A Deep Learning-Based System and Method for Precise Synthesis of Spinal MRI Images into CT Images in Axial Spondyloarthritis

Abstract

This application discloses a system and method for processing precise synthetic CT images of the spine based on deep learning algorithms for axial spondyloarthritis. The system includes: a data acquisition unit for acquiring three-dimensional T1-weighted multi-gradient echo sequences, T2-weighted sequences, and water-lipid separation sequences obtained from magnetic resonance imaging of the target user's spine; a spatial registration unit for inputting multi-sequence data into a deformable registration network to perform spatial registration of multimodal features; a feature processing unit for fusing the registered multimodal features and performing dual-domain splitting based on an anatomical attention gating mechanism to obtain skeletal features and soft tissue features; and an image fusion unit for synthesizing synthetic CT images with HU values ​​corresponding to the target user's spine based on skeletal features and soft tissue features, achieving radiation-free, high-precision CT image acquisition, which can meet the diagnostic and grading requirements of axial spondyloarthritis while reducing the impact of radiation on the target user.

Description

Technical Field

[0001] This disclosure generally relates to the interdisciplinary fields of medical image processing and artificial intelligence, and specifically to a system and method for processing spinal MRI images of axial spondyloarthritis based on deep learning algorithms. Background Technology

[0002] Axial spondyloarthritis (axSpA) is a chronic inflammatory arthritis characterized primarily by arthritis of the axial spine.

[0003] While Magnetic Resonance Imaging (MRI), the mainstream imaging method, has the advantage of being sensitive to fibrous tissue, it is not sensitive to the bone cortex, making it difficult to clearly identify early-stage osteophyte signals. Furthermore, it is prone to signal overlap between osteophytes and surrounding fibers, leading to difficulties in quantitative assessment. Although combining MRI with computed tomography (CT) can solve the problem of clear identification, the risk of ionizing radiation from CT limits repeat examinations and makes follow-up of patients difficult.

[0004] Therefore, how to provide a CT image that can clearly display structural damage such as osteophytes and calcifications without radiation scanning has become an urgent problem to be solved. Summary of the Invention

[0005] In view of the above-mentioned defects or deficiencies in the existing technology, it is desirable to provide a system and method for precise synthesis of CT images of spinal MRI for axial spondyloarthritis based on deep learning algorithms, so as to achieve high-precision CT image acquisition without radiation, which can not only meet the diagnostic and grading requirements of axial spondyloarthritis, but also reduce the impact of radiation on the target user.

[0006] In a first aspect, embodiments of this application provide a system for precise synthesis of CT images from spinal MRI for axial spondyloarthritis based on deep learning algorithms, comprising:

[0007] The data acquisition unit is used to acquire three-dimensional T1-weighted multigradient echo sequences, T2-weighted sequences, and water-lipid separation sequences obtained from magnetic resonance imaging of the target user's spine.

[0008] A spatial registration unit is used to input the three-dimensional T1-weighted multi-gradient echo sequence, the T2-weighted sequence, and the water-lipid separation sequence into a deformable registration network to perform spatial registration of multimodal features.

[0009] The feature processing unit is used to fuse the registered multimodal features and perform dual-domain splitting based on the anatomical attention gating mechanism to obtain skeletal features and soft tissue features.

[0010] An image fusion unit is used to synthesize a synthetic CT image with a HU value corresponding to the spine of the target user based on the skeletal features and the soft tissue features.

[0011] In some embodiments, the feature processing unit further includes a biomechanical prior guidance subunit:

[0012] The biomechanical prior guidance subunit is used to analyze the biomechanical stress weights between the vertebrae of the target user.

[0013] The feature processing unit is further configured to:

[0014] When performing dual-domain segmentation based on the anatomical attention gating mechanism, the skeletal features are determined in response to the biomechanical stress weights between the vertebrae.

[0015] In some embodiments, the feature processing unit is further configured to:

[0016] The biomechanical stress weights are concatenated with the multimodal fusion features and used as input features for the skeletal feature decomposition branch in the dual-domain splitting, so as to obtain the skeletal features through dual-domain splitting.

[0017] In some embodiments, the image fusion unit is further configured to:

[0018] High-stress regions between the vertebrae are identified where the biomechanical stress weight is greater than or equal to a preset weight, and the HU value corresponding to the high-stress region is enhanced.

