3D CT Imaging Method and Device Based on Dual X-ray Projections

Abstract

The present application provides a three-dimensional CT imaging method and apparatus based on X-ray double projection. The method includes: acquiring two DR images with a perpendicular relationship for an object to be imaged; obtaining a three-dimensional volume of the object to be imaged corresponding to the two DR images with a perpendicular relationship based on a preset three-dimensional volume reconstruction model; slicing the three-dimensional volume to obtain a three-dimensional CT image of the object to be imaged; wherein the generation of the preset three-dimensional volume reconstruction model includes: obtaining a training data set; wherein the training data set is composed of DR images of a reference object collected circumferentially; using the DR images with a perpendicular relationship in the training data set to train a three-dimensional volume reconstruction model to be trained, and obtaining the preset three-dimensional volume reconstruction model. This method 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.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to a three-dimensional CT imaging method and apparatus based on dual X-ray projections. Background Art

[0002] To obtain three-dimensional computed tomography (CT) imaging, the three-dimensional volume reconstruction problem is usually solved first. Three-dimensional volume reconstruction methods in related technologies, such as using algorithms like FDK for three-dimensional volume reconstruction, single-projection three-dimensional volume reconstruction methods based on deep learning, etc.;

[0003] Although using algorithms like FDK for three-dimensional volume reconstruction can ensure imaging quality, due to the large number of DR images required for circumferential acquisition, the cost of acquired data is high and the imaging speed is slow;

[0004] Using a single-projection three-dimensional volume reconstruction method based on deep learning, the detailed structures in the reconstruction results of devices with rich internal information are not well restored, that is, the imaging quality cannot be guaranteed.

[0005] Therefore, in related technologies, a technical solution that can both ensure imaging quality and improve the three-dimensional volume reconstruction speed has not been provided, and thus a high-quality three-dimensional CT image cannot be obtained quickly. Summary of the Invention

[0006] In view of this, the present application provides a three-dimensional CT imaging method, apparatus, electronic device, and storage medium based on dual X-ray projections, which can improve the three-dimensional volume reconstruction speed on the basis of ensuring imaging quality, and thus quickly obtain a high-quality three-dimensional CT image.

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

[0008] In one embodiment, a three-dimensional CT imaging method based on dual X-ray projections is provided, and the method includes:

[0009] Acquire two DR images with a perpendicular relationship for the object to be imaged;

[0010] Obtain the three-dimensional volume of the object to be imaged corresponding to the two DR images with a perpendicular relationship based on a preset three-dimensional volume reconstruction model;

[0011] Slice the three-dimensional volume to obtain the three-dimensional CT image of the object to be imaged;

[0012] Among them, the generation of the preset three-dimensional volume reconstruction model includes:

[0013] Obtain a training data set; wherein, the training data set is composed of DR images of a reference object acquired circumferentially;

[0014] Use the DR images with a vertical relationship in the training dataset to train the three-dimensional volume reconstruction model to be trained, and obtain a preset three-dimensional volume reconstruction model.

[0015] Wherein, when obtaining the training dataset, the method further includes:

[0016] Obtain the reference three-dimensional volume and the reference pseudo-sine map of each reference object based on the training dataset; wherein, the reference pseudo-sine map is obtained by stacking the DR images of the same reference object collected in the acquisition order in a three-dimensional matrix, and adding the sine maps in each layer column by column and normalizing.

[0017] Divide the DR images of each reference object in the training dataset into multiple DR image groups, and each DR image group includes two DR images with a vertical relationship.

[0018] The step of using the DR images with a vertical relationship in the training dataset to train the three-dimensional volume reconstruction model to be trained and obtain a preset three-dimensional volume reconstruction model includes:

[0019] Input multiple DR image groups into the three-dimensional volume reconstruction model to be trained, train the three-dimensional volume reconstruction model to be trained; and obtain a training pseudo-sine map and a training three-dimensional volume respectively.

[0020] Calculate the loss value based on the training pseudo-sine map, the training three-dimensional volume, the reference three-dimensional volume, and the reference pseudo-sine map in the training order.

[0021] If any calculated loss value is less than the target loss value, or the number of training times reaches the target number of times, end the training and obtain a preset three-dimensional volume reconstruction model.

[0022] Wherein, the three-dimensional volume reconstruction model to be trained includes a feature enhancement network model and a reconstruction network model;

[0023] The enhancement network model is a model based on the encoder-decoder network architecture; it is used to input DR image groups with a vertical relationship and output corresponding training pseudo-sine images.

[0024] The reconstruction network model is an autoencoder network model for two-dimensional to three-dimensional volume mapping; it is used to input the training pseudo-sine map and output the corresponding training three-dimensional volume.

