An anterior cruciate ligament reconstruction surgery full-process simulation system

Abstract

The application provides a knee anterior cruciate ligament reconstruction surgery whole-process simulation system, and constructs an end-to-end automatic simulation system composed of a main control program, four-level function execution units, resource monitoring and adaptive scheduling units. The system is based on a task state driving mechanism to organize process execution. The main control unit periodically pulls tasks through an API interface and distributes them to the corresponding units for processing. After the units complete the task, the state is automatically updated, forming a closed-loop control link. Each unit runs as an independent executable program and manages hot plug through a unified configuration file. It supports multi-case parallel scheduling. The resource monitoring and adaptive scheduling unit assesses the use of CPU, memory, network bandwidth, etc. in real time, dynamically adjusts the number of instances and the scheduling rhythm, effectively improves the system throughput efficiency and stability, reduces the dependence on personnel operation experience, and provides efficient and reliable technical support for personalized preoperative planning.

Description

Technical Field

[0001] This invention belongs to the field of biomedical technology, specifically, it relates to a full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint. Background Technology

[0002] The anterior cruciate ligament (ACL) of the knee joint is a core tissue for maintaining knee joint stability and motor control. Once damaged, it usually requires surgical reconstruction. Preoperative simulation, as an important tool for decision support, can assess postoperative biomechanical response based on individualized medical imaging data of patients, thereby optimizing graft path design and implantation location selection, and improving the accuracy of postoperative stability prediction. It plays an increasingly important role in clinical diagnosis and treatment.

[0003] The preoperative simulation process for ACL typically involves multiple stages, including medical image format conversion, anatomical structure segmentation, 3D model reconstruction, finite element modeling, and biomechanical simulation. However, this process generally relies on multiple heterogeneous software systems with fragmented functions, requiring frequent switching between different software platforms and interfaces. There is a lack of unified scheduling mechanisms and data standards between the various stages, often necessitating manual data import and export between multiple tools, and completion by personnel with different professional backgrounds (such as radiologists, 3D modeling engineers, and biomechanical analysts) in stages. This fragmented process and multi-role dependency result in low overall efficiency, operational complexity, poor coordination, and a high dependence on personnel experience, making it prone to human error and data inconsistencies. In scenarios involving batch case processing, the processing efficiency is particularly low, failing to meet the clinical demand for rapid and concurrent preoperative assessment of protocols.

[0004] More importantly, existing technologies only achieve partial automation for certain steps and lack a unified scheduling mechanism and state tracking framework covering the entire process. This makes it difficult to form a closed-loop control from image input to simulation output, and results in a lack of end-to-end consistency, controllability, and engineering usability. Summary of the Invention

[0005] The purpose of this invention is to provide a full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint. The entire process from image input to simulation output does not require manual intervention, realizing true one-click, closed-loop, and resource-efficient preoperative simulation automation, and meeting the actual needs of automated deployment in clinical environments.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] A full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint is proposed, including:

[0008] The front-end submission and initialization unit receives DICOM metadata uploaded from the web client, creates a task, and loads it locally.

[0009] The image preprocessing unit names the working directory according to the patient ID and timestamp, preprocesses the slice images, and exports them in batches to the working directory; it parses DICOM metadata to obtain the keywords required for 3D model reconstruction, writes the keywords into the index file, and stores them in the working directory;

[0010] The deep learning image segmentation unit reads slice image data from the working directory and uses a pre-trained FastSAM network to perform multi-label segmentation of key anatomical structures in the knee joint.

[0011] The 3D model reconstruction unit calls the 3D reconstruction program to reconstruct the anterior cruciate ligament of the knee joint in 3D and export the finite element mesh file in BDF format.

[0012] The finite element simulation unit calls the simulation preprocessing program to load the BDF mesh file output by the 3D reconstruction program as the basic model into the simulation environment. After defining the assembly and loads, it outputs the input file required by the implicit nonlinear solver. The implicit nonlinear solver is started to perform nonlinear finite element analysis on the input file. The analysis results are analyzed, and the key indicators of the ACL region are extracted to generate ACL stress and strain cloud maps.

[0013] The main control program performs state-driven scheduling of the image preprocessing unit, deep learning image segmentation unit, 3D model reconstruction unit and finite element simulation unit based on a task state-driven mechanism.

[0014] The resource monitoring and adaptive scheduling unit runs in parallel with the main control program and operates on the image preprocessing unit, deep learning image segmentation unit, 3D model reconstruction unit, and finite element simulation unit. It acquires the CPU utilization, memory utilization, network bandwidth usage, and disk I / O latency of each running process instance to construct a resource vector and implements adaptive scheduling based on the resource vector.