[0019] In some embodiments, the feature processing unit further includes a osteophyte propagation prediction subunit:

[0020] The osteophyte diffusion prediction subunit is used to predict the probability of future osteophyte growth based on the historical osteophyte volume and / or current stress value in the magnetic resonance imaging of the target user.

[0021] The feature processing unit is further configured to:

[0022] When performing dual-domain splitting based on the anatomical attention gating mechanism, the soft tissue feature weights are adjusted in response to the osteophyte growth probability.

[0023] In some embodiments, the soft tissue feature weights are inversely proportional to the osteophyte growth probability.

[0024] In some embodiments, the image fusion unit is further configured to:

[0025] In the process of synthesizing a synthetic CT image with a HU value corresponding to the spine of the target user based on the skeletal features and the soft tissue features, the synthesis of the synthetic CT is dynamically adjusted based on the regional changes between the skeletal features and the soft tissue features.

[0026] In some embodiments, the image fusion unit is further configured to:

[0027] Based on multimodal fusion features, the non-uniformity of each feature point is determined;

[0028] Based on the non-uniformity of each feature point, the image contribution correction coefficient of the soft tissue feature is determined;

[0029] Based on the skeletal features, the soft tissue features, and the image contribution correction coefficient of the soft tissue features, the HU value corresponding to each feature point is determined.

[0030] The synthesized CT image is synthesized based on the HU value.

[0031] In some embodiments, the data acquisition unit is further configured to:

[0032] Identify the age correction amount for the target user, and adjust the flip angle of the three-dimensional T1-weighted multigradient echo sequence based on the age correction amount.

[0033] Secondly, embodiments of this application provide a method for accurately synthesizing CT images of spinal MRI in axial spondyloarthritis based on deep learning algorithms, including:

[0034] Three-dimensional T1-weighted multigradient echo sequences, T2-weighted sequences, and water-lipid separation sequences were obtained from magnetic resonance imaging of the target user's spine;

[0035] The three-dimensional T1-weighted multi-gradient echo sequence, the T2-weighted sequence, and the water-lipid separation sequence are input into a deformable registration network to perform spatial registration of multimodal features.

[0036] The registered multimodal features are fused and then split into two domains based on the anatomical attention gating mechanism to obtain skeletal features and soft tissue features.

[0037] Based on the skeletal features and the soft tissue features, a synthetic CT image with a HU value corresponding to the target user's spine is synthesized.

[0038] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in embodiments of this application.

[0039] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in embodiments of this application.

[0040] Fifthly, embodiments of this application provide a computer program product, including a computer program, characterized in that, when the computer program is executed by a processor, it implements the method described in embodiments of this application.

[0041] The spinal MRI-based accurate synthetic CT image processing system and method for axial spondyloarthritis based on deep learning algorithms provided in this application can spatially register three-dimensional T1-weighted multi-gradient echo sequences, T2-weighted sequences, and water-lipid separation sequences obtained from magnetic resonance imaging through a deformable registration network. This improves the spatial consistency of multi-sequence feature data and provides a reliable data foundation for subsequent synthesis of high-precision synthetic CT images. Then, based on the anatomical attention gating mechanism, the multimodal fusion features are split into two domains to obtain skeletal and soft tissue features. This ensures that the obtained skeletal and soft tissue features highly satisfy anatomical characteristics and can be mutually verified. Finally, a synthetic CT image with a HU value is synthesized based on reliable skeletal and soft tissue features, making the synthesized CT image closer to the real CT image corresponding to the target user's spine.

[0042] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0043] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0044] Figure 1 This paper illustrates a flowchart of a method for accurately synthesizing CT images of spinal MRI in axial spondyloarthritis based on deep learning algorithms, according to an embodiment of this application.

[0045] Figure 2 This illustration shows a schematic diagram of the structure of a deep learning algorithm-based spinal MRI-to-CT image processing system for axial spondyloarthritis, according to an embodiment of this application.

