A reconfigurable magnetic resonance image reconstruction system and dynamic control method thereof

Abstract

The application discloses a reconfigurable magnetic resonance image reconstruction system and a dynamic control method thereof. The system comprises an input adapter, a module manager, a pipeline manager, an execution engine, a plurality of processing modules and an output adapter. The system realizes the registration and instantiation of the processing modules through the module manager, analyzes a declarative configuration file to build an execution topology through the pipeline manager, and dynamically schedules a unified data container to flow through each processing module according to the topology through the execution engine, thereby completing a reconstruction process from k-space data to a diagnostic image. The system and the dynamic control method thereof realize flexible recombination and runtime dynamic control of the reconstruction process, solve the problems of rigidity, expansion difficulty and insufficient cross-language compatibility in the prior art, and significantly improve the adaptability, expandability and reconstruction efficiency of the system.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a reconstructable magnetic resonance image reconstruction system and its dynamic control method. Background Technology

[0002] Magnetic Resonance Imaging (MRI) is an important medical diagnostic tool, and the final image quality of it highly depends on the reconstruction process from raw k-space data to image domain data. This process typically involves a series of complex steps, including preprocessing (such as phase rearrangement and zero-filling), image reconstruction (such as Fourier transform, parallel imaging, and compressed sensing), and post-processing (such as uniformity correction and filtering).

[0003] Currently, mainstream magnetic resonance imaging reconstruction systems mainly employ the following two technical solutions in terms of architecture:

[0004] The first type is the fixed-flow pipeline. This type of system tightly integrates each processing step in the reconstruction process into a unified software framework using hard-coding. The order of processing steps, parameters, and algorithms are determined during the software compilation phase, forming a fixed execution chain. This architecture leads to extreme system rigidity. When the processing flow needs to be adjusted for different scan sequences, clinical needs, or research objectives (e.g., adjusting the order of steps, replacing algorithm modules, or modifying parameters), developers must deeply modify the underlying source code of the system, and then recompile and deploy the entire system. This not only brings huge development and maintenance costs but also makes it difficult for the system to quickly adapt to diverse clinical scenarios, severely limiting its flexibility and application scope.

[0005] The second type is script-driven configuration pipelines. To overcome some of the shortcomings of fixed processes, some systems attempt to use scripts (such as Python scripts) to call underlying processing functions, thereby adjusting the process order and parameters at the script level. However, this approach typically lacks strict modular encapsulation; different algorithm modules have different interfaces and data formats, resulting in high coupling between modules. Furthermore, the flexibility at the script level and the lack of unified architecture management for the underlying execution units limit the system's scalability, stability, and performance optimization. Frequent format conversions and copies during data exchange also incur additional performance overhead.

[0006] In summary, existing technologies generally suffer from the following technical defects:

[0007] Rigid processes and poor adaptability: The system cannot dynamically reconstruct the processing pipeline according to different imaging sequences or clinical needs during runtime, resulting in insufficient flexibility in clinical applications.

[0008] High module coupling makes expansion and maintenance difficult: adding or replacing algorithm modules requires modifying the core framework, resulting in high integration costs and hindering technology iteration and algorithm fusion.

[0009] Insufficient cross-language compatibility: Existing frameworks are mostly limited to a single programming language, making it difficult to seamlessly integrate advantageous modules implemented in different languages ​​within the same pipeline, thus limiting the overall performance and development efficiency of the system.

[0010] Data and control are coupled, lacking dynamic control capabilities: the data flow and control flow are not separated, making it difficult to implement advanced control strategies such as conditional execution and dynamic parameter adjustment based on runtime states (such as image quality and system resources).

[0011] Therefore, there is an urgent need in this field for a novel magnetic resonance image reconstruction system architecture that can standardize and decouple processing modules, support dynamic definition and modification of processing flows at runtime through declarative configuration, and possess cross-language compatibility and runtime intelligent control capabilities, thereby significantly improving the system's flexibility, scalability, and intelligence. Summary of the Invention

[0012] To address the technical problems existing in the background art, this invention proposes a reconfigurable magnetic resonance image reconstruction system and its dynamic control method.