[0025] Wherein, the enhancement network model is used to input DR image groups with a vertical relationship and output corresponding training pseudo-sine images, including:

[0026] Perform pattern and feature extraction on the input DR image group, and perform a cascading operation on the extraction results;

[0027] Perform latent space feature encoding on the features after the cascading operation to obtain the encoded latent space features;

[0028] Decode the encoded latent feature space to obtain the training pseudo-sine map.

[0029] Among them, the reconstruction network model is used to input the training pseudo-sine map and output the corresponding training three-dimensional volume, including:

[0030] Perform feature capture and enhancement on the input training pseudo-sine map;

[0031] Encode the feature enhancement result, and continuously reduce the spatial resolution while keeping the number of channels unchanged;

[0032] Decode the encoded result, and continuously restore the spatial resolution to the spatial resolution of the reference three-dimensional volume to obtain the training three-dimensional volume.

[0033] Among them, calculating the loss value based on the training pseudo-sine map, the training three-dimensional volume, the reference three-dimensional volume, and the reference pseudo-sine map in the training order includes:

[0034] Calculate the loss value using the preset loss function based on the training pseudo-sine map, the training three-dimensional volume, the reference three-dimensional volume, and the reference pseudo-sine map;

[0035] Among them, the preset loss function is a function obtained by weighted summation of a first loss function and a second loss function; the first loss function is used to calculate the mean square error loss between the training pseudo-sine map and the reference pseudo-sine map; the second loss function is used to calculate the mean square error loss between the training three-dimensional volume and the reference three-dimensional volume.

[0036] In another embodiment, a three-dimensional CT imaging device based on X-ray double projection is provided. The device includes:

[0037] A storage unit configured to execute storing a preset three-dimensional volume reconstruction model. The generation of the preset three-dimensional volume reconstruction model includes: obtaining a training data set; among them, the training data set is composed of DR images of a reference object collected circumferentially; using the DR images with a vertical relationship in the training data set to train the three-dimensional volume reconstruction model to be trained to obtain a preset three-dimensional volume reconstruction model;

[0038] An acquisition unit configured to execute acquiring two DR images with a vertical relationship for the object to be imaged;

[0039] An acquisition unit configured to obtain a three-dimensional volume of an object to be imaged corresponding to the two DR images with a vertical relationship based on a preset three-dimensional volume reconstruction model;

[0040] An imaging unit configured to perform slicing on the three-dimensional volume to obtain a three-dimensional CT image of the object to be imaged.

[0041] Wherein,

[0042] The storage unit is specifically used for storing the generated three-dimensional volume reconstruction unit including: when obtaining a training data set, obtaining a reference three-dimensional volume and a reference pseudo-sine diagram of each reference object based on the training data set; wherein, the reference pseudo-sine diagram is obtained by stacking DR images of the same reference object collected in the acquisition order in a three-dimensional matrix, adding the sine diagrams in each layer column by column and normalizing; dividing the DR images of each reference object in the training data set into multiple DR image groups, each DR image group including two DR images with a vertical relationship; inputting the multiple DR image groups into the three-dimensional volume reconstruction model to be trained, training the three-dimensional volume reconstruction model to be trained; and respectively obtaining a training pseudo-sine diagram and a training three-dimensional volume; calculating a loss value based on the training pseudo-sine diagram, the training three-dimensional volume, the reference three-dimensional volume and the reference pseudo-sine diagram in the training order; if any calculated loss value is less than a target loss value, or the number of training times reaches a target number, then end the training to obtain a preset three-dimensional volume reconstruction model.

[0043] In another embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, a three-dimensional CT imaging method based on dual X-ray projection is implemented.

[0044] In another embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, a three-dimensional CT imaging method based on dual X-ray projection is implemented.

[0045] As can be seen from the above technical solutions, in the above embodiments, through the established preset three-dimensional volume reconstruction model, a three-dimensional volume of the object to be imaged can be obtained using two DR images with a vertical relationship, and the three-dimensional volume is sliced to obtain a three-dimensional CT image of the object to be imaged. This solution 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. Description of the Drawings

[0046] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0047] Figure 1 Schematic diagram of the three-dimensional CT imaging process based on X-ray double projection in the embodiments of the present application;

[0048] Figure 2 Schematic diagram of the generation process of the preset three-dimensional volume reconstruction model in the embodiments of the present application;

[0049] Figure 3 Schematic diagram of obtaining a pseudo-sine diagram based on 360 DR images in the embodiments of the present application;

[0050] Figure 4 Schematic diagram of the process of obtaining a training pseudo-sine image through an enhancement network model in the embodiments of the present application;

[0051] Figure 5 Schematic diagram of the structure of the enhancement network model in the embodiments of the present application;

[0052] Figure 6 Schematic diagram of the process of obtaining a training three-dimensional volume through a reconstruction network model in the embodiments of the present application;

[0053] Figure 7 Schematic diagram of the structure of the reconstruction network model in the embodiments of the present application;

