A multi-scale visualization simulation method and device for brain material transport
By simulating brain material transport through transmembrane coupling of LBM and DEM, the problem of scale fragmentation and insufficient accuracy in the observation and simulation of brain material transport in existing technologies is solved. This enables multi-scale coupled simulation and personalized simulation, supporting disease mechanism research and personalized medicine.
Patent Information
- Application Number
- CN202611333774.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-31
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies are insufficient for real-time, high-resolution dynamic observation of brain material transport processes. They suffer from problems such as scale fragmentation, lack of pathological specificity, insufficient visualization capabilities, and low simulation accuracy. They cannot fully reflect the multi-scale coupling mechanism of brain material transport, thus limiting disease mechanism research and clinical applications.
The flow of interstitial fluid and cerebrospinal fluid in porous media was simulated using LBM, and the motion of discrete phase particles at the BBB interface was simulated using DEM. Transmembrane coupling between LBM and DEM was achieved through bidirectional force transmission. A multi-scale computational grid was constructed, multi-source pathological data were integrated, and the transmembrane transport behavior of particles was calculated in real time, followed by dynamic visualization and quantitative analysis.
It achieves seamless coupling simulation from subcellular to tissue scale, provides personalized simulation of brain material transport, assists in disease mechanism research and personalized medicine, improves simulation accuracy and visualization capabilities, and supports drug development and surgical planning.
Smart Images

Figure CN123120337A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the interdisciplinary fields of computational biomedicine, biofluid dynamics, machine learning and computer simulation, and more specifically, to a multi-scale visualization simulation method and device for brain material transport. Background Technology
[0002] The brain, as the most intricate organ in the human body, relies on the proper functioning of its internal transport processes (including nutrient delivery, drug penetration, and metabolic waste removal) to maintain homeostasis. The blood-brain barrier (BBB) and the glial-lymphatic system are core structures regulating these processes. The BBB, as a selectively permeable interface between the brain and systemic circulation, is closely associated with abnormal permeability in various diseases, including brain tumors and neurodegenerative diseases. The glial-lymphatic system, a recently discovered lymphoid clearance pathway in the brain, removes metabolic waste (such as Aβ protein aggregates) through the exchange of cerebrospinal fluid and tissue fluid. Its dysfunction has been confirmed as a significant pathogenic mechanism in diseases such as Alzheimer's disease. Therefore, accurately elucidating the permeability mechanisms of the BBB and the dynamic characteristics of the glial-lymphatic system is of significant scientific and clinical value for research into the pathogenesis of brain diseases and the development of treatment strategies.
[0003] Currently, research in this field faces numerous insurmountable technical challenges, as follows: 1. Experimental observation has significant limitations, making it difficult to capture microscopic dynamic processes. In vivo, processes such as BBB transcellular transport and interstitial flow in brain tissue are characterized by their microscopic and dynamic nature. Existing observation techniques (such as two-photon microscopy and optical coherence tomography (OCT)) are insufficient for real-time, high-resolution dynamic observation, failing to fully capture the entire process and key details of substance transport. While in vitro BBB models (such as 3D microvascular models and cell co-culture models) can simulate some of the barrier functions of the BBB, they differ significantly from the in vivo situation in terms of the complexity of vascular network structure, dynamic intercellular interactions, and microenvironment regulation. They cannot replicate the heterogeneous pathological microenvironment in vivo, resulting in limited clinical translational value of experimental results.
[0004] 2. Computational simulations suffer from scale fragmentation and lack a unified multi-scale framework. Existing computational models for brain substance transport mostly focus on a single scale, failing to achieve seamless integration across subcellular, cellular, and tissue scales. For example, some studies use the LBM (Liquid Boundary Method) to simulate macroscopic infiltration processes in brain tissue as a porous medium, but cannot characterize the interaction between drug carriers and cell membranes at the microscopic scale. Other studies use a coupled LBM-immersion boundary method (IBM)-finite element method (FEM) model to simulate cell deformation and transport in blood vessels, but this model only covers the cellular scale and part of the tissue scale, failing to account for both subcellular scales (such as the role of carrier proteins on the BBB endothelial cell membrane and changes in tight junction dynamics) and whole-brain tissue scales (such as the overall vascular network and the global flow of the glial lymphatic system). This scale fragmentation results in simulations that cannot fully reflect the multi-scale coupling mechanisms of brain substance transport, making it difficult to reveal the relationship between microscopic behavior and macroscopic function.
[0005] 3. The model lacks personalization and pathological specificity, resulting in insufficient simulation accuracy. Most existing simulation models are based on idealized geometry and homogenized parameter settings, failing to integrate real, heterogeneous patient pathological data and thus unable to simulate the differences in substance transport at specific pathological stages. In recent years, although some studies have constructed submicron resolution three-dimensional pathological atlases of mouse gliomas, revealing the "center-periphery" gradient of BBB disruption and the heterogeneity of vascular remodeling, how to transform this high-precision static anatomical data into dynamic, predictive computational models of substance transport functions remains a gap. This idealized modeling approach leads to significant deviations between simulation results and clinical reality, failing to meet the needs of personalized medicine and pathological mechanism research.
