MRI (Magnetic Resonance Imaging) synthesized CT (Computed Tomography) image processing system and method for sacroiliac joint of central axis type spondyloarthritis

Abstract

The invention discloses a central axis type spinal arthritis sacroiliac joint MRI synthesis CT image processing system and method based on a deep learning algorithm, and the system comprises a data obtaining unit which is used for obtaining magnetic resonance sequence data obtained through the magnetic resonance imaging of a sacroiliac joint of a target user; the deep learning model unit is used for inputting the magnetic resonance sequence data into a deep learning algorithm model trained by federal learning to obtain bone cortex features and soft tissue features of sacroiliac joints; the image synthesis unit is used for synthesizing a synthesized CT image corresponding to the sacroiliac joint of the target user based on the bone cortex features and the soft tissue features, non-radiation high-precision CT image acquisition is achieved, the diagnosis requirement for the axial spine joint in the sacroiliac joint can be met, and the influence of radiation on the target user can be reduced.

Description

Technical Field

[0001] The present disclosure generally relates to the field of medical image processing technology, and more particularly to a system and method for processing MRI-to-CT images of sacroiliac joints of axial spondyloarthritis based on a deep learning algorithm. Background Art

[0002] Axial spondyloarthritis (axSpA) is a chronic inflammatory arthritis characterized by axial spondyloarthritis. The disease often affects the sacroiliac joints and can cause bamboo-like changes in the spine, severely impacting patients' quality of life. Accurate detection of sacroiliac joint structural damage is crucial for the radiographically positive stage of axial spondyloarthritis (i.e., ankylosing spondylitis).

[0003] Currently, clinical interpretation mainly relies on the combined use of computed tomography (CT) and magnetic resonance imaging (MRI). Specifically, CT clearly displays structural damage such as cortical bone erosion, sclerosis, and ankylosis through its high X-ray attenuation coefficient. However, its ionizing radiation risk limits repeated examinations, especially in pediatric patients. Although MRI is radiation-free, it has insufficient resolution of bone erosion edges and low sensitivity.

[0004] Based on this, how to provide a CT image that can clearly show structural damage such as cortical erosion and sclerosis without the need for 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 desired to provide a sacroiliac joint MRI synthesized CT image processing system and method for axial spondyloarthritis based on deep learning algorithm to achieve radiation-free high-precision CT image acquisition, which can not only meet the diagnostic needs of sacroiliac joints of axial spinal joints, but also reduce the impact of radiation on target users.

[0006] In a first aspect, an embodiment of the present application provides a system for processing MRI-to-CT images of sacroiliac joints in axial spondyloarthritis based on a deep learning algorithm, comprising:

[0007] a data acquisition unit, configured to acquire magnetic resonance sequence data obtained by performing magnetic resonance imaging on the sacroiliac joint of a target user;

[0008] a deep learning model unit, configured to input the magnetic resonance sequence data into a deep learning algorithm model trained using federated learning to obtain cortical bone features and soft tissue features of the sacroiliac joint;

[0009] An image synthesis unit is used to synthesize a synthetic CT image corresponding to the sacroiliac joint of the target user based on the bone cortex feature and the soft tissue feature.

[0010] In some embodiments, the magnetic resonance sequence data includes a three-dimensional T1-weighted multi-gradient echo sequence, a T2-weighted sequence, and a water-fat separation sequence;

[0011] The setting parameters of the three-dimensional T1-weighted multi-gradient echo sequence include TR=5.2ms, TE=2.5ms, and a slice thickness of 0.8mm; the setting parameters of the T2-weighted sequence are TR3000ms and TE=80ms.

[0012] In some embodiments, the data acquisition unit is further configured to:

[0013] An age correction value of a target user is identified, and a flip angle of the three-dimensional T1-weighted multi-gradient echo sequence is adjusted based on the age correction value.

[0014] In some embodiments, the deep learning algorithm model unit includes an anatomical specific attention subunit,

[0015] The anatomically specific attention subunit includes a spatial-channel dual-path attention mechanism.

[0016] In some embodiments, deformable convolution is used to dynamically adjust sampling points in the spatial path, so that the convolution operation adapts to the geometric deformation of the sacroiliac joint anatomical structure.

[0017] In some embodiments, the image synthesis unit includes an anatomical prior constraint subunit,

[0018] The anatomical prior constraint subunit is used to provide an anatomical loss function for the deep learning algorithm model during the deep learning algorithm model training phase, and the anatomical prior constraints include at least cortical bone continuity loss constraint, joint gap distance loss constraint and bone erosion area resistance loss constraint.

