Method, program product and device for 3D CT image reconstruction based on multi-view projection

Abstract

The application discloses a three-dimensional CT image reconstruction method based on multi-view projection, a program product and equipment, and the method comprises the following steps: acquiring a plurality of projection images of a three-dimensional target to be reconstructed collected under different views and initial volume data corresponding to the three-dimensional target to be reconstructed; expanding the channel dimension of the projection images by using a dimension expansion network in a pre-trained three-dimensional reconstruction model to obtain an expansion projection image corresponding to each of the projection images; taking each of the expansion projection images as an input of a feature extraction network in the three-dimensional reconstruction model to extract multi-scale feature data corresponding to each of the projection images; based on the initial volume data, obtaining multi-scale implicit neural representation data corresponding to the three-dimensional target to be reconstructed by using an implicit neural learning network in the three-dimensional reconstruction model; and based on the multi-scale feature data corresponding to each of the projection images and the multi-scale implicit neural representation data, performing three-dimensional image reconstruction by using a feature reconstruction network in the three-dimensional reconstruction model to output a three-dimensional reconstruction target.

Description

Technical Field

[0001] This invention relates to the field of computer tomography imaging technology, and in particular to a method and computing device for three-dimensional CT image reconstruction based on multi-view projection. Background Technology

[0002] Computed Tomography (CT) technology, with its non-destructive, efficient, and non-contact characteristics, has broad application prospects in industry. CT technology can accurately describe the internal information of the object being measured. Currently, CT imaging technology is mainly divided into two categories: traditional reconstruction and deep learning reconstruction. Traditional reconstruction methods can be broadly classified into two types: analytical reconstruction algorithms and iterative reconstruction algorithms. Through hundreds of X-ray projections, traditional reconstruction algorithms can approximately reconstruct the CT volume. Due to the limitations of the Nyquist sampling theorem, the key to high-quality reconstruction by traditional reconstruction algorithms is obtaining complete projections, which greatly limits the reconstruction speed and application scenarios. Subsequently, researchers introduced compressed sensing into CT reconstruction and developed a regularization-based reconstruction method, integrating prior information into the reconstruction model, successfully reducing the number of required projection images. However, in some application scenarios with limited conditions, only a very limited number of projection images can be obtained, such as online inspection of industrial products and transient damage dynamic testing. Relying on traditional methods and regularization methods cannot obtain high-quality results. Therefore, reconstructing images from ultra-sparsely sampled projections to accelerate the CT imaging process has become a hot research topic.

[0003] Recently, with the rapid development of deep learning, it has been applied to sparse and ultra-sparse view CT reconstruction. By using artificially designed neural networks to directly learn the mapping function of X-ray projection onto the 3D CT volume, very good results have been achieved. Compared with traditional regularization-based methods, deep learning methods extract prior information for predicting CT through data-driven end-to-end networks, thus enabling more feasible CT reconstruction.

[0004] However, deep learning-based reconstruction methods also face common limitations in application. First, it is difficult to obtain large-scale training datasets, which may constitute a significant bottleneck in certain specific application scenarios. Furthermore, deep learning-based CT reconstruction technology performs poorly in generalization, making it difficult to adapt to diverse imaging objects and limiting its wide applicability. Moreover, current 3D CT reconstruction algorithms generally use 3D convolution as the core computational unit, leveraging the spatial dimension-aware characteristics of 3D convolution to achieve feature mapping from projection data to voxel space. However, cascaded 3D convolution leads to a sharp increase in the number of parameters, significant memory consumption, severely restricting the reconstruction voxel dimension and reconstruction speed, and exacerbating computational resource consumption, placing higher demands on memory. Summary of the Invention

[0005] To address the existing technical problems, this invention provides a method and computing device for three-dimensional CT image reconstruction based on multi-view projection, which can quickly and with high quality reconstruct 3D CT images from projection images with fewer viewpoints, thereby improving the speed and quality of three-dimensional reconstruction.

[0006] In a first aspect, a method for reconstructing three-dimensional CT images based on multi-view projection is provided, comprising: acquiring multiple projected images of a three-dimensional target to be reconstructed from different viewpoints and initial volume data corresponding to the three-dimensional target to be reconstructed; using a dimensionality expansion network in a pre-trained three-dimensional reconstruction model to expand the channel dimensions of the projected images to obtain expanded projected images corresponding to each projected image; using each expanded projected image as input to a feature extraction network in the three-dimensional reconstruction model to extract multi-scale feature data corresponding to each projected image; based on the initial volume data, using an implicit neural learning network in the three-dimensional reconstruction model to obtain multi-scale implicit neural representation data corresponding to the three-dimensional target to be reconstructed; and based on the multi-scale feature data corresponding to each projected image and the multi-scale implicit neural representation data, using a feature reconstruction network in the three-dimensional reconstruction model to reconstruct the three-dimensional image and output the three-dimensional reconstructed target.

[0007] In a second aspect, a computing device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform a three-dimensional CT image reconstruction method based on multi-view projection as described in any embodiment of this application.

[0008] Thirdly, a computer program product is provided, including a computer program that, when executed by a processor, implements the three-dimensional CT image reconstruction method based on multi-view projection as described in any embodiment of this application.

