensional volume data to obtain a three-dimensional CT image of the object to be imaged. The method can improve the three-dimensional volume reconstruction speed on the basis of ensuring the imaging quality, and then quickly obtain a three-dimensional CT image with high quality.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the technical field of computer, in particular to a three-dimensional CT imaging method and device, electronic equipment and storage medium. BACKGROUND

[0002] To obtain three-dimensional CT (Computed Tomography) imaging, it is usually necessary to solve the problem of three-dimensional volume reconstruction. The three-dimensional volume reconstruction method in the related art, such as using FDK algorithm or other algorithms for three-dimensional volume reconstruction, and the single-projection three-dimensional volume reconstruction method based on deep learning;

[0003] Although using FDK algorithm or other algorithms for three-dimensional volume reconstruction can ensure imaging quality, it requires a large number of DR images collected in the circumferential direction, resulting in high cost of data acquisition and slow imaging speed;

[0004] The single-projection three-dimensional volume reconstruction method based on deep learning cannot restore the details of the reconstruction result of the device with rich internal information, that is, it cannot guarantee the imaging quality.

[0005] Therefore, the related art has not yet provided a technical solution that can guarantee imaging quality and improve the speed of three-dimensional volume reconstruction, and thus cannot quickly obtain high-quality three-dimensional CT images. SUMMARY

[0006] Therefore, the present application provides a three-dimensional CT imaging method, device, electronic equipment and storage medium, which can improve the speed of three-dimensional volume reconstruction while ensuring imaging quality, and thus quickly obtain high-quality three-dimensional CT images.

[0007] To solve the above technical problems, the technical solution of the present application is as follows:

[0008] In one embodiment, a three-dimensional CT imaging method is provided, which comprises:

[0009] Two DR images with a vertical relationship are collected for the object to be imaged, and three-dimensional coordinate information of the object to be imaged is obtained;

[0010] The two DR images with a vertical relationship and the three-dimensional coordinate information are input into a preset three-dimensional volume reconstruction model to obtain three-dimensional volume data of the object to be imaged;

[0011] The three-dimensional volume data is sliced to obtain a three-dimensional CT image of the object to be imaged;

[0012] The preset three-dimensional volume reconstruction model includes a first volume model, a second volume model and a fusion model; the first volume model is used to obtain first volume data based on the three-dimensional coordinate information, and the second volume model is used to obtain second volume data based on the two DR images having a vertical relationship; and the fusion model is used to obtain three-dimensional volume data of the object to be imaged based on the first volume data and the second volume data.

[0013] The generation of the preset three-dimensional volume reconstruction model includes:

[0014] An initial three-dimensional volume reconstruction model is established;

[0015] A training data set is obtained; the training data set includes three-dimensional coordinate information of a reference object, two DR images having a vertical relationship and three-dimensional volume data;

[0016] The initial three-dimensional volume reconstruction model is trained using the training data set, and a loss value of a loss function set is calculated according to a calculation value and a reference value; the calculation value is three-dimensional volume data obtained by the three-dimensional volume reconstruction model in the training process, and the reference value is the three-dimensional volume data in the training data set;

[0017] In response to the loss function being less than a preset loss threshold, the preset three-dimensional volume reconstruction model is obtained.

[0018] The loss function is a weighted sum of a first loss, a second loss and a third loss.

[0019] The first loss is a 2D projection loss of the calculation value and the reference value, the second loss is a structural similarity loss of the calculation value and the reference value, and the third loss is a 3D reconstruction loss of the calculation value and the reference value.

[0020] The weights of the first loss and the third loss are set to be the same and greater than the weight of the second loss.

[0021] The first volume model is used to obtain first volume data based on the three-dimensional coordinate information, including:

[0022] The first volume model encodes the three-dimensional coordinate information through Fourier feature mapping, inputs the encoded three-dimensional coordinate information into a perception neural network, performs implicit neural representation learning, and obtains the first volume data.

[0023] The second volume model is used to obtain second volume data based on the two DR images having a vertical relationship, including:

[0024] The second volume model splices the two DR images with a vertical relationship in a width dimension, and inputs the spliced result into the coding and decoding network to obtain the second volume data.

[0025] The fusion model is configured to obtain the three-dimensional volume data of the object to be imaged based on the first volume data and the second volume data, and includes:

[0026] The fusion model splices the first volume data and the second volume data in a channel dimension, and obtains the three-dimensional volume data of the object to be imaged through a three-dimensional convolution.