[0006] 4. Insufficient visualization and analysis tools limit clinical and research applications. Existing cerebral blood flow or metabolic maps are mostly generated based on statistical analysis of medical images, which are static analysis results and cannot provide interactive "digital twin" simulations and dynamic visualizations of hypothetical substance transport mechanisms. Furthermore, the lack of effective quantitative analysis tools prevents the accurate output of key biophysical parameters reflecting substance transport function (such as spatial permeability distribution, solute clearance efficiency, and vascular wall shear stress), making it difficult to directly compare simulation results with experimental observation data. This limits their application in areas such as pathological mechanism exploration, preoperative prediction, and drug screening.
[0007] In summary, current research methods for brain material transport (experimental observation and computational simulation) are significantly inadequate. Developing a computational simulation platform that can integrate multi-source pathological data, span multiple spatial and temporal scales, and achieve high-fidelity dynamic visualization and precise quantitative analysis has become a key issue that urgently needs to be addressed in this field, and has significant scientific significance and clinical translational potential. Summary of the Invention
[0008] This application provides a multi-scale visualization simulation method and apparatus for brain substance transport. This method and apparatus are suitable for high-fidelity multi-scale simulation and dynamic visualization of transmembrane transport of substances across the blood-brain barrier (BBB), and the hydrodynamics and solute clearance processes of the glial-lymphatic system within the brain parenchyma. This method addresses the technical problems of existing technologies, such as scale fragmentation, lack of pathological specificity, insufficient visualization capabilities, and low simulation accuracy. It achieves seamless coupling simulation from the subcellular scale to the tissue scale, integrates real pathological data to construct personalized models, and provides dynamic visualization and precise quantitative analysis functions, offering an efficient and accurate computational tool for brain disease mechanism research, drug development, and personalized medicine.
[0009] The technical solution provided in this application is as follows: Firstly, a multi-scale visualization simulation method for brain material transport is provided, including: Acquire multi-source datasets and construct a multi-scale computational grid that covers three scales: subcellular, cellular, and tissue, with heterogeneity. The flow of interstitial fluid and cerebrospinal fluid in porous media was simulated using LBM, and the motion and interaction of discrete phase particles at the BBB interface were simulated using DEM. Transmembrane coupling between LBM and DEM was achieved through bidirectional force transmission, and a deep learning surrogate model was associated with each DEM particle to calculate the transmembrane transport behavior of the particles in real time. Define the pulsatile pressure boundary at the arterial inlet, the pressure boundary at the venous outlet, and the impermeable boundary of the brain tissue; set pathological parameters spatially differentiated according to pathological characteristics. Parallel LBM-DEM coupled calculations and transmembrane transport behavior were performed to obtain computational data including real-time flow field data and particle data. The calculated data is rendered dynamically in real time to visualize the material transport process; key biophysical parameters are extracted, quantitatively analyzed, and a standardized simulation report is generated, which is then verified and calibrated using experimental data.
[0010] In one possible implementation, the multi-source dataset includes MRI data, DSA data, pathological slide images, tissue transparency imaging data, pathological data, and experimental data.
[0011] In one possible implementation, the LBM simulation of the flow of interstitial fluid and cerebrospinal fluid in a porous medium includes: modeling the brain parenchyma as a porous medium, using the D3Q19 lattice model, and numerically solving the unsteady Navier-Stokes equations for the interstitial fluid and cerebrospinal fluid using the BGK collision model to obtain high-precision flow field information across the entire domain. The mechanical parameters of the DEM particles include elasticity, adhesion, and coefficient of friction, which are set according to the actual characteristics of the discrete entity. The bidirectional force transfer is achieved through momentum exchange. Fluid forces drive the movement of DEM particles, and the particle reaction forces are fed back to the LBM flow field as momentum sources. A trilinear interpolation algorithm is used to achieve continuous global data interaction between the Euler LBM fixed grid and the Lagrange DEM free particles.
[0012] In one possible implementation, the deep learning agent model employs a CNN-LSTM hybrid architecture, using local microenvironment parameters as input, to predict the transmembrane transport probability, rate, and transport mode of particles in real time; the transport modes include endocytosis, carrier-mediated transport, passive diffusion, and tight junction pathways.
[0013] In one possible implementation, the spatially differentiated pathological parameters include: In the glioma simulation process, permeability is set according to the center-edge gradient of BBB disruption; In the Alzheimer's disease simulation, the driving force of the glial lymphatic system was weakened, reducing the efficiency of cerebrospinal fluid and tissue fluid exchange.
[0014] In one possible implementation, the computational data is dynamically rendered in real time to visualize the simulated material transport process; key biophysical parameters are extracted, quantitatively analyzed, and a standardized simulation report is generated, which is then verified and calibrated using experimental data, including: The calculated data is rendered dynamically in real time to obtain visualization results including flow field cloud map, particle motion trajectory, transmembrane transport heat map, and multi-scale structure fusion visualization map; The biophysical parameters of the visualization results were extracted and quantitatively analyzed to obtain quantitative results including spatial permeability, solute clearance efficiency, vascular wall shear force, drug concentration change curve, and transmembrane transport efficiency. The quantification results are compared with experimental data for verification and calibration.