[0009] This application acquires projection images from multiple viewpoints and expands these images using a dimensionality expansion network. This gradually embeds the depth information of the CT volume of the target to be reconstructed into different channels of the feature map, resulting in expanded projection images corresponding to each projection image. A feature extraction network extracts feature maps at different levels from the expanded projection images, obtaining scale feature data at multiple scales. Based on the initial volume data, an implicit neural learning network is used to represent the CT volume of the target to be reconstructed, thus incorporating the spatial geometric information of the target into the 3D reconstruction model, facilitating accurate reconstruction of the 3D image of the target. Then, the feature data and implicit neural representation data at each scale are used as inputs to each layer of the feature reconstruction network. Through continuous feature reconstruction and fusion at each layer of the feature reconstruction network, volumetric details are gradually restored, and finally, the 3D reconstructed target is output. This application can quickly and with high quality reconstruct 3D CT images using projection images from fewer viewpoints, improving both the speed and quality of 3D reconstruction. Attached Figure Description

[0010] Figure 1 This is an application environment diagram of a three-dimensional CT image reconstruction method based on multi-view projection in one embodiment;

[0011] Figure 2 This is a flowchart of a three-dimensional CT image reconstruction method based on multi-view projection in one embodiment;

[0012] Figure 3 This is a network diagram of a 3D reconstruction model based on multi-view projection in one embodiment;

[0013] Figure 4 This is an example diagram of the network structure of a three-dimensional reconstruction model in one embodiment;

[0014] Figure 5 This is an example network diagram for training a 3D reconstruction model in one embodiment;

[0015] Figure 6 This is a schematic diagram of a three-dimensional CT image reconstruction device based on multi-view projection in one embodiment;

[0016] Figure 7 This is a schematic diagram of a computing device in one embodiment. Detailed Implementation

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0019] In the following description, the expression “some embodiments” refers to a subset of all possible embodiments. However, it should be understood that “some embodiments” can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.

[0020] See Figure 1 This diagram illustrates the application environment of a multi-view projection-based 3D CT image reconstruction method in one embodiment. The multi-view projection-based 3D CT image reconstruction method is applied in a computing device 10, which can acquire projection images from multiple perspectives. These projection images are obtained through a detection device that collects data from the 3D target to be reconstructed. The X-ray source in the detection device rotates around the object under test, acquiring projection images of the object from different angles. Circumferential X-ray projection images of the 3D target to be reconstructed are acquired, ensuring that relevant parameters such as tube voltage, tube current, and amplification ratio remain constant. Projection data from at least two perspectives are selected from the circumferential projection to form multiple projection images. It can be understood that a single projection image represents the projection data collected from the 3D target to be reconstructed from one perspective. The computing device 10 can reconstruct the 3D target based on the multi-view projection-based 3D CT image reconstruction method using projection images from multiple perspectives.

[0021] Please see Figure 2 This is a flowchart illustrating a three-dimensional CT image reconstruction method based on multi-view projection according to an embodiment of this application. The three-dimensional CT image reconstruction method based on multi-view projection is applied in a computing device and includes the following steps:

[0022] S20. Acquire multiple projected images of the 3D target to be reconstructed from different perspectives and the initial volume data corresponding to the 3D target to be reconstructed.

[0023] In this embodiment, there can be at least two projected images. For example, when there are two projected images, two mutually orthogonal viewpoints can be selected, namely the first projected image and the second projected image. The initial volume data refers to the initial volume data of the 3D target to be reconstructed. For example, if a volume with a resolution of 128 needs to be reconstructed, then 128 points are uniformly sampled between -1 and 1 in the X-axis, Y-axis and Z-axis directions.

[0024] S21. Using the dimension expansion network in the pre-trained 3D reconstruction model, the channel dimensions of the projected image are expanded to obtain the expanded projected image corresponding to each of the projected images.

[0025] In this embodiment, the 3D reconstruction model is obtained through a training dataset. For example... Figure 3 As shown, the 3D reconstruction model includes a dimensionality expansion network, a feature extraction network, an implicit neural learning network, and a feature reconstruction network. The dimensionality expansion network expands the number of channels in the image's channel dimension. As the number of channels gradually increases in a stepwise manner, the depth information of the CT volume of the target to be reconstructed is gradually embedded into different channels of the feature map. The feature extraction network extracts feature maps of the target to be reconstructed at different scales from the projected image. The implicit neural learning network learns the CT volume representation of the target to be reconstructed through implicit neural representations, thus incorporating the spatial geometric information of the target into the 3D reconstruction model, fully learning prior information, and improving the model's generalization ability. The feature reconstruction network reconstructs the results of the implicit neural representation learning into results that match the dimensions of each stage of the decoder, and then fuses them with the decoder's results and the results of the feature reconstruction module itself. Through continuous feature reconstruction and fusion, volumetric details are gradually restored, and finally, the 3D reconstructed target is output.

[0026] The dimension expansion network, encoder, decoder, implicit neural learning network, and feature reconstruction network are respectively F de F encoder F decoder F inr F fr The mapping function F from multiple projected images fitted to the 3D reconstruction model to the 3D CT volume can be described by the following formula:

[0027]

[0028] in This indicates a cascaded network model. This indicates that the network model is connected in parallel.

[0029] In this embodiment, since multiple projected images are acquired from different viewpoints, the directions of the acquired rays are different. Therefore, one viewpoint can be selected as a reference viewpoint, and then, with this reference viewpoint as a reference, the channel dimensions of the projected images from other viewpoints are expanded, and an alignment operation is performed to make the expanded projected images corresponding to each projected image in the same coordinate system, thereby achieving dimensional uniformity.

[0030] S22. Use each extended projection image as input to the feature extraction network in the three-dimensional reconstruction model to extract multi-scale feature data corresponding to each projection image.

[0031] In this embodiment, the projected images from each viewpoint correspond to their respective dimension expansion networks and feature extraction networks, which output multi-scale feature data for each projected image. Then, the multi-scale feature data for each projected image is used as input to different layers in the feature reconstruction network. The multi-scale feature data includes feature data at multiple scales.

[0032] S23. Based on the initial volume data, the implicit neural learning network in the 3D reconstruction model is used to obtain the multi-scale implicit neural representation data corresponding to the 3D target to be reconstructed.

[0033] In this embodiment, the multi-scale implicit neural representation data includes implicit neural representation data at multiple scales. Each scale implicit neural representation data represents the implicit neural representation data of the volume data of the 3D target to be reconstructed at that scale.

[0034] S24. Based on the multi-scale feature data and multi-scale implicit neural representation data corresponding to each projection image, the feature reconstruction network in the three-dimensional reconstruction model is used to reconstruct the three-dimensional image and output the three-dimensional reconstruction target.