[0027] In another embodiment, a three-dimensional CT imaging device is provided, which includes:

[0028] A storage unit configured to store a preset three-dimensional volume reconstruction model, wherein the preset three-dimensional volume reconstruction model includes a first volume model, a second volume model, and a fusion model; the first volume model is configured to obtain first volume data based on three-dimensional coordinate information, the second volume model is configured to obtain second volume data based on two DR images with a vertical relationship; and the fusion model is configured to obtain three-dimensional volume data of an object to be imaged based on the first volume data and the second volume data.

[0029] An acquisition unit configured to acquire two DR images with a vertical relationship for an object to be imaged, and obtain three-dimensional coordinate information of the object to be imaged.

[0030] A processing unit configured to input the two DR images with a vertical relationship and the three-dimensional coordinate information into the preset three-dimensional volume reconstruction model to obtain three-dimensional volume data of the object to be imaged.

[0031] A slicing unit configured to slice the three-dimensional volume data to obtain a three-dimensional CT image of the object to be imaged.

[0032] In another embodiment, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the three-dimensional CT imaging method when executing the program.

[0033] In another embodiment, a computer readable storage medium is provided, which stores a computer program executable by a processor to implement the three-dimensional CT imaging method.

[0034] It can be seen from the above technical solutions that in the above embodiment, only two orthogonal projection images of the object to be imaged and three-dimensional coordinate information of the object to be imaged are needed to be input into a preset three-dimensional volume reconstruction model to obtain three-dimensional volume data of the object to be imaged, and then a three-dimensional CT image of the object is obtained; wherein the preset three-dimensional volume reconstruction model first obtains first volume data corresponding to the three-dimensional coordinate information and second volume data corresponding to the two orthogonal projection images, and then fuses the first volume data and the second volume data to obtain the three-dimensional volume of the object to be imaged. The scheme is realized by using two orthogonal projection images, reduces the size of the training data and the data collected from the object to be imaged, improves the generality of the model by introducing the three-dimensional coordinate information of the object to be imaged, improves the three-dimensional volume reconstruction speed on the basis of ensuring the imaging quality, and then quickly obtains a high-quality three-dimensional CT image. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0036] Figure 1 A flowchart of the preset three-dimensional volume reconstruction model in the embodiments of the present application is established;

[0037] Figure 2 A structural diagram of the three-dimensional volume reconstruction model in the embodiments of the present application is provided;

[0038] Figure 3 A schematic diagram of the MLP network architecture used in the embodiments of the present application is provided;

[0039] Figure 4 A schematic diagram of the second volume data structure model in the embodiments of the present application is provided;

[0040] Figure 5 A schematic diagram of the three-dimensional CT imaging process in the embodiments of the present application is provided;

[0041] Figure 6 A schematic diagram of the three-dimensional CT imaging device structure in the embodiments of the present application is provided;

[0042] Figure 7 A schematic diagram of the physical structure of the electronic device provided in the embodiments of the present application is provided. DETAILED DESCRIPTION

[0043] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0044] The terms "first", "second", "third", "fourth" and the like in the description and claims of the present application and above drawings, if any, are used for distinguishing between similar objects, and do not necessarily have to follow the sequences or priority of the objects. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprise" and "have" and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0045] The technical solutions of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0046] X-ray based imaging methods produce different degrees of attenuation by passing through the measured object to obtain its internal information, including DR and CT. DR technology is very efficient, but the results observed from the 2D perspective are often insufficient to represent the 3D structure, and the most direct and accurate method to accurately describe the internal information of the measured object is to construct a complete 3D volume. Compared with DR, CT technology can accurately reconstruct the complete 3D structure of the measured object.

[0047] Traditional CT reconstruction uses specific reconstruction algorithms such as analytical reconstruction and iterative reconstruction, but this requires a large amount of calculation, and at the same time, due to the limitation of the Shannon-Nyquist sampling theorem, the key to high-quality reconstruction is to obtain complete projections, which to a large extent limits the reconstruction speed. The key measure to improve the reconstruction speed is to reduce the projection images and the amount of calculation required for reconstruction. Researchers introduce compressive sensing into CT reconstruction and develop regularization methods based on compressive sensing. This kind of method uses the prior information of the image, which can reduce the required projection while ensuring the reconstruction quality. Recent research introduces deep learning technology into sparse view CT reconstruction, which uses artificially designed neural networks to let the network directly learn the mapping function from 2D X-ray projection to 3D CT volume. Compared with traditional regularization-based methods, deep learning methods can more feasibly realize CT reconstruction by using data-driven end-to-end networks to extract prior information of the predicted CT, and currently such methods have achieved very good results.