[0015] In some embodiments of the present invention, the simulation system further includes:

[0016] The hot-swappable management unit based on configuration files describes and manages the image preprocessing unit, deep learning image segmentation unit, 3D model reconstruction unit, and finite element simulation unit in a unified manner through independent configuration files. The main control program parses the configuration files and constructs a unit scheduling table during system initialization. It calls the corresponding units for execution according to the task status flow. The addition, deletion, upgrading, or replacement of each unit is achieved by updating the configuration files.

[0017] In some embodiments of the present invention, the simulation system is configured with an exception handling mechanism. During the execution of the image preprocessing unit, the deep learning image segmentation unit, the three-dimensional model reconstruction unit, and the finite element simulation unit, the main control program captures runtime exceptions through a try-catch structure. For recoverable exceptions, retry processing is implemented, and for unrecoverable exceptions, the task status is set to an error-marked state and the task is skipped.

[0018] In some embodiments of the present invention, the simulation system is configured with a hierarchical log management strategy, which synchronously writes all abnormal events and the execution status of each unit into the log system;

[0019] Configure a log rotation policy: When the number of lines in a single log file exceeds the limit, automatically archive the current log and create a new log file.

[0020] In some embodiments of the present invention, the resource monitoring and adaptive scheduling unit implements adaptive scheduling, including:

[0021] S1 measures the overall resource utilization intensity of an instance in the current period using the Euclidean norm of the resource vector. ;in, , For CPU utilization, For memory utilization, Due to network bandwidth usage, For disk I / O latency, Indicates the module type. Indicates the first One running instance; For normalized weights;

[0022] S2, the number of polling instances within the same period. and the actual number of tasks received Define the idle ratio of instances. ;

[0023] S3 integrates resource occupancy intensity with idle state. , ;in, and The weighted requirements are for resource overload and idle states, respectively.

[0024] S4, the number of concurrent instances of the same type of unit. Average load during the calculation period According to average load With the preset target load range Perform concurrency control operations and calculate the target concurrency for the next cycle; calculate the number of concurrent instances for the next cycle based on adaptive concurrency control.

[0025]

[0026] in, The `clip` function is used to adjust the number of concurrent connections, acting as a dynamic step size coefficient. Limited to a safe range ;when Higher than The number of concurrent users automatically decreases when it falls below a certain threshold. The number of concurrent users increases automatically, thereby increasing throughput within the limits of system resources and limiting the flow in a timely manner when the load surges.

[0027] In some embodiments of the present invention, the resource monitoring and adaptive scheduling unit implements adaptive scheduling, further comprising:

[0028] To suppress empty polling, an idle counter is designed for each instance. If the load metrics of this instance Continuously above the threshold and resource consumption Below the threshold Da At each monitoring cycle, a hibernation operation is triggered, moving the instance to the idle pool as a hibernation instance and releasing its threads and memory; the system every [period]... Check queue length When the number of tasks reaches or exceeds the set wake-up threshold The dormant instance is woken up and put back into the scheduling loop.

[0029] In some embodiments of the present invention, the image preprocessing unit performs image preprocessing including:

[0030] Image bias field correction: (1) Bias field initialization and modeling based on N4: Based on the N4 Bias FieldCorrection method, a sparse B-Spline mesh is constructed on the three-dimensional volume data to initialize the bias field function. (2) EM iterative optimization of bias field coefficients: based on the current Estimating the true signal distribution (Usually obtained through local histogram smoothing or Gaussian mixture model), minimize

[0031]

[0032]

[0033] Update the B-Spline coefficients, where , To smooth the regularization parameters, the B-Spline mesh is refined stepwise from coarse to fine and the EM optimization is repeated until the bias field converges. The generated corrected image;

[0034] Grayscale mapping and normalization: Linear mapping and truncation based on window width (WW) and window level (WL) are used to linearly map the corrected image Icorr according to the DICOM WindowCenter and WindowWidth.

[0035]

[0036] WC and WW correspond to WindowCenter and WindowWidth respectively, and the mapping result is truncated to [0, 65535]. A 16-bit grayscale value is output to ensure compatibility with subsequent network inputs.

[0037] Noise suppression and image enhancement: Gaussian filtering is applied to each slice image to remove salt-and-pepper noise and high-frequency artifacts; adaptive histogram equalization is used to enhance soft tissue contrast; and bilinear or cubic B-Spline interpolation is performed on non-equidistant slice images using pose information from the DICOM head, and the images are resampled to isotropic voxels.

