A sparse projection CBCT reconstruction method, device, equipment and readable storage medium
By constructing a neural field network model and using X-ray projection and mask prediction projection for supervised training, the artifact problem in sparse projection CBCT reconstruction was solved, and efficient 3D image reconstruction was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- WUHAN UNIV
- Filing Date
- 2023-11-30
- Publication Date
- 2026-07-24
AI Technical Summary
Under sparse projection conditions, existing medical image reconstruction methods cannot effectively solve the artifact problem in CBCT reconstruction, and deep learning methods have limited effectiveness in medical image reconstruction.
A neural rendering-based approach was adopted. A neural field network model was constructed, and supervised training was performed using X-ray projection and mask prediction projection to generate a 3D image. The image was then reconstructed by combining grayscale data from multiple tissue units.
It improves the effectiveness and interpretability of neural network models in reconstructing 3D images and enhances the reconstruction quality under sparse projection conditions.
Smart Images
Figure CN117653162B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of CBCT, specifically to a sparse projection CBCT reconstruction method, apparatus, device, and readable storage medium. Background Technology
[0002] Cone-beam computed tomography (CBCT) is a medical imaging technique used to acquire three-dimensional images of the human body or objects. Its imaging component mainly consists of an X-ray source and a planar detector. During image acquisition, these two components rotate simultaneously around the patient. The X-ray source emits numerous cone-shaped X-ray beams, and the detector captures the remaining X-ray signals after they have been absorbed by the patient from multiple angles. These captured X-rays are then projected to reconstruct a volumetric image. CBCT is widely used in clinical practice in dentistry, significantly improving the efficiency of diagnosis and treatment of various diseases. However, frequent CBCT scans can pose potential radiation hazards to patients. Reducing the radiation dose per image or limiting the number of original projections can reduce the overall radiation dose. However, reducing the radiation dose per image or the number of projections inevitably leads to a significant loss of information about the target object, resulting in reconstructions containing numerous artifacts.
[0003] Due to the ill-posed nature of sparse projection reconstruction, classic medical image reconstruction methods such as filtered back-propagation (FBP) and algebraic reconstruction technique (ART) cannot achieve reliable reconstruction results. Traditional solutions typically involve adding regularization constraints or applying denoising techniques during the reconstruction optimization process, but these manually designed methods have certain limitations.
[0004] With the development of deep learning, deep learning methods have gradually been applied to sparse projection CT reconstruction problems. However, encoder-decoder type end-to-end deep learning struggles to capture the physical processes of imaging, resulting in such methods typically requiring extremely large amounts of data for pre-training, and limiting reconstruction performance. Therefore, many deep learning methods are still based on improvements to classical medical image reconstruction methods.
[0005] In recent years, coordinate-based neural rendering methods for 3D vision have developed rapidly in the field of computer science. These methods learn 3D spatial information by simulating the imaging and rendering process and fitting a coordinate-based neural field to supervise a finite 2D viewpoint. Such methods achieve good learning results under sparse input conditions. However, most of these methods are designed for natural scenes. For example, some models attempt to incorporate physical constraints such as depth and visibility into the 3D reconstruction process, which can significantly improve learning performance under sparse viewpoints. However, this technique is currently unable to effectively match the image reconstruction process in the medical field. Therefore, it is essential to construct a method based on neural rendering and medical image constraints to address the technical challenge of poor reconstruction results for medical images under sparse projection conditions. Summary of the Invention
[0006] This application provides a sparse projection CBCT reconstruction method, apparatus, device, and readable storage medium, which can solve the technical problems existing in the background art.
[0007] In a first aspect, embodiments of this application provide a sparse projection CBCT reconstruction method, employing the following technical solution:
[0008] A sparse projection CBCT reconstruction method includes the following steps:
[0009] Obtain X-ray projections of the target patient from multiple angles;
[0010] Based on the preset tissue projection prediction model and the X-ray projection, the overall tissue mask prediction projection and hard tissue mask prediction projection of the target patient are obtained, which are used as the overall tissue mask prediction projection and hard tissue mask prediction projection for reconstruction.