[0048] However, the deep learning method also faces the limitations that exist universally in application. First, it is difficult to obtain a large-scale training data set, which may constitute a significant bottleneck in some specific application scenarios, because collecting a sufficient amount of data often accompanies high costs. In addition, deep learning-based CT reconstruction technology performs poorly in generalization ability and is difficult to adapt to diversified imaging objects, thereby limiting its wide applicability. Furthermore, most current deep learning-driven CT reconstruction methods not only have a large number of model parameters, but also have a complex architecture design, which not only aggravates the consumption of computing resources, but also puts higher requirements on memory.

[0049] Therefore, the related art has not yet provided a technical solution that guarantees imaging quality and improves three-dimensional volume reconstruction speed, and thus a high-quality three-dimensional CT image cannot be quickly obtained.

[0050] Based on the above problems, a three-dimensional CT imaging method is provided in the embodiments of the present application, which does not require full-view projection, only needs two orthogonal projection images of the object to be imaged and three-dimensional coordinate information of the object to be imaged, and inputs a preset three-dimensional volume reconstruction model to obtain three-dimensional volume data of the object to be imaged and then obtain a three-dimensional CT image of the object. The preset three-dimensional volume reconstruction model first obtains first volume data corresponding to the three-dimensional coordinate information and second volume data corresponding to the two orthogonal projection images, and then fuses the first volume data and the second volume data to obtain a three-dimensional volume of the object to be imaged. This scheme reduces the scale of the training data and the data collected from the object to be imaged through the implementation of the two orthogonal projection images, and improves the generalization of the model by introducing the three-dimensional coordinate information of the object to be imaged, which can improve the three-dimensional volume reconstruction speed on the basis of ensuring the imaging quality, and thus quickly obtain a high-quality three-dimensional CT image.

[0051] In the embodiment of the present application, before three-dimensional CT imaging is performed, a preset three-dimensional volume reconstruction model needs to be established. Referring to Figure 1 , Figure 1 A flowchart for establishing the preset three-dimensional volume reconstruction model in the embodiment of the present application is shown. The specific steps are as follows:

[0052] Step 101, an initial three-dimensional volume reconstruction model is established.

[0053] In the embodiment of the present application, the initial three-dimensional volume reconstruction model established and the preset three-dimensional volume reconstruction model obtained through training have the same structure, both of which include three models, namely, a first volume model, a second volume model and a fusion model.

[0054] The first volume model is used to obtain first volume data based on three-dimensional coordinate information, the second volume model is used to obtain second volume data based on two DR images having a vertical relationship, and the fusion model is used to obtain three-dimensional volume data of an object to be imaged based on the first volume data and the second volume data.

[0055] Step 102, a training data set is obtained; the training data set includes three-dimensional coordinate information of a reference object, two DR images having a vertical relationship and three-dimensional volume data.

[0056] The obtained training data set can be a pre-stored data set or a collected data set.

[0057] The implementation of collecting the training data set is as follows:

[0058] DR images are collected in a circumferential direction for the reference object; the collected DR images of the reference object are 360 DR images in different directions.

[0059] In the collection process, the related parameters such as tube voltage, tube current, exposure time and position of the object to be measured are kept unchanged.

[0060] The collected DR images can be pre-processed, and the pre-processing includes:

[0061] The collected DR images are processed through size cropping, logarithm taking, normalization and the like based on chip parameters and mechanical parameters.

[0062] Then, FDK algorithm is used to realize 3D volume reconstruction for each reference object, so as to obtain one-to-one corresponding three-dimensional volume data and 360 DR images in different directions.

[0063] Then, three-dimensional position coordinate information is obtained for each reference object. The three-dimensional position coordinate information is generated according to the volume size of the reference object; the number of coordinates is the volume size. If the volume size is 128x128x128, the number of coordinates is 128 cubed.

[0064] Thus, there is a set of corresponding data for each reference object: three-dimensional volume data, three-dimensional position coordinate information, and 360 DR images in different directions;

[0065] Select multiple sets of two DR images with vertical relationship from the 360 DR images in different directions; each set of two DR images with vertical relationship is combined with corresponding three-dimensional volume data and three-dimensional position coordinate information as a training sample;

[0066] Two DR images with vertical relationship for different reference objects can generate a training sample, and then form a training data set.

