Bone imaging method and system based on double-view depth enhancement CT (Computed Tomography)

Abstract

The invention discloses a double-view-based deep enhancement CT bone imaging method and system, high-quality image recovery and bone microstructure analysis are realized by using deep learning and multi-strategy fusion technologies, and the system comprises the following modules: a Cascade-SwinUNETR trunk module used for feature extraction and enhancement and realizing efficient multi-scale feature aggregation; and the double-view feature fusion module is used for extracting and integrating complementary information by combining CT image data from different visual angles, so that the detail representation and the structural accuracy of the image are improved. And the unsupervised domain adaptation module (UDA) is used for adapting to different data distributions, does not need additional annotation, realizes cross-domain learning and model generalization, and ensures the stable performance of the system in various medical scenes. According to the method, by improving the definition and accuracy of CT imaging, the decision-making quality and efficiency of doctors in bone health assessment can be remarkably improved, and meanwhile dependence on invasive sampling is reduced.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging, and in particular to a dual-view depth-enhanced CT bone imaging method and system. Background Art

[0002] As the global population ages, bone-related diseases such as osteoporosis are becoming a major health concern. Conventional CT imaging technology is widely used in clinical practice to obtain three-dimensional structural images of bones. However, the spatial resolution of conventional CT images is limited, making it difficult to fully display the details of microstructures such as trabecular bone, which are crucial for assessing the mechanical properties and pathological risk of bones. Although microCT can provide higher resolution, its scope of application is limited and is typically only used for in vitro specimens or small animal experiments.

[0003] In recent years, with the rapid development of deep learning technology, image restoration and super-resolution techniques have been gradually applied to medical imaging to enhance the quality and detail of CT images. However, existing methods mostly focus on single-view image processing and fail to fully utilize the complementary information from different perspectives. Furthermore, the difficulty of obtaining large-scale annotated data has limited the widespread application of supervised learning models.

[0004] Therefore, a deep learning method combining dual-view data and unsupervised domain adaptation is proposed, which can make full use of CT data from different perspectives for feature fusion and enhancement, and realize adaptive adjustment of the model under different data distributions, which is of great significance for improving the resolution of medical imaging and the accuracy of diagnosis. Summary of the Invention

[0005] The present invention proposes a dual-view depth-enhanced CT bone imaging method and system. Addressing the limitations of existing CT imaging technology in detail display and resolution, especially in the analysis of complex bone structures such as trabecular bone, the present invention combines deep learning technology with unsupervised domain adaptation strategies to significantly improve the quality and detail display of CT images.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] According to a first aspect of an embodiment of the present invention, a method for deep-enhanced CT bone imaging based on dual views is provided. By combining a deep learning model to perform feature extraction and deep feature fusion of dual-view data, unsupervised domain adaptation is implemented to adapt to different input data distributions, thereby achieving high-resolution image restoration and detail enhancement. The method specifically includes the following steps:

[0008] (1) Feature extraction using the Cascade-SwinUNETR backbone network: The Cascade-SwinUNETR backbone network is used for efficient feature extraction. Combining the win-Transformer module with the feature aggregation technology of U-Net++, this backbone network can capture complex image details and feature distribution through layer-by-layer feature fusion and deep supervision mechanism, thereby achieving higher resolution and enhancement of subtle structures. The network architecture uses multi-scale feature representation to ensure stability and robustness in various imaging noise environments.

[0009] (2) Using dual-view data for deep feature fusion and aggregation: Through dual-view data input, including CT scan data from different perspectives, to improve the overall clarity and accuracy of bone imaging, the dual-view feature fusion module uses a deep feature fusion strategy to jointly process the data from the two perspectives and extract the complementary information in each perspective. This process uses a deep fusion method based on the attention mechanism to ensure the integrity and detail clarity of the image through global and local feature weighting. The advantage of dual-view fusion is that it compensates for the information loss or artifacts that may exist in a single view, thereby generating a more comprehensive bone structure imaging;

[0010] (3) Implement unsupervised domain adaptation (UDA) to adapt to different input distributions. The domain adaptation module adopts an unsupervised learning strategy to adapt to different input data distributions, enabling the model to be trained on unlabeled data, expanding its adaptability and application scope. By aligning the feature distributions of the target domain and the source domain, the unsupervised domain adaptation module can effectively reduce the distribution differences and improve the generalization ability of the model, thereby maintaining efficient imaging quality in different environments. This module adopts adversarial training and self-supervised loss function to ensure the stable performance of the model on diverse data sets.