[0046] Figure 3 A schematic diagram of the structure of a spinal MRI-based precise synthetic CT image processing system for axial spondyloarthritis based on a deep learning algorithm, provided in another embodiment of this application, is shown. Detailed Implementation

[0047] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0048] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0049] axSpA is a chronic inflammatory rheumatic disease that severely impacts patients' quality of life, and the spine is a key site for early diagnosis and treatment assessment. MRI primarily uses magnetic fields and radio waves for imaging, emphasizing contrast and detail in soft tissues. MRI is highly effective when detecting changes in soft tissues (such as inflammation and edema). However, MRI is less effective at imaging bone structures because minerals in bone have a relatively small impact on the MRI signal, resulting in insufficient sensitivity in detecting bone destruction. In MRI images, bone tissue typically appears as a low signal intensity, making it difficult to separate from surrounding tissues, thus further complicating the interpretation of bone destruction.

[0050] CT's high spatial resolution and bone density-sensitive imaging characteristics make it more accurate in identifying minute bone damage, better revealing subtle changes in bone, including bone destruction and sclerosis. However, CT carries the risk of ionizing radiation and is not suitable for frequent repeat examinations, especially for young patients. Many patients may have already undergone MRI scans, and repeating CT scans not only increases radiation risk but also leads to wasted time and effort. Therefore, developing a radiation-free imaging method that can accurately reconstruct bone details has significant clinical value. Deep learning has shown great potential in medical image synthesis and reconstruction. By constructing accurate cross-modal image generation models, it is hoped that non-destructive conversion from MRI to CT can be achieved, providing innovative solutions for the early diagnosis and treatment monitoring of axSpA.

[0051] To further illustrate the technical solutions provided in the embodiments of this application, a detailed description is provided below in conjunction with the accompanying drawings and specific implementation methods. Although the embodiments of this application provide method operation instruction steps as shown in the following embodiments or drawings, the method may include more or fewer operation instruction steps based on conventional or non-creative effort. In steps where there is no logically necessary causal relationship, the execution order of these steps is not limited to the execution order provided in the embodiments of this application. In actual processing or when the device executes the method, it may be executed sequentially or in parallel according to the method shown in the embodiments or drawings.

[0052] It should be noted that the acquisition or use of data in the embodiments of this application requires the user's consent. The relevant data can only be obtained after the user's authorization, and the acquisition or use of the data complies with the provisions of relevant laws and regulations.

[0053] Please refer to Figure 1 , Figure 1 This illustration shows a flowchart of a method for accurately synthesizing CT images of spinal MRI in axial spondyloarthritis based on a deep learning algorithm, according to an embodiment of this application. Figure 1 As shown, the method includes:

[0054] Step 101: Obtain the three-dimensional T1-weighted multigradient echo sequence, T2-weighted sequence, and water-lipid separation sequence obtained by magnetic resonance imaging of the target user's spine.

[0055] Specifically, magnetic resonance imaging was performed on the spine of the target user. By setting the parameters TR=5.2ms, TE=2.5ms, and slice thickness to 0.8mm, a three-dimensional T1-weighted multi-gradient echo sequence was obtained. By setting the parameters TR=3000ms and TE=80ms, a T2-weighted sequence was obtained, as well as a water-fat separation sequence of the target user's spine was acquired.

[0056] In magnetic resonance imaging (MRI), the three-dimensional T1-weighted multi-gradient echo sequence is abbreviated as 3D-T1-MGE. 3D refers to three-dimensional volumetric imaging, which achieves arbitrary plane reconstruction (e.g., sagittal, coronal, axial) by continuously acquiring isotropic voxels, avoiding the gaps between slices in traditional 2D layering. T1-weighting highlights the T1 relaxation differences of tissues and displays anatomical structures by adjusting repetition time (TR) and echo time (TE). Multi-gradient echo (MGE) sequence acquires signals with different TE values ​​using multiple gradient echoes after a single radiofrequency excitation, improving scanning efficiency and supporting quantitative analysis, providing a data basis for subsequent synthesis of synthetic CT with HU (Hounsfield Hnit) values.

[0057] In a preferred embodiment, before performing magnetic resonance imaging on the target user, the age correction amount of the target user is further identified, and the inversion angle of the three-dimensional T1-weighted multigradient echo sequence is adjusted based on the age correction amount.

