An automated cerebral blood flow analysis system based on MRI data
Through an automated cerebral blood flow analysis system based on MRI data, it uses skeleton extraction, meshing and solver algorithms to package it into a 'black box', solving the complexity and high threshold problems of blood flow three-dimensional physical simulation in the prior art. It realizes automated calculations and result output, simplifies clinician operations, and improves the accuracy and reliability of the simulation process.
Patent Information
- Application Number
- CN202011094283.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2020-07-02
- Filing Date
- 2020-10-14
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2040-10-14
AI Technical Summary
In the prior art, the three-dimensional physical simulation analysis process of blood flow is complex, requiring a large amount of human-computer interaction, high professional skills requirements, and difficult to widely promote in clinical practice. The quality of grid division directly affects the accuracy of the result, and the threshold is relatively high.
Provides an automated cerebral blood flow analysis system based on MRI data, including a skeleton extraction algorithm, meshing method and solver algorithm, encapsulating the simulation process into a 'black box', and only the clinician needs to provide the vascular STL model and boundary conditions. Automatically execute blood flow field results to lower the simulation threshold.
It significantly lowers the technical threshold for blood flow simulation, automatically calculates hemodynamic parameters, simplifies model production and boundary definition, improves the accuracy and reliability of results, and facilitates clinicians to use physical simulation technology to assist surgical solutions.
Smart Images

Figure CN113887147B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of subsequent processing and analysis of medical images, and relates to an automated analysis system for cerebral blood flow, and specifically to three-dimensional model processing of blood vessels and three-dimensional fluid dynamics simulation, and in particular to an automated analysis system for cerebral blood flow based on MRI data. The present invention proposes a three-dimensional hemodynamics automated calculation process for clinicians, and improves and encapsulates the algorithms involved in the blood flow calculation in the field of non-clinical doctors, requiring only the clinicians to provide necessary medical parameters, and then the entire calculation process will automatically execute the calculation and give three-dimensional field quantity results such as blood flow velocity and shear stress, which greatly facilitates clinicians to use physical simulation technology to assist surgical plans. Background Art
[0002] Current research suggests that hemodynamic factors play an important role in the occurrence, development and rupture of aneurysms. The industry consensus is that understanding the hemodynamic parameters of aneurysms through physical simulation technology can effectively guide surgical plans. The existing three-dimensional physical simulation technology of blood flow is based on computational fluid dynamics (CFD).
[0003] The basic idea of CFD is to replace the field of physical quantities originally connected in the time domain and space domain, such as velocity field and pressure field, with a set of variable values at a finite number of discrete points. By forming the variables at these discrete points into a set of algebraic equations according to the control equation, the distribution of basic physical quantities (such as velocity, pressure, etc.) at various positions in the flow field of extremely complex problems and the changes of these physical quantities over time are obtained through this kind of numerical simulation. At present, the simulation of hemodynamics using CFD software is mainly divided into the following processes: first, using medical image processing software such as Mimics, the DICOM format images collected from the clinic are converted, and the three-dimensional geometry of the patient's blood vessels (STL format) is reconstructed and saved; then the three-dimensional blood vessel geometry is imported into the computational fluid meshing software such as ANSYS-ICEM to generate the computational grid for numerical simulation; then the boundary conditions need to be set for the inlet and outlet of the model, such as blood pressure, blood flow velocity, flow, etc.; set the property parameters of blood and blood vessel walls, such as density, viscosity, elasticity, etc.; finally, start the calculation to extract the visualization parameters after convergence. The reliability of the numerical simulation study depends on whether the above steps are carried out completely correctly. Practice shows that the typical program for CFD-based blood flow analysis is SimVascular[1], which requires a lot of human-computer interaction to obtain results.
[0004] In the prior art, numerical simulation is an interdisciplinary research. Blood flow calculation is based on CFD program, and mechanical analysis of blood vessel wall is based on CAE program. The whole analysis process involves fluid mechanics, mathematics, computer software, computer simulation, finite element method, materials science, topology, medicine, etc. The process of analysis using commercial software is biased towards mechanics. Therefore, the above research methods have not been widely promoted in clinical work. For example, the most important regional discretization process in the simulation process is far from the professional skills of clinicians. Professional CAE engineers are required to divide the fluid area or solid area into computational grids before calculation and solution can be performed. The quality of the grid directly determines the accuracy of the results.