[0015] Secondly, a multi-scale visualization simulation device for brain material transport is provided, comprising: The module is used to acquire multi-source datasets and construct a multi-scale computational grid that covers the three scales of subcellular, cellular, and tissue heterogeneity. The coupling module is used to simulate the flow of interstitial fluid and cerebrospinal fluid in porous media using LBM and to simulate the motion and interaction of discrete phase particles at the BBB interface using DEM. The transmembrane coupling of LBM and DEM is achieved through bidirectional force transmission, and a deep learning surrogate model is associated with each DEM particle to calculate the transmembrane transport behavior of the particles in real time. The settings module is used to set the pulsatile pressure boundary at the arterial inlet, the pressure boundary at the venous outlet, and the impermeable boundary of brain tissue; and to set pathological parameters spatially differentiated according to pathological characteristics. The execution module is used to perform LBM-DEM coupled calculations and transmembrane transport behavior in parallel, and obtain computational data including real-time flow field data and particle data. The output module is used to perform real-time dynamic rendering of the calculated data, visualize the material transport process, extract key biophysical parameters, perform quantitative analysis, generate standardized simulation reports, and verify and calibrate them through experimental data.
[0016] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the multi-scale brain material transport visualization simulation method as described in the first aspect.
[0017] Fourthly, a non-transitory computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the multi-scale brain material transport visualization simulation method as described in the first aspect.
[0018] Fifthly, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the multi-scale brain material transport visualization simulation method as described in the first aspect.
[0019] This application has the following beneficial effects: 1. Scientific research level: Providing new tools for studying brain substance transport mechanisms The method provided in this application realizes multi-scale coupled simulation from the subcellular to the tissue scale, which can accurately simulate the entire process of BBB transmembrane transport, glial lymphatic system fluid dynamics and solute clearance, revealing the relationship between microscopic mechanisms and macroscopic functions. By simulating changes in material transport under different pathological conditions (such as BBB destruction and glial lymphatic system dysfunction), it can provide fluid dynamic explanations for the pathogenesis research of diseases such as Alzheimer's disease and brain tumors, and help discover new pathological mechanisms. At the same time, it can serve as a "computational laboratory" to verify different scientific hypotheses, reduce experimental costs, and accelerate research progress.
[0020] 2. Clinical Applications: Facilitating Personalized Medicine and Surgical Planning The method provided in this application can integrate patient-specific multimodal data to construct personalized simulation models. It can predict the penetration and distribution of chemotherapy drugs and targeted drugs in the brain before surgery, and assist doctors in formulating personalized dosing plans (such as drug dosage and route of administration). For patients with brain tumors, it can simulate the material transport characteristics of the tumor area, assist in the formulation of precise surgical boundary planning, reduce surgical damage, and improve treatment effects. At the same time, it can evaluate the intervention effects of novel therapies such as focused ultrasound open BBB and cerebrospinal fluid shunt, provide data support for therapy optimization, and promote the development of precision medicine.
[0021] 3. Industrial Application Level: Accelerating the Development of Brain-Targeted Drugs and Bio-instruments The method provided in this application can be used for virtual screening and optimization of brain-targeted drug delivery systems. By simulating the efficiency of nanocarriers of different sizes, shapes, and surface modifications across specific pathological brain-body junctions (BBBs), the optimal drug carrier design scheme can be screened, significantly reducing the cost of in vitro and in vivo experiments and shortening the drug development cycle. At the same time, it can be used as an evaluation tool for novel biological instruments (such as brain drug delivery instruments and cerebrospinal fluid detection instruments), simulating changes in brain material transport under instrument intervention, assisting in the design and optimization of instruments, and promoting technological progress in related industries.
[0022] 4. Technological Level: Promoting interdisciplinary integration and technological innovation The method provided in this application integrates multidisciplinary technologies such as computational fluid dynamics (LBM, DEM), deep learning, medical image processing, pathological analysis, and computer visualization, and constructs a multidisciplinary computational simulation platform. The improved LBM segmentation algorithm, LBM-DEM bidirectional coupling mechanism, and deep learning surrogate model can not only be applied to brain material transport simulation, but also extended to material transport simulation of other biological tissues (such as liver and kidney), promoting technological innovation and development in the fields of computational biomedicine and biofluid dynamics. Attached Figure Description
[0023] Figure 1 A flowchart illustrating the multi-scale brain material transport visualization simulation method provided in this application embodiment; Figure 2 A schematic diagram of a multi-scale coupled computational grid provided for an embodiment of this application; wherein, a-macro-tissue-scale grid (whole brain vascular network), b-micro-cellular-scale grid (BBB interface and cellular region), c-subcellular-scale grid (cell membrane and carrier proteins), showing the hierarchical structure and grid connection relationship from macro to micro; Figure 3This is a schematic diagram of the principle of the LBM-DEM bidirectional coupling mechanism provided in the embodiments of this application; wherein, the left side is the LBM fluid grid (for calculating fluid velocity and pressure), the right side is the DEM particles (drug carrier), and the arrows indicate the transmission process of fluid action force and particle reaction force, realizing bidirectional coupling between fluid and particles; Figure 4 A flowchart of particle trans-BBB transport decision-making based on a deep learning surrogate model is provided for embodiments of this application; it shows the entire process by which the surrogate model predicts the probability, rate and mode of particle transmembrane transport based on local microenvironment parameters (pH, shear force, acceptor density, etc.); Figure 5 This is a visualization example of simulation results for high-grade glioma provided in the embodiments of this application; wherein, a-the distribution of nanomedicine in the traditional homogeneous BBB model, b-the distribution of nanomedicine in the heterogeneous pathological model of the present invention, and c-the heatmap of the distribution difference between the two, clearly showing that the model of the present invention can more accurately simulate the heterogeneous distribution of drugs in the tumor region. Figure 6 This is a comparison diagram of the accumulation of metabolic waste particles in the brain parenchyma before and after the weakening of the glial lymphatic system function, provided for the embodiments of this application; wherein, a-waste distribution when the glial lymphatic system function is normal, b-waste distribution when the glial lymphatic system function is weakened, and c-waste accumulation amount changes over time, intuitively showing the influence of glial lymphatic system function on waste clearance. Figure 7 A schematic flowchart of the multi-scale brain material transport visualization simulation device provided in the embodiments of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0025] In the description of this application, 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 indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.