[0011] As a preferred embodiment, in the step (1), the deep learning architecture DVSR3D utilizes an attention mechanism and autoencoder-style reconstruction to achieve better noise reduction and feature clarity. The architecture adopts an autoencoder-style design and performs image reconstruction by introducing an attention mechanism and deep feature aggregation. The architecture includes an encoder, a feature fusion module, and a decoder module. The encoder part is responsible for extracting multi-scale features from dual-view data and weighting them through the attention mechanism to obtain the most relevant feature representation. The fusion module combines local and global attention to ensure the effective integration of input features. The decoder module performs reconstruction and restores the extracted deep features to a high-resolution image. It also reduces image artifacts and noise through a specific denoising strategy to improve the clarity and detail of the image.

[0012] As a preferred embodiment, the dual-view data fusion in step (2) enhances the details and accuracy of the trabecular bone image by combining orthogonal CT views. By inputting CT scan data from orthogonal perspectives (such as horizontal and vertical directions) to capture the microstructural details of the trabecular bone, the combination of orthogonal perspectives ensures that the data contains more dimensional information, reduces artifacts and information redundancy problems, and through the feature fusion module, the system can complement and weigh information from different perspectives, thereby enhancing the structural consistency and detail display of the image. This fusion technology is particularly suitable for medical image analysis that requires high resolution and multi-level details, such as the diagnosis of osteoporosis and bone health assessment.

[0013] As a preferred embodiment, the unsupervised domain adaptation method in step (3) enables the model to be trained with limited labeled data, thereby adapting to new medical data sets and enabling the model to be efficiently trained on limited labeled data or unlabeled data. The specific steps of this method include:

[0014] (31) Feature alignment: By calculating the statistical difference (e.g., distribution distance) between the source domain and target domain feature distributions and reducing this difference through the domain adaptation loss function, the robustness of the model under the new data distribution is ensured;

[0015] (32) Adversarial training: Using generative adversarial networks (GANs) or their variants, the model can adaptively adjust to the image characteristics of the new environment, reducing the problem of the model overfitting to a single data source;

[0016] (33) Self-supervised learning strategy: The model is trained in the absence of labels by using self-supervised mechanisms such as pseudo-label generation or image transformation adversarial training to enhance the generalization ability of new data. In practical applications, it is particularly suitable for the field of medical image analysis, such as the consistency of imaging data across hospitals or devices, which improves the applicability and reliability of the model in different medical environments.

[0017] According to the second aspect of an embodiment of the present invention, a dual-view deep-enhanced CT bone imaging system is also provided for CT image restoration tasks, and includes a deep learning model framework for CT image restoration to realize the functions of data acquisition, image registration, and data preprocessing, including Cascade-SwinUNETR, a single-view restoration architecture, a deep feature fusion module, and a dual-view restoration architecture. Each module performs a specific function, and these modules can be combined and reused to improve training efficiency and expand generalization capabilities. Cascade-SwinUNETR, as a basic module, provides a robust architectural foundation to support the construction of subsequent methods. The dual-view restoration architecture integrates a deep feature fusion module to process dual-view input data.

[0018] As a preferred embodiment, the Cascade-SwinUNETR acts as a basic model in the framework. As a deep feature recovery module, it can be integrated with different input and output modules. The network adapts to the 3D Swin-Transformer module, U-Net++ structure and DropKey technology. The single-view recovery architecture is an independent model for the single-view task. Before dual-view learning, pre-training is performed to accelerate training and enhance generalization ability. After the pre-training is completed, the trained recovery model is extracted and integrated into the overall architecture. The deep feature fusion module, CT scans come from two orthogonal perspectives. In order to allow the model to learn, recognize and combine features from two sources, deep feature fusion technology is adopted. This deep feature fusion technology combines deeply refined features instead of directly merging low-level images to improve performance. The dual-view recovery architecture is a dual-view architecture DVSR3D, including an encoder, a restorer and a decoder. The encoder partially upsamples and fuses deep features, the restorer generates an output image, and the decoder prevents overfitting through regularization.

[0019] As a preferred embodiment, the data acquisition: the organized dataset is designed to support dual-view tasks, including clinical CT scans and Micro-CT scans obtained from four different human bone specimens. The experimental setup includes acquiring CT scans from horizontal and vertical directions, while Micro-CT scans are acquired from only a single direction. The horizontal and vertical directions refer to manually rotating the human bone blocks after the scanning procedure is completed. In practical applications, this process can be achieved by having the patient lie in different directions, such as supine or prone, or using different CT scanning equipment. It should be noted that all original CT scans are initially unregistered because they are collected under different real-world environmental conditions and therefore contain difficult-to-handle noise distributions and different spatial information.