[0058] In other words, the inversion angle of the 3D T1-weighted gradient echo sequence can be adjusted based on the user's age. Specifically, the target user's age is obtained. If the target user's age is greater than or equal to a preset age, it means the target user is an adult and no adjustment is needed; the corresponding age correction is 0. If the target user's age is less than the preset age, it means the target user is a child and adjustment is needed. The age correction is then obtained, and the inversion angle of the 3D T1-weighted gradient echo sequence is adjusted based on the age correction.

[0059] Optionally, the age correction amount can be a fixed value, such as 5°, or a value that is linearly related to age; this application does not impose any specific limitations.

[0060] For example, the age correction is a fixed value of 5°, and when the target user is identified as a child, the inversion angle of the three-dimensional T1-weighted gradient echo sequence is adjusted from 15° to 10°.

[0061] Therefore, this application reduces the specific absorption ratio (SAR) by adjusting the inversion angle of the three-dimensional T1-weighted gradient echo sequence to meet the safety requirements for pediatric scanning.

[0062] Step 102: Input the three-dimensional T1-weighted multi-gradient echo sequence, T2-weighted sequence, and water-lipid separation sequence into a deformable registration network to perform spatial registration of multimodal features.

[0063] It should be noted that, due to the large spatial extent of the spine, especially including the area of ​​respiratory motion of the target user, spatial deviations between different sequences are easily caused during magnetic resonance imaging. Based on this, this application proposes using a deformable registration network to spatially register the features of three-dimensional T1-weighted multi-gradient echo sequences, T2-weighted sequences, and water-lipid separation sequences. This eliminates sequence misalignment caused by respiratory motion or changes in the target user's position, ensuring spatial consistency of multimodal features and enabling accurate superposition of pathological features through the aligned multi-sequence feature data.

[0064] For example, spatial registration can be performed using the following formula:

[0065]

[0066] Where φ is the deformation field, representing the non-rigid deformation of the sequence feature image, φ id For unit deformation field, i.e., the initial state without displacement, λ1 is the regularization coefficient used to balance deformation smoothness, λ2 is the regularization coefficient used to balance multimodal alignment accuracy, and NCC represents the calculation of feature images I of two images (three-dimensional T1-weighted multigradient echo sequences). T1 Feature image I of T2 weighted sequence T2Local similarity of ) The operator for the deformation field domain image represents the image I corresponding to the T2 weighted sequence. T2 Perform non-rigid deformation.

[0067] Step 103: The registered multimodal features are fused and then split into two domains based on the anatomical attention gating mechanism to obtain skeletal features and soft tissue features.

[0068] It should be noted that the registered multimodal features are fused to obtain the multimodal fused features corresponding to magnetic resonance imaging.

[0069] It should also be noted that the dual-domain splitting of multimodal fusion features based on the anatomical attention gating mechanism is essentially to ensure that the skeletal features obtained from the skeletal feature splitting conform to biomechanical stress, and the soft tissue features obtained from the soft tissue feature splitting conform to the osteophyte diffusion prediction results, so that the skeletal features split by the anatomical attention gating mechanism conform to the biomechanical development trend of osteophyte formation, i.e. the pathological mechanism of osteophyte calcification, while the soft tissue features can effectively avoid the osteophyte diffusion prediction results, reduce the probability of overlap between soft tissue features and skeletal features, and reduce feature conflicts in the subsequent synthesized CT.

[0070] Specifically, in some embodiments, since the cortical bone signal is low in multiple sequences of magnetic resonance imaging, this application further proposes a biomechanical prior-guided dynamic modeling of the mechanical transmission path of the spine, so as to increase the anatomical attention mechanism of the osteophyte formation area based on the biomechanical stress condition.

[0071] Preferably, the biological stress weights between vertebrae are calculated using an adaptive adjacency matrix:

[0072]

[0073] Among them, A ij Stress(V) represents the biomechanical stress weight between vertebrae i and j. i V j ) represents the stress value between the two vertebrae, and α is the biomechanical weighting coefficient. These are the attention query vector and key vector, used to calculate the feature similarity between cones.

[0074] Furthermore, after obtaining the biomechanical weights between the vertebrae, the biomechanical stress weights are used as the anatomical attention mechanism corresponding to the multimodal fusion feature map. Specifically, the biomechanical stress weights are concatenated with the multimodal fusion features, and the concatenated features are used as the input features for the skeletal feature decomposition branch in the dual-domain splitting process, so as to obtain skeletal features through decomposition. In other words, the biomechanical stress weights are used as an attention mask to select skeletal features in the skeletal feature decomposition branch, thereby improving the accuracy of skeletal feature recognition.