[0067] Establish an initial three-dimensional volume reconstruction model, which can be executed simultaneously with the implementation of obtaining the training data set, or can be executed in sequence, and the execution of steps 101 and 102 is not in sequence.

[0068] Step 103, training the initial three-dimensional volume reconstruction model using the training data set; and calculating the loss value of the set loss function according to the calculated value and the reference value.

[0069] Wherein, the calculated value is the three-dimensional volume data obtained by the three-dimensional volume reconstruction model in the training process, and the reference value is the three-dimensional volume data in the training data set.

[0070] The loss function in this embodiment is the weighted sum of the first loss, the second loss and the third loss; wherein the first loss is the 2D projection loss (Projection Loss) of the calculated value and the reference value, the second loss is the structural similarity loss (SSIM Loss) of the calculated value and the reference value; the third loss is the 3D reconstruction loss (Reconstruction Loss) of the calculated value and the reference value.

[0071] The 3D reconstruction loss measures the geometric difference between the predicted 3D structure and the real structure, and this loss ensures that the model reconstructs the most accurate 3D model in space; the structural similarity loss focuses on the visual quality and structural consistency of the image, which usually evaluates the similarity between the predicted image and the real image, and can better capture the sensitivity of the human eye to image quality. The 2D projection loss can further improve the training efficiency. By combining the three losses together and properly weighting, the model performance can be optimized from different aspects, which can effectively improve the robustness and reconstruction accuracy of the model. The specific loss calculation formula is as follows:

[0072] L(Y truth ,Y pred )=λ3L re +λ1L pl +λ2LSSIM

[0073] wherein Y truth represents a reference value, Y pred represents a calculated value;

[0074] represents a 3D reconstruction loss of the calculated value and the reference value;

[0075]

[0076] L pl represents a 2D projection loss of the calculated value and the reference value; P1(Y truth ), P2(Y truth ) and P3(Y truth ) represent the projection reference values in the depth, width and height directions respectively, and P1(Y pred ), P2(Y pred ) and P3(Y pred ) represent the projection calculated values in the depth, width and height directions respectively;

[0077] L SSIM (Ypred, Ytruth) = 1 - SSIM(Ypred, Ytruth)

[0078] L SSIM (Ypred, Ytruth) represents a structural similarity loss of the calculated value and the reference value; generally, the larger the SSIM value is, the better, and the maximum value is 1, since the structural similarity loss is calculated together with the other two losses, therefore, 1-SSIM is used here to represent the structural similarity loss.

[0079] wherein λ1, λ2 and λ3 control the relative importance of different loss terms. In the dual-view CT reconstruction task, the global shape consistency is given priority, based on which, in specific use, the weights of the first loss and the third loss can be set to be the same, and greater than the weight of the second loss. For example, λ1 = λ3 = 1, λ2 = 0.1, but the specific implementation is not limited thereto.

[0080] Through the above loss function, a loss value can be calculated for the three-dimensional volume reconstruction model in each training.

[0081] Step 104, in response to the calculated loss value being less than a preset loss threshold, obtaining a preset three-dimensional volume reconstruction model.

[0082] In response to the calculated loss value being not less than the preset loss threshold, the three-dimensional volume reconstruction model in training is trained again; before the model training, an initial learning rate and a learning decay rate are set; when training again, the learning rate obtained from the last training is decayed according to the set decay rate, and the three-dimensional volume reconstruction model is trained again.

[0083] The specific structural implementation of the preset three-dimensional volume reconstruction model and the initial three-dimensional volume reconstruction model is given below in combination with the accompanying drawings.

[0084] Referring to Figure 2 , Figure 2 The structure diagram of the three-dimensional volume reconstruction model in the embodiments of the present application is shown. Figure 2 The three-dimensional volume reconstruction model in the embodiments of the present application includes a first volume model, a second volume model and a fusion model; the first volume model is used to obtain first volume data based on three-dimensional coordinate information, that is, to process the input three-dimensional coordinate information and output the first volume data; the second volume model is used to obtain second volume data based on two DR images having a vertical relationship, that is, to process the input two DR images having a vertical relationship and output the second volume data; and the fusion model is used to obtain three-dimensional volume data of an object to be imaged based on the first volume data and the second volume data, that is, to fuse the first volume data output by the first model and the second volume data output by the second model and output the three-dimensional volume data of the object to be imaged.