[0011] Based on the X-ray projection, the predicted projection of the overall tissue mask for reconstruction, and the predicted projection of the hard tissue mask for reconstruction, a neural field network initial model is trained under supervision to obtain a neural field network model corresponding to the target patient; wherein, the neural field network model is configured to generate grayscale data of multiple tissue units, including grayscale data of the overall tissue mask and grayscale data of the hard tissue mask, based on the coordinate information of any coordinate point in three-dimensional space, and to obtain the overall tissue grayscale data at the coordinate point based on the grayscale data of the multiple tissue units.
[0012] Based on the neural field network model and the coordinate information of each coordinate point in the three-dimensional image to be reconstructed for the target patient, the overall tissue grayscale data of each coordinate point in the three-dimensional image to be reconstructed is obtained, and the three-dimensional image to be reconstructed is formed.
[0013] In conjunction with the first aspect, in one embodiment, the step of supervising and training an initial neural field network model based on the X-ray projection, the predicted projection using the overall tissue mask for reconstruction, and the predicted projection using the hard tissue mask for reconstruction, to obtain a neural field network model corresponding to the target patient, includes:
[0014] During iterative training, the loss function of the neural field network model after each iteration is obtained based on the X-ray projection, the predicted projection using the overall tissue mask for reconstruction, the predicted projection using the hard tissue mask for reconstruction, the grayscale data of the overall tissue, the grayscale data of the overall tissue mask, and the grayscale data of the hard tissue mask.
[0015] Determine whether the loss function is less than a set value;
[0016] If the value is less than the target patient's value, the initial model of the neural field network is trained to obtain the neural field network model corresponding to the target patient.
[0017] In conjunction with the first aspect, in one implementation, during the iterative training process, based on the X-ray projection, the predicted projection using the overall tissue mask for reconstruction, the predicted projection using the hard tissue mask for reconstruction, the grayscale data of the overall tissue, the grayscale data of the overall tissue mask, and the grayscale data of the hard tissue mask, the loss function of the neural field network model after each iteration is obtained, using the following calculation formula:
[0018]
[0019] Where L1 is the mean absolute error and L2 is the mean squared error. ∑ gt These are the overall tissue grayscale data at the corresponding coordinate points and the grayscale data of the X-ray projection, respectively. A pred These are the overall tissue grayscale data at the corresponding coordinate points and the grayscale data of the predicted projection using the overall tissue mask for reconstruction, respectively. B pred λ(t) represents the grayscale data of the hard tissue mask at the corresponding coordinate point and the grayscale data of the predicted projection of the hard tissue mask for reconstruction, respectively. λ(t) is an influence parameter function that is negatively correlated with the number of iterations.
[0020] In conjunction with the first aspect, the influencing parameter function λ(t) is calculated using the following formula:
[0021]
[0022] Where y and T represent the current iteration and the target total number of iterations, respectively, λ0 is the initial value of the parameter, and k∈(0,1).
[0023] In conjunction with the first aspect, in one implementation, the tissue projection prediction model is obtained according to the following steps:
[0024] Multiple CBCT 3D image data from different objects were acquired as training 3D image data.
[0025] Based on the training 3D image data, the training X-ray projection and the true projection of the whole tissue mask and the true projection of the hard tissue mask corresponding to the training X-ray projection are obtained.
[0026] A tissue projection prediction model is established based on supervised training of the actual projection of the overall tissue mask and the actual projection of the hard tissue mask; wherein, the tissue projection prediction model is configured to obtain the predicted projection of the overall tissue mask and the predicted projection of the hard tissue mask based on the input X-ray projection.