[0019] In some embodiments, the image synthesis unit includes a feature alignment subunit,

[0020] The feature alignment subunit is used to make the distribution of bone cortical features and soft tissue features in the synthetic CT image closer to the distribution of bone cortical features and soft tissue features in the real CT image during the deep learning algorithm model training stage.

[0021] In some embodiments, the image synthesis unit is further configured to:

[0022] HU values ​​are mapped in the synthetic CT image based on the soft tissue features.

[0023] In some embodiments, quantitative analysis is also included.

[0024] The quantitative analysis unit is used to determine the bone erosion volume and maximum depth in the synthetic CT image according to the HU value mapped in the synthetic CT image.

[0025] In a second aspect, the present invention provides a method for processing MRI-synthesized CT images of sacroiliac joints in axial spondyloarthritis based on a deep learning algorithm, comprising:

[0026] Acquiring magnetic resonance sequence data obtained by performing magnetic resonance imaging on the sacroiliac joint of the target user;

[0027] Inputting the magnetic resonance sequence data into a deep learning algorithm model trained using federated learning to obtain cortical bone features and soft tissue features of the sacroiliac joint;

[0028] A synthetic CT image corresponding to the sacroiliac joint of the target user is synthesized based on the cortical bone features and the soft tissue features.

[0029] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the embodiment of the present application when executing the program.

[0030] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements the method described in the embodiment of the present application.

[0031] In a fifth aspect, an embodiment of the present application provides 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 the embodiment of the present application.

[0032] The embodiment of the present application provides a MRI-synthesized CT image processing system and method for sacroiliac joints of axial spondyloarthritis based on a deep learning algorithm. The system and method can use the magnetic resonance sequence data of the sacroiliac joints of the target user to obtain the cortical bone features and soft tissue features of the sacroiliac joints, and then use the cortical bone features and soft tissue features to synthesize a synthetic CT image corresponding to the sacroiliac joints of the target user, thereby achieving radiation-free high-precision CT image acquisition, which can not only meet the diagnostic needs of the sacroiliac joints of axial spinal joints, but also reduce the impact of radiation on the target user.

[0033] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0035] Figure 1 An MRI image of the sacroiliac joint is shown;

[0036] Figure 2 A flowchart of a method for processing MRI-synthesized CT images of sacroiliac joints of axial spondyloarthritis based on a deep learning algorithm according to an embodiment of the present application is shown;

[0037] Figure 3 A schematic diagram of the structure of a sacroiliac joint MRI-to-CT image processing system for axial spondyloarthritis based on a deep learning algorithm provided in one embodiment of the present application is shown;

[0038] Figure 4 A structural schematic diagram of a sacroiliac joint MRI-to-CT image processing system for axial spondyloarthritis based on a deep learning algorithm provided in another embodiment of the present application is shown. DETAILED DESCRIPTION

[0039] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the invention are shown in the accompanying drawings.

[0040] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0041] AxSpA is a chronic inflammatory rheumatic disease that seriously affects the quality of life of patients. The sacroiliac joint is a key site for its early diagnosis and efficacy evaluation. MRI mainly uses magnetic fields and radio waves to image, emphasizing the contrast and details of soft tissues. MRI is very effective when detecting changes in soft tissues (such as inflammation, edema, etc.). Figure 1 As shown by the arrows in (a) and (b). However, MRI has a weak imaging capability for bone structure because the minerals in the bones have little effect on the MRI signal, resulting in insufficient sensitivity in detecting bone destruction. In MRI images, bone tissue usually has low signal intensity and is difficult to separate from surrounding tissues, making the interpretation of bone destruction more difficult, as shown in Figure 2. Figure 1 Indicated by * in (a) and (b).

[0042] The high spatial resolution and bone density-sensitive imaging characteristics provided by CT make it more accurate in identifying tiny bone destruction and better able to reveal subtle changes in bone, including bone destruction and sclerosis. However, CT carries the risk of ionizing radiation and is not suitable for frequent repeated examinations, especially in young patients. Many patients may have undergone MRI examinations before, and repeating CT scans will not only increase the radiation risk but also lead to wasteful repeated examinations. Therefore, the development of a radiation-free imaging method that can accurately reconstruct bone details has important clinical value. Deep learning has shown great potential in the field of medical image synthesis and reconstruction. By constructing an accurate cross-modal image generation model, it is expected to achieve lossless conversion from MRI to CT, providing innovative solutions for the early diagnosis and treatment monitoring of axSpA.