[0054] Figure 8 Schematic diagram of the structure of the three-dimensional CT imaging device based on X-ray double projection in the embodiments of the present application;

[0055] Figure 9 Schematic diagram of the physical structure of the electronic device provided in the embodiments of the present invention. Detailed implementation manners

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0057] In the description, claims and the above-mentioned drawings of the present invention, terms such as "first", "second", "third", "fourth", etc. (if any) are used to distinguish similar objects and are not necessarily used to describe the order or sequence of the target. It should be understood that such data used can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" 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 need not be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0058] The technical solution of the present invention will be described in detail below with specific embodiments. The following several specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0059] To obtain three-dimensional CT imaging, usually the problem of three-dimensional volume reconstruction is solved first. Three-dimensional volume reconstruction methods in related technologies, such as using algorithms like FDK for three-dimensional volume reconstruction, single-projection three-dimensional volume reconstruction methods based on deep learning, etc.;

[0060] Although using algorithms like FDK for three-dimensional volume reconstruction can ensure the imaging quality, due to the large number of digital radiography (DR) images that need to be collected circumferentially, the cost of collecting data is high and the imaging speed is slow;

[0061] Using the single-projection three-dimensional volume reconstruction method based on deep learning, the detailed structures in the reconstruction results of devices with rich internal information are not well restored, that is, the imaging quality cannot be guaranteed.

[0062] Based on the above technical problems, in the embodiments of the present application, a three-dimensional CT imaging method based on X-ray double projection is proposed. By establishing a preset three-dimensional volume reconstruction model, the three-dimensional volume of the object to be imaged can be obtained using two DR images with a vertical relationship, and the three-dimensional CT imaging of the object to be imaged can be obtained by slicing the three-dimensional volume. This solution can improve the three-dimensional volume reconstruction speed on the basis of ensuring the imaging quality, and thus quickly obtain high-quality three-dimensional CT imaging.

[0063] The object to be imaged for three-dimensional volume reconstruction and the reference object for model training in the embodiments of the present application can be various types of semiconductor devices, etc.

[0064] The following will describe in detail the process of realizing three-dimensional CT imaging based on X-ray double projection in the embodiments of the present application with reference to the drawings.

[0065] See Figure 1 , Figure 1 which is a schematic diagram of the three-dimensional CT imaging process based on X-ray double projection in the embodiment of the present application. The specific steps are as follows:

[0066] Step 101, acquire two DR images with a perpendicular relationship for the object to be imaged.

[0067] In the embodiment of the present application, when acquiring DR images for the object to be imaged, it is only necessary to acquire two DR images with a perpendicular relationship, and there is no need to obtain 360 DR images by circumferentially acquiring 360 degrees.

[0068] Step 102, obtain the three-dimensional volume of the object to be imaged corresponding to the two DR images with a perpendicular relationship based on a preset three-dimensional volume reconstruction model.

[0069] Input the two acquired DR images with a perpendicular relationship into the three-dimensional volume reconstruction model, and output the three-dimensional volume of the object to be imaged.

[0070] Among them, the generation of the preset three-dimensional volume reconstruction model includes:

[0071] Obtain a training data set; among them, the training data set is composed of DR images of a reference object acquired circumferentially;

[0072] Use the DR images with a perpendicular relationship in the training data set to train the three-dimensional volume reconstruction model to be trained, and obtain the preset three-dimensional volume reconstruction model.

[0073] Step 103, slice the three-dimensional volume to obtain the three-dimensional CT image of the object to be imaged.

[0074] The present application embodiment does not limit how to slice a three-dimensional volume to obtain a three-dimensional CT image.

[0075] In the embodiment of the present application, through the established preset three-dimensional volume reconstruction model, the three-dimensional volume of the object to be imaged can be obtained by using two DR images with a perpendicular relationship, and the three-dimensional volume is sliced to obtain the three-dimensional CT image of the object to be imaged. This solution can improve the three-dimensional volume reconstruction speed on the basis of ensuring the imaging quality, and thus quickly obtain a three-dimensional CT image with high quality.

[0076] Next, in conjunction with the accompanying drawings, the process of generating the three-dimensional volume reconstruction model in the embodiment of the present application is described in detail.

[0077] See Figure 2 , Figure 2 which is a schematic diagram of the generation process of the preset three-dimensional volume reconstruction model in the embodiment of the present application. The specific steps are as follows:

[0078] Step 201: Obtain a training data set, where the training data set consists of DR images of a reference object collected circumferentially.

[0079] The obtained training data set can be a pre-stored data set or a collected data set.

[0080] The implementation of collecting and obtaining the training data set is as follows:

[0081] Collect circumferential DR images of semiconductor devices (reference objects) of multiple defect types; for each semiconductor device, 360 DR images in different directions are collected.