[0027] In conjunction with the first aspect, in one embodiment, the step of obtaining the training X-ray projection and the overall tissue mask real projection and hard tissue mask real projection corresponding to the training X-ray projection based on the training three-dimensional image data,
[0028] The training 3D image data is segmented by observing the numerical distribution of the volumetric image and setting two thresholds to obtain the true projection of the overall tissue mask and the true projection of the hard tissue mask.
[0029] In conjunction with the first aspect, in one embodiment, the neural field network model is configured to generate grayscale data of the overall tissue mask, grayscale data of the hard tissue mask, hard tissue texture value, and soft tissue texture value at any coordinate point in three-dimensional space, and to obtain the overall tissue grayscale data at the coordinate point based on the grayscale data of the overall tissue mask, the grayscale data of the hard tissue mask, the hard tissue texture value, and the soft tissue texture value.
[0030] Secondly, embodiments of this application provide a sparse projection CBCT reconstruction device, which adopts the following technical solution:
[0031] A sparse projection CBCT reconstruction apparatus, the sparse projection CBCT reconstruction apparatus comprising:
[0032] The acquisition module is configured to acquire the X-ray projections of the target patient at multiple angles.
[0033] The training module is configured to obtain the overall tissue mask prediction projection and hard tissue mask prediction projection of the target patient based on a preset tissue projection prediction model and the X-ray projection, which serve as the overall tissue mask prediction projection and hard tissue mask prediction projection for reconstruction; based on the X-ray projection, the overall tissue mask prediction projection for reconstruction, and the hard tissue mask prediction projection for reconstruction, it supervises the training of a set initial model of the neural field network to obtain a neural field network model corresponding to the target patient; wherein, the neural field network model is configured to generate grayscale data of multiple tissue units, including the grayscale data of the overall tissue mask and the hard tissue mask at any coordinate point, based on the coordinate information of any coordinate point in three-dimensional space, and obtain the overall tissue grayscale data at the coordinate point based on the grayscale data of the multiple tissue units.
[0034] The reconstruction module is configured to obtain the overall tissue grayscale data of each coordinate point in the three-dimensional image to be reconstructed based on the neural field network model and the coordinate information of each coordinate point in the three-dimensional image to be reconstructed for the target patient, and to form the three-dimensional image to be reconstructed.
[0035] Thirdly, embodiments of this application provide a sparse projection CBCT reconstruction device, which adopts the following technical solution:
[0036] A sparse projection CBCT reconstruction device is characterized in that the sparse projection CBCT reconstruction device includes a processor, a memory, and a sparse projection CBCT reconstruction program stored in the memory and executable by the processor, wherein when the sparse projection CBCT reconstruction program is executed by the processor, it implements the steps of the sparse projection CBCT reconstruction method as described above.
[0037] Fourthly, embodiments of this application provide a readable storage medium, employing the following technical solution:
[0038] A readable storage medium storing a sparse projection CBCT reconstruction program, wherein when the sparse projection CBCT reconstruction program is executed by a processor, it implements the steps of the sparse projection CBCT reconstruction method as described above.
[0039] The beneficial effects of the technical solutions provided in this application include:
[0040] Because the neural field network model used in this application to construct the required reconstructed 3D image is trained from three aspects during the training process—based on the X-ray projection of the target patient, and the overall tissue mask prediction projection and hard tissue mask prediction projection obtained from the X-ray projection—and during execution, the neural field network model is established based on the grayscale data of multiple tissue units to build the overall tissue grayscale data, that is, by modeling different tissue features in CBCT through multiple specialized unit networks, the interpretability and controllability of the neural field network model are improved, which has strong practical significance and ultimately effectively improves the effectiveness of the neural network model in reconstructing 3D images. Attached Figure Description
[0041] Figure 1 This is a schematic flowchart of an embodiment of the sparse projection CBCT reconstruction method of this application;
[0042] Figure 2 This is a schematic diagram of the functional modules of an embodiment of the sparse projection CBCT reconstruction device of this application;
[0043] Figure 3 This is a schematic diagram of the hardware structure of the sparse projection CBCT reconstruction device involved in the embodiments of this application. Detailed Implementation
[0044] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.