[0035] In this embodiment, feature data at each scale and implicit neural representation data at each scale are used as inputs to each layer in the feature reconstruction network. By continuously reconstructing and fusing features through each layer in the feature reconstruction network, volumetric details are gradually restored, and finally, the three-dimensional reconstruction target is output.

[0036] In the above embodiments, by acquiring projection images from multiple perspectives, the projection images are expanded using a dimensionality expansion network, gradually embedding the depth information of the CT volume of the target to be reconstructed into different channels of the feature map, thus obtaining expanded projection images corresponding to each projection image; feature maps at different levels in the expanded projection images are extracted using a feature extraction network, resulting in scale feature data at multiple scales; and based on the initial volume data, the CT volume representation of the target to be reconstructed is represented using an implicit neural learning network, thereby adding the spatial geometric information of the target to be reconstructed into the 3D reconstruction model, facilitating the accurate reconstruction of the 3D image of the target to be reconstructed; then, the feature data at each scale and the implicit neural representation data at each scale are used as inputs to each layer in the feature reconstruction network, and the features are continuously reconstructed and fused through each layer in the feature reconstruction network, gradually restoring volume details, and finally outputting the 3D reconstructed target. This application can quickly and with high quality reconstruct 3D CT images using projection images from fewer perspectives, improving the speed and quality of 3D reconstruction.

[0037] In some embodiments, the dimension expansion network includes a gradient calculation layer, a gradient stitching layer, and a dimension expansion layer. The step of using the dimension expansion network in a pre-trained 3D reconstruction model to expand the channel dimensions of the projected images to obtain expanded projected images corresponding to each of the projected images includes:

[0038] Based on the projected image, the gradients in the horizontal and vertical directions are calculated using a gradient calculation layer to obtain the gradient image corresponding to the projected image.

[0039] Using a gradient stitching layer, the projected image and the gradient image corresponding to the projected image are summed to obtain the composite image corresponding to the projected image. The projected image, the gradient image corresponding to the projected image, and the composite image corresponding to the projected image are then stitched together to obtain the three-channel initial image corresponding to the projected image.

[0040] Based on multiple cascaded residual blocks in the dimension expansion layer, feature maps are extracted step by step from the initial three-channel image to obtain the expanded projection image corresponding to the projection image.

[0041] In this embodiment, for example, the projected image can be an image formed from unprocessed projected data, or an image formed from projected data after taking the logarithm. The projected image is denoted by {I1,I2}∈R. H×W×1 Using the Schrödinger gradient, we obtain {I3,I4}∈R H×W×1 The formula for calculating the Scharr gradient is as follows:

[0042]

[0043] Where A represents the image to be processed. G represents convolution. x (i,j) represents the gradient in the horizontal direction, G y (i,j) represents the gradient in the vertical direction, G(i,j) represents the gradient image, θ represents the gradient direction, i represents the coordinate in the horizontal direction, and j represents the coordinate in the vertical direction. The Scharr gradient can more accurately approximate the true gradient of an image and considers more neighborhood information when calculating the gradient, thus exhibiting some robustness to noise. Using the above method, the gradient image I3 corresponding to I1 and the gradient image I4 corresponding to I2 can be calculated. Then, the projected image and the gradient images are summed to obtain the synthesized image {I1+I3,I2+I4}∈R. H×W×1 Finally, {I1,I2}∈R H×W×1 {I3,I4}∈R H×W×1 and {I1+I3,I2+I4}∈R H×W×1 The images are stitched together along the channel dimension to generate the initial image {I5,I6}∈R.H×W×3 Next, the initial three-channel image undergoes channel expansion, encoding the channel dimension as a depth dimension, and the CT volume depth is inferred using the feature channels. Channel expansion consists of multiple cascaded ResBlock blocks. As the ResBlock blocks are executed, the number of channels in the feature map increases stepwise. During this process, the network continuously explores depth information and progressively embeds 3D information into different channels of the feature map, resulting in an expanded projection image corresponding to the projected image. For example, channel expansion consists of four cascaded ResBlock blocks. As the ResBlock blocks are executed, the number of channels in the feature map increases stepwise, successively from 3 to 16 to 32 to 64 to 128. After expanding the initial three-channel image, for non-reference projection images (projection images acquired from non-reference viewpoints), an alignment network is then used for alignment, resulting in an expanded projection image corresponding to the projected image.

[0044] In the above embodiments, the gradient image of the projection image is obtained, and a three-channel initial image is obtained based on the gradient image and the projection image. The three-channel initial image is then expanded in dimension. In this process, the depth information of the CT volume is gradually embedded into different channels of the feature map, which makes it easier to reconstruct the three-dimensional object more accurately and improve the reconstruction accuracy.

[0045] In some embodiments, the feature extraction network includes an encoder and a decoder, the decoder is connected to the feature reconstruction network, and the implicit neural learning network is connected to the feature reconstruction network;

[0046] The encoder includes multiple encoding layers, and the decoder includes multiple decoding layers. Encoding layers located in the same scale layer are skip-connected to their corresponding decoding layers. The feature reconstruction network includes multiple reconstruction layers with the same number of layers as the decoder and reconstruction output layers connected to the reconstruction layers. The input of the decoding layer located in the same scale layer serves as the input of the corresponding reconstruction layer. The feature reconstruction network includes a multi-scale fusion network. The implicit neural learning network is connected to the multi-scale fusion network. The multi-scale fusion network is connected to each layer of the feature reconstruction network. The multi-scale fusion network is used to output implicit neural representation data at each scale.