[0026] In the description of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "set up," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this technology based on the specific circumstances.
[0027] In the description of this application, spatial relation terms such as "below," "under," "below," "below," "above," "over," etc., are used herein to describe the relationship between one element or feature shown in the figures and other elements or features. It should be understood that, in addition to the orientation shown in the figures, spatial relation terms also include different orientations of the device in use and operation. For example, if the device in the figures is flipped, an element or feature described as "below" or "under" or "below" of other elements or features will be oriented "above" other elements or features. Therefore, the exemplary terms "below" and "under" can include both upper and lower orientations. Furthermore, the device may also include other orientations (e.g., rotated 90 degrees or other orientations), and the spatial descriptive terms used herein are interpreted accordingly.
[0028] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.
[0029] Existing visualization simulation methods for the transmembrane transport of substances across the blood-brain barrier (BBB), the hydrodynamics of the glial lymphatic system in the brain parenchyma, and solute clearance processes suffer from technical problems such as scale fragmentation, lack of pathological specificity, insufficient visualization capabilities, and low simulation accuracy.
[0030] In view of this, this application provides a method and apparatus for visual simulation of multi-scale brain material transport.
[0031] This application provides a multi-scale visualization simulation method for brain material transport, comprising: S100. Obtain multi-source datasets and construct a multi-scale computational grid that covers the three scales of subcellular, cellular, and tissue heterogeneity. S200 uses LBM to simulate the flow of interstitial fluid and cerebrospinal fluid in porous media, and DEM to simulate the motion and interaction of discrete phase particles at the BBB interface; the transmembrane coupling of LBM and DEM is realized through bidirectional force transmission, and a deep learning surrogate model is associated with each DEM particle to calculate the transmembrane transport behavior of the particles in real time. S300: Set the pulsating pressure boundary at the arterial inlet, the pressure boundary at the venous outlet, and the impermeable boundary of the brain tissue; set pathological parameters spatially differentiated according to pathological characteristics. S400 performs parallel LBM-DEM coupled calculations and transmembrane transport behavior calculations to obtain computational data including real-time flow field data and particle data; S500: The calculated data is rendered dynamically in real time to visualize the material transport process; key biophysical parameters are extracted, quantitative analysis is performed, a standardized simulation report is generated, and the simulation is verified and calibrated using experimental data.
[0032] In one possible implementation, in S100, the multi-source dataset includes MRI data (magnetic resonance imaging data), DSA data (digital subtraction angiography data), pathological slide images, tissue transparency imaging data, pathological data, and experimental data.
[0033] Furthermore, the experimental data includes: in vitro cell experimental data, nanomedicine transport across the blood-brain barrier experimental data, nanocarrier physical property characterization data, tumor-bearing animal brain in vivo detection data, and glial lymphatic waste clearance control experimental data.
[0034] Furthermore, the pathological data are all purely numerical data and do not contain image structure information. Specifically, they include tumor location, tumor size, pathological grade (disease degree), tumor morphological boundary, lesion extent, and BBB rupture rate.
[0035] The parameters are defined as follows: Tumor location: The three-dimensional spatial coordinates corresponding to the tumor centroid, used to accurately determine the spatial location of the tumor in the unified coordinate system of the whole brain.
[0036] Tumor size: This refers to the quantitative dimensions of the tumor, such as its three-dimensional diameter and total volume, used to characterize the macroscopic size of the tumor.
[0037] Pathological grading (disease severity): This is a quantitative grade value assigned according to the clinical WHO grading standard, used to characterize the malignancy and progression of the tumor; in this example, high-grade glioma corresponds to WHO grade IV, with a quantitative value of 4.
[0038] Tumor morphological boundary: This is a set of three-dimensional discrete coordinate points corresponding to the tumor outline, stored in the form of a standardized numerical array. It is used to accurately delineate the geometric boundary outline of the tumor, supporting subsequent pathological zoning and parameter differentiation assignment.
[0039] Lesion extent: This refers to the volume percentage occupied by the tumor lesion area, the maximum infiltration radius, and other quantitative statistical values used to characterize the spatial coverage and infiltration extent of the tumor lesion.