[0038] In some embodiments of the present invention, in the 3D model reconstruction unit, the main control program calls the 3D reconstruction program to perform the following processing on the ACL region:

[0039] Based on multi-channel mask data and slice image spacing information, the Marching Cubes algorithm is used to reconstruct the initial surface model of bones, menisci and ligaments, and the start and end regions are automatically identified at the ACL insertion point to set a higher local mesh density.

[0040] The Quadric Error Metrics are applied to simplify the overall mesh, significantly reducing the number of facets while maintaining the accuracy of the boundary shape.

[0041] The ACL surface is refined using the Laplacian smoothing algorithm to eliminate sharp abrupt changes and preserve key curvature features along the ligament's direction.

[0042] The global mesh is optimized based on geometric quality indices. It automatically detects and repairs self-penetration, mesh overlap, and narrow and long degenerate triangles in local small regions, and exports the BDF format finite element mesh file required by the implicit nonlinear solver.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are:

[0044] 1. This invention achieves end-to-end fully automated closed-loop control of the ACL simulation process. Through a unified master control program, it implements state-driven scheduling for the four stages of image preprocessing, deep learning segmentation, 3D reconstruction, and finite element simulation. This constructs an end-to-end closed-loop control system from medical image import to simulation result output, eliminating errors and delays caused by fragmented manual operations and significantly improving the standardization and consistency of the preoperative simulation process.

[0045] 2. Significantly improved efficiency and system throughput for multi-case processing. The main control program uses a task-state driven mechanism to schedule the image preprocessing unit, deep learning image segmentation unit, 3D model reconstruction unit, and finite element simulation unit in a state-driven manner. Multiple units execute in parallel, supporting concurrent processing of multiple patient tasks at different stages with a hypothetical asynchronous parallel architecture, achieving pipelined scheduling and eliminating serial bottlenecks. Experimental results show that compared to traditional serial workflows, this system can reduce the overall processing time by more than 40% under multi-case concurrent conditions, significantly improving simulation efficiency and meeting the needs of high-throughput clinical assessment.

[0046] 3. A resource-aware adaptive scheduling mechanism was constructed. This invention introduces a unified control index that integrates resource load intensity and instance idle ratio. Combined with a resource model based on Euclidean norm and idle polling rate, and through a concurrent dynamic adjustment formula and instance sleep / wake-up mechanism, it achieves system rate limiting control under high load conditions and resource expansion under low load conditions, improving CPU and memory utilization by approximately 22%, reducing idle thread occupancy by more than 40%, and increasing system throughput by nearly 30%.

[0047] 4. Highly modular and hot-swappable, adaptable to various clinical configuration needs. This system adopts a configuration file-based module registration and scheduling strategy, supporting flexible replacement and upgrades of image segmentation models, mesh reconstruction algorithms, and simulation solvers without requiring changes to the main control logic or system recompilation, significantly reducing maintenance costs and possessing excellent scalability and engineering implementation capabilities.

[0048] 5. The entire process strictly follows DICOM metadata for geometric correction and image standardization preprocessing, combined with image segmentation and mesh reconstruction strategies specifically for the ACL region, to ensure that the 3D model is highly consistent with the anatomical structure. It is compatible with multiple image sources such as CT and MRI, providing a high-precision basis for personalized preoperative assessment.

[0049] 6. Possesses a stable and reliable fault tolerance recovery and log management mechanism. The system embeds anomaly monitoring, automatic retry, and task isolation strategies into each functional module. Combined with multi-level log tracking and round-robin archiving mechanisms, it can ensure stable system operation under abnormal conditions such as batch processing or network fluctuations, effectively reducing the probability of task interruption due to single point of failure.

[0050] In summary, this invention not only integrates key modules such as deep learning segmentation, geometric reconstruction, and finite element simulation in terms of technical approach, but also establishes a highly automated, scalable, and engineering-practical preoperative simulation platform for ACL through multi-dimensional innovations such as resource scheduling, state control, and system architecture design. It breaks through the bottleneck problems of traditional process fragmentation, low efficiency, and resource waste, and has significant technological progress and clinical application value.

[0051] Other features and advantages of the present invention will become clearer after reading the detailed description of the embodiments of the present invention in conjunction with the accompanying drawings. Attached Figure Description

[0052] Figure 1 This invention proposes an architecture for a full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint;

[0053] Figure 2 This is the sagittal plane segmentation result of the knee joint image obtained by the deep learning image segmentation unit in this embodiment of the invention;

[0054] Figure 3 This is the result of coronal plane segmentation of a knee joint image obtained by a deep learning image segmentation unit in this embodiment of the invention.