[0005] Based on the current status of the prior art, the inventor of this application intends to provide an automated calculation process for hemodynamic problems, which mainly includes a skeleton extraction algorithm, a meshing method and a solver algorithm; specifically, it relates to an automated analysis system for cerebral blood flow based on MRI data, especially an automated calculation process for three-dimensional hemodynamics for clinicians. The present invention does not consider the stress factor of the vascular wall for the time being, and mainly analyzes hemodynamics by assuming solid wall conditions. Starting from the stl model, the mesh production process and the calculation process are packaged into a "black box" with minimal user interaction. The entire simulation process becomes three stages: image extraction, model production, and boundary definition. The settings of these three stages are all within the professional scope of clinicians, and the simulation threshold will be greatly reduced.
[0006] References related to the present invention include:
[0007] [1]Updegrove,A.,Wilson,N.,Merkow,J.,Lan,H.,Marsden,ALand Shadden,SC,SimVascular-An open source pipeline for cardiovascular simulation,AnnalsofBiomedical Engineering(2016).DOI:10.1007 / s10439-016-1762-8
[0008] [2] Liu Fengrui, Huang Jianping, Li Zhong, Wang Naxiu, et al. Two-stage gravity ball algorithm for generating single-size particle accumulation in cylindrical container[J]. Nuclear Technology, 2017. Summary of the invention
[0009] The purpose of the present invention is to provide an automated calculation process for hemodynamic problems based on the current status of the prior art, which mainly includes a skeleton extraction algorithm, a meshing method and a solver algorithm; specifically, it relates to an automated cerebral blood flow analysis system based on MRI data, especially an automated three-dimensional hemodynamic calculation process for clinicians.
[0010] The present invention does not consider the stress factors of the blood vessel wall for the time being. It mainly analyzes the hemodynamics by assuming solid wall conditions. Starting from the STL model, the grid production process and the calculation process are combined into a "black box" with minimal user interaction. The entire simulation process is transformed into three stages: image extraction, model production, and boundary definition. The settings of these three stages are all within the professional scope of clinicians, and the simulation threshold will be greatly reduced.
[0011] Specifically, the cerebral blood flow automatic analysis system based on MRI data of the present invention mainly includes an automatic process for calculating three-dimensional hemodynamics, which only requires the operator to provide a blood vessel stl model and specify the inlet and outlet surfaces of the blood flow in the model and the corresponding boundary conditions, and the program can automatically execute and output the blood flow field quantity results;
[0012] The blood flow automatic calculation program in the automatic process includes a model preprocessing module, a tracer particle module, a track statistics module, a skeleton extraction module, a gridding module and a fluid solution module.
[0013] In the present invention, the tracer particle module extracts the topological information of the blood vessel model based on the particle dynamics method, thereby avoiding the instability problem of the traditional skeleton extraction algorithm based on topology.
[0014] In the present invention, the track statistics module injects tracer particles based on the Monte Carlo concept, the initial positions of the particles are randomly distributed on the end surface, and the initial velocities of the particles are randomly distributed. The continuous repetition of the injection process enhances the particle's ability to capture the vascular geometry, and the injection process ends when the track category is stable.
[0015] In the present invention, the skeleton extraction module performs maximum inscribed sphere "correction" on the initial skeleton obtained by track statistics, and the corrected skeleton can conveniently provide the location and volume information of the hemangioma.
[0016] In the present invention, the mesh division module adopts the idea of segmentation, divides the model into sub-models according to the skeleton, and meshes each sub-model separately, and there is no topological constraint between the sub-grids.
[0017] In the present invention, in the fluid solving module, the fluid solver performs coupled solving for different blood flow areas.
[0018] In the present invention, the dynamic parameters of blood flow are automatically calculated with little interaction with the clinician, and the clinician is required to provide a vascular model in STL format based on the DICOM image; the program in the present invention re-triangulates the STL, and makes the distribution of triangular facets as uniform as possible under the premise that the shape of the model file remains unchanged, which is convenient for subsequent mesh division and particle-vascular wall collision detection; for the sharp areas in the model, the program appropriately smoothes the model to approach the real characteristics of the blood vessel; based on the VTK library, a model interaction interface is provided to the clinician for specifying the inlet and outlet of the vascular cluster and the related blood flow boundary conditions, which is also the only link that requires interaction with the user in this program.