[0020] As a preferred embodiment, the image registration: Since the data were originally acquired under different settings, accurate image correspondence is required when creating the dataset. To achieve this goal, the open source software ITK-SNAP and SimpleITK Python library are used for image registration. ITK-SNAP is used for manual and automatic registration, with 16 coarse registration levels and 8 fine registration levels. Subsequently, the registration algorithm in Simple ITK is used to optimize the resulting transformation parameters to generate registered data for dual-view CT scanning.

[0021] As a preferred embodiment, the data preprocessing: Since the original registration data is large in size, a preprocessing step is required to make it suitable for model training, verification and testing. The preprocessing includes random cropping from the registration data. Each cropping unit contains three samples, corresponding to the visual Figure 1 CT scan, Figure 2CT scans and Micro-CT scans. The super-resolution scale used in this task is four times, so the pixel dimension is aligned with this scale. The random cropping process is restricted to specific boundaries. The cropping range is limited to 100 pixels on each side of the width, height, and depth of the registered data. The first 60 samples correspond to the first three bone samples and constitute the training dataset. The next 20 samples are associated with the last bone sample and used as the validation dataset. In addition, 20 more samples are extracted from the same bone sample to create an independent test dataset. Ideally, if more bone samples are available, it is better to select different bone samples as the test dataset. However, due to the challenges in obtaining the original data, this method has been the relatively optimal solution in this study. In addition, the dynamic range of all data is adjusted to the range of 0 to 255 and stored in grayscale image format.

[0022] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is further provided, on which computer-executable instructions are stored. When the executable instructions are executed by a processor, the dual-view depth-enhanced CT bone imaging method described above is implemented.

[0023] The technical principle of this invention: A dual-view deep-enhanced CT bone imaging method and system, combined with deep learning technology, is used to achieve high-precision image restoration and detail enhancement of complex bone structures such as trabecular bone, aiming to improve imaging quality and reliability in medical diagnosis and bone health assessment. The system comprises a Cascade-SwinUNETR backbone network for feature extraction, a dual-view feature fusion module for integrating data from different viewpoints, a DVSR3D architecture for image reconstruction and noise reduction, and an unsupervised domain adaptation module for adaptive adjustment of cross-domain features. The entire system works together to achieve high-resolution restoration of CT images.

[0024] Beneficial effects: The present invention significantly improves the detail display and structural fidelity of CT images through the deep fusion of dual-view data and the unsupervised domain adaptation strategy, and effectively solves the problems of insufficient resolution and artifacts in traditional CT images. In addition, the technical solution of the present application does not rely on large-scale annotated data, and can achieve consistent and robust high-quality image output in different devices and environments, and is suitable for osteoporosis diagnosis and other bone health assessments. The technical solution can also reduce dependence on high-cost and invasive sampling, shorten diagnosis time, and improve diagnostic efficiency, allowing hospitals and clinics to quickly obtain high-precision bone images in daily work, helping doctors make diagnostic decisions faster and optimize patient treatment plans, thereby improving the quality and efficiency of overall medical services. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The specific embodiments of the present invention are further described in detail below with reference to the accompanying drawings, wherein:

[0026] Figure 1 FIG. 4 is a flow chart of the single view recovery architecture of the present invention.

[0027] Figure 2 Schematic diagram of the dual-view super-resolution 3D (DVSR3D) model architecture of the present invention.

[0028] Figure 3 Schematic diagram of the overall process of the unsupervised domain adaptation (UDA) method of the present invention.

[0029] Figure 4 The figure is a schematic diagram of the overall process of the dual-view depth-enhanced CT bone imaging system of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, working principle and technical solution of the present invention clearer, the present invention is further described below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0031] The dual-view depth-enhanced CT bone imaging proposed by the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] like Figure 1 As shown in Figure 1, the single-view restoration architecture is one of the core components of this embodiment. Its function is to restore high-resolution bone structure images from single-view CT data through a deep learning model. The architecture design consists of three main components: the input layer, the feature extraction backbone network, and the output layer.