[0013] In a first aspect, the present invention proposes a reconstructable magnetic resonance image reconstruction system for reconstructing acquired magnetic resonance k-space data into diagnostic images, comprising:

[0014] An input adapter is used to encapsulate the received raw magnetic resonance k-space data and its metadata into a unified data container;

[0015] The module manager is used to maintain the mapping relationship between the types of available processing modules and their creation functions, and supports dynamic loading of plugins; the module manager is also used to obtain and parse external declarative configuration files, and instantiate multiple processing modules according to the module configuration information corresponding to the magnetic resonance image reconstruction process in the external declarative configuration files.

[0016] The pipeline manager is used to obtain and parse external declarative configuration files, and construct the execution topology based on the execution topology information corresponding to the magnetic resonance image reconstruction process in the external declarative configuration files.

[0017] The execution engine is used to obtain a unified data container from the input adapter and, according to the execution topology, sequentially schedule the unified data container to the corresponding processing module to execute the magnetic resonance image reconstruction process.

[0018] Multiple processing modules, managed by a module manager, are used to receive a unified data container from the execution engine, perform assigned image processing operations on the raw magnetic resonance k-space data or intermediate image data converted from it in the unified data container, and update the unified data container.

[0019] The output adapter is used to obtain the final unified data container processed by all processing modules from the execution engine and output the final magnetic resonance reconstructed image.

[0020] Preferably, the unified data container is a data structure specifically designed to carry magnetic resonance imaging data. The unified data container includes a magnetic resonance k-space data array, a magnetic resonance image domain data array, a magnetic resonance metadata dictionary derived from the device acquisition, and a quality control information dictionary generated during the reconstruction process.

[0021] Preferably, the pipeline manager specifically includes:

[0022] The configuration acquisition and parsing unit is used to acquire external declarative configuration files and parse them to extract execution topology information. The execution topology information includes at least the unique identifier of each processing module and the order or dependency of each processing module in the magnetic resonance image reconstruction process.

[0023] The topology construction unit, connected to the configuration acquisition and parsing unit, is used to logically connect the processing module instances created by the module instantiation unit according to the execution topology information, forming a data flow graph corresponding to the magnetic resonance image reconstruction process, i.e., the execution topology.

[0024] Preferably, the module manager specifically includes:

[0025] A module information storage unit is used to store multiple registration records. Each registration record maps a unique module identifier to the corresponding module type and module creation function. The module configuration information includes at least the identifier, type, and initialization parameters of the processing module used for magnetic resonance image processing.

[0026] The module registration interface unit is connected to the module information storage unit. It is used to receive module registration requests from static link libraries or dynamic plugins when the system starts up or during operation, and add the module identifier, module type and module creation function contained in the module registration request as a new registration record to the module information storage unit.

[0027] The module instantiation unit, connected to the configuration acquisition and parsing unit, is used to query the corresponding module creation function from the module manager based on the module identifier and type obtained from the module configuration information, so as to dynamically create instances of the multiple processing modules;

[0028] When instantiating a processing module, the module manager dynamically creates a processing module instance by querying and obtaining the corresponding module creation function from the module information storage unit using the module identifier parsed from the external declarative configuration file.

[0029] Preferably, the execution engine is connected to the input adapter, the pipeline manager, and the module manager, respectively; the execution engine includes:

[0030] A data interface unit, connected to the input adapter, is used to receive a unified data container from the input adapter;

[0031] The topology interface unit, connected to the pipeline manager, is used to receive the execution topology from the pipeline manager;

[0032] The module interface unit connects to the module manager and is used to obtain instance references, status information, or metadata of instantiated processing modules from the module manager.

[0033] The process scheduler is connected to the data interface unit, topology interface unit, and module interface unit, respectively. It is used to obtain a unified data container from the data interface unit, obtain the execution topology from the topology interface unit, and obtain instance references of the processing modules from the module interface unit. Then, it schedules the unified data container to the corresponding processing module in sequence according to the execution topology.