[0019] Based on the current cutting and naming operation, clinicians are asked to qualitatively make a section at the inlet and outlet to mark the inlet and outlet positions. For the inlet boundary, clinicians can specify on the program input card that the inlet surface is extended a certain distance along the external normal direction. In order to ensure the full development of fluid flow, at each inlet and outlet end face of the vascular cluster, the program starts the tracer particle injection operation, and injects different batches of particles with different speeds according to the Monte Carlo idea. After the particle trajectory is stable, the tracer particle trajectory is statistically analyzed to mark the vascular branches and the areas with vortex trajectories inside. Then, the statistically obtained vascular skeleton is further corrected for the maximum inscribed sphere based on the tracer particle expansion algorithm. The corrected skeleton and the maximum The distribution of the radius of the inscribed sphere can be used clinically. After the skeleton information is obtained, the model is segmented based on the vortex area, and the segmented model set is meshed with a boundary layer separately. For the area where the segmentation is unsuccessful, it degenerates into a mesh without a boundary layer. However, as a compromise for the degradation of the mesh quality, the mesh size without a boundary layer is set to be small. The mesh division requires the user to provide the inlet blood flow conditions in advance, so as to determine the mesh size parameters such as the thickness of the mesh boundary layer based on the blood flow Reynolds number. Finally, the boundary conditions specified by the clinician and the generated partitioned mesh are combined for fluid calculation. Its specific execution requires calling a dedicated fluid solver for multi-zone meshes. The entire program execution process is as follows: Figure 3 As shown;
[0020] In the present invention, the core algorithm is divided into four modules: statistical tracer particle trajectory / vascular skeleton extraction / model grid division and fluid solver.
[0021] In the present invention, for a given stl model, the program first automatically identifies all the head and end faces of the vascular cluster, removes the end faces to be processed later so that the model is in an open state at the end faces (subsequent injection of tracer particles), and then the program traverses the head and end faces of each vascular cluster, and uses the Monte Carlo idea to randomly throw some spherical particles from the end faces, allowing these particles to move forward along the end face entrance under the action of random initial velocity, and sets the external force field of the entire system to drive the particle movement (the effect is similar to the pressure difference in the fluid flow equation). At this time, the system only has particle-particle forces and particle-triangular unit forces (i.e., particle-vascular wall effects), tracks particle trajectories, and uses statistical methods to summarize the topological characteristics of the particle group, providing basic information for subsequent skeleton extraction and meshing.
[0022] The present invention provides an automated cerebral blood flow analysis system based on MRI data, especially an automated three-dimensional hemodynamics calculation process for clinicians. The calculation program of the present invention includes a model preprocessing module, a tracer particle module, a track statistics module, a skeleton extraction module, a gridding module, and a fluid solution module. The main program of this analysis system is based on the python language, and the complex function module is encapsulated as a dynamic library based on C++ for the main program to call. The process only requires the clinician to provide a blood vessel stl model and specify the inlet and outlet surfaces of the blood flow in the model and the corresponding boundary conditions. The program can automatically execute and output the blood flow field quantity results. The present invention can significantly reduce the technical threshold of blood flow simulation, facilitate clinicians to simulate and observe blood flow details before surgery, and provide reference information for surgical plans. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 This is the execution flow of the discrete element method in the present invention.
[0024] Figure 2 It is a flow chart of the fluid calculation in the present invention.
[0025] Figure 3 This is the overall execution flow of the main program in the present invention. DETAILED DESCRIPTION
[0026] Example 1
[0027] The particle discrete element method (DEM) is used to calculate the motion of particles under external forces. Figure 1 As shown in the figure, the forces between particles and between particles and patches are equivalent to spring-damper models. The spring and damper are applied in the normal direction n and tangential direction t respectively. For two contact particles i and j, the radii are Ri and Rj respectively. The control equation of the angular velocity of particle i is:
[0028]
[0029] Where mi, Ii represent the mass and moment of inertia of particle i, respectively, and Nc is the number of neighboring particles in contact with particle i. The program uses a simplified Hertz–Mindlin–Deresiewicz contact model to calculate the contact force between particles:
[0030]
[0031] Where k and γ represent the stiffness and damping coefficient; represents the relative velocity between particles; μ represents the friction coefficient, and the tangential deformation and normal deformation between particles are:
[0032]
[0033] The stiffness coefficient k and damping coefficient γ in the tangential and normal directions are:
[0034]
[0035] Where R*, m* and E* are:
[0036]
[0037] E, G and υ are Young's modulus, shear modulus and Poisson's ratio, respectively, and ε is the coefficient of restitution;
[0038] The maximum time step size is determined by the Rayleigh time step:
[0039]
[0040] The position and velocity of the particle evolve over time and are iteratively updated using the Leap-Frog algorithm, which has third-order accuracy and is time-reversible. The position and velocity are updated using the position of the current time step, the velocity of the first half of the time step, and the force on the particle at the current moment.