[0085] The first volume model receives the input three-dimensional coordinate information, encodes the three-dimensional coordinate information through Fourier feature mapping, inputs the encoded three-dimensional coordinate information into a perception neural network, performs implicit neural representation learning, and obtains the first volume data.

[0086] Since Fourier features are effective for network learning of high-frequency functions, we use Fourier feature mapping to encode spatial coordinates, and the specific calculation formula is as follows:

[0087] γ(c)=[cos(2πBc),sin(2πBc)] T

[0088] Where B represents the coefficient of Fourier feature transformation, B is a matrix sampled from a Gaussian distribution N(0,σ 2 ), σ is a hyperparameter representing the standard deviation of the prior distribution, and in specific implementation, σ can be 4, but is not limited thereto, and c is the coordinate information before position encoding, that is, the input three-dimensional coordinate information. γ(c) is the coordinate information after Fourier feature embedding, that is, the encoded three-dimensional coordinate information;

[0089] The encoded three-dimensional coordinate information is taken as an input of the perception neural network, implicit neural representation learning is performed on the perception neural network, and first volume data is obtained.

[0090] The perception neural network can be implemented by a multi-layer perception (MLP), see Figure 3 , Figure 3 FIG. 1 is a schematic diagram of an MLP network architecture used in an embodiment of the present application. Figure 3 The MLP in FIG. 1 has a depth of 8 and is implemented by 8 linear. The input of the MLP is the encoded three-dimensional coordinate information, and the output is the first volume data. When the MLP is implemented with a depth of 8, a network with a width of 256 neural points is constructed for implicit neural representation learning. Such implementation requires that the size of the Fourier feature embedding is 256. The MLP implementation herein is an example implementation, and the specific implementation is not limited thereto.

[0091] The second volume model splices two DR images with a vertical relationship in the width dimension, and inputs the spliced result into the encoding and decoding network to obtain second volume data.

[0092] See Figure 4 , Figure 4 FIG. 2 is a schematic diagram of a second volume data structure model in an embodiment of the present application. The input two DR images with a vertical relationship (a first DR image and a second DR image) are spliced in the width dimension by a connector C, and then input into an encoder (Encoder)-reshape (Reshape) module-decoder (Decoder) to output second volume data.

[0093] The encoder-decoder convolutional neural network is used to extract features from 2D images, and the features of the input images are aligned with voxels in 3D geometry. The encoder is 2D, and the decoder is 3D. A modified ResNet can be used as a 2D encoder to extract local and global information from the projection image; then the extracted features are reshaped into 3D volume features through the Reshape operation; and then the decoder is used for successive up-sampling processing, and the decoder restores the reshaped 3D volume features to a 3D feature map aligned with the true value volume.

[0094] The decoder includes four up-sampling modules and a bilinear interpolation module. Each up-sampling module includes a 3D transpose convolution layer, a 3D convolution layer, a 3D BN layer, and a 3D activation (ReLU) layer.

[0095] The first volume data output by the first volume model and the second volume data output by the second volume model both have the same shape as the target result (three-dimensional volume data of the object to be imaged).

[0096] The fusion model splices the first volume data and the second volume data in a channel dimension, and obtains three-dimensional volume data of the object to be imaged through a three-dimensional convolution.

[0097] So far, the architecture of the preset three-dimensional volume reconstruction model and the detailed implementation process of the training process are given. The preset three-dimensional volume reconstruction model is stored, and can be directly obtained and used when three-dimensional volume reconstruction and three-dimensional CT imaging are needed.

[0098] The three-dimensional CT imaging process of the embodiment of the present application is given in detail below in combination with the drawings.

[0099] Referring to Figure 5 , Figure 5 is a three-dimensional CT imaging flowchart in the embodiment of the present application. The specific steps are as follows:

[0100] Step 501, two DR images with a vertical relationship are collected for the object to be imaged, and three-dimensional coordinate information of the object to be imaged is obtained.

[0101] In the embodiment of the present application, when collecting DR images for the object to be imaged, two DR images with a vertical relationship can be collected, without the need for 360 DR images to be obtained in a circumferential direction of 360 degrees.