[0034] Preferably, the processing module includes:

[0035] A configuration unit is used to receive and store initialization parameters from an external declarative configuration file when a module is instantiated;

[0036] The execution unit, connected to the configuration unit, is used to receive a unified data container from the execution engine, execute a preset image processing algorithm on the k-space data, image domain data, or metadata in the unified data container by calling the parameters stored in the configuration unit, and update the corresponding data segments in the unified data container.

[0037] Preferably, the processing module includes at least one of the following types:

[0038] The preprocessing module is used to perform phase rearrangement, zero-filling, or undersampling recovery operations on the k-space data array in the unified data container;

[0039] The image reconstruction module is used to perform Fourier transform, parallel imaging, or compressed sensing reconstruction algorithms on k-space data arrays or image domain data arrays in a unified data container.

[0040] The post-processing module is used to perform amplitude calculation, uniformity correction, or image filtering operations on the image domain data array in the unified data container.

[0041] Preferably, the execution topology is a directed acyclic graph structure, which defines the conversion path from k-space data to image domain data and the dependencies between processing modules in the magnetic resonance image reconstruction process.

[0042] Preferably, it further includes: a control plane, which is communicatively connected to the execution engine and processing modules, for receiving external control commands during the execution engine's operation according to the executable topology, and dynamically enabling, disabling, or updating any processing module based on the external control commands.

[0043] Secondly, the present invention proposes a dynamic control method for a reconfigurable magnetic resonance image reconstruction system, comprising the following steps:

[0044] The raw magnetic resonance k-space data and its metadata are encapsulated into a unified data container through an input adapter;

[0045] The module manager maintains the mapping relationship of processing modules and dynamically loads plugins, while instantiating multiple processing modules based on the module configuration information in the external declarative configuration file.

[0046] The pipeline manager constructs the execution topology based on the execution topology information in the external declarative configuration file;

[0047] The execution engine schedules the unified data containers to the corresponding processing modules in sequence according to the execution topology.

[0048] Each processing module performs the assigned image processing operation on the received unified data container.

[0049] The final magnetic resonance reconstructed image, processed by all processing modules, is output via the output adapter.

[0050] Thirdly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the dynamic control method of the magnetic resonance image reconstruction system as described above.

[0051] Fourthly, the present invention provides a computer device comprising a memory and a processor connected to the memory; the memory is used to store a computer program; the processor is used to run the computer program stored in the memory to execute the dynamic control method of the magnetic resonance image reconstruction system as described above.

[0052] This invention presents a reconfigurable magnetic resonance imaging (MRI) image reconstruction system and its dynamic control method. Through a modular and configurable pipeline architecture, it achieves flexible reconfiguration and dynamic control of the MRI image reconstruction process, significantly improving the system's adaptability, scalability, and intelligence. Modular design reduces algorithm expansion and maintenance costs, while the separation of data flow and control flow provides a foundation for real-time strategy adjustment and multi-algorithm fusion. This results in overall synergistic optimization of reconstruction efficiency and image quality, providing strong support for the application of MRI technology. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the system architecture of a reconfigurable magnetic resonance image reconstruction system proposed in this invention.

[0054] Figure 2 This is a flowchart illustrating the dynamic control method for a reconfigurable magnetic resonance image reconstruction system proposed in this invention.

[0055] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0056] Reference Figure 1 and Figure 2 This invention proposes a reconstructable magnetic resonance image reconstruction system for reconstructing diagnostic images from acquired magnetic resonance k-space data, comprising:

[0057] The input adapter is used to encapsulate the received raw magnetic resonance k-space data and its metadata into a unified data container.

[0058] In this embodiment, the unified data container is a data structure specifically designed to carry magnetic resonance imaging data. The unified data container includes a magnetic resonance k-space data array, a magnetic resonance image domain data array, a magnetic resonance metadata dictionary derived from the device acquisition, and a quality control information dictionary generated during the reconstruction process.