[0045] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.
[0046] In a first aspect, embodiments of this application provide a sparse projection CBCT reconstruction method.
[0047] In one embodiment, reference is made to Figure 1 , Figure 1 This is a schematic flowchart of the first embodiment of the sparse projection CBCT reconstruction method of this application. Figure 1 As shown, sparse projection CBCT reconstruction methods include:
[0048] S100: Obtain X-ray projections of the target patient from multiple angles;
[0049] S200. Based on the preset tissue projection prediction model and the X-ray projection, obtain the overall tissue mask prediction projection and hard tissue mask prediction projection of the target patient, which are used as the overall tissue mask prediction projection and hard tissue mask prediction projection for reconstruction.
[0050] S300. Based on the X-ray projection, the predicted projection of the overall tissue mask for reconstruction, and the predicted projection of the hard tissue mask for reconstruction, a neural field network initial model is trained under supervision to obtain a neural field network model corresponding to the target patient; wherein, the neural field network model is configured to generate grayscale data of multiple tissue units, including grayscale data of the overall tissue mask and grayscale data of the hard tissue mask, based on the coordinate information of any coordinate point in three-dimensional space, and to obtain the overall tissue grayscale data at the coordinate point based on the grayscale data of the multiple tissue units.
[0051] S400. Based on the neural field network model and the coordinate information of each coordinate point in the three-dimensional image to be reconstructed for the target patient, obtain the overall tissue grayscale data of each coordinate point in the three-dimensional image to be reconstructed, and form the three-dimensional image to be reconstructed.
[0052] In this embodiment, the neural field network model used to construct the three-dimensional image to be reconstructed in this application is trained from three aspects during the training process: the X-ray projection of the target patient, and the overall tissue mask prediction projection and hard tissue mask prediction projection obtained from the X-ray projection. In addition, during the execution process, the neural field network model is established based on the grayscale data of multiple tissue units to build the overall tissue grayscale data. That is, different tissue features in CBCT are modeled through multiple specialized unit networks, which improves the interpretability and controllability of the neural field network model and has strong practical significance. Ultimately, it effectively improves the effectiveness of the neural network model in reconstructing three-dimensional images.
[0053] Further, in one embodiment, step S300, which involves supervising and training an initial neural field network model based on the X-ray projection, the predicted projection using the overall tissue mask for reconstruction, and the predicted projection using the hard tissue mask for reconstruction, to obtain a neural field network model corresponding to the target patient, includes:
[0054] S310. During iterative training, the loss function of the neural field network model after each iteration is obtained based on the X-ray projection, the predicted projection using the overall tissue mask for reconstruction, the predicted projection using the hard tissue mask for reconstruction, the grayscale data of the overall tissue, the grayscale data of the overall tissue mask, and the grayscale data of the hard tissue mask.
[0055] S320. Determine whether the loss function is less than a set value;
[0056] S330. If it is less than, complete the training of the initial model of the neural field network to obtain the neural field network model corresponding to the target patient.
[0057] Specifically, step S310 uses the following calculation formula:
[0058]
[0059] Where L1 is the mean absolute error and L2 is the mean squared error. ∑ gt These are the overall tissue grayscale data at the corresponding coordinate points and the grayscale data of the X-ray projection, respectively. A pred These are the overall tissue grayscale data at the corresponding coordinate points and the grayscale data of the predicted projection using the overall tissue mask for reconstruction, respectively. B pred λ(t) represents the grayscale data of the hard tissue mask at the corresponding coordinate point and the grayscale data of the predicted projection of the hard tissue mask for reconstruction, respectively. λ(t) is an influence parameter function that is negatively correlated with the number of iterations.