[0102] In the process of collecting the two DR images with a vertical relationship, the related parameters such as tube voltage, tube current, exposure time, and position during the measurement period are kept unchanged, and are the same as the parameters for collecting the data set.

[0103] Whether the collected DR images are processed and how to process them can be the same as the processing method of the DR images in the training data set, so as to ensure that the desired output can be obtained after being input into the preset three-dimensional volume reconstruction model.

[0104] If the DR images in the sample data set during the training of the model have been preprocessed as follows, the collected DR images are also preprocessed as follows in this step:

[0105] The collected DR images are size-cropped, logarithmic, normalized, and processed based on chip parameters and mechanical parameters.

[0106] Three-dimensional position coordinate information is obtained for the object to be imaged. The three-dimensional position coordinate information is generated according to the volume size of the object to be imaged, and the number of coordinates is the volume size. If the volume size is 128x128x128, the number of coordinates is 128.

[0107] Step 502, input the two DR images with a vertical relationship and the three-dimensional coordinate information into the preset three-dimensional volume reconstruction model, and obtain three-dimensional volume data of the object to be imaged.

[0108] Step 503, slice the three-dimensional volume data to obtain a three-dimensional CT image of the object to be imaged.

[0109] The three-dimensional CT image is obtained by slicing the three-dimensional volume in the embodiments of the present application, which is not limited.

[0110] In a specific implementation, if only the three-dimensional CT image of the object to be imaged is obtained, the implementation of step 503 can also be encapsulated in a preset three-dimensional volume reconstruction model as a three-dimensional CT imaging model; if the three-dimensional CT image of the object to be imaged is obtained at the same time as the three-dimensional volume data of the object to be imaged, the above-mentioned method disclosed in the embodiments of the present application can be implemented.

[0111] In the embodiments of the present application, full-view projection is not required, only two orthogonal projection images of the object to be imaged and three-dimensional coordinate information thereof are required to be input into the preset three-dimensional volume reconstruction model to obtain the three-dimensional volume data of the object to be imaged, and then the three-dimensional CT image of the object is obtained. Through the implementation of the two orthogonal projection images, the size of the training data and the data collected for the object to be imaged are reduced; by introducing the three-dimensional coordinate information of the object to be imaged, the generalization of the model is improved, the three-dimensional volume reconstruction speed is improved on the basis of ensuring the imaging quality, and then the three-dimensional CT image with high quality is quickly obtained.

[0112] Based on the same inventive concept, the embodiments of the present application also provide a three-dimensional CT imaging device. Referring to Figure 6 , Figure 6 is a structural schematic diagram of the three-dimensional CT imaging device in the embodiments of the present application. The device comprises:

[0113] The storage unit 601 is configured to store a preset three-dimensional volume reconstruction model; wherein the preset three-dimensional volume reconstruction model comprises: a first volume model, a second volume model and a fusion model; the first volume model is configured to obtain first volume data based on three-dimensional coordinate information, the second volume model is configured to obtain second volume data based on two DR images having a vertical relationship; and the fusion model is configured to obtain three-dimensional volume data of the object to be imaged based on the first volume data and the second volume data.

[0114] The acquisition unit 602 is configured to collect two DR images having a vertical relationship for the object to be imaged, and acquire three-dimensional coordinate information of the object to be imaged.

[0115] The processing unit 603 is configured to input the two DR images having a vertical relationship and the three-dimensional coordinate information into the preset three-dimensional volume reconstruction model to obtain the three-dimensional volume data of the object to be imaged.

[0116] The slicing unit 604 is configured to slice the three-dimensional volume data to obtain a three-dimensional CT image of the object to be imaged.

[0117] In another example, the generation of the preset three-dimensional volume reconstruction model comprises:

[0118] establishing an initial three-dimensional volume reconstruction model;

[0119] obtaining a training data set; wherein the training data set comprises three-dimensional coordinate information of a reference object, two DR images having a vertical relationship, and three-dimensional volume data;

[0120] training the initial three-dimensional volume reconstruction model using the training data set; and calculating a loss value of a preset loss function according to a calculated value and a reference value; wherein the calculated value is three-dimensional volume data obtained by the three-dimensional volume reconstruction model in the training process, and the reference value is the three-dimensional volume data in the training data set;

[0121] in response to the loss function being less than a preset loss threshold, obtaining the preset three-dimensional volume reconstruction model.