[0040] BBB rupture rate: A quantitative value representing the percentage of volume / area of blood-brain barrier rupture in different regions of the tumor, used to characterize the degree of damage to the BBB barrier function.
[0041] In one possible implementation, S100 includes: S101. Obtain multi-source datasets; S102. Denoise and enhance the image data (MRI data, DSA data, pathological slide images and tissue transparency imaging data), and spatially align it with the numerical data (pathological data) and normalize it to a unified coordinate system. S103. Perform image segmentation on the preprocessed MRI and DSA data to extract the three-dimensional geometric structure of the whole brain vascular network, brain region contours, and tumor regions; and eliminate segmentation errors by manually correcting the segmentation results to ensure the segmentation accuracy of the vascular network (especially the malformed blood vessels in the tumor region); and obtain segmentation results that include macroscopic vascular and brain region structures. S104. The segmentation results are fused with pathological data. Taking the tumor lesion as the core, the key regions of interest within the tumor area and the surrounding preset range are finely reconstructed in three dimensions to generate a cell-level micro-geometric structure with sub-micron precision. The characteristic microstructure of glioma is accurately restored, including large-diameter malformed blood vessels, vascular co-selected regions, tumor cell clusters, astrocyte terminale, and perivascular cells. S105. Construct computational grids at the macroscopic tissue scale, microscopic cell scale, and subcellular scale. Connect the grids at each scale through grid mapping to ensure smooth data interaction.
[0042] Furthermore, in step S102, image data and numerical data are spatially aligned using a rigid registration algorithm, uniformly normalized to the same three-dimensional coordinate system, eliminating the spatial offset problem of multi-source data, and achieving accurate matching of structural information and pathological numerical information.
[0043] Furthermore, in S103, the image segmentation method is an improved geometric active contour model or GAC model based on a fully parallelized LBM.
[0044] Furthermore, the improved geometric active contour model based on fully parallelized LBM is obtained by making targeted improvements to the traditional GAC (Geometric Active Contour) model in three aspects: computational architecture, contour evolution criteria, and convergence determination mechanism. The specific improvement methods are as follows: 1. Improved computational architecture: The traditional GAC model's pixel-by-pixel serial computation mode is replaced by a fully parallelized LBM lattice iterative computation architecture. The image domain is discretized into uniform lattice units, and the synchronous iterative computation of all pixels in the image is achieved by relying on the natural parallel characteristics of LBM particle collision migration, thus solving the problem of low efficiency in segmenting large-size medical images by traditional serial algorithms.
[0045] 2. Improved contour evolution criteria: The single image gradient driving rule of the traditional GAC model is replaced by a composite driving potential field of gray-level features and texture features; at the same time, the contour curvature constraint coefficient is adaptively adjusted; the curvature constraint is weakened in the small blood vessels of tumors and blurred boundary areas to preserve fine structure, while the curvature constraint is strengthened in the background area to suppress noise artifacts, which significantly improves the segmentation accuracy of complex pathological structures.
[0046] It should be noted that, based on the improved contour evolution criteria, curvature constraints are weakened in thin-walled and fine-branched regions of blood vessels, preserving delicate vascular structures; while curvature constraints are strengthened in the brain parenchyma background region to suppress noise-induced false edges, effectively improving the segmentation accuracy of complex pathological vascular structures. This step can address the pain points of blurred edges, weak gradients, and easy over-segmentation / under-segmentation of tumor-derived malformed blood vessels and fine vascular branches in brain MRI and DSA images.
[0047] 3. Improved convergence determination mechanism: The traditional GAC model's fixed iteration number convergence method is replaced by the LBM iterative residual adaptive convergence rule. The contour convergence status is determined in real time based on the gradient change of the target boundary and the iteration is automatically terminated, avoiding over-segmentation and under-segmentation, and adapting to the accurate segmentation of fuzzy boundary structures such as glioma malformed blood vessels.
[0048] It should be noted that the improved convergence determination mechanism replaces the fixed number of iterations or single global residual convergence method of the traditional GAC model with a dual-condition joint convergence control strategy for hemodynamic simulation. It relies on the LBM iterative residual to determine the overall evolution state of the contour in real time, and combines the gradient change of the blood vessel boundary to verify the detail accuracy of key areas. It adaptively terminates the iteration only when the global evolution and the local boundary simultaneously meet the convergence conditions, thus balancing the efficiency of segmentation calculation and the physical accuracy of the boundary.
[0049] Specifically, the improvement methods include: 1. Construction of dual convergence criteria (1) Macro-global convergence condition (LBM iterative residual) The L2 norm relative residual of the level set function after each iteration is used as the macroscopic residual index of the LBM iteration to characterize the stability of the overall profile evolution. The calculation formula is:
[0050] In the formula, For the first The symbolic distance function of the level set in step iteration; when At that time, the global evolution of the determination contour reaches macroscopic stability.
[0051] (2) Convergence conditions of boundary physical quantities (gradient change at the blood vessel boundary) Extract key target regions such as the vessel wall and tumor boundary corresponding to the current evolution profile, calculate the mean gradient of the composite driving potential field on the boundary line (corresponding to the core physical feature of wall shear stress in subsequent blood flow simulation), and use the relative change between adjacent iterations as the boundary convergence index. The calculation formula is:
[0052] In the formula, For the first The average gradient of the composite potential field along the contour line in step iteration; when At that time, the physical characteristics of the critical boundary are determined to be stable in detail.