[0082] During the collection process, ensure that relevant parameters such as tube voltage, tube current, exposure time, and the position of the device under test remain unchanged.

[0083] The collected DR images can be pre-processed first. The pre-processing includes:

[0084] Perform size cropping, taking logarithms, normalization, etc. on the collected DR images based on chip parameters and mechanical parameters.

[0085] Step 202: Obtain the reference three-dimensional volume and reference pseudo-sine diagram of each reference object based on the training data set, where the reference pseudo-sine diagram is obtained by stacking the DR images of the same reference object collected in sequence in a three-dimensional matrix, and adding the sine diagrams in each layer column by column and normalizing them.

[0086] Obtaining the reference three-dimensional volume of each reference object includes:

[0087] Mature algorithms such as FDK can be used to obtain the reference three-dimensional volume of the reference object based on 360 DR images of each reference object.

[0088] Obtaining the reference pseudo-sine diagram of each reference object includes:

[0089] Stack 360 DR images of each reference object in sequence in a three-dimensional matrix, where each layer is a sine diagram at this time.

[0090] Add the sine diagrams of all layers column by column and normalize them to obtain a pseudo-sine diagram.

[0091] See Figure 3 , Figure 3 which is a schematic diagram of obtaining a pseudo-sine diagram based on 360 DR images in an embodiment of this application. Figure 3 On the left side in is the aligned stacking of 360 DR images. Add each column and perform normalization processing to obtain the pseudo-sine diagram on the right side.

[0092] Divide the 360 DR images corresponding to each object to be referenced into multiple DR image groups, where each DR image includes two DR images with a vertical relationship.

[0093] Multiple DR image groups of the same reference object correspond to a reference three-dimensional volume and a pseudo-sine graph.

[0094] Step 203: Divide the DR images of each reference object in the training dataset into multiple DR image groups, where each DR image group includes two DR images with a vertical relationship; and associate the DR image groups with the reference pseudo-sine graph and the reference three-dimensional volume through the reference object.

[0095] Each reference object corresponds to multiple DR image groups, a reference pseudo-sine graph, and a reference three-dimensional volume. Each DR image group can be associated with a reference pseudo-sine graph and a reference three-dimensional volume through the reference object, that is, for each DR image group, there is a pseudo-sine graph label and a three-dimensional volume label, which are used for subsequent loss value calculation.

[0096] Step 204: Input multiple DR image groups into the three-dimensional volume reconstruction model to be trained, and train the three-dimensional volume reconstruction model to be trained; and obtain the training pseudo-sine graph and the training three-dimensional volume respectively.

[0097] The three-dimensional volume reconstruction model to be trained and the preset three-dimensional volume model in the embodiments of the present application both include a feature enhancement network model and a reconstruction network model;

[0098] Among them, the enhancement network model is a model based on the encoder and decoder network architectures; it is used to input DR image groups with a vertical relationship and output the corresponding training pseudo-sine images;

[0099] The reconstruction network model is an autoencoder network model for two-dimensional to three-dimensional volume mapping; it is used to input the training pseudo-sine graph and output the corresponding training three-dimensional volume.

[0100] See Figure 4 , Figure 4 is a schematic flowchart of obtaining the training pseudo-sine image through the enhancement network model in the embodiments of the present application. The specific steps are as follows:

[0101] Step 401: Extract the patterns and features of the input DR image group, and perform a concatenation operation on the extraction results.

[0102] Step 402: Perform latent space feature encoding on the features after the concatenation operation to obtain the encoded latent space features.

[0103] Step 403: Decode the encoded latent feature space to obtain the training pseudo-sine graph.

[0104] In specific implementation, the extraction of patterns and features from the input DR image group can be achieved through three dense residual blocks. The encoding of latent space features for the features after the cascading operation can be achieved through an encoder including four downsamplers. The decoding of the encoded latent feature space to obtain the training pseudo-sine map can be achieved through a decoder including four upsampling modules. At the same time, the role of connecting context information is realized by using residual links, and finally the acquisition of the training pseudo-sine map is realized.

[0105] See Figure 5 , Figure 5 which is a schematic diagram of the enhanced network model structure in the embodiment of the present application. Figure 5 In the enhanced network model in [reference], two DR images in the input DR image group are respectively received and processed by two groups of dense residual blocks 501. Each group includes three dense residual blocks, and the three dense residual blocks are used to extract complex patterns and features. Then, the cascading operation 502 is performed to share features. The two groups of dense residual blocks 501 and the cascading operation 502 realize the retention of the original information and the enhancement of the features.

[0106] Next, the enhanced features are input into the encoder 503 composed of four downsampling blocks. The encoding process of the encoder is used to encode the latent space features. At this time, the input data is mapped to a low-dimensional space, and the main features are retained while ensuring that the number of feature channels remains unchanged, so that accurate reconstruction can be performed using the extracted spatial information and global information.