[0043] Based on this, this application proposes a MRI-synthesized CT image processing system and method for sacroiliac joints of axial spondyloarthritis based on a deep learning algorithm to achieve radiation-free CT image acquisition and provide high-precision examination images for more accurate diagnosis and treatment decisions.

[0044] In order to further illustrate the technical solutions provided by the embodiments of the present application, this is described in detail below with reference to the accompanying drawings and specific implementation methods. Although the embodiments of the present application provide the method operation instruction steps shown in the following embodiments or drawings, more or fewer operation instruction steps may be included in the method based on conventional or no creative labor. In the steps where there is no necessary causal relationship logically, the execution order of these steps is not limited to the execution order provided by the embodiments of the present application. The method may be executed in the order of the methods shown in the embodiments or drawings or in parallel during the actual processing process or when the device is executed.

[0045] 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 permission, and the acquisition or use of the data complies with the provisions of relevant laws and regulations.

[0046] Please refer to Figure 2 , Figure 2 The flowchart of the method for processing MRI-synthesized CT images of sacroiliac joints of axial spondyloarthritis based on a deep learning algorithm provided in one embodiment of the present application is shown. Figure 2 As shown, the method includes:

[0047] Step 201 : Acquire magnetic resonance sequence data obtained by performing magnetic resonance imaging on the sacroiliac joint of a target user.

[0048] It should be noted that the target users are users who are to undergo axSpA diagnosis and evaluation of the sacroiliac joint.

[0049] Specifically, magnetic resonance imaging of the target user's sacroiliac joint is performed, and by setting the parameters TR = 5.2ms, TE = 2.5ms, and a layer thickness of 0.8mm, a three-dimensional T1-weighted multi-gradient echo sequence is obtained. By setting the parameters TR 3000ms, TE = 80ms, a T2-weighted sequence is obtained, and a water-fat separation sequence of the target user's sacroiliac joint is obtained.

[0050] The three-dimensional T1-weighted multi-gradient echo sequence in magnetic resonance imaging (MRI) is abbreviated as 3D-T1-MGE. 3D refers to three-dimensional volumetric imaging, which achieves arbitrary plane reconstruction (e.g., sagittal, coronal, and axial) by continuously acquiring isotropic voxels, avoiding the inter-slice gaps of traditional 2D stratification. T1 weighting emphasizes T1 relaxation differences in tissues and displays anatomical structures by adjusting repetition time (TR) and echo time (TE). Multi-gradient echo (MGE) sequences utilize multiple gradient echoes to acquire signals at different TE values ​​after a single radiofrequency excitation, improving scanning efficiency and supporting quantitative analysis, providing the data foundation for subsequent synthesis of synthetic CT with Hounsfield Hnit (HU) values.

[0051] In a preferred embodiment, before performing magnetic resonance imaging on the target user, an age correction value of the target user is further identified, and the inversion angle of the three-dimensional T1-weighted multi-gradient echo sequence is adjusted based on the age correction value.

[0052] In other words, the reversal 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, the target user is an adult and no adjustment is required. The corresponding age correction value is 0. If the target user's age is less than the preset age, the target user is a child and adjustment is required. The age correction value is then further obtained and the reversal angle of the 3D T1-weighted gradient echo sequence is adjusted based on the age correction value.

[0053] Optionally, the age correction amount may be a fixed value, such as 5°, or a value linearly related to age, which is not specifically limited in this application.

[0054] For example, the age correction amount is a fixed value of 5°. 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°.

[0055] Therefore, the present application adjusts the inversion angle of the three-dimensional T1-weighted gradient echo sequence to reduce the specific absorption ratio (SAR) to meet the safety requirements of children's scanning.

[0056] In some embodiments, after acquiring the magnetic resonance sequence data, a rigid registration algorithm based on mutual information is further used to align the three-dimensional T1-weighted multi-gradient echo sequence, T2-weighted sequence and water-fat separation sequence with the reference CT, with a registration error of less than 0.5 mm, so as to facilitate high-precision feature extraction based on the multi-sequence magnetic resonance sequence data.

[0057] Preferably, the aligned magnetic resonance sequence data is further processed using a Nyul histogram matching algorithm to eliminate signal differences among multi-center magnetic resonance devices.