[0047] like Figure 4 As shown, Figure 4This is an example diagram of the network structure of a 3D reconstruction model in one embodiment. Multiple projected images include a first projection image and a second projection image. The first projection image corresponds to a dimensionality expansion network and a feature extraction network. The second projection image also corresponds to a dimensionality expansion network and a feature extraction network. The first projection image is a projected image acquired from a reference viewpoint. The second projection image also corresponds to an alignment network, used to transform the expanded image and align it with the reference viewpoint, achieving dimensionality unification. The encoder includes multiple downsampling layers, for example... Figure 4 The diagram shows four downsampling layers, designated A1, A2, A3, and A4. The decoder consists of four decoding layers, each including an upsampling layer and a connection layer. Figure 4 The algorithm includes four decoding layers: B4, B3, B2, and B1. Adjacent upsampling layers are connected by connection layers. Since the feature maps output by A1 and B1 are at the same scale, A1 and B1 are in the same scale layer. Similarly, A2 and B2, A3 and B3, and A4 and B4 are in the same scale layer. The encoding layers are skip-connected to their corresponding decoding layers. Specifically, the input of the encoding layer at the same scale layer is skip-connected to the output of the downsampling layer in the corresponding decoding layer. For example, the input of A1 is skip-connected to the output of the downsampling layer in B1. The network structures of the dimension expansion network and feature extraction network corresponding to each projection are the same, only their inputs differ.

[0048] The encoder takes the expanded projection image (after channel number expansion) as input and progressively compresses and encodes the high-dimensional feature map. As downsampling proceeds, the size of the feature map is continuously compressed, but the number of feature channels remains constant, consistently matching the depth of the target CT volume data. During this process, the receptive field gradually expands, allowing it to perceive more low-frequency information. During encoding, skip connections are used to directly pass feature information from different levels to the decoder, sharing feature information to enhance information utilization. Figure 4 As shown, the encoder consists of four cascaded downsampling layers. Each downsampling layer comprises a 2D residual block ResBlock layer followed by a pooling layer. Through progressive downsampling operations, the feature map size is gradually reduced.

[0049] The decoder takes the encoded low-dimensional high-dimensional features as input, gradually reconstructs the high-dimensional features, and restores image details. For example... Figure 4 As shown, the decoder includes four concatenation and fusion operations in the connection layer and four upsampling layers. The connection layer is used to skip-connect the input of the coding layer at the same scale to the output of the corresponding decoding layer. For example, the input of A1 and the output of the downsampling layer in B1 are concatenated and fused through the connection layer in B1.

[0050] Each upsampling layer consists of a set of "2D deconvolution layer → 2D ResBlock layer → 2DResBlock layer". Through progressive upsampling operations, the feature map size gradually increases, but the number of feature map channels remains unchanged. As upsampling proceeds, the model gradually fuses the information from the decoder itself with the feature information from the corresponding layers of the encoder, making full use of multi-scale information to gradually restore the structure and details of the target image.

[0051] In this embodiment, the implicit neural learning network is connected to the multi-scale fusion network in the feature reconstruction network. Initial volume data serves as the input to the implicit neural learning network. First, the spatial coordinates of the volume to be reconstructed are determined. Then, Fourier feature mapping is used to encode the spatial coordinates. The result of the positional encoding is then input into a perceptron neural network (MLP) for implicit neural representation learning. This learns the continuous implicit neural representation of the CT volume, embedding prior information from the image into the network parameters. Finally, the implicit neural representation learning and the decoder are combined through the feature reconstruction module. The perceptron neural network includes multiple sequentially connected linear interpolation modules. By learning the CT volume representation of the object to be reconstructed through implicit neural representation, the spatial geometric information of the object to be reconstructed is added to the 3D reconstruction model. This allows for sufficient learning of prior information, thereby improving the model's generalization ability.

[0052] The feature reconstruction network consists of a multi-scale feature fusion network and multiple reconstruction layers. The multi-scale feature fusion block is constructed using bilinear interpolation, which reconstructs the results of implicit neural representation learning into feature maps that match the dimensions of each stage of the decoder. A reconstruction layer is an upsampling block, which includes a set of connection layers: Concat → 2D convolutional layer (Conv layer) → 2D deconvolution layer (Deconvolution layer) → 2D convolutional layer (Conv layer) → 2D normalization layer (Batch Normalization layer) → ReLU activation layer. The reconstruction output layer includes a concatenation operation and two 2D convolutions.

[0053] like Figure 4 As shown, the feature reconstruction network consists of four reconstruction layers. The feature maps output by B4 and C4 have the same scale and are located on the same scale layer. Similarly, B3 and C3 are located on the same scale layer, B2 and C2 are located on the same scale layer, and B1 and C1 are located on the same scale layer. For example, the feature maps output by the multi-scale fusion network are D4, D3, D2, D1, and D0, where the scale of D4 is the same as the scale of the feature map input by C4, and similarly, the scale of D3 is the same as the scale of the feature map input by C3, and so on, with the scale of D0 being the same as the scale of the feature map output by C1.

[0054] The inputs of B4 corresponding to the first projection map, the input of B4 corresponding to the second projection map, and the scale implicit neural representation data at the same scale as the input of B4 are used as the inputs to the C4 reconstruction layer, and this process is repeated layer by layer until the reconstruction layer C1 outputs the maximum scale feature data. Then, the outputs of B1 corresponding to the first projection map, the output of B1 corresponding to the second projection map, the final reconstructed feature data output by the final reconstruction layer C1, and the scale implicit neural representation data D0 corresponding to the reconstruction output layer are used as the inputs to the reconstruction output layer, and finally, the 3D reconstructed target is output.

[0055] Optionally, the step of reconstructing the 3D image using the feature reconstruction network in the 3D reconstruction model based on the multi-scale feature data corresponding to each of the projected images and the multi-scale implicit neural representation data, and outputting the 3D reconstruction target, includes:

[0056] For the current reconstruction layer at the current layer number, obtain the current scale feature input data corresponding to each projected image, where the current scale feature input data is the input of the current decoding layer connected to the current reconstruction layer in the decoder corresponding to each projected image;

[0057] Obtain the output of the previous reconstruction layer connected to the current reconstruction layer;

[0058] Based on the initial volume data, obtain the current scale implicit neural representation data corresponding to the current layer number output by the multi-scale fusion network;

[0059] Based on the current scale feature input data corresponding to each projected image, the output of the previous reconstruction layer, and the current scale implicit neural representation data, the current reconstruction feature data output by the current reconstruction layer is obtained, and this process is repeated until the final reconstruction feature data of the last reconstruction layer is obtained.