[0055] Figure 4 This is a diagram showing the sagittal plane segmentation effect of a single knee joint image in an embodiment of the present invention;

[0056] Figure 5 This is the result of coronal plane segmentation of a single knee joint image in an embodiment of the present invention;

[0057] Figure 6 The three-dimensional reconstruction result obtained by the three-dimensional model reconstruction unit in the embodiment of the present invention;

[0058] Figure 7 The results obtained by the implicit nonlinear solver in the finite element simulation unit of this invention are shown below.

[0059] Figure 8 The present invention provides simulation results of the anterior cruciate ligament of the knee joint using the system of this invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] In the description of this invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0062] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0063] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0064] Preoperative simulation for anterior cruciate ligament (ACL) reconstruction of the knee mainly includes medical image acquisition, image preprocessing, deep learning segmentation, 3D model reconstruction, and finite element simulation analysis. Currently, each step is handled on separate platforms, resulting in a fragmented workflow that relies on multidisciplinary collaboration involving radiologists, modeling engineers, and simulation engineers. This process is cumbersome and inefficient. This invention aims to propose a fully automated, end-to-end system for preoperative simulation of ACL reconstruction, achieving closed-loop control from medical image input to simulation output, eliminating the need for manual intervention throughout the entire process.

[0065] Specifically, such as Figure 1 As shown, the full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint proposed in this invention includes:

[0066] The front-end submission and initialization unit receives DICOM (Digital Imaging and Communications in Medicine) metadata uploaded from the web client, creates a task, and loads it locally.

[0067] The image preprocessing unit names the working directory according to the patient ID and timestamp, preprocesses the slice images, and exports them in batches to the working directory; it parses DICOM metadata to obtain the keywords required for 3D model reconstruction, writes the keywords into the index file, and stores them in the working directory;

[0068] The deep learning image segmentation unit reads slice image data from the working directory and uses a pre-trained FastSAM network to perform multi-label segmentation of key anatomical structures in the knee joint.

[0069] The 3D model reconstruction unit calls the 3D reconstruction program to reconstruct the anterior cruciate ligament of the knee joint in 3D and export the finite element mesh file in BDF format.

[0070] The finite element simulation unit calls the simulation preprocessing program to load the BDF mesh file output by the 3D reconstruction program as the basic model into the simulation environment. After defining the assembly and loads, it outputs the input file required by the implicit nonlinear solver. The implicit nonlinear solver is started to perform nonlinear finite element analysis on the input file. The analysis results are analyzed, and the key indicators of the ACL region are extracted to generate ACL stress and strain cloud maps.

[0071] In the front-end submission and initialization unit, the doctor fills in the patient ID, examination sequence number, and DICOM storage path (to establish a working directory) on the web interface. The back-end interface writes this record to the database and initializes the patient status to "0" (to be processed by the image). The system retrieves the patient ID and its working directory information with the status "0" from the back-end interface, obtains the patient information, and loads the task into the local environment.

[0072] In the image preprocessing unit, all DICOM files are read from the patient's working directory. Key fields required for constructing 3D volumetric data, such as PixelSpacing, SliceThickness, ImageOrientationPatient, ImagePositionPatient, WindowCenter, and WindowWidth, are extracted and an index file is created. These key fields are then written into the index file. Next, the sliced ​​images undergo preprocessing, including image bias field correction, grayscale mapping and image normalization, and noise suppression.

[0073] Image bias field correction includes: (1) N4-based bias field initialization and modeling: Based on the N4 Bias FieldCorrection method, a sparse B-Spline mesh is constructed on the three-dimensional volume data to initialize the bias field function. (2) EM iterative optimization of bias field coefficients: based on the current Estimating the true signal distribution (Usually obtained through local histogram smoothing or Gaussian mixture model), minimize

[0074]

[0075]

[0076] Update the B-Spline coefficients, where , To smooth the regularization parameters, the B-Spline mesh is refined stepwise from coarse to fine and the EM optimization is repeated until the bias field converges. This is the generated corrected image.

[0077] Grayscale mapping and normalization processing includes: linear mapping and truncation based on window width (WW) and window level (WL), and linear mapping of the corrected image Icorr according to DICOM WindowCenter and WindowWidth:

[0078]

[0079] WC and WW correspond to WindowCenter and WindowWidth respectively, and the mapping result is truncated to [0, 65535]. The output is a 16-bit grayscale value to ensure compatibility with subsequent network inputs.