[0058] In step 202 , the magnetic resonance sequence data is input into a deep learning algorithm model trained using federated learning to obtain the cortical bone features and soft tissue features of the sacroiliac joint.

[0059] It should be noted that the deep learning algorithm model is a feature analysis model trained using a federated learning framework. Specifically, the federated learning framework allows each medical institution to train the model independently using local data, sharing only encrypted gradient parameters. This ensures the generalization of model training while addressing privacy issues across multiple medical centers.

[0060] In some embodiments, the deep learning algorithm model is a hybrid model of an adversarial network and a U-Net. That is, the backbone network of the deep learning algorithm model adopts a U-Net architecture, which fuses multi-scale features through nested dense skip links to preserve the local details and global anatomical structure of the sacroiliac joint. Preferably, a residual block is introduced between the encoder and decoder to alleviate the gradient vanishing problem while strengthening the low-frequency feature transmission of the cortical bone signal, thereby improving the accuracy of the synthesized CT image in the post-synthesis.

[0061] In one feasible embodiment, the deep learning algorithm model includes an anatomical-specific attention subunit, wherein the anatomical-specific attention subunit includes a spatial-channel dual-path attention mechanism.

[0062] It should be noted that the spatial-channel dual-path attention mechanism includes a spatial path and a channel path. The spatial path is used to introduce deformable convolution to capture the irregular edges of the sacroiliac joint and dynamically weight key anatomical regions using the spatial attention map. The channel path uses channel attention to enhance the signal response characteristics of the cortical bone and sclerotic areas. It should be understood that the outputs of the two paths are fused through the adaptive fusion module.

[0063] Specifically, the multi-scale features after multimodal feature fusion of the three-dimensional T1-weighted multi-gradient echo sequence, T2-weighted sequence and water-fat separation sequence can be input into the spatial path and channel path respectively, so as to extract the deformation feature y through the deformable convolution in the spatial path.def , generate the spatial attention map A s , to generate channel weights A through channel attention in the channel path c Finally, the output results of the two paths are weighted fused to obtain the enhanced fusion feature y out .

[0064] Firstly, the sacroiliac joint is complex in structure. Normally, it consists of an uneven, ear-like surface covered with fibrocartilage and hyaline cartilage. During inflammation or degeneration, the cartilage is detached, exposing the cortical bone and magnifying this natural irregularity. Ankylosing spondylitis can cause sacroiliac joint inflammation, leading to subchondral bone erosion, followed by fibrosis and osteophyte formation. The joint surface gradually hardens, but the surface remains uneven.

[0065] Based on this, this application uses deformable convolution to dynamically adjust sampling points in the spatial path, fit the geometric deformation of the joint surface, and improve the expression of key anatomical features in MRI data, such as the articular surface of the sacroiliac joint, bone erosion edges, sclerosis areas and ankylosis changes.

[0066] Specifically, this application uses deformable convolution to introduce a dynamic offset Δp:

[0067]

[0068] Among them, y def (p) is the deformable convolution at the current position p of the feature map to extract the deformation feature, p k is the predefined convolution kernel offset, Δp k (p) is the dynamic offset of the kth sampling point.

[0069] It should be understood that the offset Δp is entirely determined by the content of the input feature map, rather than being predefined. Therefore, different offset patterns will be generated in different position areas in the multi-scale feature map generated based on sacroiliac joint MRI data. For example, the Δp of the hardened area is close to 0, maintaining the standard convolution behavior, indicating that the edge of the area is regular, and the Δp of the joint surface edge is significantly non-zero, so that the sampling point is aligned with the joint anatomical structure, indicating that the edge of the area is irregular.

[0070] Furthermore, by introducing a dynamic offset Δp using deformable convolution, the sampling points at the sacroiliac joint surface can be concentrated towards the protruding part of the edge of the joint surface, avoiding the problem of difficult alignment in traditional convolution. At the same time, the sampling points can be moved deeper into the recessed area in the bone erosion area, avoiding the problem that small recesses are easily ignored by smooth convolution.

[0071] Therefore, this application introduces a dynamic offset Δp in the spatial path through deformable convolution, so that the convolution operation can adapt to the geometric deformation of the anatomical structure of the sacroiliac joint, thereby accurately capturing irregular structures such as the sacroiliac joint surface and bone erosion, improving edge alignment capabilities, and providing more accurate image data support for post-synthesis of CT images, thereby compensating for the defect of insensitivity of MRI data itself.