[0059] The module manager is used to maintain the mapping relationship between the types of available processing modules and their creation functions, and supports dynamic loading of plugins. The module manager is also used to obtain and parse external declarative configuration files, and instantiate multiple processing modules according to the module configuration information corresponding to the magnetic resonance image reconstruction process in the external declarative configuration files.

[0060] In this embodiment, the module manager specifically includes:

[0061] The module information storage unit is used to store multiple registration records. Each registration record maps a unique module identifier to the corresponding module type and module creation function. The module configuration information includes at least the identifier, type, and initialization parameters of the processing module used for magnetic resonance image processing.

[0062] The module registration interface unit is connected to the module information storage unit. It is used to receive module registration requests from static link libraries or dynamic plugins when the system starts up or during operation, and add the module identifier, module type and module creation function contained in the module registration request as a new registration record to the module information storage unit.

[0063] The module instantiation unit, connected to the configuration acquisition and parsing unit, is used to query the corresponding module creation function from the module manager based on the module identifier and type obtained from the module configuration information, so as to dynamically create instances of multiple processing modules;

[0064] When instantiating a processing module, the module manager retrieves the corresponding module creation function from the module information storage unit by parsing the module identifier from the external declarative configuration file, thereby dynamically creating a processing module instance.

[0065] The pipeline manager is used to obtain and parse external declarative configuration files, and construct the execution topology based on the execution topology information corresponding to the magnetic resonance image reconstruction process in the external declarative configuration files.

[0066] Specifically, the execution topology is a directed acyclic graph structure, which defines the conversion path from k-space data to image domain data and the dependencies between processing modules in the magnetic resonance image reconstruction process.

[0067] In this embodiment, the pipeline manager specifically includes:

[0068] The configuration acquisition and parsing unit is used to acquire external declarative configuration files and parse them to extract execution topology information. The execution topology information includes at least the unique identifier of each processing module and the order or dependency of each processing module in the magnetic resonance image reconstruction process.

[0069] The topology construction unit, connected to the configuration acquisition and parsing unit, is used to logically connect the processing module instances created by the module instantiation unit according to the execution topology information, forming a data flow graph corresponding to the magnetic resonance image reconstruction process, i.e., the execution topology.

[0070] The execution engine is used to obtain a unified data container from the input adapter and, according to the execution topology, sequentially schedule the unified data container to the corresponding processing module to execute the magnetic resonance image reconstruction process.

[0071] In this embodiment, the execution engine is connected to the input adapter, the pipeline manager, and the module manager, respectively; the execution engine includes:

[0072] A data interface unit, connected to the input adapter, is used to receive a unified data container from the input adapter;

[0073] The topology interface unit, connected to the pipeline manager, is used to receive the execution topology from the pipeline manager;

[0074] The module interface unit connects to the module manager and is used to obtain instance references, status information, or metadata of instantiated processing modules from the module manager.

[0075] The process scheduler is connected to the data interface unit, topology interface unit, and module interface unit, respectively. It is used to obtain a unified data container from the data interface unit, obtain the execution topology from the topology interface unit, and obtain instance references of the processing modules from the module interface unit. Then, it schedules the unified data container to the corresponding processing module in sequence according to the execution topology.

[0076] Multiple processing modules, managed by a module manager, are used to receive a unified data container from the execution engine, perform assigned image processing operations on the raw magnetic resonance k-space data or intermediate image data converted from it in the unified data container, and update the unified data container.

[0077] In this embodiment, the processing module includes:

[0078] A configuration unit is used to receive and store initialization parameters from an external declarative configuration file when a module is instantiated;

[0079] The execution unit, connected to the configuration unit, is used to receive a unified data container from the execution engine, execute a preset image processing algorithm on the k-space data, image domain data, or metadata in the unified data container by calling the parameters stored in the configuration unit, and update the corresponding data segments in the unified data container.

[0080] Specifically, the processing module includes at least one of the following types:

[0081] The preprocessing module is used to perform phase rearrangement, zero-filling, or undersampling recovery operations on the k-space data array in the unified data container;

[0082] The image reconstruction module is used to perform Fourier transform, parallel imaging, or compressed sensing reconstruction algorithms on k-space data arrays or image domain data arrays in a unified data container.