[0107] Each downsampling block is composed of three residual blocks and a max pooling layer, so that the spatial resolution is reduced to half of the original.

[0108] The encoded latent space features output by the encoder are input into the decoder 504. The decoder 504 is used to decode the latent space features to obtain the training pseudo-sine map. The decoder includes four upsampling modules for performing upsampling operations. At the same time, the role of connecting context information is realized by using residual connections, and the spatial dimension is gradually restored to the spatial resolution of the original input data space to obtain the output of the training pseudo-sine map. The spatial resolution of the original input data is the spatial resolution of the reference pseudo-sine map.

[0109] See Figure 6 , Figure 6 which is a schematic diagram of the process of obtaining the training three-dimensional volume through the reconstruction network model in the embodiment of the present application. The specific steps are as follows:

[0110] Step 601, perform feature capture and enhancement on the input training pseudo-sine map.

[0111] Step 602, encode the feature enhancement result, and continuously reduce the spatial resolution while ensuring that the number of channels remains unchanged.

[0112] Step 603: Decode the encoding result, continuously restore the spatial resolution to the spatial resolution of the reference three-dimensional volume, and obtain the training three-dimensional volume.

[0113] In specific implementation, feature capture and enhancement are achieved for the input training pseudo-sine diagram through three dense residual blocks; encoding the feature enhancement result can be implemented through an encoder including four downsamplers; decoding the encoding result can be implemented through a decoder including four upsampling modules.

[0114] See Figure 7 , Figure 7 which is a schematic structural diagram of the reconstruction network model in the embodiment of the present application. Figure 7 In the reconstruction network model in [reference], a group of dense residual blocks 701 first receives and processes the input training pseudo-sine diagram. This group of dense residual blocks includes three dense residual blocks, and feature capture and enhancement are achieved through these three dense residual blocks. Since the feature output of each channel in the residual dense group is directly related to the intermediate feature map, it has significant advantages in feature extraction in the deep neural network. During this processing period, the spatial resolution size remains unchanged, and the number of channels is continuously increased.

[0115] Then, the learned feature enhancement result is input into the encoder 702 composed of four downsampling modules. At this time, the number of channels remains unchanged, and the spatial resolution is continuously reduced. In this stage, the features are better expressed.

[0116] Finally, the encoding result is decoded through the decoder 703 composed of four upsampling modules, continuously restoring the spatial resolution to the spatial resolution of the reference three-dimensional volume, and finally obtaining the three-dimensional volume to be trained.

[0117] Step 205: Calculate the loss value based on the training pseudo-sine diagram, training three-dimensional volume, reference three-dimensional volume, and reference pseudo-sine diagram in the training order.

[0118] Each DR image group will obtain a corresponding training pseudo-sine diagram and training three-dimensional volume when input into the preset three-dimensional volume reconstruction model; calculate the loss value based on the obtained training pseudo-sine diagram and training three-dimensional volume, as well as the corresponding reference three-dimensional volume and reference pseudo-sine diagram of this DR image group.

[0119] In specific implementation, the number of DR image groups input into the preset three-dimensional volume reconstruction model each time is not limited. Whether it is parallel processing or serial processing is determined by the device capabilities.

[0120] Regardless of how the DR image groups are input, in the specific implementation of the present application, when obtaining the training pseudo-sine diagram and training three-dimensional volume for each DR image group, the loss value corresponding to this DR image group is directly calculated.

[0121] The specific process of calculating the loss value is as follows:

[0122] Calculate the loss value based on the training pseudo-sine graph, the training three-dimensional volume, the reference three-dimensional volume, and the reference pseudo-sine graph using a preset loss function;

[0123] Among them, the preset loss function is a function obtained by weighted summation of the first loss function and the second loss function;

[0124] The first loss function is used to calculate the mean squared error loss between the training pseudo-sine graph and the reference pseudo-sine graph;

[0125] The second loss function is used to calculate the mean squared error loss between the training three-dimensional volume and the reference three-dimensional volume.

[0126] The mean squared error (MSE) loss function is often used to calculate the gap between the input data and the reconstructed data of the model, and the calculation formula is as follows.

[0127]

[0128] Among them, n is the total number of data points, y i is the i-th actual observed value, is the i-th predicted value.

[0129] Based on the mean squared error loss, the preset loss function in the embodiments of the present application can be expressed as:

[0130] a×loss 1 + b×loss 2;

[0131] --

[0132] Among them, represents the first loss function, S represents the reference pseudo-sine graph, represents the training pseudo-sine graph;

[0133] represents the second loss function, V represents the reference three-dimensional volume, represents the training three-dimensional volume;

[0134] a and b represent weighted values, which can also be called proportionality coefficients.