[0080] Noise suppression and image enhancement include: applying Gaussian filtering to each slice image to remove salt-and-pepper noise and high-frequency artifacts; using adaptive histogram equalization to enhance soft tissue contrast; and utilizing pose information from the DICOM head to perform bilinear or cubic B-Spline interpolation on non-equidistant slice images, resampling to isotropic voxels to ensure the accuracy of subsequent 3D reconstruction.

[0081] After the above image processing, the working directory is named according to the patient ID and timestamp, and the processed slice images (16-bit TIFF) are exported in batches. At the same time, an index file is generated to record the original DICOM path, resampled spatial coordinates and processing parameters of each slice image. After completion, the patient status is updated to "1" through the Web API interface.

[0082] In the deep learning image segmentation unit, the FastSAM network is used to perform multi-label segmentation of nine key anatomical structures in the knee joint in a single operation. These structures include: femur, tibia, anterior cruciate ligament (ACL), posterior cruciate ligament (PCL), medial collateral ligament (MCL), lateral collateral ligament (LCL), medial meniscus (MM), lateral meniscus (LM), and fibula. Specifically, the femur, tibia, ACL, and PCL are segmented from sagittal images, while the medial collateral ligament, lateral collateral ligament, medial meniscus, lateral meniscus, and fibula are segmented from sagittal images. After pre-training on the large knee joint DICOM dataset, the FastSAM network was used to fine-tune the network parameters for the nine tissues mentioned above, taking into account the morphological and grayscale features of each tissue. Inputting a pre-processed sequence of sliced ​​images, FastSAM outputs a nine-channel probability map in a single forward inference, with each channel corresponding to a structure class. An adaptive threshold is used to generate a binary mask for each channel probability map, and these masks are saved to the current patient's working directory according to tissue classification. Figure 2 The image shown is the sagittal segmentation result of the knee joint image. Figure 3 The image shown is the coronal plane segmentation result of a knee joint image. Different colors are used to annotate the identification results of the nine tissue types in each image. Figure 4 This provides accurate multi-tissue input for the next step of 3D mesh generation. The segmentation results of this unit can be directly used for subsequent 3D reconstruction and finite element mesh generation without manual correction. Through the above segmentation, the ACL and its surrounding bone and cartilage multi-tissue masks can be output in a single inference, providing a high-quality, continuous, and consistent image-driven foundation for subsequent simulation processes. After image segmentation is completed, the process status is updated to status "2" using the Web API's Post method.

[0083] In the 3D model reconstruction unit, the main control program calls the 3D reconstruction program. During this process, the ACL region is processed as follows: First, based on multi-channel mask data and slice image spacing information, the Marching Cubes algorithm is used to reconstruct the initial surface model of the skeleton, meniscus, and ligaments. The start and end regions are automatically identified at the ACL insertion point to set a higher local mesh density. Then, Quadric Error Metrics is applied to simplify the overall mesh, significantly reducing the number of facets while ensuring boundary shape accuracy. Next, the Laplacian smoothing algorithm is used to refine the ACL surface, eliminating sharp abrupt changes and preserving key curvature features at the ligament's direction. Finally, based on geometric quality indicators, the global mesh is optimized, automatically detecting and repairing problems such as self-penetration, mesh overlap, and narrow, long degenerate triangles in local areas, and exporting the BDF format finite element mesh file required by the implicit nonlinear solver. The 3D reconstruction result is as follows: Figure 6 As shown. After this step is completed, the main control program updates the patient status to "3" via the Web API's Post method.

[0084] In the finite element simulation processing unit, for the "case to be simulated," the main control program first calls the simulation preprocessing program to load the BDF mesh file output by the 3D reconstruction program as the basic model into the simulation environment. The system assigns preset material parameters to the bone, cartilage, and ACL regions (e.g., bone elastic modulus 17000 MPa, ACL elastic modulus 500 MPa, cartilage viscoelastic parameter 10 MPa), applies rigid constraints to the bone-ligament insertion node set of the ACL, and applies equivalent motion constraints simulating a 30° buckling state to the node set in the middle of the ACL. After completing the above assembly and load definition, the integrated mesh, material properties, boundary conditions, and load information are output as the input file required by the implicit nonlinear solver.

[0085] The main control program then automatically launches the implicit nonlinear solver module to perform nonlinear finite element analysis on the generated input file. During the solution process, the large deformation geometry option is enabled, and iterations are performed according to the set convergence tolerance to capture the stress-strain response of the ACL under buckling loading. The system can output intermediate results at predefined intervals for monitoring or subsequent analysis.