[0083] The post-processing module is used to perform amplitude calculation, uniformity correction, or image filtering operations on the image domain data array in the unified data container.

[0084] Specifically, the magnetic resonance image processing workflow is decomposed into three main modules: preprocessing, image reconstruction, and postprocessing. Each module can be encapsulated and called as an independent class, providing flexible replacement, expansion, and dynamic configuration capabilities. A data flow mechanism is provided to enable configurable connections between different modules and to support dynamic adjustment of module execution order and parameters at runtime, achieving rapid adaptation to different reconstruction needs.

[0085] The output adapter is used to obtain the final unified data container processed by all processing modules from the execution engine and output the final magnetic resonance reconstructed image.

[0086] In this embodiment, it also includes a control plane, which is communicatively connected to the execution engine and processing modules, and is used to receive external control commands during the execution engine's operation according to the executable topology, and to dynamically enable, disable, or update any processing module based on the external control commands.

[0087] This application provides a complete reconfigurable MRI reconstruction pipeline: through input adapters, module managers, pipeline managers, execution engines, multiple processing modules, and output adapters, it enables free combination of modules, runtime control, cross-language extension, and high-performance execution.

[0088] In this embodiment, the architecture is generally organized using a "large module + key step refinement" approach. This maintains the clarity of the overall architecture while allowing for optional expansion of key steps (such as phase rearrangement, undersampled restoration, zero-filling, Fourier transform, and layer rearrangement) to broaden the protection scope. Each module uses a unified input / output interface definition, supporting the transfer of multi-dimensional data (k-space data, image domain data, masks, weights, etc.) to ensure interoperability and scalability between modules. The architecture is not limited to specific reconstruction methods such as GRAPPA and SENSE; it supports the integration of various reconstruction algorithms and image post-processing methods, and allows for combined invocation within the framework. Users can flexibly customize the pipeline through configuration files or external control interfaces. Modules are pluggable, facilitating rapid deployment and upgrades in different scenarios.

[0089] Example 1: Basic Modular Production Line

[0090] This embodiment's pipeline includes: a pre-processing module: receiving raw magnetic resonance k-space data and its metadata, performing noise removal, phase rearrangement, and data normalization; an image reconstruction module: performing Fast Fourier Transform (FFT) on the pre-processed data, and performing undersampling restoration or zero-padding operations to obtain a preliminary reconstructed image; and a post-processing module: performing layer rearrangement, filtering, artifact removal, etc., on the reconstructed image, and outputting the final image. The modules are connected through a unified data interface, and both input and output use normalized tensor format.

[0091] The pipeline configuration can be set via an external declarative configuration file, allowing users to choose whether to enable certain steps (such as whether to perform zero-padding). This embodiment emphasizes the basic modular division and standardized interface of the invention, ensuring the integrity of the framework.

[0092] This embodiment demonstrates the most basic modular components and workflow of the system of this application, reflecting its core idea of ​​achieving flexibility through configuration. When the system starts running, the input adapter first reads the raw magnetic resonance k-space data and its metadata from the specified path and encapsulates it into an initial unified data container. This container serves as the sole data carrier, passed throughout the pipeline.

[0093] Subsequently, the pipeline manager reads and parses an external declarative configuration file, which not only defines the input and output paths, but more importantly, it specifies which processing modules and their parameters need to be enabled for this rebuild task through a series of key-value pairs, and instantiates the corresponding processing modules through the module manager.

[0094] In this embodiment, the system constructs a linear processing sequence: preprocessing module → image reconstruction module → postprocessing module.

[0095] The execution engine drives the unified data container to flow through each module in sequence:

[0096] Preprocessing module: Receives a unified data container and performs operations such as phase rearrangement and data normalization on the k-space data within it.