[0041]
[0042] in They represent the velocity, position and force of the particle at time t respectively.
[0043] The DEM algorithm uses a neighbor list to search for particle neighbors. After completing the neighbor search, it performs strict collision detection and applies particle-particle and particle-vascular wall interaction forces based on the relative positions between the particles and their neighbors. Finally, Newton's laws are used to update the displacement, velocity, angular velocity and other dynamic information of each particle. The process is as follows: Figure 1 shown.
[0044] The first characteristic of the tracer particle trajectory is the connectivity characteristics of different inlets and outlets, which is similar to the definition of the divergence of field quantities in differential equations. Multiple trajectories may intersect in a certain area or separate at a certain point. The particle mass flow rate on a certain section can be used as a sign of line division and merging in topology. The place where the mass flow rate increases represents the merging point of the blood flow branch, and vice versa. The tracer particle enters the vascular network from a certain end face, and moves forward under the action between particles and vascular walls until it flows out of the vascular network from a certain end face. During this period, a complete particle trajectory can represent a vascular pathway. By traversing all end faces, a set of vascular pathways can be obtained. By merging the vascular pathways, a preliminary vascular model skeleton can be formed.
[0045] The model skeleton should be able to reflect the overall topological structure of the model in a concise way, while also maintaining the branching structure of the model well. In order to improve the quality of model skeleton extraction, it is necessary to ensure that the model skeleton points can be well concentrated at the center axis of the model. The set of pathways obtained by statistically analyzing the trajectories of the tracer particles depends on the tracing method in the Monte Carlo process, and is not necessarily close to the center of the blood vessel. Therefore, it is necessary to "correct" it based on the three-dimensional maximum inscribed sphere principle. The specific method is to place virtual particles of a certain radius on the pathway at intervals, gradually increase the particle radius, cancel the particle-particle interaction, and simply consider the particle-wall interaction force. The particle radius is increased until it moves under the action of the blood vessel wall and reaches stability. The corrected vascular pathway can be obtained. At this time, the radius information of the maximum inscribed sphere at the position of the virtual particle on the vascular pathway has been obtained, and the topological information of the vascular system can be used for clinical application. The algorithm is similar to the particle expansion algorithm in the "two-stage gravity ball algorithm for generating single-size particle accumulation in a cylindrical container". The particle dynamics algorithm is used to avoid some limitations of the pure topological algorithm [2]. The robustness of the algorithm is increased and it is more suitable for embedding into the automation process.
[0046] Another characteristic of the tracer particle trajectory is the vortex characteristic in the local area, which can be regarded as the connectivity characteristic when the inlet and outlet inside the blood vessel overlap (such as the blood flow vortex inside the hemangioma); similar to the definition of the curl of the field quantity, the tracer particles sometimes flow out from a certain position on the blood vessel wall and then flow in, and their trajectory is a closed curve. At this time, the blood flow divergence is 0 but the curl is not 0; the mass flow rate of the tracer particles inside the vortex is not 0, and the mass flow rate of the tracer particles in the upstream and downstream is unchanged, and the existence of the vortex cannot be determined. However, at this time, the corrected vascular skeleton will degenerate into a relatively short branch line segment in the vortex area. Therefore, for these non-user-marked end faces, the tracer particle mass flow rate can be detected in the area represented by the branch line segment to determine the vortex area.
[0047] Example 2
[0048] The vortex region based on the prior art is often the area of clinical concern and is usually the key position that affects the quality of the mesh. It is difficult to properly divide all regions by traditionally dividing the entire vascular model into meshes. It usually takes repeated compromises and attempts to obtain a relatively complete overall mesh, which requires a lot of human-computer interaction.