[0033] The input layer receives preprocessed single-view CT scan data. The CT data undergoes normalization, denoising, and cropping to ensure data consistency and input quality. Normalization ensures that the grayscale values ​​of the data are within a certain range, preventing noise from affecting subsequent model training. Cropping removes irrelevant areas from the original image, allowing the network to focus on the region of interest and improving computational efficiency.

[0034] In this embodiment, the backbone network for feature extraction adopts the Cascade-SwinUNETR structure, which combines the advantages of Swin-Transformer and U-Net++. Swin-Transformer is good at processing multi-scale features and can capture the details and global information in the image; while U-Net++ enhances the network's ability to fuse information of different scales and levels through its hierarchical aggregation capability, thereby better restoring the subtle features of the bone structure. The backbone network gradually refines the important information in the image while extracting features through multi-level convolution operations. In particular, for the subtle parts of the bone structure (such as the microscopic features of trabecular bone), more refined image restoration can be achieved through multi-scale convolution operations. Through inter-layer skip connections and deep supervision mechanisms, the model can more effectively restore the details and structure of the image during training.

[0035] The output layer is responsible for generating the restored high-resolution image. During training, the model optimizes multiple loss functions, including the mean squared error (MSE) loss and the structural similarity index (SSIM) loss, to ensure that the output image is as close as possible to the true label in terms of quality and detail. The MSE loss helps minimize pixel-level errors, while the SSIM loss focuses on the fidelity of the image structure, better preserving texture and details.

[0036] like Figure 2 A deep feature fusion module for dual-view data fusion is demonstrated. This module further enhances the accuracy of image restoration by effectively integrating CT scan data from different perspectives (e.g., horizontal and vertical directions).

[0037] The deep feature fusion module first stitches feature maps from different viewpoints. This process fuses the data from different perspectives and provides richer information for subsequent processing. The stitched feature maps contain both local details and global structural information from different perspectives, which is particularly important for complex bone structure images (especially trabecular bone with its microstructure).

[0038] The concatenated feature maps are then processed through convolution and nonlinear activation functions. This process not only helps extract local features from the image but also improves the representation of global features. Convolution allows the network to extract more refined features, while nonlinear activation functions enhance the model's nonlinear modeling capabilities, making it more adaptable to complex medical image restoration tasks.

[0039] To further optimize the distribution of feature importance, this embodiment introduces the attention mechanism from deep learning. In the fusion module, local and global attention mechanisms are used to weight features from different perspectives. This mechanism automatically adjusts the weights of different features in the final output based on their contribution, thereby improving image restoration accuracy and detail presentation.

[0040] Through this method, this embodiment can preserve the details of the input data while leveraging the complementary information from different perspectives to improve the accuracy and detail of image restoration. In the field of medical image processing, particularly, minute features in bone structure imaging are often key to diagnosing diseases such as osteoporosis, and dual-view fusion technology can significantly improve the resolution of these features.

[0041] like Figure 3 As shown, the unsupervised domain adaptation method of this embodiment aims to improve the model's generalization ability under different data distributions. The UDA module inputs source and target domain data into a pre-trained single-view restoration architecture to generate restored images and uses a domain adaptation strategy to compare feature distributions between different domains. Figure 3 The UDA pipeline in

[15] involves parallel input of source and target domain data. The restored images are then passed to two discriminators to assess whether the images belong to the same domain. The first discriminator is dedicated to domain discrimination of the restored images, while the second discriminator performs domain discrimination on the reconstructed input images. These two discriminators employ adversarial training to reduce the distribution difference between the source and target domains.

[0042] During the training process, the output image intensity of the target domain data may vary significantly due to different acquisition environments. To this end, a simple correction method is adopted, which is to perform mean correction on the target domain restored image to ensure that the output image is consistent with the input data in intensity. The output of the discriminator is used to guide the update of the model to achieve alignment of the feature distribution. Through this unsupervised domain adaptation, the model of this embodiment can learn the feature distribution of data in different environments without the need for additional labels, thereby improving the adaptability and stability of the model in cross-device and cross-environment applications.

[0043] like Figure 4As shown in the figure, in the specific implementation, the CT images are first normalized and cropped through a preprocessing step to ensure the consistency of the input data. The single-view CT images are then input into the Cascade-SwinUNETR backbone network for pre-training to extract multi-scale features. During training, the model is optimized using MSE and SSIM as loss functions to ensure that the generated images are highly consistent with the ground-truth labels in terms of structure and quality. Next, the dual-view data is integrated using a deep feature fusion module to extract complementary information from different viewpoints. The fused feature maps are then input into a deep restoration architecture to generate high-resolution output.