[0097] Image reconstruction module: Receives preprocessed data and performs Fast Fourier Transform (FFT) for image reconstruction. Zero-padding is skipped due to a zero-padding skip setting in the configuration.

[0098] Post-processing module: Performs layer rearrangement on the reconstructed image data and uses a Tukey filter for filtering to improve image quality.

[0099] Finally, the output adapter exports the processed image data from the container as a DICOM format file for clinical diagnostic use.

[0100] This embodiment achieves basic configurability by flexibly controlling the activation and parameters of the externally configured module, avoiding the hassle of writing hard code for different processes.

[0101] Example 2: Dynamically Reconfigurable Pipeline

[0102] This embodiment, based on Embodiment 1, further introduces a dynamic scheduling and reconfigurable mechanism:

[0103] Each module is encapsulated as a class, supporting dynamic adjustment of the execution order via configuration files or runtime control interfaces. For example, in some applications, zero-padding can be performed before reconstruction, while in others, it can be performed after reconstruction. The system can add or remove modules at runtime, such as adding additional phase correction or deep learning enhancement modules. Pipeline execution uses a dataflow approach, with modules exchanging data only through input / output ports, avoiding hard-coded dependencies. This embodiment emphasizes dynamic reconfigurability, ensuring that this application can adapt to different scenarios and algorithms.

[0104] In this embodiment, the core of the system lies in a flow configuration that defines the execution order of modules. The pipeline manager parses JSON configurations in the following form:

[0105] {"pipeline": [ {"id": "preprocess", "steps": ["phase_reorder", "normalize"]},{"id":"reconstruction","steps":["fft","zero_filling"]},{"id": "postprocess", "steps": ["filter", "slice_reorder"]}]}.

[0106] This system allows users to change the entire processing logic with simple modifications. For example, if a user finds that zero-padding works better for a certain type of image after reconstruction, they can simply modify the configuration as follows:

[0107] The system will automatically execute in the new order the next time it runs, for example, moving zero_filling to postprocess.

[0108] Furthermore, this embodiment supports hot-swappable expansion. If a user needs to introduce a deep learning-based noise reduction module, they only need to add a module entry to the pipeline array in the configuration file.

[0109] During system runtime, the module manager dynamically loads plugin libraries containing the cnn_denoising module implementation and automatically inserts them into the appropriate locations in the pipeline, without requiring recompilation or deployment of the main system. This greatly improves the system's agility in adapting to new algorithms and scenarios.

[0110] Example 3: Multi-algorithm fusion and expansion pipeline

[0111] The pipeline structure in this embodiment is similar to that in Embodiments 1 and 2, but it supports multiple algorithms in the image reconstruction module: GRAPPA algorithm for parallel imaging reconstruction can be selected; SENSE algorithm for sensitivity-weighted reconstruction can be selected; and traditional FFT reconstruction or deep learning-based reconstruction methods can be selected. The system supports cascading multiple algorithm modules within the same pipeline, for example, performing GRAPPA processing first, followed by artifact removal using a deep learning module. Users can select these modules through configuration files, and different algorithm modules share a unified interface to ensure interoperability. The post-processing module can further incorporate medical task-related functions (such as segmentation and registration). This embodiment highlights the algorithm independence and scalability of this application, protecting not only traditional reconstruction methods but also covering the integration of future new algorithms. Algorithm options: FFT, SENSE, GRAPPA, deep_learning.

[0112] When deep learning is enabled, the system loads the specified model file. Users can specify multiple algorithm combinations, for example: perform GRAPPA first, then perform deep learning enhancements; perform SENSE first, then perform artifact removal.

[0113] This embodiment focuses on demonstrating the algorithm independence and multi-algorithm fusion capability of the architecture of this invention, while providing a rich selection of algorithms in key stages such as image reconstruction. This embodiment proves that the system of this application can not only accommodate traditional reconstruction methods (such as FFT, GRAPPA, SENSE), but also smoothly integrate new algorithms based on cutting-edge technologies such as deep learning, providing a powerful and universal framework for the continuous evolution of magnetic resonance image reconstruction technology.