[0135] Step 206, if any of the calculated loss values is less than the target loss value, or the number of training times reaches the target number of times, then end the training and obtain a preset three-dimensional volume reconstruction model.

[0136] For each DR image group, a loss value is obtained. Once the loss value of a DR image group is less than the target loss value, the training is ended;

[0137] If the loss value less than the target loss value has not been obtained all the time, but the number of training times reaches the target number, the training ends;

[0138] In specific implementation, if the loss value less than the target loss value has not been obtained all the time, it may be that the target loss value is set too small, or the initial parameters of the model are set unreasonably. In this case, the training needs to end; it can be directly considered that the model has been trained, or the parameters can be reset and trained again. In the embodiments of the present application, taking the initial parameters of the training as the most reasonable parameters and the training ends as an example, a preset three-dimensional volume reconstruction model is directly determined.

[0139] In the embodiments of the present application, the concept of a pseudo-sine graph is proposed for training the model. The proposed pseudo-sine graph has regular feature details and a shape structure that is more conducive to learning. Transforming it into the supervision of the intermediate step from two-dimensional projection to three-dimensional volume is more conducive to extracting effective representations of features. This soft transition using the pseudo-sine graph as mapping from two-dimensional data to three-dimensional volume can reconstruct a cross-dimensional generation network with high detail accuracy, and thus can establish a preset three-dimensional volume reconstruction model with high accuracy; further ensuring the accuracy of obtaining the three-dimensional volume;

[0140] And in the embodiments of the present application, the method of dual perspectives (two DR images with a vertical direction) is adopted to 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.

[0141] All the above optional technical solutions can be combined arbitrarily to form optional embodiments of the present disclosure, which will not be elaborated here one by one.

[0142] Based on the same inventive concept, embodiments of the present application also provide a three-dimensional CT imaging device based on X-ray dual projection. Refer to Figure 8 , Figure 8 which is a schematic structural diagram of the three-dimensional CT imaging device based on X-ray dual projection in the embodiments of the present application. The device includes:

[0143] A storage unit 801, configured to store a preset three-dimensional volume reconstruction model. The generation of the preset three-dimensional volume reconstruction model includes: obtaining a training data set; wherein, the training data set is composed of DR images of a reference object collected circumferentially; using the DR images with a vertical relationship in the training data set to train the three-dimensional volume reconstruction model to be trained, and obtaining a preset three-dimensional volume reconstruction model;

[0144] An acquisition unit 802, configured to acquire two DR images with a vertical relationship for the object to be imaged;

[0145] An obtaining unit 803, configured to obtain the three-dimensional volume of the object to be imaged corresponding to the two DR images with a vertical relationship based on the preset three-dimensional volume reconstruction model;

[0146] An imaging unit 804, configured to perform three-dimensional CT imaging of an object to be imaged by slicing a three-dimensional volume.

[0147] In one embodiment,

[0148] A storage unit 801, specifically used for storing a three-dimensional volume reconstruction unit to generate: when obtaining a training data set, obtaining a reference three-dimensional volume and a reference pseudo-sine diagram of each reference object based on the training data set; wherein, the reference pseudo-sine diagram is obtained by stacking DR images of the same reference object collected in the acquisition order in a three-dimensional matrix, adding the sine diagrams in each layer column by column and normalizing; dividing the DR images of each reference object in the training data set into multiple DR image groups, and each DR image group includes two DR images with a vertical relationship; inputting the multiple DR image groups into a three-dimensional volume reconstruction model to be trained, training the three-dimensional volume reconstruction model to be trained; and respectively obtaining a training pseudo-sine diagram and a training three-dimensional volume; calculating a loss value based on the training pseudo-sine diagram, the training three-dimensional volume, the reference three-dimensional volume, and the reference pseudo-sine diagram in the training order; if any calculated loss value is less than the target loss value, or the number of training times reaches the target number of times, then end the training to obtain a preset three-dimensional volume reconstruction model.

[0149] In another embodiment, the three-dimensional volume reconstruction model to be trained includes a feature enhancement network model and a reconstruction network model;

[0150] The enhancement network model is a model based on an encoder-decoder network architecture; it is used to input DR image groups with a vertical relationship and output corresponding training pseudo-sine images;

[0151] The reconstruction network model is an autoencoder network model for two-dimensional to three-dimensional volume mapping; it is used to input a training pseudo-sine diagram and output a corresponding training three-dimensional volume.

[0152] In another embodiment, the enhancement network model is used to input DR image groups with a vertical relationship and output corresponding training pseudo-sine images, including:

[0153] Performing pattern and feature extraction on the input DR image groups, and performing a concatenation operation on the extraction results;

[0154] Performing latent space feature encoding on the features after the concatenation operation to obtain encoded latent space features;

[0155] Decoding the encoded latent feature space to obtain a training pseudo-sine diagram.