[0044] To further enhance the model's generalization capabilities, an unsupervised domain adaptation module is introduced. Both source and target domain data are fed into the model simultaneously. After restoration, the resulting image is passed to the discriminator for domain discrimination. Through adversarial training, the model gradually learns and aligns features from different data distributions, maintaining consistent image output quality across various environments.

[0045] The technical solution of this invention has important application value in the field of medical imaging, particularly in the diagnosis of osteoporosis and bone structure analysis. By enhancing the resolution and detail of CT images, doctors can obtain high-quality bone images without invasive sampling, thereby improving diagnostic efficiency and accuracy.

[0046] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for bone imaging based on dual-view depth enhancement CT, characterized in that: By combining deep learning models for feature extraction and deep feature fusion of dual-view data, unsupervised domain adaptation is implemented to adapt to different input data distributions, achieving high-resolution restoration and detail enhancement of images. The specific steps include: (1) Feature extraction using the Cascade-SwinUNETR backbone network: The Cascade-SwinUNETR backbone network is used for efficient feature extraction. The feature aggregation technology of the win-Transformer module and U-Net++ is combined. Through layer-by-layer feature fusion and deep supervision mechanism, the backbone network can capture complex image details and feature distribution, thereby achieving higher resolution and enhancement of subtle structures. The network architecture uses multi-scale feature representation to ensure stability and robustness in various imaging noise environments. (2) Using dual-view data for deep feature fusion and aggregation: Through dual-view data input, including CT scan data from different perspectives, to improve the overall clarity and accuracy of bone imaging, the dual-view feature fusion module uses a deep feature fusion strategy to jointly process the data from the two perspectives and extract complementary information from each perspective. This process uses a deep fusion method based on the attention mechanism to ensure the integrity and detail clarity of the image through global and local feature weighting. The advantage of dual-view fusion is that it compensates for the information loss or artifacts that may exist in a single view, thereby generating a more comprehensive bone structure imaging; (3) Implement unsupervised domain adaptation (UDA) to adapt to different input distributions. The domain adaptation module adopts an unsupervised learning strategy to adapt to different input data distributions, so that the model can be trained on unlabeled data, expanding its adaptability and application scope. By aligning the feature distributions of the target domain and the source domain, the unsupervised domain adaptation module can effectively reduce the distribution difference and improve the generalization ability of the model, thereby maintaining efficient imaging quality in different environments. This module adopts adversarial training and self-supervised loss function to ensure the stable performance of the model on diverse data sets.

2. A method for bone imaging based on dual-view depth enhancement CT according to claim 1, characterized in that: In the step (1), the deep learning architecture DVSR3D uses an attention mechanism and autoencoder-style reconstruction to achieve better noise reduction and feature clarity. The architecture adopts an autoencoder-style design and reconstructs images by introducing an attention mechanism and deep feature aggregation. The architecture includes an encoder, a feature fusion module, and a decoder module. The encoder part is responsible for extracting multi-scale features from dual-view data and weighting them through an attention mechanism to obtain the most relevant feature representation. The fusion module combines local and global attention to ensure the effective integration of input features. The decoder module reconstructs and restores the extracted deep features to a high-resolution image, and reduces image artifacts and noise through a specific noise reduction strategy to improve image clarity and detail performance.

3. The method for bone imaging based on dual-view depth enhancement CT according to claim 1, characterized in that: The dual-view data fusion in step (2) enhances the details and accuracy of the trabecular bone image by combining orthogonal CT views. The CT scan data from orthogonal perspectives (such as horizontal and vertical directions) are input to capture the microstructural details of the trabecular bone. The combination of orthogonal perspectives ensures that the data contains more dimensional information, reduces artifacts and information redundancy problems, and through the feature fusion module, the system can complement and weigh information from different perspectives, thereby enhancing the structural consistency and detail display of the image.

4. The method for bone imaging based on dual-view depth enhancement CT according to claim 1, characterized in that: The unsupervised domain adaptation method in step (3) enables the model to be trained with limited labeled data, thereby adapting to new medical data sets, and enables the model to be efficiently trained on limited labeled data or unlabeled data. The specific steps of the method include: (31) Feature alignment: By calculating the statistical difference (such as distribution distance) between the source domain and target domain feature distributions and reducing this difference through the domain adaptation loss function, the robustness of the model under the new data distribution is ensured; (32) Adversarial training: Using a generative adversarial network (GAN) or its variants, the model can adaptively adjust to the image features of the new environment, reducing the problem of the model overfitting to a single data source; (33) Self-supervised learning strategy: The model is trained in the absence of labels by using self-supervised mechanisms such as pseudo-label generation or image transformation adversarial training to enhance the generalization ability of new data. In practical applications, it is particularly suitable for the field of medical image analysis, such as the consistency of imaging data across hospitals or devices, which improves the applicability and reliability of the model in different medical environments.