[0060] The influencing parameter function λ(t) is calculated using the following formula:
[0061]
[0062] Where t and T represent the current iteration and the target total number of iterations, respectively, λ0 is the initial value of the parameter, and k∈(0,1).
[0063] Furthermore, in one embodiment, the tissue projection prediction model is obtained according to the following steps:
[0064] F100: Acquire multiple CBCT 3D image data from different objects as training 3D image data;
[0065] F200. Based on the training three-dimensional image data, obtain the training X-ray projection and the true projection of the whole tissue mask and the true projection of the hard tissue mask corresponding to the training X-ray projection.
[0066] F300, a tissue projection prediction model set by supervised training based on the actual projection of the overall tissue mask and the actual projection of the hard tissue mask; wherein, the tissue projection prediction model is configured to obtain the predicted projection of the overall tissue mask and the predicted projection of the hard tissue mask based on the input X-ray projection.
[0067] Furthermore, in step F200, in obtaining the training X-ray projection and the corresponding whole tissue mask real projection and hard tissue mask real projection based on the training three-dimensional image data,
[0068] The training 3D image data is segmented by observing the numerical distribution of the volumetric image and setting two thresholds to obtain the true projection of the overall tissue mask and the true projection of the hard tissue mask.
[0069] Specifically, several CBCT 3D image data sets are acquired clinically, and the corresponding X-ray projections are first established. Then, by observing the numerical distribution of volumetric images in the CBCT 3D image data, two thresholds are selected to segment the 3D images to obtain the true projections of the overall tissue mask and the hard tissue mask corresponding to the CBCT 3D image data.
[0070] Furthermore, in one embodiment, the neural field network model is configured to generate grayscale data of the overall tissue mask, grayscale data of the hard tissue mask, hard tissue texture value, and soft tissue texture value at any coordinate point in three-dimensional space, and to obtain the overall tissue grayscale data at the coordinate point based on the grayscale data of the overall tissue mask, the grayscale data of the hard tissue mask, the hard tissue texture value, and the soft tissue texture value.
[0071] Specifically, the CBCT image's radiation-absorbing field is first decoupled into four different intensity fields and a global scalar, as shown in the following formula.
[0072] σ(x)=(α(x)+ε)(β(x)v b (x)+v s (x)),
[0073] Where α represents the grayscale data of the overall tissue mask, the grayscale data of the hard tissue mask, and v b and v s α represents the texture values for hard and soft tissues, respectively. σ is a small, learnable, position-independent global scalar used to ensure that low-intensity values (such as air) outside of α are well defined.
[0074] This decoupling representation has the following advantages and features: the formula can divide the intensity field into four context-specific components: α, β, v b and v s These correspond to the separation of soft and hard tissues, as well as shape and material, respectively, increasing the interpretability of the network and making it easier to control the output content, thereby better controlling the reconstruction process. In local v b ,v s Under relatively flat conditions, bone tissue generally exhibits a higher strength value (σ) than soft tissue in certain areas. α＝1，β＝1 >σ α＝1，β＝0 This is because the bone tissue can be considered as superimposed on the soft tissue material, σ α＝1，β＝0When this happens, the above formula degenerates into one that only includes the soft tissue portion. This representation conforms to the numerical distribution of grayscale data in both soft and hard tissues, making it easier to obtain the ideal grayscale distribution characteristics.
[0075] Meanwhile, assuming the tissue segmentation is binary and ideal segmentation results can be obtained, the material representation values will be relatively flat. Compared to directly predicting σ, predicting the binary classification output (α,β) and the low-frequency signal (v) is more effective. b ,v s It is easier to learn online.
[0076] In particular, in clinical practice, X-ray cephalography is one of the routine tools in the process of orthodontic diagnosis using lateral radiographs. It can produce clear and detailed hard tissue images and is often used to assess bony structures. Tissue decoupling makes full use of this clinical experience and advantage. By decoupling soft and hard tissues, additional supervision can be provided for the training of neural networks. Especially under sparse input conditions, it can significantly improve the learning efficiency of the model and the quality of image reconstruction.