[0156] In another embodiment, the reconstruction network model is used to input a training pseudo-sine diagram and output a corresponding training three-dimensional volume, including:

[0157] Capture and enhance the features of the input training pseudo-sine diagram;

[0158] Encode the feature enhancement result, and continuously reduce the spatial resolution while keeping the number of channels unchanged;

[0159] Decode the encoded result, continuously restore the spatial resolution to the spatial resolution of the reference three-dimensional volume, and obtain the training three-dimensional volume.

[0160] In another embodiment, calculate the loss value based on the training pseudo-sine diagram, the training three-dimensional volume, the reference three-dimensional volume, and the reference pseudo-sine diagram in the training order, including:

[0161] Calculate the loss value using a preset loss function based on the training pseudo-sine diagram, the training three-dimensional volume, the reference three-dimensional volume, and the reference pseudo-sine diagram;

[0162] Wherein, the preset loss function is a function after weighted summation of a first loss function and a second loss function; the first loss function is used to calculate the mean square error loss between the training pseudo-sine diagram and the reference pseudo-sine diagram; the second loss function is used to calculate the mean square error loss between the training three-dimensional volume and the reference three-dimensional volume.

[0163] The units in the above embodiments can be integrated into one body or separated and deployed; they can be combined into one unit or further split into multiple sub-units.

[0164] In another embodiment, an electronic device is further provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a three-dimensional CT imaging method based on X-ray double projection.

[0165] In another embodiment, a computer-readable storage medium is further provided, on which computer instructions are stored. When the instructions are executed by a processor, a three-dimensional CT imaging method based on X-ray double projection is implemented.

[0166] Figure 9 This is a schematic diagram of the physical structure of the electronic device provided by the embodiments of the present invention. As Figure 9 shown, the electronic device may include: a processor (Processor) 910, a communication interface (Communications Interface) 920, a memory (Memory) 930, and a communication bus 940. Among them, the processor 910, the communication interface 920, and the memory 930 communicate with each other through the communication bus 940. The processor 910 can call the logical instructions in the memory 930 to execute the following method:

[0167] Acquire two DR images with a vertical relationship for the object to be imaged;

[0168] Obtain the three-dimensional volume of the object to be imaged corresponding to two DR images with a vertical relationship based on a preset three-dimensional volume reconstruction model;

[0169] Slice the three-dimensional volume to obtain a three-dimensional CT image of the object to be imaged;

[0170] Among them, the generation of the preset three-dimensional volume reconstruction model includes:

[0171] Obtain a training data set; among them, the training data set is composed of DR images of a reference object collected circumferentially;

[0172] Use the DR images with a vertical relationship in the training data set to train the three-dimensional volume reconstruction model to be trained, and obtain a preset three-dimensional volume reconstruction model.

[0173] In addition, when the logical instructions in the above-mentioned memory 930 can be implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0174] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0175] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including 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 parts of the embodiments.

[0176] The flowcharts and block diagrams in the accompanying drawings of the present application illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments disclosed in the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of the code, and the above module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in the order marked in different drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, as well as the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0177] Those skilled in the art can understand that the features described in the various embodiments and / or claims of the present application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present application. In particular, without departing from the spirit and teachings of the present application, the features described in the various embodiments and / or claims of the present application can be combined and / or combined in various ways, and all such combinations and / or combinations fall within the scope disclosed in the present application.

[0178] In this article, specific embodiments are used to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention, and is not used to limit the present application. For those skilled in the art, according to the idea, spirit, and principle of the present invention, changes can be made in the specific implementation manner and application scope, and any modifications, equivalent replacements, improvements, etc. made by them shall be included within the scope protected by the present application.

Claims

1. A three-dimensional CT imaging method based on X-ray double projection, characterized in that: The method comprises: Acquire two DR images with a perpendicular relationship for the object to be imaged; Obtaining the three-dimensional volume of the object to be imaged corresponding to the two DR images having a vertical relationship based on a preset three-dimensional volume reconstruction model; Slicing the three-dimensional volume to obtain a three-dimensional CT image of the object to be imaged; Wherein, the generation of the preset three-dimensional volume reconstruction model includes: Acquire a training data set; wherein the training data set is composed of DR images of a reference object collected circumferentially; Using the DR images with a vertical relationship in the training data set to train the 3D volume reconstruction model to be trained, to obtain a preset 3D volume reconstruction model; Wherein, when obtaining the training data set, the method further comprises: Based on the training data set, a reference three-dimensional volume and a reference pseudo-sinusoidal graph of each reference object are obtained; wherein the reference pseudo-sinusoidal graph is obtained by stacking the acquired DR images of the same reference object in a three-dimensional matrix in the order of acquisition, and adding and normalizing the sinusoidal graphs in each layer by column; Dividing the DR image of each reference object in the training data set into a plurality of DR image groups, wherein the DR image group includes two DR images having a vertical relationship; The method of training the to-be-trained three-dimensional volume reconstruction model using the DR images having a vertical relationship in the training data set to obtain a preset three-dimensional volume reconstruction model includes: Inputting the plurality of DR image groups into the three-dimensional volume reconstruction model to be trained, training the three-dimensional volume reconstruction model to be trained; and obtaining a training pseudo-sinusoidal graph and a training three-dimensional volume respectively; Calculating a loss value based on the training pseudo-sinusoidal graph, the training three-dimensional volume, the reference three-dimensional volume, and the reference pseudo-sinusoidal graph in a training order; If any calculated loss value is less than the target loss value, or the number of training times reaches the target number, the training is terminated and the preset three-dimensional volume reconstruction model is obtained.