[0060] Obtain the maximum scale feature data of the final decoding layer output corresponding to each projection image, and thus obtain the maximum scale feature data of each projection image.

[0061] Based on the maximum scale feature data corresponding to each projected image, the scale implicit neural representation data corresponding to the reconstruction output layer, and the final reconstruction feature data, the predicted three-dimensional reconstruction target is output through the reconstruction output layer.

[0062] The results from the two decoders are concatenated with the results from the implicit neural learning network using a multi-scale feature fusion block and the results from the feature reconstruction module itself, along with the concatenation in the channel dimension. This concatenation is then applied via a 2D convolution operation to achieve effective feature information fusion. Through gradual, simple upsampling operations, the feature map size is progressively increased, while the number of channels remains constant. Finally, a concatenation operation and two 2D convolutions are used to output the 3D reconstructed target.

[0063] In this embodiment, the current reconstruction feature data output by the current reconstruction layer is used as the input of the next reconstruction layer. In this way, the next layer is updated to the current reconstruction layer, and so on, until the final reconstruction feature data of the last reconstruction layer is obtained.

[0064] In the above embodiments, by splicing the results of the two decoders with the results of the implicit neural learning network through the multi-scale feature fusion block and the results of the feature reconstruction module itself in the channel dimension, the fusion of features of different scales in the projected image can be realized, which facilitates the more accurate reconstruction of the three-dimensional object and improves the reconstruction accuracy.

[0065] In some embodiments, the method further includes:

[0066] Obtain a training dataset, wherein each training sample in the training dataset includes multiple sample projection images of a 3D sample target acquired from multiple perspectives, initial sample volume data corresponding to the 3D sample target, and 3D reconstructed volume labels corresponding to the 3D sample target;

[0067] Based on the training dataset, obtain the input training sample for the current iteration, iteratively train the initial 3D reconstruction model, obtain the 3D reconstruction sample for the current iteration, and calculate the current total loss value based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample for the current iteration.

[0068] Based on the current total loss value, determine whether the current iteration meets the iteration termination condition. If the current iteration does not meet the iteration termination condition, continue to acquire input training samples for iterative training until the iteration termination condition is met. The 3D reconstruction model after meeting the iteration termination condition is used as the pre-trained 3D reconstruction model.

[0069] In this embodiment, the 3D reconstructed volume labels are obtained through traditional reconstruction algorithms, such as the FDK algorithm, reconstructing circumferential projection images. It can be understood that one 3D sample target corresponds to a set of sample projection images and 3D reconstructed volume labels. For example... Figure 5 As shown, Figure 5 This is an example network diagram for training a 3D reconstruction model in one embodiment. During training, the network of the 3D reconstruction model is the same as the pre-trained network. However, the network parameters in the 3D reconstruction model are not yet fully trained and need to be iteratively updated. For example, if a 3D sample target includes two sample projection images, the two sample projection images and the initial sample volume data are input into the trained 3D reconstruction model to obtain the 3D reconstruction sample corresponding to the current iteration. The network processing during training is the same as the process of reconstructing the 3D target in the actual scenario described above, i.e., it is similar to... Figure 2The process of obtaining the 3D reconstructed target from the mid-scale reconstruction is the same and will not be repeated here. The initial sample volume data is an initial representation of the volume of the 3D sample target. The iteration termination conditions include, but are not limited to, the number of iterations exceeding a preset number, and the current total loss value being less than a preset error value, etc.

[0070] In this embodiment, to effectively improve reconstruction quality and accuracy and enhance model robustness, this application constructs a multi-dimensional constrained loss function. This loss function includes reconstruction loss, structural similarity loss, gradient loss, and a projection loss with explicit physical meaning. Reconstruction loss measures the geometric difference between the predicted 3D structure and the real structure, ensuring that the model reconstructs the most accurate 3D model possible in space. Structural similarity loss focuses on the visual quality and structural consistency of the image; it typically assesses the similarity between the predicted image and the real image, better capturing the human eye's sensitivity to image quality. Gradient loss focuses on the edge and texture information of the image, helping the model better recover the structural information and edge details of the image, avoiding blurring. The projection loss with explicit physical meaning considers the specific physical processes of CT imaging, providing clear guidance for network learning, allowing the network to learn the physical processes of CT imaging, thereby enhancing the model's generalization and interpretability.

[0071] Optionally, based on the 3D reconstruction volume labels corresponding to the input training samples and the 3D reconstruction samples of the current iteration, the current total loss value is calculated as follows:

[0072] Based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration, the current reconstruction loss value is calculated using the first loss function, wherein the current reconstruction loss value is used to indicate the structural difference between the predicted 3D reconstruction sample and the 3D reconstruction volume label;

[0073] Based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration, the current structural similarity loss value is calculated using the second loss function, wherein the current structural similarity loss value is used to indicate the similarity difference between the predicted 3D reconstruction sample and the 3D reconstruction volume label.

[0074] Based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration, the current gradient loss is calculated using the third loss function, wherein the current gradient loss is used to indicate the difference in image detail information between the predicted 3D reconstruction sample and the 3D reconstruction volume label;

[0075] Based on the 3D reconstruction volume labels corresponding to the input training samples and the 3D reconstruction samples of the current iteration, the current projection loss is calculated using the fourth loss function, wherein the current projection loss is used to indicate the physical process loss of CT imaging.

[0076] The current total loss value is obtained by weighting the current reconstruction loss value, the current structural similarity loss value, the current gradient loss, and the current projection loss.