[0077] To fully utilize the advantages of decoupled representation and facilitate supervision of its training process, the neural field network model proposed in this invention is mainly based on the following design principles: First, the network needs to output four elements, namely α, β, and v. b and v s These correspond to the separation of soft and hard tissues, as well as shape and material, respectively. However, simply adjusting the last layer of an MLP to a 4-channel output and adding tissue-based supervision is insufficient to produce optimal results. This is because different shapes and tissue materials are inherently different, and fully connected MLPs struggle to learn such diverse representations; all features are always shared between layers. In particular, when using strongly parameterized positional encoding, high-frequency details are learned more within the positional encoding, making it difficult for the network to learn semantically diverse representations. Furthermore, the network only supervises the shape branches α and β, neglecting the texture branch v. b and v s Without any constraints, the network can produce predictions on these branches that may contradict low-frequency results. Therefore, this method employs four separate, embedded, small MLPs to construct the network, each with its own set of weights and errors, to learn decoupled features independently. Furthermore, introducing smaller networks reduces network complexity, the number of network parameters, and accelerates training. In particular, grid-based representations can reduce network size, accelerate training, and produce better reconstruction results in neural implicit scene representation. Therefore, this method uses a hash grid as the positional encoding, and compared to other related methods, it converges faster and performs better under sparse view conditions.
[0078] The constructed neural field network model is specifically deconstructed as follows: a four-element network with a depth of 4 and a width of 32, using Sigmoid as the final activation function. The output of this four-element network has 4 channels, which are the decoupled α, β, and v. b and v s The structure of the model constructed in this invention is as follows: Based on a 4-channel feature hash grid position encoder, it is connected to a ModuleList containing four Sequential modules. Each module is a sub-network used to process different feature types. The four-dimensional features at different scales output by the hash grid are input into the four modules respectively. The first two Sequential modules each contain two parts: an MLP and a Sigmoid layer. The MLP is a multilayer perceptron used to map input features to an output value. The Sigmoid layer is an activation function used to compress the output value to between 0 and 1, representing probability or confidence. The latter two Sequential modules each contain only two single layers: a Linear layer and a Sigmoid layer. The Linear layer is a linear transformation used to map input features to a one-dimensional output value. The Sigmoid layer serves the same purpose. Finally, there is a combination module, which contains only global weights ε, used to combine the output values of the four sub-networks into a final output value to output the prediction result of the entire network.
[0079] Secondly, embodiments of this application also provide a sparse projection CBCT reconstruction device.
[0080] In one embodiment, reference is made to Figure 2 , Figure 2 This is a functional module diagram of an embodiment of the sparse projection CBCT reconstruction device of this application. Figure 2 As shown, the sparse projection CBCT reconstruction device includes:
[0081] The acquisition module is configured to acquire the X-ray projections of the target patient at multiple angles.
[0082] The training module is configured to obtain the overall tissue mask prediction projection and hard tissue mask prediction projection of the target patient based on a preset tissue projection prediction model and the X-ray projection, which serve as the overall tissue mask prediction projection and hard tissue mask prediction projection for reconstruction; based on the X-ray projection, the overall tissue mask prediction projection for reconstruction, and the hard tissue mask prediction projection for reconstruction, it supervises the training of a set initial model of the neural field network to obtain a neural field network model corresponding to the target patient; wherein, the neural field network model is configured to generate grayscale data of multiple tissue units, including the grayscale data of the overall tissue mask and the hard tissue mask at any coordinate point, based on the coordinate information of any coordinate point in three-dimensional space, and obtain the overall tissue grayscale data at the coordinate point based on the grayscale data of the multiple tissue units.