[0072] Furthermore, a spatial attention map is generated based on the deformation features in the spatial path. The specific address can be expressed as follows:

[0073] A s =σ(Conv(ReLU(Conv(y def ))))

[0074] Among them, A s is the spatial weight attention map corresponding to the deformation feature, Conv is the convolution layer, ReLU is the activation function, and σ is the normalized weight.

[0075] On the other hand, this application utilizes the SE-Net channel attention mechanism to dynamically enhance the high-frequency feature channels related to the continuity of the bone cortex.

[0076] Specifically, the feature channel weights related to the target area are enhanced through the dependency between channels. For example, the following formula can be used:

[0077] A c =σ(W u ·δ(W d ·GAP(y channel )))

[0078] Among them, A c is the channel weight, GAP(y channel ) is the channel descriptor obtained based on global average pooling and compressed spatial dimension, W d and W u is the fully connected layer for dimensionality reduction and dimensionality increase, δ is the activation function, and σ is the normalized weight.

[0079] Furthermore, a gating mechanism is used to weightedly fuse the dual-path outputs to adaptively fuse the attention information of the spatial path and the channel path, giving priority to improving the reconstruction weights of key areas such as the sacroiliac joint surface and sclerosis area.

[0080] For example, the following formula may be used:

[0081] y out =α·(A s ⊙y def )+(1-α)·(A c ⊙y channel )

[0082] Among them, y out is the output fusion feature, α is the contribution coefficient of dynamic balance space and channel attention, y def Extract deformation features for deformable convolution, y channel is the channel feature, A s is the spatial attention weight, A c is the channel attention weight.

[0083] Therefore, the present application can amplify the details of key anatomical parts and enhance the channel response of specific pathological features through the anatomical-specific attention sub-unit composed of the spatial-channel dual-path attention mechanism.

[0084] Preferably, after obtaining the fusion features, the present application further classifies the fusion features to obtain the cortical bone features and soft tissue features of the sacroiliac joint, so as to synthesize a CT image based on the cortical bone features and soft tissue features.

[0085] Step 203 : synthesizing a synthetic CT image corresponding to the target user's sacroiliac joint based on the bone cortex features and the soft tissue features.

[0086] That is to say, after using the magnetic resonance sequence data for feature analysis to obtain more accurate cortical bone features and soft tissue features, the cortical bone features and soft tissue features are further used to synthesize CT images to meet the diagnostic needs of sacroiliac joint axSpA.

[0087] In some embodiments, an anatomical loss function is provided to the deep learning algorithm model during its training phase, so that the CT images generated by the trained deep learning algorithm model strictly conform to the anatomical features of the sacroiliac joint. The anatomical prior constraints include at least a cortical bone continuity loss constraint, a joint space distance loss constraint, and a bone erosion area resistance loss constraint.

[0088] It should be understood that the anatomical loss constraint can ensure that the generated CT images strictly conform to the anatomical structural characteristics of the sacroiliac joint, including cortical bone continuity, joint space distance, and bone erosion morphology, thereby effectively improving the usability of the synthesized CT images in clinical diagnosis.

[0089] For example, the cortical bone continuity loss constraint may be formulated as follows:

[0090]

[0091] in, is the bone cortex continuity loss constraint function, Dice is the similarity coefficient, which is used to calculate the overlap between the real CT and the synthetic CT in the bone cortex, M real To use real CT in training, Mfake is the synthetic CT in training, λ g is the weight coefficient of gradient difference loss, is a gradient operator used to extract the edge gradient of the bone cortex.

[0092] Therefore, the cortical bone continuity loss constraint can be used to ensure that the cortical bone region of the synthetic CT is spatially aligned with the real CT, while the gradient difference is used to enhance the continuity of the cortical bone edge to avoid generating broken or blurred edges.

[0093] For example, the joint gap distance loss constraint may be formulated as follows:

[0094]

[0095] Hausdorff(A, B)=max(sup infd(a, b), sup infd(a, b))

[0096] in, is the joint gap distance loss constraint function, d(a, b) is the Euclidean distance between points a and b, sup is the upper bound, inf is the lower bound, S real is the real CT joint space segmentation, S fake Joint space segmentation for synthetic CT.

[0097] Therefore, the joint gap distance loss constraint can be used to constrain the anatomical rationality of the joint gap and ensure that the width and shape of the joint gap in the synthetic CT image conform to the real anatomical structure.