[0086] After the solution is completed, the main control program parses the binary result file and text report output by the implicit nonlinear solver, and automatically extracts key indicators of the ACL region, including the maximum principal stress, maximum principal strain, and stiffness slope of the stress-strain curve. For example... Figure 7 The results shown are from the implicit nonlinear solver. The system generates stress and strain contour plots of the ACL, which intuitively reflect the stress distribution of the ligament. Figure 8 The image shows the simulation results of the anterior cruciate ligament of the knee joint.

[0087] After all the solving and post-processing operations are completed, the main control program updates the patient's status to "4" using the Web API's Post method and saves the simulation results in the specified directory, preparing for the final result summary and closed-loop steps.

[0088] After all stages are completed, the final simulation results are stored in the specified directory, and the end-to-end closed loop of a case is completed by notifying the backend.

[0089] In the full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint of this invention, a resource monitoring and adaptive scheduling unit is designed. This unit serves as the core control module, running in parallel with the main control program, and operates on all functional units, including image preprocessing, deep learning image segmentation, 3D model reconstruction, and finite element simulation. This unit operates at a fixed monitoring cycle. Internally, the CPU utilization of each running process instance can be obtained in real time through the operating system interface or a lightweight agent. Memory utilization Network bandwidth usage and disk I / O latency .in Indicates the module type. Indicates the first One running example. The above four metrics are then weighted and normalized. After processing, a resource vector is formed.

[0090]

[0091] Then, adaptive scheduling of the entire system is implemented using the following steps:

[0092] S1: The overall resource utilization intensity of an instance in the current period is measured by the Euclidean norm of the resource vector.

[0093]

[0094] This value reflects the resource consumption of a single instance in the current period; the larger the value, the heavier the resource load.

[0095] S2: Number of polling instances within the same period. and the actual number of tasks received Define the idle ratio of instances.

[0096]

[0097] in A value close to 1 indicates that the instance has not received new tasks for a long time, resulting in wasted time and energy from idle polling. A higher idle ratio indicates a more severe problem of instances running idle.

[0098] S3: Integrate resource occupancy intensity and idle status, and use linear fusion to obtain a unified comprehensive control index.

[0099]

[0100]

[0101] in and This indicator reflects the weighted requirements for resource overload and idle waste, and can be used uniformly to measure whether an instance is in a high-load or idle state, avoiding resource overload or computational idleness.

[0102] S4: For units of the same type Current number of concurrent instances Average load during the calculation period

[0103]

[0104] Based on average load With the preset target load range Perform concurrency control operations and calculate the target concurrency for the next cycle; calculate the number of concurrent instances for the next cycle according to the following adaptive concurrency formula.

[0105]

[0106] in The `clip` function is used to adjust the number of concurrent connections, acting as a dynamic step size coefficient. Limited to a safe range .when Higher than The number of concurrent users automatically decreases; when it falls below a certain threshold... The number of concurrent users increases automatically, thereby increasing throughput within the limits of system resources and limiting the flow in a timely manner when the load surges.

[0107] To further reduce resource waste caused by empty polling, this unit maintains an idle counter for each instance. If the load metrics of this instance Continuously above the threshold And resource consumption Below the threshold Da At the beginning of each monitoring cycle, a hibernation operation is triggered, moving the instance to the idle pool and releasing its threads and memory. The system performs this operation every [period]. Check queue length When the number of tasks reaches or exceeds the set wake-up threshold When the time comes, the dormant instance is woken up and re-entered into the scheduling loop.

[0108] Through the above unified comprehensive control indicators Along with the concurrent adaptive formula, this invention simultaneously resolves two conflict issues—resource overload and ineffective idle running—within the same scheduling framework, achieving dynamic scaling up and down for all processing stages. Experimental results show that in a simulation scenario of 100 consecutive cases, this scheduling strategy can improve the average utilization of CPU and memory by approximately 22%, reduce idle thread usage by more than 40%, and increase overall throughput by approximately 30%, effectively ensuring the high efficiency of parallel processing of multiple cases and the long-term stable operation of the system.

[0109] In the full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint of this invention, a hot-swappable management unit is constructed. A hot-swappable module management mechanism based on configuration files enables flexible addition, removal, and dynamic replacement of system functional modules. Each functional unit in the system (including image preprocessing, image segmentation, 3D model reconstruction, and finite element simulation) is uniformly described and managed through an independent configuration file. The configuration file adopts JSON format and records the metadata information of each module, including module name, execution program path, input / output path template, trigger status code, completion status code, command line parameter template, concurrency control parameters, and fault tolerance strategy.