[0083] The reconstruction module is configured to obtain the overall tissue grayscale data of each coordinate point in the three-dimensional image to be reconstructed based on the neural field network model and the coordinate information of each coordinate point in the three-dimensional image to be reconstructed for the target patient, and to form the three-dimensional image to be reconstructed.
[0084] The functions of each module in the sparse projection CBCT reconstruction device correspond to the steps in the sparse projection CBCT reconstruction method embodiment, and their functions and implementation processes will not be described in detail here.
[0085] Thirdly, embodiments of this application provide a sparse projection CBCT reconstruction device, which can be a personal computer (PC), laptop computer, server, or other device with data processing capabilities.
[0086] Reference Figure 3 , Figure 3 This is a schematic diagram of the hardware structure of the sparse projection CBCT reconstruction device involved in the embodiments of this application. In the embodiments of this application, the sparse projection CBCT reconstruction device may include a processor, a memory, a communication interface, and a communication bus.
[0087] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.
[0088] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting devices within the sparse projection CBCT reconstruction equipment, as well as interfaces used for interconnecting the sparse projection CBCT reconstruction equipment with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.
[0089] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.
[0090] The processor can be a general-purpose processor, which can call the sparse projection CBCT reconstruction program stored in memory and execute the sparse projection CBCT reconstruction method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the sparse projection CBCT reconstruction program is called can be referred to in the various embodiments of the sparse projection CBCT reconstruction method of this application, and will not be repeated here.
[0091] Those skilled in the art will understand that Figure 3 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.
[0092] Fourthly, embodiments of this application also provide a readable storage medium.
[0093] This application has a sparse projection CBCT reconstruction program stored on a readable storage medium, wherein when the sparse projection CBCT reconstruction program is executed by a processor, it implements the steps of the sparse projection CBCT reconstruction method as described above.
[0094] The method implemented when the sparse projection CBCT reconstruction procedure is executed can be referred to in various embodiments of the sparse projection CBCT reconstruction method of this application, and will not be repeated here.
[0095] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0096] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.
[0097] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.
[0098] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.
[0099] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.
[0100] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.
[0101] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.
Claims
1. A sparse projection CBCT reconstruction method, characterized in that, It includes the following steps: Obtain X-ray projections of the target patient from multiple angles; Based on the preset tissue projection prediction model and the X-ray projection, the overall tissue mask prediction projection and hard tissue mask prediction projection of the target patient are obtained, which are used as the overall tissue mask prediction projection and hard tissue mask prediction projection for reconstruction. Based on the X-ray projection, the predicted projection of the overall tissue mask for reconstruction, and the predicted projection of the hard tissue mask for reconstruction, a neural field network initial model is trained under supervision to obtain a neural field network model corresponding to the target patient. The neural field network model is configured to generate grayscale data of the overall tissue mask, grayscale data of the hard tissue mask, hard tissue texture values, and soft tissue texture values at any coordinate point in three-dimensional space, and to obtain the overall tissue grayscale data at the coordinate point based on the grayscale data of the overall tissue mask, the grayscale data of the hard tissue mask, the hard tissue texture values, and the soft tissue texture values. Based on the neural field network model and the coordinate information of each coordinate point in the three-dimensional image to be reconstructed for the target patient, the overall tissue grayscale data of each coordinate point in the three-dimensional image to be reconstructed is obtained, and the three-dimensional image to be reconstructed is formed. The step of supervising and training an initial neural field network model based on the X-ray projection, the predicted projection using the overall tissue mask for reconstruction, and the predicted projection using the hard tissue mask for reconstruction, to obtain a neural field network model corresponding to the target patient, includes: During iterative training, the loss function of the neural field network model after each iteration is obtained based on the X-ray projection, the predicted projection using the overall tissue mask for reconstruction, the predicted projection using the hard tissue mask for reconstruction, the grayscale data of the overall tissue, the grayscale data of the overall tissue mask, and the grayscale data of the hard tissue mask. Determine whether the loss function is less than a set value; If it is less than, complete the training of the initial model of the neural field network to obtain the neural field network model corresponding to the target patient; The loss function is calculated using the following formula: in, The mean absolute error, The mean squared error. , These are the overall tissue grayscale data at the corresponding coordinate points and the grayscale data of the X-ray projection, respectively. , These are the overall tissue grayscale data at the corresponding coordinate points and the grayscale data of the predicted projection using the overall tissue mask for reconstruction, respectively. , These are the grayscale data of the hard tissue mask at the corresponding coordinate points and the grayscale data of the predicted projection using the hard tissue mask for reconstruction, respectively. This is an influence parameter function that is negatively correlated with the number of iterations.