[0098] For example, the bone erosion area resistance loss constraint can be expressed as follows:

[0099]

[0100] in, is the bone erosion area loss constraint function, D patch is a local discriminator used to perform true synthesis discrimination on the bone erosion region (ROI), G(x) ROI is the local image block of the bone erosion area in the synthetic CT image, y ROI It is the local image block of the bone erosion area in the real CT image. The ROI is determined by manual calibration during the training process.

[0101] Therefore, by using the loss constraint against the bone erosion area, we can enhance the authenticity of local texture by targeting the surrounding features of the bone erosion area, such as irregular bone destruction and edge sharpness, and avoid ignoring local pathological details when the global image quality meets the standard during the synthetic CT process.

[0102] In a preferred embodiment, during the deep learning algorithm model training stage, the distribution of the bone cortical features and soft tissue features in the synthetic CT image is made closer to the distribution of the bone cortical features and soft tissue features in the real CT image.

[0103] Specifically, an adversarial network is used for adversarial training during the training process, so that the deep feature distribution of the synthetic CT and the real CT tends to be consistent.

[0104] In a preferred embodiment, during the process of synthesizing the synthetic CT image, HU values ​​are also mapped in the synthetic CT image based on soft tissue features.

[0105] It should be noted that the HU value is the core physical property of CT images and directly reflects the difference in tissue density. In the task of synthesizing synthetic CT, ensuring that the synthesized CT conforms to the HU calibration range of the real CT is one of the indicators for judging whether the synthetic CT synthesis is successful or not.

[0106] It should also be noted that traditional generative adversarial networks typically use the Tanh activation function to constrain the output to the range [-1, 1]. However, the HU value range of real CT images is relatively wide. For example, the HU value corresponding to air is -1000HU, the HU value corresponding to water is 0HU, the HU value corresponding to fat is -70 to -90HU, and the HU value corresponding to cortical bone is +1000HU. Based on this, to ensure the reliability of the HU values ​​of the synthesized synthetic CT images, this application uses a linear activation function, allowing the generator to directly output an unrestricted range of values, and then adapts the data to the physical range of the real HU values ​​through data normalization and denormalization steps.

[0107] Specifically, in the training phase, the HU value of the real CT is first standardized, and then the last layer of the training generator uses linear activation to directly predict the standardized HU value. In the application phase, the output of the synthetic CT generator is denormalized to restore the HU value corresponding to the synthetic CT, thereby ensuring that the HU value corresponding to the synthesized synthetic CT is consistent with the HU value of the real CT.

[0108] In a preferred embodiment, based on mapping the HU value when synthesizing the synthetic CT image, the bone erosion volume and maximum depth in the synthetic CT image may be further determined according to the mapped HU value in the synthetic CT image.

[0109] Specifically, region segmentation is performed based on the HU value in the synthetic CT image according to a preset HU value range, and the bone erosion volume and maximum depth are calculated based on the image region that meets the bone erosion area.

[0110] Preferably, the HU value corresponding to the bone erosion area is greater than 400, that is, the area with HU greater than 400 in the synthetic CT image is determined as the synthetic bone erosion area, and the range of pixels occupied by the bone erosion area is calculated, and then the volume of the bone erosion area is determined. In this way, the calculation error is limited to ±0.15cm on the basis of being able to automatically calculate the bone erosion area. 3 Inside.

[0111] Therefore, the embodiment of the present application provides a method for processing MRI-synthesized CT images of the sacroiliac joint of axial spondyloarthritis based on a deep learning algorithm. The method can use the magnetic resonance sequence data of the sacroiliac joint of the target user to obtain the cortical bone characteristics and soft tissue characteristics of the sacroiliac joint, and then use the cortical bone characteristics and soft tissue characteristics to synthesize the synthetic CT image corresponding to the sacroiliac joint of the target user, thereby realizing radiation-free high-precision CT image acquisition, which can not only meet the diagnostic needs of the sacroiliac joint of axial spinal joints, but also reduce the impact of radiation on the target user.

[0112] It should be noted that although the operations of the present method are described in a particular order in the drawings, this does not require or imply that the operations must be performed in this particular order, or that all illustrated operations must be performed to achieve desirable results.

[0113] Figure 3 A structural schematic diagram of a sacroiliac joint MRI-to-CT image processing system for axial spondyloarthritis based on a deep learning algorithm provided in one embodiment of the present application is shown.