[0053] 2. Adaptive Termination of Iteration Execution Rules A strict convergence logic that simultaneously satisfies two conditions is adopted, which does not rely on a single global residual threshold. It combines the overall convergence of the flow field with the convergence of key boundary physical quantities. The termination rule is as follows: (1) The iteration is officially terminated only when the LBM macroscopic iteration residual and the change of physical quantity of blood vessel boundary simultaneously satisfy their respective convergence thresholds; (2) If only the macroscopic residual meets the standard but the boundary physical quantities still fluctuate, it indicates that the overall profile has become stable, but key boundary details such as the blood vessel wall have not yet converged. Continue to iterate and optimize the local boundary accuracy. (3) If the boundary physical quantities are stable but the global residual does not meet the standard, it means that the local boundary has been formed but the overall contour evolution is insufficient. Continue iterating until the global shape is stable.
[0054] 3. Exemplary parameters and application effects Taking the blood vessel segmentation scenario in brain DSA imaging as an example, a preset macroscopic residual threshold is used. Boundary physical quantity threshold .
[0055] Compared to the traditional GAC model with a fixed 200 iterations, this adaptive convergence strategy can achieve accurate segmentation in an average of 120-150 iterations. This avoids the problems of boundary ambiguity and undersegmentation caused by insufficient iterations, while also eliminating the computational redundancy and oversegmentation risks associated with excessive iterations. Furthermore, since the convergence determination is directly coupled with the boundary gradient, a key physical quantity in blood flow simulation, the segmented vascular geometry model has stronger physical consistency with subsequent LBM hemodynamic calculations, effectively reducing boundary errors in subsequent simulations.
[0056] The following section details the composite driving potential field and the adaptive adjustment of the profile curvature constraint coefficient.
[0057] Specifically, the composite driving potential field is constructed by taking a single preprocessed medical image as input and performing feature extraction, normalization, and adaptive weighted fusion in three steps. The construction method is as follows: 1. Local grayscale feature extraction and normalization (1) For the input MRI / DSA grayscale image, use The sliding neighborhood window traverses the entire image, calculating the local gray-level mean, gray-level variance, and gray-level gradient magnitude for each pixel. (2) Perform global maximum-minimum normalization on the grayscale gradient image, and map the gradient magnitude to... Range, eliminating grayscale value differences between different imaging devices; (3) Substitute the grayscale edge stopping term formula to generate the grayscale feature edge stopping map of the whole image. This completes the quantization and edge mapping of grayscale features.
[0058] 2. Local texture feature extraction and normalization (1) Use uniform size A sliding window is used to calculate the local texture coding value of each pixel using the uniform mode local binary mode (LBP) algorithm, generating an initial texture energy map; (2) Perform Gaussian smoothing to denoise the initial texture energy map, and then calculate the gradient magnitude of the texture energy map to obtain the texture gradient distribution map; (3) Perform normalization processing on the texture gradient map in the same way as the grayscale gradient, substitute it into the texture edge stopping term formula, and generate the full-image texture feature edge stopping map. This completes the quantization of texture features and edge mapping.
[0059] 3. Adaptive weighted fusion of dual features to generate a composite driving potential field. (1) Calculate the global gray-level gradient map of the original input image, and use the local gradient intensity as the basis for weight allocation to calculate the gray-level feature weights pixel by pixel. Texture feature weights ; (2) According to the composite driving potential field fusion formula, the gray-scale edge stopping map and the texture edge stopping map under the same coordinate are weighted and summed at the pixel level to obtain the initial composite edge stopping map; (3) Perform the initial composite diagram Gaussian smoothing eliminates local numerical jumps caused by feature fusion, ultimately resulting in a smooth and continuous composite driving potential field, which directly serves as the driving core for contour evolution.
[0060] 4. Linkage between composite potential field and adaptive curvature constraint Using the pixel values of the composite driving potential field as the basis for edge confidence, the adaptive curvature constraint coefficient is calculated pixel by pixel. By substituting the composite potential field and curvature constraints into the level set evolution equation, a complete improved contour evolution criterion is formed, replacing the single gradient-driven evolution rule of the original GAC model.
[0061] Specifically, the composite driving potential field uses the single gradient edge stopping function of the traditional GAC model as an improved basis, and is constructed by adaptive weighted fusion of local gray-level feature terms and local texture feature terms to replace the original single image gradient driving term. At the same time, it is linked with the adaptive curvature constraint coefficient to jointly form a complete contour evolution criterion.
[0062] Specifically, the relevant formula is as follows: 1. Basic Definitions and Expressions for Single Characteristic Terms Assume the input medical image is a two-dimensional grayscale image. , These are the image pixel space coordinates; centered on the pixel. The neighborhood window is the feature extraction unit.
[0063] Local grayscale feature term: Characterizes the grayscale difference between the target area and the background area, extracts local grayscale gradient and grayscale contrast information, and generates grayscale edge stopping term after normalization:
[0064] in, This represents the local grayscale gradient magnitude. This is the grayscale feature adjustment coefficient, used to control the sensitivity of the grayscale gradient; The range of values is The smaller the value, the higher the marginal confidence.