[0049] The program in the present invention divides the stl model of the blood vessel according to the vortex area, and each subset after division is independently meshed. The mesh consistency is not required between different blood vessel units. Therefore, the topological shape and blood flow characteristics of each mesh unit are relatively simple. A fluid mesh that conforms to its flow characteristics can be generated for each blood vessel subset, which is the key to the automation of the entire simulation process.
[0050] After the grid subset is generated for the blood vessel subset, the fluid solver needs to support the information transmission of different sub-vessel interfaces (the traditional CFD typical algorithm PISO / SIMPLE is only for a single flow area); the fluid solver in the program of the present invention adopts the PISO algorithm to iteratively calculate the flow in each sub-grid, and interpolate the velocity, pressure and other field quantities at the interface of different sub-vessels. Based on computational fluid dynamics (CFD), the fluid dynamics information is calculated, and the control equations followed mainly include the continuity equation:
[0051]
[0052] For a flow with constant density, the above equation simplifies to:
[0053]
[0054] Momentum conservation equation:
[0055]
[0056] The first term on the left represents the rate of change of the particle velocity; the second term on the left represents the convection term; the first term on the right represents the pressure gradient term; the second term on the right represents the external volume force acting on the particle; the third term on the right represents the particle stress; and the last term represents gravity. For isotropic fluids, the fluid relationship is:
[0057]
[0058] The program adopts the turbulence model based on the Reynolds-averaged NS equations:
[0059]
[0060] Before the calculation begins, the clinician needs to specify the vascular boundary conditions. This is the only place where the algorithm needs to interact with the user. The user needs to move the interface to move / scale / rotate the section, and cut the rectangular section with the stl model to obtain the section as the blood flow inlet and outlet surface, and then enter the surface type (inlet / outlet), inlet boundary (blood flow / flow velocity distribution model / stretching length), outlet boundary (outlet pressure) and other flow information in the save window; the fluid solver performs calculations based on the user-specified boundaries and physical parameters. For each subnet area, the momentum equation is first solved with the given initial pressure or the pressure of the previous iteration step. Because the obtained velocity variable does not necessarily satisfy the continuity equation, it is necessary to solve the pressure Poisson equation to obtain the pressure, and use the obtained pressure to correct the velocity so that it can satisfy the continuity equation. If the velocity does not satisfy the momentum equation, the cycle is repeated until convergence; the program flow is as follows Figure 2 shown.
[0061] The present invention can significantly lower the technical threshold of blood flow simulation, making it easier for clinicians to simulate and observe blood flow details before surgery, thus providing reference information for surgical plans.
Claims
1. An automated cerebral blood flow analysis system based on MRI data, characterized in that: It includes an automated process for calculating three-dimensional hemodynamics, in which an operator provides a blood vessel stl model and specifies the inlet and outlet surfaces of the blood flow in the model and corresponding boundary conditions, and the blood flow automated calculation program in the automated process automatically executes and outputs the blood flow field quantity results; The blood flow automation calculation program in the automation process includes a model preprocessing module, a tracer particle module, a track statistics module, a skeleton extraction module, a grid division module and a fluid solution module; The tracer particle module extracts topological information of the vascular model based on a particle dynamics method; The track statistics module injects tracer particles based on the Monte Carlo concept, and the initial positions of the particles are randomly distributed on the end surface, and the initial velocities of the particles are randomly distributed; the DEM particle discrete element method is used to calculate the movement of the particles under the action of external forces; The skeleton extraction module performs maximum inscribed sphere correction on the initial skeleton obtained by track statistics based on the tracer particle expansion algorithm, and the corrected skeleton can provide the location and volume information of the hemangioma; The meshing module adopts the idea of segmentation, divides the model into sub-models according to the skeleton, and meshes each sub-model separately to generate a fluid mesh that meets its flow characteristics, and there is no topological constraint between the sub-grids; In the fluid solving module, the fluid solver performs coupled solving for different blood flow areas; the fluid solver adopts the PISO algorithm to perform iterative calculation on the flow in each sub-grid respectively.
2. The automated cerebral blood flow analysis system based on MRI data according to claim 1, characterized in that: In the track statistics module, the injection process is continuously repeated to enhance the particle's ability to capture the blood vessel geometry, and the injection process is terminated when the track category is stable.