[0114] Reference Figure 1 and Figure 2 The present invention proposes a dynamic control method for a reconfigurable magnetic resonance image reconstruction system, comprising the following steps:

[0115] The raw magnetic resonance k-space data and its metadata are encapsulated into a unified data container through an input adapter;

[0116] The module manager maintains the mapping relationship of processing modules and dynamically loads plugins, while instantiating multiple processing modules based on the module configuration information in the external declarative configuration file.

[0117] The pipeline manager constructs the execution topology based on the execution topology information in the external declarative configuration file;

[0118] The execution engine schedules the unified data containers to the corresponding processing modules in sequence according to the execution topology.

[0119] Each processing module performs the assigned image processing operation on the received unified data container.

[0120] The final magnetic resonance reconstructed image, processed by all processing modules, is output via the output adapter.

[0121] This application provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed by a processor, they are used to implement the technical solution shown in the above method embodiments. The implementation principle and beneficial effects are similar and will not be repeated here.

[0122] Figure 3 A schematic diagram of the structure of a computer device provided in an embodiment of this application. Please refer to... Figure 3 The computer device 10 may include a memory 11 and a processor 12. Exemplarily, the memory 11 and the processor 12 are interconnected via a bus 13.

[0123] Memory 11 is used to store computer programs;

[0124] The processor 12 is used to run a computer program stored in the memory, so that the computer device 10 performs the method shown in the above method embodiment.

[0125] The computer device provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be described again here.

[0126] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A reconstructable magnetic resonance image reconstruction system for reconstructing acquired magnetic resonance k-space data into diagnostic images, characterized in that, include: An input adapter is used to encapsulate the received raw magnetic resonance k-space data and its metadata into a unified data container; The module manager is used to maintain the mapping relationship between the types of available processing modules and their creation functions, and supports dynamic loading of plugins; the module manager is also used to obtain and parse external declarative configuration files, and instantiate multiple processing modules according to the module configuration information corresponding to the magnetic resonance image reconstruction process in the external declarative configuration files. The pipeline manager is used to obtain and parse external declarative configuration files, and construct the execution topology based on the execution topology information corresponding to the magnetic resonance image reconstruction process in the external declarative configuration files. The execution engine is used to obtain a unified data container from the input adapter and, according to the execution topology, sequentially schedule the unified data container to the corresponding processing module to execute the magnetic resonance image reconstruction process. Multiple processing modules, managed by a module manager, are used to receive a unified data container from the execution engine, perform assigned image processing operations on the raw magnetic resonance k-space data or intermediate image data converted from it in the unified data container, and update the unified data container. The output adapter is used to obtain the final unified data container processed by all processing modules from the execution engine and output the final magnetic resonance reconstructed image.

2. The reconfigurable magnetic resonance image reconstruction system according to claim 1, characterized in that, The unified data container is a data structure specifically designed to carry magnetic resonance imaging data. The unified data container includes a magnetic resonance k-space data array, a magnetic resonance image domain data array, a magnetic resonance metadata dictionary derived from the device acquisition, and a quality control information dictionary generated during the reconstruction process.

3. The reconfigurable magnetic resonance image reconstruction system according to claim 1, characterized in that, The pipeline manager specifically includes: The configuration acquisition and parsing unit is used to acquire external declarative configuration files and parse them to extract execution topology information. The execution topology information includes at least the unique identifier of each processing module and the order or dependency of each processing module in the magnetic resonance image reconstruction process. The topology construction unit, connected to the configuration acquisition and parsing unit, is used to logically connect the processing module instances created by the module instantiation unit according to the execution topology information, forming a data flow graph corresponding to the magnetic resonance image reconstruction process, i.e., the execution topology.