2. The sparse projection CBCT reconstruction method as described in claim 1, characterized in that, The influence parameter function The following calculation formula is used: in , These represent the current iteration and the target total number of iterations, respectively. This is the initial value for the parameter. .
3. The sparse projection CBCT reconstruction method as described in claim 1, characterized in that, The tissue projection prediction model is obtained according to the following steps: Multiple CBCT 3D image data from different objects were acquired as training 3D image data. Based on the training 3D image data, the training X-ray projection and the true projection of the whole tissue mask and the true projection of the hard tissue mask corresponding to the training X-ray projection are obtained. A tissue projection prediction model is established based on supervised training of the actual projection of the overall tissue mask and the actual projection of the hard tissue mask; wherein, the tissue projection prediction model is configured to obtain the predicted projection of the overall tissue mask and the predicted projection of the hard tissue mask based on the input X-ray projection.
4. The sparse projection CBCT reconstruction method as described in claim 3, characterized in that, The step involves obtaining the training X-ray projection and the corresponding overall tissue mask real projection and hard tissue mask real projection based on the training X-ray image data. The training 3D image data is segmented by observing the numerical distribution of the volumetric image and setting two thresholds to obtain the true projection of the overall tissue mask and the true projection of the hard tissue mask.
5. A sparse projection CBCT reconstruction device, employing the sparse projection CBCT reconstruction method as described in claim 1, characterized in that, The sparse projection CBCT reconstruction device includes: The acquisition module is configured to acquire the X-ray projections of the target patient at multiple angles. The training module is configured to obtain the overall tissue mask prediction projection and hard tissue mask prediction projection of the target patient based on a preset tissue projection prediction model and the X-ray projection, which serve as the overall tissue mask prediction projection and hard tissue mask prediction projection for reconstruction; based on the X-ray projection, the overall tissue mask prediction projection for reconstruction, and the hard tissue mask prediction projection for reconstruction, it supervises the training of a set initial model of the neural field network to obtain a neural field network model corresponding to the target patient; wherein, the neural field network model is configured to generate grayscale data of multiple tissue units, including the grayscale data of the overall tissue mask and the hard tissue mask at any coordinate point, based on the coordinate information of any coordinate point in three-dimensional space, and obtain the overall tissue grayscale data at the coordinate point based on the grayscale data of the multiple tissue units. The reconstruction module is configured to obtain the overall tissue grayscale data of each coordinate point in the three-dimensional image to be reconstructed based on the neural field network model and the coordinate information of each coordinate point in the three-dimensional image to be reconstructed for the target patient, and to form the three-dimensional image to be reconstructed.
6. A sparse projection CBCT reconstruction device, characterized in that, The sparse projection CBCT reconstruction device includes a processor, a memory, and a sparse projection CBCT reconstruction program stored in the memory and executable by the processor, wherein when the sparse projection CBCT reconstruction program is executed by the processor, it implements the steps of the sparse projection CBCT reconstruction method as described in any one of claims 1 to 4.
7. A readable storage medium, characterized in that, The readable storage medium stores a sparse projection CBCT reconstruction program, wherein when the sparse projection CBCT reconstruction program is executed by a processor, it implements the steps of the sparse projection CBCT reconstruction method as described in any one of claims 1 to 4.