[0114] like Figure 3 As shown, the embodiment of the present application proposes a deep learning algorithm-based sacroiliac joint MRI synthesis CT image processing system 10 for axial spondyloarthritis, including:

[0115] A data acquisition unit 11 is used to acquire magnetic resonance sequence data obtained by performing magnetic resonance imaging on the sacroiliac joint of the target user;

[0116] a deep learning model unit 12, configured to input the magnetic resonance sequence data into a deep learning algorithm model trained using federated learning to obtain cortical bone features and soft tissue features of the sacroiliac joint;

[0117] The image synthesis unit 13 is configured to synthesize a synthetic CT image corresponding to the sacroiliac joint of the target user based on the bone cortex feature and the soft tissue feature.

[0118] In some embodiments, the magnetic resonance sequence data includes a three-dimensional T1-weighted multi-gradient echo sequence, a T2-weighted sequence, and a water-fat separation sequence;

[0119] The setting parameters of the three-dimensional T1-weighted multi-gradient echo sequence include TR=5.2ms, TE=2.5ms, and a slice thickness of 0.8mm; the setting parameters of the T2-weighted sequence are TR3000ms and TE=80ms.

[0120] In some embodiments, the data acquisition unit 11 is further configured to:

[0121] An age correction value of a target user is identified, and a flip angle of the three-dimensional T1-weighted multi-gradient echo sequence is adjusted based on the age correction value.

[0122] In some embodiments, as Figure 4 As shown, the deep learning algorithm model unit 12 includes an anatomical specific attention subunit 121,

[0123] The anatomical-specific attention subunit 121 includes a spatial-channel dual-path attention mechanism.

[0124] In some embodiments, deformable convolution is used to dynamically adjust sampling points in the spatial path, so that the convolution operation adapts to the geometric deformation of the sacroiliac joint anatomical structure.

[0125] In some embodiments, as Figure 4 As shown, the image synthesis unit 13 includes an anatomical prior constraint subunit 131,

[0126] The anatomical prior constraint subunit 131 is used to provide an anatomical loss function for the deep learning algorithm model during the deep learning algorithm model training phase, and the anatomical prior constraints include at least cortical bone continuity loss constraint, joint gap distance loss constraint and bone erosion area resistance loss constraint.

[0127] In some embodiments, as Figure 4 As shown, the image synthesis unit 13 includes a feature alignment subunit 132,

[0128] The feature alignment subunit 132 is used to make the distribution of the bone cortex features and soft tissue features in the synthetic CT image closer to the distribution of the bone cortex features and soft tissue features in the real CT image during the deep learning algorithm model training stage.

[0129] In some embodiments, as Figure 4 As shown, the image synthesis unit 13 is further used for:

[0130] HU values ​​are mapped in the synthetic CT image based on the soft tissue features.

[0131] In some embodiments, as Figure 4 As shown, it also includes quantitative analysis 14,

[0132] The quantitative analysis unit 14 is configured to determine the bone erosion volume and maximum depth in the synthetic CT image according to the HU value mapped in the synthetic CT image.

[0133] It should be understood that the modules or modules described in the axial spondyloarthritis sacroiliac joint MRI synthesis CT image processing system 10 based on deep learning algorithm are the same as those in the reference Figure 2 The various steps in the described method correspond to each other. Therefore, the operations and features described above for the method are also applicable to the axial spondyloarthritis sacroiliac joint MRI synthetic CT image processing system 10 based on deep learning algorithm and the modules contained therein, and will not be repeated here. The axial spondyloarthritis sacroiliac joint MRI synthetic CT image processing system 10 based on deep learning algorithm can be pre-implemented in the browser or other security applications of the electronic device, and can also be loaded into the browser or its security application of the electronic device by downloading or other means. The corresponding modules in the axial spondyloarthritis sacroiliac joint MRI synthetic CT image processing system 10 based on deep learning algorithm can cooperate with the modules in the electronic device to implement the solution of the embodiment of the present application.

[0134] The several modules or units mentioned in the detailed description above are not necessarily divided into one module or unit. In fact, according to the embodiments of the present 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 into multiple modules or units to be embodied.

[0135] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operating instructions of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the aforementioned module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than the order marked in the accompanying drawings. For example, the boxes represented by two connections can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operating instruction, or can be implemented using a combination of dedicated hardware and computer instructions.