4. The reconfigurable magnetic resonance image reconstruction system according to claim 1, characterized in that, The module manager specifically includes: A module information storage unit is used to store multiple registration records. Each registration record maps a unique module identifier to the corresponding module type and module creation function. The module configuration information includes at least the identifier, type, and initialization parameters of the processing module used for magnetic resonance image processing. The module registration interface unit is connected to the module information storage unit. It is used to receive module registration requests from static link libraries or dynamic plugins when the system starts up or during operation, and add the module identifier, module type and module creation function contained in the module registration request as a new registration record to the module information storage unit. The module instantiation unit, connected to the configuration acquisition and parsing unit, is used to query the corresponding module creation function from the module manager based on the module identifier and type obtained from the module configuration information, so as to dynamically create instances of the multiple processing modules; When instantiating a processing module, the module manager dynamically creates a processing module instance by querying and obtaining the corresponding module creation function from the module information storage unit using the module identifier parsed from the external declarative configuration file.

5. The reconfigurable magnetic resonance image reconstruction system according to claim 1, characterized in that, The execution engine is connected to the input adapter, the pipeline manager, and the module manager, respectively; the execution engine includes: A data interface unit, connected to the input adapter, is used to receive a unified data container from the input adapter; The topology interface unit, connected to the pipeline manager, is used to receive the execution topology from the pipeline manager; The module interface unit connects to the module manager and is used to obtain instance references, status information, or metadata of instantiated processing modules from the module manager. The process scheduler is connected to the data interface unit, topology interface unit, and module interface unit, respectively. It is used to obtain a unified data container from the data interface unit, obtain the execution topology from the topology interface unit, and obtain instance references of the processing modules from the module interface unit. Then, it schedules the unified data container to the corresponding processing module in sequence according to the execution topology.

6. The reconfigurable magnetic resonance image reconstruction system according to claim 1, characterized in that, The processing module includes: A configuration unit is used to receive and store initialization parameters from an external declarative configuration file when a module is instantiated; The execution unit, connected to the configuration unit, is used to receive a unified data container from the execution engine, execute a preset image processing algorithm on the k-space data, image domain data, or metadata in the unified data container by calling the parameters stored in the configuration unit, and update the corresponding data segments in the unified data container.

7. The reconfigurable magnetic resonance image reconstruction system according to claim 1, characterized in that, The processing module includes at least one of the following types: The preprocessing module is used to perform phase rearrangement, zero-filling, or undersampling recovery operations on the k-space data array in the unified data container; The image reconstruction module is used to perform Fourier transform, parallel imaging, or compressed sensing reconstruction algorithms on k-space data arrays or image domain data arrays in a unified data container. The post-processing module is used to perform amplitude calculation, uniformity correction, or image filtering operations on the image domain data array in the unified data container.

8. The reconfigurable magnetic resonance image reconstruction system according to claim 1, characterized in that, The execution topology is a directed acyclic graph structure, which defines the conversion path from k-space data to image domain data and the dependencies between processing modules in the magnetic resonance image reconstruction process.

9. The reconfigurable magnetic resonance image reconstruction system according to claim 1, characterized in that, Also includes: The control plane communicates with the execution engine and processing modules. It is used to receive external control commands during the execution engine's operation according to the executable topology, and to dynamically enable, disable, or update any processing module based on the external control commands.

10. A dynamic control method for a reconfigurable magnetic resonance image reconstruction system, characterized in that, Includes the following steps: The raw magnetic resonance k-space data and its metadata are encapsulated into a unified data container through an input adapter; The module manager maintains the mapping relationship of processing modules and dynamically loads plugins, while instantiating multiple processing modules based on the module configuration information in the external declarative configuration file. The pipeline manager constructs the execution topology based on the execution topology information in the external declarative configuration file; The execution engine schedules the unified data containers to the corresponding processing modules in sequence according to the execution topology. Each processing module performs the assigned image processing operation on the received unified data container. The final magnetic resonance reconstructed image, processed by all processing modules, is output via the output adapter.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the dynamic control method of the magnetic resonance image reconstruction system as described in claim 10.

12. A computer device, characterized in that, The computer device includes a memory and a processor connected to the memory; the memory is used to store computer programs; the processor is used to run the computer programs stored in the memory to execute the dynamic control method of the magnetic resonance image reconstruction system as described in claim 10.

Images