[0075] As a specific example, based on the pathological mechanism of osteophyte formation, it is known that the osteophyte formation area experiences higher stress. Therefore, high-stress areas, such as A, are... ij Regions with a value >0.7 are designated as osteophyte regions. During the dual-domain splitting stage, these regions are assigned to skeletal branches. Enhancing the corresponding HU value during image synthesis can further improve the matching degree between the osteophyte region and the real CT image.

[0076] In another feasible embodiment, a osteophyte diffusion prediction model is used to predict the probability of future osteophyte growth based on the historical osteophyte volume and / or current stress value in the target user's magnetic resonance imaging.

[0077] It should be understood that the osteophyte propagation prediction model was trained using a large number of expert-annotated MRI images and real CT images.

[0078] Optionally, the growth of osteophytes for a target user can be predicted based on the historical osteophyte volume recorded in at least one historical feature image of the target user, or based on the aforementioned vertebral stress value, or when the information is relatively abundant, both the historical osteophyte volume and the current stress value can be used for prediction. This application does not make any specific limitations.

[0079] It should be noted that the predicted probability of future osteophyte growth is the same as the predicted probability of osteophyte growth at the current time of the target user's MRI scan. It should also be noted that the osteophyte growth probability is also used to represent the probability that the feature location is an osteophyte.

[0080] In other words, after obtaining the osteophyte growth probability corresponding to each feature location point, the soft tissue feature weights can be corrected in response to the osteophyte growth probability when performing dual-domain splitting based on the anatomical attention gating mechanism.

[0081] Specifically, the weight of soft tissue features is inversely proportional to the probability of osteophyte growth.

[0082] For example, when the probability of osteophyte growth is P(ΔV) t When ), 1-P(ΔV) can be used. t ) is used as the weight corresponding to the soft tissue feature.

[0083] In other words, before performing dual-domain splitting, the corrected result of the osteophyte growth probability is concatenated with the multimodal fusion feature as the corresponding weight of the soft tissue feature. The concatenated feature is then used as the input feature for the soft tissue feature splitting branch in the dual-domain splitting process, so as to obtain soft tissue features through splitting. In other words, the corrected result of the osteophyte growth probability is used as an attention mask to select soft tissue features in the soft tissue feature splitting branch, thereby improving the accuracy of soft tissue feature recognition.

[0084] Therefore, this application embodiment uses a dual-domain segmentation based on anatomical attention gating, which can effectively segment skeletal features based on the biomechanical stress of osteophyte formation and soft tissue features based on the osteophyte growth probability. This ensures the accuracy of skeletal and soft tissue feature segmentation, reduces the risk of overlap, and uses different weights to segment the two types of features, verifying each other and improving the accuracy of feature segmentation.

[0085] Step 104: Based on skeletal and soft tissue features, synthesize a synthetic CT image with HU value corresponding to the target user's spine.

[0086] Optionally, CT images can be synthesized by separately determining the HU values ​​corresponding to skeletal features and soft tissue features, and then using the HU values.

[0087] Specifically, the synthesis of synthetic CT images can be dynamically adjusted based on regional changes between skeletal and soft tissue features.

[0088] In a feasible embodiment, the non-uniformity of each feature point can be determined based on multimodal fusion features, and the image contribution correction coefficient of soft tissue features can be determined based on the non-uniformity of each feature point; the HU value corresponding to each feature point can be determined based on the image contribution correction coefficient of bone features, soft tissue features and soft tissue features, and the HU value can be used to synthesize a CT image.

[0089] Specifically, in the embodiments of this application, the non-uniformity of multimodal fusion features can be determined by using feature variance. Based on the image discrimination principle, the high variance region is the inflammation region and the low variance region is the osteophyte region. Based on this, the dynamic correction weight, i.e. the image contribution correction coefficient, is determined when generating the synthetic CT image.