[0065] Local texture features: Characterizing the differences in microscopic texture structure of tumor blood vessels and lesion areas, extracting local texture variance and texture gradient information, and generating texture edge stopping terms after normalization.
[0066] in, This is a local texture energy map calculated using Local Binary Pattern (LBP). The gradient magnitude of the texture energy map; This is the texture feature adjustment coefficient, used to control the weight ratio of texture features.
[0067] 2. Expression of the composite driving potential field with dual-feature fusion The grayscale feature term and the texture feature term are weighted and fused by adaptive dynamic weights to generate a composite edge stopping function (i.e., a composite driving potential core). The fusion formula is as follows:
[0068]
[0069] The adaptive weight calculation rule is as follows:
[0070] In the formula The minimum regularization constant (values) ( ), to prevent the denominator from being zero.
[0071] 3. Complete Evolution Criteria for Linked Adaptive Curvature Constraints Substituting the composite driving potential field into the GAC level set evolution equation, and replacing the original fixed curvature coefficients with spatial adaptive curvature constraint coefficients, the final improved profile evolution equation is as follows:
[0072] in: The sign distance function for the level set; For the contour curvature; The contour expansion coefficient; The adaptive curvature constraint coefficient satisfies In regions with high edge confidence, the curvature constraint is automatically weakened and the fine structure is preserved; in regions with background noise, the curvature constraint is automatically strengthened and false edges are suppressed. for The gradient magnitude.
[0073] Specifically, the adaptive adjustment of the contour curvature constraint coefficients is based on the pixel-level edge confidence output by the composite driving potential field. It achieves spatial adaptive dynamic adjustment of the curvature constraint strength through a linear inverse mapping rule, replacing the globally fixed curvature coefficients in the traditional GAC model, while simultaneously ensuring both fine structure preservation and background noise suppression. The specific construction method is as follows: 1. Core mathematical expressions Adaptive curvature constraint coefficient Using the composite driving potential field as the mapping reference, the calculation is performed pixel-by-pixel using linear interpolation, and the expression is:
[0074] In the formula: The minimum curvature constraint coefficient (lower limit of curvature constraint) is preset, corresponding to the region with the highest edge confidence, and is usually taken as 0.1~0.3; The maximum curvature constraint coefficient (curvature constraint upper limit) corresponds to a homogeneous background region without edges, typically with a value of 1.0 to 1.5; This represents the composite driving potential value for the corresponding pixel, with a value range of [value range missing]. The smaller the value, the higher the confidence that the location is the edge of the real target.
[0075] 2. Step-by-step construction and adaptive adjustment process Step 1: Preset global upper and lower limits for curvature constraints Based on the modal type and target structural features of the image to be segmented, upper and lower thresholds for global curvature constraints are pre-set. For brain MRI / DSA vascular and tumor segmentation scenarios, [the following is taken]: , While ensuring that the delicate structures such as malformed blood vessels and fine branches of the tumor are not lost, it effectively suppresses background noise and artifact interference in brain tissue.
[0076] Step 2: Establish edge confidence-curvature intensity mapping rules Establish an inverse linear mapping rule where "higher edge confidence corresponds to weaker curvature constraints, and lower edge confidence corresponds to stronger curvature constraints": (1) When the pixel is the edge of the real target, the composite potential field value is low and the edge confidence is high. The curvature constraint coefficient is automatically reduced, allowing the contour to deform with the fine structure of the edge. (2) When the pixel is a homogeneous background region, the composite potential field value is high and the edge confidence is low. The curvature constraint coefficient is automatically adjusted upward, and the contour is forced to remain smooth to filter out false edges.
[0077] Using the above linear interpolation formula, The composite potential values of the interval are mapped one by one to The curvature constraint coefficient value of the interval.
[0078] Step 3: Calculate and generate adaptive curvature coefficient field pixel by pixel By traversing all pixels in the image, and based on the composite driving potential field value corresponding to each pixel, the curvature constraint coefficient at that location is calculated by substituting it into the mapping formula point by point, thereby generating an adaptive curvature coefficient field that is continuous throughout the image space, and realizing differentiated and refined configuration of curvature constraint strength at different spatial locations.
[0079] Step 4: Embedding evolution equations to complete rule replacement The generated adaptive curvature coefficient field is substituted into the GAC level set evolution equation to replace the global fixed curvature coefficient in the original model. It works in conjunction with the composite driving potential field to drive the contour evolution, thus completing the overall improvement of the evolution criterion.
[0080] 3. Specific mechanisms for regional adaptive adjustment (1) True marginal region (regions with clear gradients such as blood vessel walls and tumor boundaries): Approaching 0, Automatically descend Near the edge, the curvature constraint is weakest, allowing the contour to bend and deform with the fine structure of the edge, avoiding excessive smoothing that could lead to the loss of small blood vessel branches and tumor lobulation structures.
[0081] (2) Homogeneous background region (normal brain parenchyma and cerebrospinal fluid region): Approaching 1, Automatically upgraded Near the edge, the curvature constraint is strongest, forcing the contour to maintain a smooth shape, effectively filtering out false edges caused by local noise and artifacts, and avoiding irregular oscillations in the contour.