[0077] Optionally, the step of calculating the current gradient loss based on the 3D reconstruction volume labels corresponding to the input training samples and the 3D reconstruction samples of the current iteration, using a third loss function, includes:

[0078] Using the mean squared error function, the horizontal gradient loss in the horizontal direction between the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration is calculated, and the vertical gradient loss in the vertical direction between the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration is calculated.

[0079] The current gradient loss is calculated based on the horizontal gradient loss and the vertical gradient loss.

[0080] Optionally, the step of calculating the current projection loss based on the 3D reconstruction volume labels corresponding to the input training samples and the 3D reconstruction samples of the current iteration, using the fourth loss function, includes:

[0081] Based on the 3D reconstruction samples of the current iteration, calculate the current projection data of the current iteration;

[0082] Calculate the projection label based on the 3D reconstructed volume label;

[0083] The current projection loss is calculated based on the current projection data and the projection label.

[0084] In this embodiment, the formula for calculating the reconstruction loss is as follows:

[0085] Reconstruction losses: Where Y pred For the predicted 3D reconstruction sample, Y truth This represents the volume label for 3D reconstruction, where L RE (Y pred ,Y truth () indicates reconstruction losses.

[0086] Structural similarity loss L RS (Y pred ,Y truth ) = 1 - SSIM(Y pred ,Y truth ), where SSIM(Y pred ,Y truth ) represents Y pred With Y truth Similarity between them.

[0087] Gradient loss MSE x MSE represents the horizontal gradient loss. y This represents the vertical gradient loss.

[0088]

[0089] Projection loss

[0090] P(x)=∫σ(x)dx

[0091] Total loss L total =λ1L RE +λ2L SSIM +λ3L GL +λ4L PL Where λ1, λ2, λ3, and λ4 are weighting coefficients, which control the relative importance of different loss terms, and σ(x) is the density value. P(Y pred P(Y) represents the projection data of the 3D reconstructed sample. truth ) represents the projection label corresponding to the 3D reconstruction volume label, where D, H, and W represent length, height, and width.

[0092] In the above embodiments, 3D CT images can be rapidly and reconstructed with high quality through multi-view projection. Simultaneously, coupling implicit neural representations and introducing projection loss consistent with the physical principles of CT scanning improves the network's generalization ability. Finally, gradient constraints are introduced to enhance the network's utilization of prior information, accelerate model convergence, and strengthen the network's edge recovery capabilities. This application has broad development prospects and application value in scenarios where full-view projection is unavailable or near real-time reconstruction is required.

[0093] In another aspect, this application provides a computer program product, including a computer program that, when executed by a processor, implements the three-dimensional CT image reconstruction method based on multi-view projection as described in any embodiment of this application.

[0094] In the computer program product, the optional implementation form of the program module architecture of the computer program that implements each step of the target recognition method can be a three-dimensional CT image reconstruction device based on multi-view projection.

[0095] Please see Figure 6This application provides an embodiment of a three-dimensional CT image reconstruction device based on multi-view projection, comprising: an acquisition module 61, used to acquire multiple projection images of a three-dimensional target to be reconstructed from different viewpoints and initial volume data corresponding to the three-dimensional target to be reconstructed; an expansion module 62, used to expand the channel dimensions of the projection images using a dimensionality expansion network in a pre-trained three-dimensional reconstruction model to obtain expanded projection images corresponding to each projection image; an extraction module 63, used to extract multi-scale feature data corresponding to each projection image by using each expanded projection image as input to a feature extraction network in the three-dimensional reconstruction model; an implicit neural learning module 64, used to obtain multi-scale implicit neural representation data corresponding to the three-dimensional target to be reconstructed based on the initial volume data and using an implicit neural learning network in the three-dimensional reconstruction model; and a reconstruction module 65, used to reconstruct the three-dimensional image based on the multi-scale feature data corresponding to each projection image and the multi-scale implicit neural representation data, using a feature reconstruction network in the three-dimensional reconstruction model, and output the three-dimensional reconstructed target.

[0096] Optionally, extension module 62 is used for:

[0097] Based on the projected image, the gradients in the horizontal and vertical directions are calculated using a gradient calculation layer to obtain the gradient image corresponding to the projected image.

[0098] Using a gradient stitching layer, the projected image and the gradient image corresponding to the projected image are summed to obtain the composite image corresponding to the projected image. The projected image, the gradient image corresponding to the projected image, and the composite image corresponding to the projected image are then stitched together to obtain the three-channel initial image corresponding to the projected image.

[0099] Based on multiple cascaded residual blocks in the dimension expansion layer, feature maps are extracted step by step from the initial three-channel image to obtain the expanded projection image corresponding to the projection image.

[0100] Optionally, the feature extraction network includes an encoder and a decoder, the decoder is connected to the feature reconstruction network, and the implicit neural learning network is connected to the feature reconstruction network;

[0101] The encoder includes multiple encoding layers, and the decoder includes multiple decoding layers. Encoding layers located in the same scale layer are skip-connected to their corresponding decoding layers. The feature reconstruction network includes multiple reconstruction layers with the same number of layers as the decoder and reconstruction output layers connected to the reconstruction layers. The input of the decoding layer located in the same scale layer serves as the input of the corresponding reconstruction layer. The feature reconstruction network includes a multi-scale fusion network. The implicit neural learning network is connected to the multi-scale fusion network. The multi-scale fusion network is connected to each layer of the feature reconstruction network. The multi-scale fusion network is used to output implicit neural representation data at each scale.

[0102] Optionally, the refactoring module 65 is also used for:

[0103] For the current reconstruction layer at the current layer number, obtain the current scale feature input data corresponding to each projected image, where the current scale feature input data is the input of the current decoding layer connected to the current reconstruction layer in the decoder corresponding to each projected image;

[0104] Obtain the output of the previous reconstruction layer connected to the current reconstruction layer;

[0105] Based on the initial volume data, obtain the current scale implicit neural representation data corresponding to the current layer number output by the multi-scale fusion network;

[0106] Based on the current scale feature input data corresponding to each projected image, the output of the previous reconstruction layer, and the current scale implicit neural representation data, the current reconstruction feature data output by the current reconstruction layer is obtained, and this process is repeated until the final reconstruction feature data of the last reconstruction layer is obtained.