2. The method according to claim 1, characterized in that The three-dimensional volume reconstruction model to be trained includes a feature enhancement network model and a reconstruction network model; The enhanced network model is a model based on an encoder and decoder network architecture; it is used to input a DR image group with a vertical relationship and output a corresponding training pseudo-sinusoidal image; The reconstruction network model is an autoencoder network model of two-dimensional to three-dimensional volume mapping; It is used to input a training pseudo-sinusoidal image and output the corresponding training 3D volume.

3. The method according to claim 2, characterized in that The enhanced network model is used to input a DR image group having a vertical relationship and output a corresponding training pseudo-sinusoidal image, including: Performing pattern and feature extraction on the input DR image group, and performing cascade operation on the extraction results; Perform latent space feature encoding on the features after the cascade operation to obtain the encoded latent space features; The encoded latent feature space is decoded to obtain the training pseudo-sinusoidal graph.

4. The method according to claim 2, characterized in that: The reconstruction network model is used to input a training pseudo-sinusoidal graph and output a corresponding training three-dimensional volume, including: Capture and enhance the features of the input training pseudo-sinusoidal graph; Encode the feature enhancement results and continuously reduce the spatial resolution while keeping the number of channels unchanged; The encoded results are decoded, and the spatial resolution is continuously restored to the spatial resolution of the reference 3D volume to obtain the training 3D volume.

5. The method according to any one of claims 1 to 4, characterized in that: The calculating the loss value based on the training pseudo-sinusoidal graph, the training three-dimensional volume, the reference three-dimensional volume and the reference pseudo-sinusoidal graph in a training order comprises: Calculating a loss value using a preset loss function based on the training pseudo-sinusoidal graph, the training three-dimensional volume, the reference three-dimensional volume, and the reference pseudo-sinusoidal graph; Among them, the preset loss function is a function that is the weighted sum of a first loss function and a second loss function; the first loss function is used to calculate the mean square error loss of a training pseudo-sinusoidal graph and a reference pseudo-sinusoidal graph; the second loss function is used to calculate the mean square error loss of a training three-dimensional volume and a reference three-dimensional volume.

6. A three-dimensional CT imaging device based on X-ray double projection, characterized in that: The device comprises: The storage unit is configured to execute and store a preset three-dimensional volume reconstruction model, wherein the generation of the preset three-dimensional volume reconstruction model comprises: acquiring a training data set; wherein the training data set is composed of DR images of a reference object collected in a circumferential direction; using the DR images having a vertical relationship in the training data set to train the three-dimensional volume reconstruction model to obtain the preset three-dimensional volume reconstruction model; A collection unit, configured to collect two DR images having a vertical relationship with respect to the object to be imaged; An acquisition unit is configured to execute, based on a preset three-dimensional volume reconstruction model, obtaining a three-dimensional volume of the object to be imaged corresponding to the two DR images having a vertical relationship; An imaging unit, configured to slice the three-dimensional volume to obtain three-dimensional CT imaging of the object to be imaged; in, The storage unit is specifically used to store a three-dimensional volume reconstruction unit, which includes: when obtaining a training data set, obtaining a reference three-dimensional volume and a reference pseudo-sine map of each reference object based on the training data set; wherein the reference pseudo-sine map is obtained by stacking the acquired DR images of the same reference object in a three-dimensional matrix in an acquisition order, adding the sinusoidal maps in each layer by column and normalizing them; dividing the DR image of each reference object in the training data set into a plurality of DR image groups, wherein the DR image group includes two DR images with a vertical relationship; inputting the plurality of DR image groups into the three-dimensional volume reconstruction model to be trained, and training the three-dimensional volume reconstruction model to be trained; and obtaining a training pseudo-sine map and a training three-dimensional volume respectively; calculating a loss value based on the training pseudo-sine map, the training three-dimensional volume, the reference three-dimensional volume and the reference pseudo-sine map in a training order; if any loss value calculated is less than a target loss value, or the number of training times reaches the target number, the training is terminated to obtain a preset three-dimensional volume reconstruction model.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

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

Images