[0090] Optionally, the image contribution correction coefficient is positively correlated with the non-uniformity of the multimodal fusion features. That is, the higher the non-uniformity of the feature distribution, the larger the image contribution correction coefficient, which is closer to 1, in order to enhance the influence of soft tissue features on image reconstruction. The lower the uniformity of the feature distribution, the smaller the image contribution correction coefficient, which is closer to 0, in order to ensure the influence of skeletal features on image reconstruction.

[0091] For example, the HU value corresponding to each feature point can be determined using the following formula:

[0092] HU = G bone (F bone-out )+β·tan(G soft (F soft-out ))

[0093] Where HU is the HU value corresponding to each feature point, and F bone-out The skeletal features output by the skeletal feature branch, G bone (F bone-out ) represents the HU value output by the skeletal feature branch, F soft-out G represents the soft tissue features output by the soft tissue feature branch. soft (F soft-out ) represents the HU value output by the soft tissue feature branch, and β is the image contribution correction coefficient.

[0094] Therefore, by correcting and adjusting the HU value generated by soft tissue features during the HU value generation process through the image contribution correction coefficient, the conflict between soft tissue information and bone information during the synthesis of CT images can be effectively prevented, thereby effectively ensuring the reliability of the synthesized CT images.

[0095] Therefore, the deep learning algorithm-based method for accurately synthesizing CT images of spinal MRI in axial spondyloarthritis provided in this application can spatially register the three-dimensional T1-weighted multi-gradient echo sequence, T2-weighted sequence, and water-lipid separation sequence obtained from MRI using a deformable registration network. This improves the spatial consistency of multi-sequence feature data and provides a reliable data foundation for subsequent synthesis of high-precision synthetic CT images. Then, based on the anatomical attention gating mechanism, the multimodal fusion features are split into two domains to obtain skeletal and soft tissue features. This ensures that the obtained skeletal and soft tissue features highly satisfy anatomical features and can be mutually verified. Finally, a synthetic CT image with a HU value is synthesized based on reliable skeletal and soft tissue features, making the synthesized CT image closer to the real CT image corresponding to the target user's spine.

[0096] It should be noted that although the operation of the method of the present invention is described in a specific order in the accompanying drawings, this does not require or imply that the operations must be performed in that specific order, or that all the operations shown must be performed in order to achieve the desired result.

[0097] Figure 2 This illustration shows a schematic diagram of the structure of a deep learning algorithm-based spinal MRI-to-CT image processing system for axial spondyloarthritis, provided in an embodiment of this application.

[0098] like Figure 2As shown, the deep learning algorithm-based spinal MRI-to-CT image processing system 10 for axial spondyloarthritis includes:

[0099] Data acquisition unit 11 is used to acquire three-dimensional T1-weighted multigradient echo sequences, T2-weighted sequences, and water-lipid separation sequences obtained by magnetic resonance imaging of the target user's spine;

[0100] Spatial registration unit 12 is used to input the three-dimensional T1-weighted multi-gradient echo sequence, the T2-weighted sequence and the water-lipid separation sequence into a deformable registration network to perform spatial registration of multimodal features;

[0101] The feature processing unit 13 is used to fuse the registered multimodal features and perform dual-domain splitting based on the anatomical attention gating mechanism to obtain skeletal features and soft tissue features.

[0102] Image fusion unit 14 is used to synthesize a synthetic CT image with HU value corresponding to the spine of the target user based on the skeletal features and the soft tissue features.

[0103] In some embodiments, such as Figure 3 As shown, the feature processing unit 13 further includes a biomechanical prior guidance subunit 131:

[0104] The biomechanical prior guidance subunit 131 is used to analyze the biomechanical stress weights between the vertebrae of the target user.

[0105] The feature processing unit 13 is further configured to:

[0106] When performing dual-domain segmentation based on the anatomical attention gating mechanism, the skeletal features are determined in response to the biomechanical stress weights between the vertebrae.

[0107] In some embodiments, the feature processing unit 13 is further configured to:

[0108] The biomechanical stress weights are concatenated with the multimodal fusion features and used as input features for the skeletal feature decomposition branch in the dual-domain splitting, so as to obtain the skeletal features through dual-domain splitting.

[0109] In some embodiments, the image fusion unit 14 is further configured to:

[0110] High-stress regions between the vertebrae are identified where the biomechanical stress weight is greater than or equal to a preset weight, and the HU value corresponding to the high-stress region is enhanced.