[0110] The main control program automatically parses the configuration file during system initialization, dynamically constructs the unit scheduling table, and calls the corresponding units for execution based on task status flow. Adding, deleting, upgrading, or replacing units can all be achieved by updating the configuration file, without requiring code modification to the main control program or system redeployment. It possesses excellent scalability and compatibility, making it suitable for personalized configuration needs in clinical applications.

[0111] Through the above technical solutions, the present invention achieves configurability and hot-swappability at the system module level, significantly reduces the system's operational complexity, and enhances the system's adaptability to multiple organizations, processes, and toolchains.

[0112] In the full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint of this invention, a robust exception handling mechanism and hierarchical log management strategy are designed to address potential operational anomalies in the multi-stage automated process, ensuring task stability and traceability. During the execution of each functional unit, the main control program captures operational anomalies through a try-catch structure, mainly including: EXE execution failure (non-zero exit code), exceeding the set timeout period, missing or incorrectly formatted output files, and API request failure or response anomalies. For recoverable anomalies (such as temporary I / O errors or network fluctuations), the system will automatically retry the process, with the number of retries not exceeding the max_retry limit set in the configuration file (default K times). If the process still cannot be successfully completed within the maximum number of retries, or if an unrecoverable fatal anomaly occurs (such as missing EXE files or resource crashes), the patient's task status is automatically set to error flag status -1, and the task is skipped.

[0113] All abnormal events and unit execution status will be synchronously written to the log system. The log includes the following information fields: timestamp, module name and executed command, patient ID, abnormality type and error message, and number of retries and final result.

[0114] To prevent log files from growing indefinitely, the system has a log rotation policy: when the number of lines in a single log file exceeds the limit (e.g., 5000 lines), the current log is automatically archived and a new log file is created, ensuring that the system has a controllable log storage mechanism during long-term operation.

[0115] Through the above-mentioned exception handling and logging mechanisms, the system of the present invention can achieve error isolation, task recovery and traceability when facing batch patient automatic processing tasks, thereby improving the robustness and engineering practicality of the overall system.

[0116] Through the above technical solutions, this invention achieves fully automated closed-loop control of the ACL preoperative simulation process, which not only significantly reduces manual intervention and improves batch processing efficiency, but also ensures the efficient and stable operation of the system through resource-aware scheduling, and provides a highly scalable modular design to meet the clinical needs for personalized, rapid and reliable assessment.

[0117] It should be noted that the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint, characterized in that, include: The front-end submission and initialization unit receives DICOM metadata uploaded from the web client, creates a task, and loads it locally. The image preprocessing unit names the working directory according to the patient ID and timestamp, and exports the slice images in batches to the working directory after preprocessing them; Parse DICOM metadata to obtain the keywords required for 3D model reconstruction, write the keywords into an index file and store them in the working directory; The deep learning image segmentation unit reads slice image data from the working directory and uses a pre-trained FastSAM network to perform multi-label segmentation of key anatomical structures in the knee joint. The 3D model reconstruction unit calls the 3D reconstruction program to reconstruct the anterior cruciate ligament of the knee joint in 3D and export the finite element mesh file in BDF format. The finite element simulation unit calls the simulation preprocessing program to load the BDF mesh file output by the 3D reconstruction program as the basic model into the simulation environment. After implementing assembly and load definitions, it outputs the input file required by the implicit nonlinear solver. The implicit nonlinear solver is started to perform nonlinear finite element analysis on the input file; the analysis results are analyzed, and key indicators of the ACL region are extracted to generate ACL stress and strain contour maps; The main control program performs state-driven scheduling of the image preprocessing unit, deep learning image segmentation unit, 3D model reconstruction unit and finite element simulation unit based on a task state-driven mechanism. The resource monitoring and adaptive scheduling unit runs in parallel with the main control program and operates on the image preprocessing unit, deep learning image segmentation unit, 3D model reconstruction unit, and finite element simulation unit. It acquires the CPU utilization, memory utilization, network bandwidth usage, and disk I / O latency of each running process instance to construct a resource vector and implements adaptive scheduling based on the resource vector.

2. The full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint according to claim 1, characterized in that, The simulation system also includes: The hot-swappable management unit based on configuration files describes and manages the image preprocessing unit, deep learning image segmentation unit, 3D model reconstruction unit, and finite element simulation unit in a unified manner through independent configuration files. The main control program parses the configuration files and constructs a unit scheduling table during system initialization. It calls the corresponding units for execution according to the task status flow. The addition, deletion, upgrading, or replacement of each unit is achieved by updating the configuration files.