[0122] In another example, the loss function is a weighted sum of a first loss, a second loss, and a third loss;

[0123] wherein the first loss is a 2D projection loss of the calculated value and the reference value, the second loss is a structural similarity loss of the calculated value and the reference value, and the third loss is a 3D reconstruction loss of the calculated value and the reference value.

[0124] In another example, the weights of the first loss and the third loss are set to be the same and greater than the weight of the second loss.

[0125] In another example, the first volume model is used to obtain first volume data based on the three-dimensional coordinate information, comprising:

[0126] The first volume model encodes the three-dimensional coordinate information through Fourier feature mapping, inputs the encoded three-dimensional coordinate information into a perception neural network, performs implicit neural representation learning, and obtains the first volume data.

[0127] In another example, the second volume model is used to obtain second volume data based on the two DR images having a vertical relationship, comprising:

[0128] The second volume model splices the two DR images having a vertical relationship in the width dimension, inputs the splicing result into a coding-decoding network, and obtains the second volume data.

[0129] In another example, the fusion model is used to obtain three-dimensional volume data of the object to be imaged based on the first volume data and the second volume data, comprising:

[0130] The fusion model splices the first volume data and the second volume data in a channel dimension, and obtains the three-dimensional volume data of the object to be imaged through a three-dimensional convolution.

[0131] The units in the above embodiments can be integrated or deployed separately, and can be combined into one unit or further split into multiple sub-units.

[0132] In another embodiment, an electronic device is also provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the three-dimensional CT imaging method when executing the program.

[0133] In another embodiment, a computer readable storage medium is also provided, which stores computer instructions, and the instructions are executed by a processor to implement the three-dimensional CT imaging method.

[0134] Figure 7 The electronic device provided by the embodiment of the present application is shown in the entity structure diagram. As shown in the figure, the electronic device can include a processor (Processor) 710, a communication interface (Communications Interface) 720, a memory (Memory) 730, and a communication bus (Communications Bus) 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call the logical instructions in the memory 730 to execute the following method: Figure 7

[0135] Two DR images with a vertical relationship are collected for the object to be imaged, and three-dimensional coordinate information of the object to be imaged is obtained;

[0136] The two DR images with a vertical relationship and the three-dimensional coordinate information are input into a preset three-dimensional volume reconstruction model to obtain three-dimensional volume data of the object to be imaged;

[0137] The three-dimensional volume data is sliced to obtain a three-dimensional CT image of the object to be imaged;

[0138] The preset three-dimensional volume reconstruction model includes a first volume model, a second volume model, and a fusion model; the first volume model is used to obtain first volume data based on the three-dimensional coordinate information, the second volume model is used to obtain second volume data based on the two DR images with a vertical relationship, and the fusion model is used to obtain the three-dimensional volume data of the object to be imaged based on the first volume data and the second volume data.

[0139] ​In addition, the logic instructions in the memory 730 described above can be implemented in the form of software functional units and sold or used as independent products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0140] The device embodiments described above are only schematic, wherein the units illustrated as separate components can or can not be physically separate, and the components illustrated as units can or can not be physical units, i.e., 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 present embodiment. Those skilled in the art can understand and implement without creative labor.

[0141] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus necessary universal hardware platforms, and of course can also be implemented by hardware. Based on such understanding, the above technical solutions essentially or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some part of the embodiment.

[0142] The computer program product of the first aspect can include a computer readable storage medium. The computer readable storage medium can be a tangible computer readable storage medium storing the program code. The computer readable storage medium can include, but is not limited to, magnetic storage, optical storage, and the like. The computer readable storage medium can be non-transitory. The computer readable storage medium can be a non-transitory computer readable medium. The above definitions of the computer program product of the first aspect apply, mutatis mutandis, to the computer readable storage medium.

[0143] Those skilled in the art can understand that the features described in various embodiments and / or claims of the present application can be combined and / or integrated in various combinations, even if such combinations or integrations are not explicitly described in the present application. In particular, the features described in various embodiments and / or claims of the present application can be combined and / or integrated in various combinations without departing from the spirit and teachings of the present application, and all such combinations and / or integrations fall within the scope of the present application.

[0144] The principles and implementation modes of the present application are described herein by applying specific embodiments. The above description of the embodiments is only for the purpose of helping to understand the method of the present application and its core idea, and is not intended to limit the present application. For those skilled in the art, any modification, equivalent replacement, improvement, etc. made according to the idea, spirit and principle of the present application in the specific implementation mode and application range shall be included in the scope of protection of the present application.