[0111] In some embodiments, such as Figure 3 As shown, the feature processing unit 13 further includes a osteophyte propagation prediction subunit 132:

[0112] The osteophyte diffusion prediction subunit 132 is used to predict the probability of future osteophyte growth based on the historical osteophyte volume and / or current stress value in the magnetic resonance imaging of the target user.

[0113] The feature processing unit 13 is further configured to:

[0114] When performing dual-domain splitting based on the anatomical attention gating mechanism, the soft tissue feature weights are adjusted in response to the osteophyte growth probability.

[0115] In some embodiments, the soft tissue feature weights are inversely proportional to the osteophyte growth probability.

[0116] In some embodiments, the image fusion unit 14 is further configured to:

[0117] In the process of synthesizing a synthetic CT image with a HU value corresponding to the spine of the target user based on the skeletal features and the soft tissue features, the synthesis of the synthetic CT is dynamically adjusted based on the regional changes between the skeletal features and the soft tissue features.

[0118] In some embodiments, the image fusion unit 14 is further configured to:

[0119] Based on multimodal fusion features, the non-uniformity of each feature point is determined;

[0120] Based on the non-uniformity of each feature point, the image contribution correction coefficient of the soft tissue feature is determined;

[0121] Based on the skeletal features, the soft tissue features, and the image contribution correction coefficient of the soft tissue features, the HU value corresponding to each feature point is determined.

[0122] The synthesized CT image is synthesized based on the HU value.

[0123] In some embodiments, the data acquisition unit 11 is further configured to:

[0124] Identify the age correction amount for the target user, and adjust the flip angle of the three-dimensional T1-weighted multigradient echo sequence based on the age correction amount.

[0125] It should be understood that the modules or modules described in the Deep Learning Algorithm-Based Spinal MRI Precision Synthetic CT Image Processing System 10 for Axial Spondyloarthritis are similar to those in the reference... Figure 1The steps in the described method correspond accordingly. Therefore, the operations and features described above for the method are also applicable to the deep learning algorithm-based spinal MRI-based precise synthetic CT image processing system 10 for axial spondyloarthritis and its included modules, and will not be repeated here. The deep learning algorithm-based spinal MRI-based precise synthetic CT image processing system 10 for axial spondyloarthritis can be pre-implemented in the browser or other secure applications of an electronic device, or can be loaded into the browser or its secure applications of an electronic device through download or other means. The corresponding modules in the deep learning algorithm-based spinal MRI-based precise synthetic CT image processing system 10 for axial spondyloarthritis can cooperate with modules in the electronic device to implement the solutions of the embodiments of this application.

[0126] The division of modules or units mentioned in the detailed description above is not mandatory. In fact, according to the embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0127] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operational instructions of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two connected blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified functions or operational instructions, or using a combination of dedicated hardware and computer instructions.

[0128] In another aspect, this application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments, or may exist independently and not assembled into the electronic device. The aforementioned computer-readable storage medium stores one or more programs that, when used by one or more processors, execute the deep learning algorithm-based method for precise synthesis of CT images from spinal MRI in axial spondyloarthritis, as described in this application.

[0129] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A deep learning algorithm-based axial spondylarthritis spine MRI precise synthetic CT image processing system, characterized in that, The method comprises the following steps: a data acquisition unit is configured to acquire a three-dimensional T1-weighted multi-echo sequence, a T2-weighted sequence, and a water-fat separation sequence obtained by performing magnetic resonance imaging on a spine of a target user; a spatial registration unit is configured to input the three-dimensional T1-weighted multi-echo sequence, the T2-weighted sequence, and the water-fat separation sequence into a deformable registration network to perform spatial registration on multi-modal features; a feature processing unit is configured to perform feature fusion on the registered multi-modal features, and perform dual-domain splitting based on an anatomical attention gate mechanism to obtain bone features and soft tissue features; an image fusion unit is configured to synthesize a synthetic CT image with HU values corresponding to the spine of the target user based on the bone features and the soft tissue features; the feature processing unit further comprises a biomechanics prior guidance subunit: the biomechanics prior guidance subunit is configured to analyze biomechanics stress weights between vertebral bodies of the target user, and determine the bone features in response to the biomechanics stress weights between the vertebral bodies when performing dual-domain splitting based on the anatomical attention gate mechanism; the biomechanics stress weights are spliced with multi-modal fusion features and used as input features of a bone feature splitting branch in the dual-domain splitting to obtain the bone features through dual-domain splitting; the feature processing unit further comprises an osteophyte diffusion prediction subunit:

2. The deep learning algorithm-based axial spondyloarthritis spine MRI to precise synthetic CT image processing system according to claim 1, characterized in that, the osteophyte diffusion prediction subunit is configured to predict a future osteophyte growth probability based on a historical osteophyte volume and / or a current stress value in the magnetic resonance imaging of the target user, and correct the soft tissue feature weights in response to the osteophyte growth probability when performing dual-domain splitting based on the anatomical attention gate mechanism. the image fusion unit is further configured to:

3. The deep learning algorithm-based axial spondyloarthritis spine MRI to precise synthetic CT image processing system of claim 1, wherein, identify a high stress region where the biomechanics stress weights between the vertebral bodies are greater than or equal to a preset weight, and perform strengthening processing on HU values corresponding to the high stress region.

4. The deep learning algorithm-based axial spondyloarthritis spine MRI to precise synthetic CT image processing system of claim 1, wherein, the soft tissue feature weights are inversely proportional to the osteophyte growth probability. the image fusion unit is further configured to:

5. The deep learning algorithm-based axial spondyloarthritis spine MRI to precise synthetic CT image processing system according to claim 4, characterized in that, during the process of synthesizing the synthetic CT image with HU values corresponding to the spine of the target user based on the bone features and the soft tissue features, dynamically adjust the synthesis of the synthetic CT based on regional changes between the bone features and the soft tissue features. the image fusion unit is further configured to: determine unevenness of each feature point based on multi-modal fusion features; determine an image contribution correction coefficient of the soft tissue features based on the unevenness of each feature point; determine HU values corresponding to each feature point based on the bone features, the soft tissue features, and the image contribution correction coefficient of the soft tissue features; 6. The deep learning algorithm-based axial spondyloarthritis spine MRI to precise synthetic CT image processing system of claim 1, wherein, synthesize the synthetic CT image based on the HU values. the data acquisition unit is further configured to:

7. A method for processing precise synthetic CT images of axial type spondylitis of the spine based on deep learning algorithms, characterized in that, identify an age correction amount of the target user, and adjust a flip angle of the three-dimensional T1-weighted multi-echo sequence based on the age correction amount. The method comprises the following steps: acquiring a three-dimensional T1-weighted multi-echo sequence, a T2-weighted sequence, and a water-fat separation sequence obtained by performing magnetic resonance imaging on a spine of a target user; input the three-dimensional T1 weighted multi- gradient echo sequence, the T2 weighted sequence and the water-fat separation sequence to a deformable registration network to spatially register multi-modal features; fuse the registered multi-modal features and perform dual-domain splitting based on an anatomical attention gate mechanism to obtain skeletal features and soft tissue features; synthesize a synthesized CT image with HU values corresponding to the target user's spine based on the skeletal features and the soft tissue features; wherein the registered multi-modal features are fused by a feature processing unit based on an anatomical attention gate mechanism to obtain skeletal features and soft tissue features; the feature processing unit further comprises a biomechanical prior guidance subunit: the biomechanical prior guidance subunit is configured to analyze the biomechanical stress weight between the target user's vertebral bodies; when performing dual-domain splitting based on the anatomical attention gate mechanism, the biomechanical stress weight between the vertebral bodies is determined in response to the biomechanical stress weight between the vertebral bodies; the biomechanical stress weight is spliced with the multi-modal fusion features and used as an input feature of a skeletal feature splitting branch in the dual-domain splitting to obtain the skeletal features through dual-domain splitting; the feature processing unit further comprises a bone hyperplasia diffusion prediction subunit: the bone hyperplasia diffusion prediction subunit is configured to predict a future bone hyperplasia growth probability based on a historical bone hyperplasia volume and / or a current stress value in the target user's magnetic resonance imaging; when performing dual-domain splitting based on the anatomical attention gate mechanism, the soft tissue feature weight is corrected in response to the bone hyperplasia growth probability.

Images