3. The full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint according to claim 1, characterized in that, The simulation system is equipped with an exception handling mechanism. During the execution of the image preprocessing unit, deep learning image segmentation unit, 3D model reconstruction unit, and finite element simulation unit, the main control program captures runtime exceptions through a try-catch structure. For recoverable exceptions, retry processing is implemented. For unrecoverable exceptions, the task status is set to an error flag state and the task is skipped.

4. The full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint according to claim 3, characterized in that, The simulation system is configured with a hierarchical log management strategy, which synchronously writes all abnormal events and the execution status of each unit into the log system. Configure a log rotation policy: When the number of lines in a single log file exceeds the limit, automatically archive the current log and create a new log file.

5. The full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint according to claim 1, characterized in that, The resource monitoring and adaptive scheduling unit implements adaptive scheduling, including: S1 measures the overall resource utilization intensity of an instance in the current period using the Euclidean norm of the resource vector. ;in, , For CPU utilization, For memory utilization, Due to network bandwidth usage, For disk I / O latency, Indicates the module type. Indicates the first One running instance; For normalized weights; S2, the number of polling instances within the same period. and the actual number of tasks received Define the idle ratio of instances. ; S3 integrates resource occupancy intensity with idle state. , ;in, and The weighted requirements are for resource overload and idle states, respectively. S4, the current number of concurrent instances for the same type of unit. Average load during the calculation period According to average load With the preset target load range Perform concurrency control operations and calculate the target concurrency for the next cycle; calculate the number of concurrent instances for the next cycle based on the adaptive concurrency formula: ； in, The `clip` function is used to adjust the number of concurrent connections, acting as a dynamic step size coefficient. Limited to a safe range ;when Higher than The number of concurrent users automatically decreases when it falls below a certain threshold. The number of concurrent users increases automatically, thereby increasing throughput within the limits of system resources and limiting the flow in a timely manner when the load surges.

6. The full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint according to claim 5, characterized in that, The resource monitoring and adaptive scheduling unit implements adaptive scheduling, and also includes: To suppress empty polling, an idle counter is designed for each instance. If the load metrics of this instance Continuously above the threshold and resource consumption Below the threshold Da At each monitoring cycle, a hibernation operation is triggered, moving the instance to the idle pool as a hibernation instance and releasing its threads and memory; the system every [period]... Check queue length When the number of tasks reaches or exceeds the set wake-up threshold The dormant instance is woken up and put back into the scheduling loop.

7. The full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint according to claim 1, characterized in that, The image preprocessing unit performs preprocessing on the image, including: Image bias field correction: (1) Bias field initialization and modeling based on N4: Based on the N4 Bias Field Correction method, a sparse B-Spline mesh is constructed on the three-dimensional volume data to initialize the bias field function. (2) EM iterative optimization of bias field coefficients: based on the current Estimating the true signal distribution , minimize ； ； Update the B-Spline coefficients, where , To smooth the regularization parameters, the B-Spline mesh is refined stepwise from coarse to fine and the EM optimization is repeated until the bias field converges. The generated corrected image; Grayscale mapping and normalization: Linear mapping and truncation based on window width and window level are used to perform linear mapping on the corrected image Icorr according to DICOMWindowCenter and WindowWidth. ； WC and WW correspond to WindowCenter and WindowWidth respectively, and the mapping result is truncated to [0, 65535]; 16-bit grayscale value is output to ensure compatibility with subsequent network input; Noise suppression and image enhancement: Gaussian filtering is applied to each slice image to remove salt-and-pepper noise and high-frequency artifacts; adaptive histogram equalization is used to enhance soft tissue contrast; and bilinear or cubic B-Spline interpolation is performed on non-equidistant slice images using pose information from the DICOM head, and the images are resampled to isotropic voxels.

8. The full-process simulation system for anterior cruciate ligament reconstruction surgery of the knee joint according to claim 1, characterized in that, In the 3D model reconstruction unit, the main control program calls the 3D reconstruction program to perform the following processing on the ACL region: Based on multi-channel mask data and slice image spacing information, the Marching Cubes algorithm is used to reconstruct the initial surface model of bones, menisci and ligaments, and the start and end regions are automatically identified at the ACL insertion point to set a higher local mesh density. The Quadric Error Metrics are applied to simplify the overall mesh, significantly reducing the number of facets while maintaining the accuracy of the boundary shape. The ACL surface is refined using the Laplacian smoothing algorithm to eliminate sharp abrupt changes and preserve key curvature features along the ligament's direction. The global mesh is optimized based on geometric quality indices. It automatically detects and repairs self-penetration, mesh overlap, and narrow and long degenerate triangles in local small regions, and exports the BDF format finite element mesh file required by the implicit nonlinear solver.

Images