Claims

1. A method of three-dimensional CT imaging, characterized by, The method comprises: acquiring two DR images with a vertical relationship for an object to be imaged and obtaining three-dimensional coordinate information of the object to be imaged; inputting the two DR images with a vertical relationship and the three-dimensional coordinate information into a preset three-dimensional volume reconstruction model to obtain three-dimensional volume data of the object to be imaged; slicing the three-dimensional volume data to obtain a three-dimensional CT image of the object to be imaged; wherein the preset three-dimensional volume reconstruction model comprises a first volume model, a second volume model and a fusion model; the first volume model is used to obtain first volume data based on the three-dimensional coordinate information, and the second volume model is used to obtain second volume data based on the two DR images with a vertical relationship; and the fusion model is used to obtain three-dimensional volume data of the object to be imaged based on the first volume data and the second volume data; wherein the first volume model is used to obtain first volume data based on the three-dimensional coordinate information, comprising: the first volume model encodes the three-dimensional coordinate information through Fourier feature mapping, inputs the encoded three-dimensional coordinate information into a perception neural network, performs implicit neural representation learning, and obtains the first volume data; the second volume model is used to obtain second volume data based on the two DR images with a vertical relationship, comprising: the second volume model splices the two DR images with a vertical relationship in the width dimension, inputs the splicing result into a coding-decoding network, and obtains the second volume data.

2. The method of claim 1, wherein, The generation of the preset three-dimensional volume reconstruction model comprises: establishing an initial three-dimensional volume reconstruction model; obtaining a training data set; wherein the training data set comprises three-dimensional coordinate information of a reference object, two DR images with a vertical relationship and three-dimensional volume data; training the initial three-dimensional volume reconstruction model using the training data set; and calculating a loss value of a loss function according to a calculated value and a reference value; wherein the calculated value is three-dimensional volume data obtained by the three-dimensional volume reconstruction model in the training process, and the reference value is the three-dimensional volume data in the training data set; in response to the loss function being less than a preset loss threshold, obtaining the preset three-dimensional volume reconstruction model.

3. The method of claim 2, wherein, The loss function is a weighted sum of a first loss, a second loss and a third loss; wherein the first loss is a 2D projection loss of the calculated value and the reference value, the second loss is a structural similarity loss of the calculated value and the reference value, and the third loss is a 3D reconstruction loss of the calculated value and the reference value.

4. The method of claim 3, wherein, The weights of the first loss and the third loss are set to be the same and greater than the weight of the second loss.

5. The method of claim 1, wherein, The fusion model is used to obtain three-dimensional volume data of the object to be imaged based on the first volume data and the second volume data, comprising: the fusion model splices the first volume data and the second volume data in the channel dimension, and obtains the three-dimensional volume data of the object to be imaged through a three-dimensional convolution.

6. A three-dimensional CT imaging apparatus, characterized by comprising: The device comprises: The storage unit is configured to store a preset three-dimensional volume reconstruction model; wherein the preset three-dimensional volume reconstruction model comprises a first volume model, a second volume model and a fusion model; the first volume model is configured to obtain first volume data based on three-dimensional coordinate information, and the second volume model is configured to obtain second volume data based on two DR images having a vertical relationship; the fusion model is configured to obtain three-dimensional volume data of an object to be imaged based on the first volume data and the second volume data; wherein the first volume model is configured to obtain first volume data based on the three-dimensional coordinate information, and comprises that the first volume model encodes the three-dimensional coordinate information through Fourier feature mapping, inputs the encoded three-dimensional coordinate information into a perception neural network, performs implicit neural representation learning, and obtains the first volume data; the second volume model is configured to obtain second volume data based on the two DR images having a vertical relationship, and comprises that the second volume model splices the two DR images having a vertical relationship in a width dimension, inputs the splicing result into a coding and decoding network, and obtains the second volume data; The acquisition unit is configured to acquire two DR images having a vertical relationship for an object to be imaged, and obtain three-dimensional coordinate information of the object to be imaged; The processing unit is configured to input the two DR images having a vertical relationship and the three-dimensional coordinate information into the preset three-dimensional volume reconstruction model, and obtain three-dimensional volume data of the object to be imaged; The slicing unit is configured to slice the three-dimensional volume data to obtain a three-dimensional CT image of the object to be imaged.

7. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the method of any one of claims 1-5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method of any one of claims 1-5.

Images