[0136] As another aspect, the present application further provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiment, or may exist independently without being incorporated into the electronic device. The computer-readable storage medium stores one or more programs, which, when used by one or more processors, execute the method for processing MRI-synthesized CT images of sacroiliac joints in axial spondyloarthritis based on a deep learning algorithm described in the present application.

[0137] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the aforementioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A deep learning algorithm-based sacroiliac joint MRI synthesis CT image processing system for axial spondyloarthritis, characterized by: include: a data acquisition unit, configured to acquire magnetic resonance sequence data obtained by performing magnetic resonance imaging on the sacroiliac joint of a target user; a deep learning model unit, configured to input the magnetic resonance sequence data into a deep learning algorithm model trained using federated learning to obtain cortical bone features and soft tissue features of the sacroiliac joint; An image synthesis unit is used to synthesize a synthetic CT image corresponding to the sacroiliac joint of the target user based on the bone cortex feature and the soft tissue feature.

2. The MRI-to-CT image processing system for sacroiliac joints of axial spondyloarthritis based on a deep learning algorithm according to claim 1 is characterized in that: The magnetic resonance sequence data includes a three-dimensional T1-weighted multi-gradient echo sequence, a T2-weighted sequence, and a water-fat separation sequence; The setting parameters of the three-dimensional T1-weighted multi-gradient echo sequence include TR=5.2ms, TE=2.5ms, and a slice thickness of 0.8mm; the setting parameters of the T2-weighted sequence are TR3000ms and TE=80ms.

3. The deep learning algorithm-based sacroiliac joint MRI-CT image processing system for axial spondyloarthritis according to claim 2, characterized in that: The data acquisition unit is further configured to: An age correction value of a target user is identified, and a flip angle of the three-dimensional T1-weighted multi-gradient echo sequence is adjusted based on the age correction value.

4. The MRI-to-CT image processing system for sacroiliac joints of axial spondyloarthritis based on a deep learning algorithm according to claim 1, characterized in that: The deep learning algorithm model unit includes an anatomical specific attention subunit, The anatomically specific attention subunit includes a spatial-channel dual-path attention mechanism.

5. The deep learning algorithm-based sacroiliac joint MRI-CT image processing system for axial spondyloarthritis according to claim 4, characterized in that: Deformable convolution is used in the spatial path to dynamically adjust the sampling points, so that the convolution operation can adapt to the geometric deformation of the sacroiliac joint anatomical structure.

6. The deep learning algorithm-based sacroiliac joint MRI-CT image processing system for axial spondyloarthritis according to claim 1, characterized in that: The image synthesis unit includes an anatomical prior constraint subunit, The anatomical prior constraint subunit is used to provide an anatomical loss function for the deep learning algorithm model during the deep learning algorithm model training phase, and the anatomical prior constraints include at least cortical bone continuity loss constraint, joint gap distance loss constraint and bone erosion area resistance loss constraint.

7. The MRI-synthesized CT image processing system for sacroiliac joints of axial spondyloarthritis based on a deep learning algorithm according to claim 1 or 6, characterized in that: The image synthesis unit includes a feature alignment subunit, The feature alignment subunit is used to make the distribution of bone cortical features and soft tissue features in the synthetic CT image closer to the distribution of bone cortical features and soft tissue features in the real CT image during the deep learning algorithm model training stage.

8. The deep learning algorithm-based sacroiliac joint MRI-CT image processing system for axial spondyloarthritis according to claim 1, characterized in that: The image synthesis unit is further configured to: HU values ​​are mapped in the synthetic CT image based on the soft tissue features.

9. The deep learning algorithm-based sacroiliac joint MRI-CT image processing system for axial spondyloarthritis according to claim 8, characterized in that: It also includes quantitative analysis, The quantitative analysis unit is used to determine the bone erosion volume and maximum depth in the synthetic CT image according to the HU value mapped in the synthetic CT image.

10. A method for processing MRI-CT images of sacroiliac joints in axial spondyloarthritis based on a deep learning algorithm, characterized in that: include: Acquiring magnetic resonance sequence data obtained by performing magnetic resonance imaging on the sacroiliac joint of the target user; Inputting the magnetic resonance sequence data into a deep learning algorithm model trained using federated learning to obtain cortical bone features and soft tissue features of the sacroiliac joint; A synthetic CT image corresponding to the sacroiliac joint of the target user is synthesized based on the cortical bone features and the soft tissue features.

Images