[0107] Obtain the maximum scale feature data of the final decoding layer output corresponding to each projection image, and thus obtain the maximum scale feature data of each projection image.

[0108] Based on the maximum scale feature data corresponding to each projected image, the scale implicit neural representation data corresponding to the reconstruction output layer, and the final reconstruction feature data, the predicted three-dimensional reconstruction target is output through the reconstruction output layer.

[0109] Optionally, a training module 66 is also included, for:

[0110] Obtain a training dataset, wherein each training sample in the training dataset includes multiple sample projection images of a 3D sample target acquired from multiple perspectives, initial sample volume data corresponding to the 3D sample target, and 3D reconstructed volume labels corresponding to the 3D sample target;

[0111] Based on the training dataset, obtain the input training sample for the current iteration, iteratively train the initial 3D reconstruction model, obtain the 3D reconstruction sample for the current iteration, and calculate the current total loss value based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample for the current iteration.

[0112] Based on the current total loss value, determine whether the current iteration meets the iteration termination condition. If the current iteration does not meet the iteration termination condition, continue to acquire input training samples for iterative training until the iteration termination condition is met. The 3D reconstruction model after meeting the iteration termination condition is used as the pre-trained 3D reconstruction model.

[0113] Optionally, training module 66 is also used for:

[0114] Based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration, the current reconstruction loss value is calculated using the first loss function, wherein the current reconstruction loss value is used to indicate the structural difference between the predicted 3D reconstruction sample and the 3D reconstruction volume label;

[0115] Based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration, the current structural similarity loss value is calculated using the second loss function, wherein the current structural similarity loss value is used to indicate the similarity difference between the predicted 3D reconstruction sample and the 3D reconstruction volume label.

[0116] Based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration, the current gradient loss is calculated using the third loss function, wherein the current gradient loss is used to indicate the difference in image detail information between the predicted 3D reconstruction sample and the 3D reconstruction volume label;

[0117] Based on the 3D reconstruction volume labels corresponding to the input training samples and the 3D reconstruction samples of the current iteration, the current projection loss is calculated using the fourth loss function, wherein the current projection loss is used to indicate the physical process loss of CT imaging.

[0118] The current total loss value is obtained by weighting the current reconstruction loss value, the current structural similarity loss value, the current gradient loss, and the current projection loss.

[0119] Optionally, training module 66 is also used for:

[0120] Using the mean squared error function, the horizontal gradient loss in the horizontal direction between the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration is calculated, and the vertical gradient loss in the vertical direction between the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration is calculated.

[0121] The current gradient loss is calculated based on the horizontal gradient loss and the vertical gradient loss.

[0122] Optionally, training module 66 is also used for:

[0123] Based on the 3D reconstruction samples of the current iteration, calculate the current projection data of the current iteration;

[0124] Calculate the projection label based on the 3D reconstructed volume label;

[0125] The current projection loss is calculated based on the current projection data and the projection label.

[0126] Please see Figure 7 In another aspect of this application, a computing device 10 is also provided, including a memory 3011 and a processor 3012. The memory 3011 stores a computer program, and when the computer program is executed by the processor, the processor 3012 performs the steps of the three-dimensional CT image reconstruction method based on multi-view projection provided in any of the above embodiments of this application. The computing device 10 can be (e.g., a desktop computer, laptop computer, tablet computer, handheld computer, smart speaker, server, etc.), a mobile phone (e.g., a smartphone, cordless phone, etc.), a wearable device (e.g., a pair of smart glasses or a smartwatch), or a similar device.

[0127] The processor 3012 is the control center, connecting various parts of the computer device via various interfaces and lines. It executes software programs and / or modules stored in the memory 3011, and calls data stored in the memory 3011 to perform various functions and process data. Optionally, the processor 3012 may include one or more processing cores; preferably, the processor 3012 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user page, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 3012.

[0128] The memory 3011 can be used to store software programs and modules. The processor 3012 executes various functional applications and data processing by running the software programs and modules stored in the memory 3011. The memory 3011 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 3011 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 3011 may also include a memory controller to provide the processor 3012 with access to the memory 3011.

[0129] In another aspect, this application also provides a storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the three-dimensional CT image reconstruction method based on multi-view projection provided in any of the above embodiments of this application.