5. A dual-view depth-enhanced CT bone imaging system, characterized in that: It is used for CT image restoration tasks and includes a deep learning model framework for CT image restoration to realize the functions of data acquisition, image registration, and data preprocessing, including Cascade-SwinUNETR, single-view restoration architecture, deep feature fusion module, and dual-view restoration architecture. Each module performs a specific function. These modules can be combined and reused to improve training efficiency and expand generalization capabilities. Cascade-SwinUNETR, as a basic module, provides a robust architectural foundation to support the construction of subsequent methods. The dual-view restoration architecture integrates a deep feature fusion module to process dual-view input data.

6. The dual-view depth-enhanced CT bone imaging system according to claim 5, characterized in that: The Cascade-SwinUNETR acts as a basic model in the framework. As a deep feature recovery module, it can be integrated with different input and output modules. The network adapts to the 3D Swin-Transformer module, U-Net++ structure and DropKey technology. The single-view recovery architecture is an independent model for single-view tasks. Before dual-view learning, pre-training is performed to accelerate training and enhance generalization ability. After pre-training, the trained recovery model is extracted and integrated into the overall architecture. The deep feature fusion module, CT scans come from two orthogonal perspectives. In order to allow the model to learn, identify and combine features from two sources, a deep feature fusion technology is used. This deep feature fusion technology combines deeply refined features instead of directly merging low-level images to improve performance. The dual-view recovery architecture is a dual-view architecture DVSR3D, including an encoder, a restorer and a decoder. The encoder partially upsamples and fuses deep features, the restorer generates an output image, and the decoder prevents overfitting through regularization.

7. The dual-view depth-enhanced CT bone imaging system according to claim 5, characterized in that: Data acquisition: The collated dataset is designed to support dual-view tasks and includes clinical CT scans and Micro-CT scans obtained from four different human bone specimens. The experimental setup includes acquiring CT scans from both horizontal and vertical directions, while Micro-CT scans are acquired from only a single direction. The horizontal and vertical directions refer to manually rotating the human bone blocks after the scanning procedure is completed. In practical applications, this process can be achieved by having the patient lie in different directions, such as supine or prone, or using different CT scanning equipment. It should be noted that all the original CT scans are initially unregistered because they are collected under different real-world environmental conditions and therefore contain difficult-to-handle noise distributions and different spatial information.

8. The dual-view depth-enhanced CT bone imaging system according to claim 5, characterized in that: Image registration: Since the data were originally acquired in different settings, accurate image correspondence was required when creating the dataset. To achieve this goal, image registration was performed using the open source software ITK-SNAP and Simple ITK Python library. ITK-SNAP was used for manual and automatic registration with 16 coarse registration levels and 8 fine registration levels. Subsequently, the registration algorithm in Simple ITK was used to optimize the resulting transformation parameters to generate registered data for dual-view CT scanning.

9. The dual-view depth-enhanced CT bone imaging system according to claim 5, characterized in that: Data preprocessing: Due to the large volume of the original registered data, a preprocessing step is required to make it suitable for training, validation, and testing of the model. The preprocessing includes random cropping from the registered data. Each cropping unit contains three samples, corresponding to view 1 CT scan, view 2 CT scan, and Micro-CT scan. The super-resolution ratio used in this task is four times, so the pixel dimension is aligned with this ratio. The random cropping process is restricted to specific boundaries. The cropping range is limited to 100 pixels on each side of the width, height, and depth of the registered data. The first 60 samples correspond to the first three bone samples and constitute the training dataset. The next 20 samples are associated with the last bone sample and used as a validation dataset. In addition, 20 more samples are extracted from the same bone sample to create an independent test dataset. Ideally, if more bone samples are available, it is better to select different bone samples as the test dataset. However, due to the challenges in obtaining the original data, this method is already the relatively optimal solution in this study. In addition, the dynamic range of all data is adjusted to the range of 0 to 255 and stored in grayscale image format.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: When the executable instructions are executed by a processor, the dual-view depth-enhanced CT bone imaging method according to any one of claims 1 to 4 is implemented.

Images