[0130] Those skilled in the art will understand that all or part of the processes in the methods provided in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0131] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for reconstructing three-dimensional CT images based on multi-view projection, characterized in that, The method includes: Acquire multiple projected images of the 3D target to be reconstructed from different perspectives and the initial volume data corresponding to the 3D target to be reconstructed; By using the dimension expansion network in the pre-trained 3D reconstruction model, the channel dimensions of the projected image are expanded to obtain the expanded projected image corresponding to each of the projected images. Each of the extended projection images is used as input to the feature extraction network in the three-dimensional reconstruction model to extract multi-scale feature data corresponding to each projection image; Based on the initial volume data, the implicit neural learning network in the three-dimensional reconstruction model is used to obtain the multi-scale implicit neural representation data corresponding to the three-dimensional target to be reconstructed. Based on the multi-scale feature data and multi-scale implicit neural representation data corresponding to each of the projected images, a 3D image reconstruction is performed using the feature reconstruction network in the 3D reconstruction model to output a 3D reconstruction target. The feature extraction network includes an encoder and a decoder. The decoder is connected to the feature reconstruction network, and the implicit neural learning network is also connected to the feature reconstruction network. The encoder includes multiple encoding layers, and the decoder includes multiple decoding layers. Encoding layers at the same scale are skipped to their corresponding decoding layers. The feature reconstruction network includes multiple reconstruction layers with the same number of layers as the decoder and reconstruction output layers connected to these layers. The input of the decoding layer at the same scale is used as the input of the corresponding reconstruction layer. The feature reconstruction network includes a multi-scale fusion network. The implicit neural learning network is connected to the multi-scale fusion network, and the multi-scale fusion network is connected to each layer of the feature reconstruction network. The multi-scale fusion network is used to output implicit neural representation data at each scale. The 3D reconstruction is performed using the feature reconstruction network based on the multi-scale feature data and multi-scale implicit neural representation data corresponding to each of the projected images. The feature reconstruction network in the model performs 3D image reconstruction and outputs the 3D reconstruction target, including: for the current reconstruction layer at the current layer number, obtaining the current scale feature input data corresponding to each projected image, wherein the current scale feature input data is the input of the current decoding layer connected to the current reconstruction layer in the decoder corresponding to each projected image; obtaining the output of the previous reconstruction layer of the current reconstruction layer; based on the initial volume data, obtaining the current scale implicit neural representation data corresponding to the current layer number output by the multi-scale fusion network; based on the current scale feature input data corresponding to each projected image, the output of the previous reconstruction layer, and the current scale implicit neural representation data, obtaining the current reconstruction feature data output by the current reconstruction layer, and performing this process sequentially until the final reconstruction feature data of the last reconstruction layer is obtained; obtaining the maximum scale feature data output by the last decoding layer corresponding to each projected image, obtaining the maximum scale feature data corresponding to each projected image; based on the maximum scale feature data corresponding to each projected image, the scale implicit neural representation data corresponding to the reconstruction output layer, and the final reconstruction feature data, outputting the predicted 3D reconstruction target through the reconstruction output layer.

2. The three-dimensional CT image reconstruction method based on multi-view projection as described in claim 1, characterized in that, The dimension expansion network includes a gradient calculation layer, a gradient stitching layer, and a dimension expansion layer. The dimension expansion network in the pre-trained 3D reconstruction model is used to expand the channel dimensions of the projected images to obtain expanded projected images corresponding to each projected image, including: Based on the projected image, the gradients in the horizontal and vertical directions are calculated using a gradient calculation layer to obtain the gradient image corresponding to the projected image. Using a gradient stitching layer, the projected image and the gradient image corresponding to the projected image are summed to obtain the composite image corresponding to the projected image. The projected image, the gradient image corresponding to the projected image, and the composite image corresponding to the projected image are then stitched together to obtain the three-channel initial image corresponding to the projected image. Based on multiple cascaded residual blocks in the dimension expansion layer, feature maps are extracted step by step from the initial three-channel image to obtain the expanded projection image corresponding to the projection image.

3. The three-dimensional CT image reconstruction method based on multi-view projection as described in claim 1 or 2, characterized in that, The method further includes: Obtain a training dataset, wherein each training sample in the training dataset includes multiple sample projection images of a 3D sample target acquired from multiple perspectives, initial sample volume data corresponding to the 3D sample target, and 3D reconstructed volume labels corresponding to the 3D sample target; Based on the training dataset, obtain the input training sample for the current iteration, iteratively train the initial 3D reconstruction model, obtain the 3D reconstruction sample for the current iteration, and calculate the current total loss value based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample for the current iteration. Based on the current total loss value, determine whether the current iteration meets the iteration termination condition. If the current iteration does not meet the iteration termination condition, continue to acquire input training samples for iterative training until the iteration termination condition is met. The 3D reconstruction model after meeting the iteration termination condition is used as the pre-trained 3D reconstruction model.

4. The three-dimensional CT image reconstruction method based on multi-view projection as described in claim 3, characterized in that, Based on the 3D reconstruction volume labels corresponding to the input training samples and the 3D reconstruction samples of the current iteration, the current total loss value is calculated as follows: Based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration, the current reconstruction loss value is calculated using the first loss function, wherein the current reconstruction loss value is used to indicate the structural difference between the predicted 3D reconstruction sample and the 3D reconstruction volume label; Based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration, the current structural similarity loss value is calculated using the second loss function, wherein the current structural similarity loss value is used to indicate the similarity difference between the predicted 3D reconstruction sample and the 3D reconstruction volume label. Based on the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration, the current gradient loss is calculated using the third loss function, wherein the current gradient loss is used to indicate the difference in image detail information between the predicted 3D reconstruction sample and the 3D reconstruction volume label; Based on the 3D reconstruction volume labels corresponding to the input training samples and the 3D reconstruction samples of the current iteration, the current projection loss is calculated using the fourth loss function, wherein the current projection loss is used to indicate the physical process loss of CT imaging. The current total loss value is obtained by weighting the current reconstruction loss value, the current structural similarity loss value, the current gradient loss, and the current projection loss.

5. The three-dimensional CT image reconstruction method based on multi-view projection as described in claim 4, characterized in that, The calculation of the current gradient loss based on the 3D reconstruction volume labels corresponding to the input training samples and the 3D reconstruction samples of the current iteration, using the third loss function, includes: Using the mean squared error function, the horizontal gradient loss in the horizontal direction between the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration is calculated, and the vertical gradient loss in the vertical direction between the 3D reconstruction volume label corresponding to the input training sample and the 3D reconstruction sample of the current iteration is calculated. The current gradient loss is calculated based on the horizontal gradient loss and the vertical gradient loss.

6. The three-dimensional CT image reconstruction method based on multi-view projection as described in claim 4, characterized in that, The calculation of the current projection loss based on the 3D reconstruction volume labels corresponding to the input training samples and the 3D reconstruction samples of the current iteration, using the fourth loss function, includes: Based on the 3D reconstruction samples of the current iteration, calculate the current projection data of the current iteration; Calculate the projection label based on the 3D reconstructed volume label; The current projection loss is calculated based on the current projection data and the projection label.

7. A computing device, characterized in that, The system includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the three-dimensional CT image reconstruction method based on multi-view projection as described in any one of claims 1 to 6.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the three-dimensional CT image reconstruction method based on multi-view projection as described in any one of claims 1 to 6.

Images