[0082] (3) Transitional fuzzy region (weak gradient region such as tumor infiltration edge and distal microvessel): The curvature coefficient transitions smoothly and continuously with the edge confidence, which suppresses local noise interference while preserving the weak edge structure, thus solving the inherent contradiction of the traditional fixed curvature coefficient: "preserving details results in more noise, while suppressing noise results in losing details".
[0083] In summary, the improved geometric active contour model based on fully parallelized LBM balances the segmentation efficiency of large-size medical images with the segmentation accuracy of complex pathological microstructures, and solves the technical defects of traditional general geometric active contour models in brain tumor vascular segmentation, such as low efficiency, loss of details, and boundary distortion.
[0084] Furthermore, in S103, for complex and easily distorted structures such as malformed blood vessels and small blood vessel branches in tumors, the segmentation results are manually corrected to eliminate automatic segmentation errors, ensuring the segmentation accuracy of the fine vascular network in the tumor region and providing a precise structural basis for subsequent three-dimensional reconstruction.
[0085] Furthermore, in S104, the segmentation results and pathological data are fused using a multi-dimensional spatial-attribute coupling fusion method, which specifically includes a three-step precise fusion process: spatial registration fusion, structural semantic fusion, and pathological parameter heterogeneity assignment fusion, to achieve precise binding between macroscopic anatomical structures and microscopic pathological features.
[0086] Specifically, in S104, fusing the segmentation result with the pathological data includes: (1) High-precision spatial registration and fusion Using the macroscopic geometric structures of the whole brain vessels and brain regions obtained from MRI and DSA image segmentation as the benchmark template, the iterative nearest point (ICP) rigid registration algorithm based on feature points is adopted to complete the sub-pixel-level spatial alignment of the microscopic high-resolution structural images corresponding to pathological sections and tissue transparency imaging with the macroscopic segmented structures. Through global coordinate transformation and local micro-offset correction, the spatial misalignment, scale deviation and perspective deviation of multi-source structural data are eliminated, ensuring that the microscopic structures of tumor malformed blood vessels, cell clusters and BBB are completely superimposed with the macroscopic brain regions and vascular networks in three-dimensional space, realizing the spatial integration of macroscopic anatomical structures and microscopic pathological structures.
[0087] (2) Structural semantic tag fusion Based on information about tumor location, morphological boundaries, and lesion extent from pathological data, semantic labels are used to reclassify and finely annotate the segmented general vascular and brain tissue structures. For the registered 3D structures, the system automatically distinguishes between tumor core areas, tumor infiltration margin areas, normal brain tissue areas, malformed vascular areas, co-selected vascular areas, intact BBB areas, and ruptured BBB areas, assigning corresponding pathological semantic attributes to different structures. This step addresses the problems of traditional structural segmentation lacking pathological semantics and the disconnect between structural and lesion information, achieving semantic fusion of "geometric structure + pathological zoning."
[0088] (3) Heterogeneity of pathological parameters assignment and fusion By combining quantitative numerical indicators such as tumor size, pathological grade, and BBB rupture rate from pathological data, a spatial Gaussian decay interpolation mapping method is used to transform discrete pathological quantitative parameters into a continuous spatial distribution field and bind them to the corresponding 3D reconstructed structure. Based on the pathological characteristics of gliomas, a "center-periphery" gradient assignment rule is constructed: the tumor core region is matched with pathological parameters of high rupture rate, high porosity, and high permeability; the tumor infiltration periphery region is matched with gradient decay pathological parameters; and normal brain regions are matched with physiological standard parameters. This step achieves deep coupling and fusion of geometric morphology, spatial distribution, and pathological quantitative attributes, providing a precise data foundation for subsequent multi-scale differentiated mesh construction and pathological parameter setting.
[0089] Furthermore, in S105, the computational grid 3D Slicer and TetGen are coupled with MeshLab grid optimization software and combined with an autonomous grid mapping algorithm to construct the grid; wherein, the macroscopic tissue scale grid resolution is 100μm, covering the whole brain; the microscopic cell scale grid resolution is 1μm, covering the tumor and surrounding area; and the subcellular scale grid resolution is 100nm, covering the BBB interface.
[0090] Furthermore, in S105, the method for constructing the computational grid includes: (1) Using the open-source medical image modeling platform 3D Slicer, multi-scale geometric model preprocessing was completed. The macroscopic whole brain blood vessels, microscopic cells and BBB fine three-dimensional models reconstructed by S104 were imported to complete model topology repair, hole filling and redundant structure removal to ensure the geometric integrity and topological continuity of the model and adapt to the subsequent mesh generation requirements.
[0091] (2) The high-precision tetrahedral mesh generation software TetGen was used to complete the layered mesh division and to match the simulation requirements of different scales: sparse uniform tetrahedral meshes were generated for the macroscopic regions of the whole brain tissue to construct the macroscopic tissue scale calculation mesh; mesh densification was carried out for key fine regions such as tumor region, BBB interface, and microvessels, and non-uniform fine meshes were generated adaptively to construct microscopic cell scale meshes and subcellular scale meshes respectively, taking into account both the global simulation efficiency and the microscopic transmembrane simulation accuracy.
[0092] (3) Import the MeshLab software to complete mesh smoothing, quality optimization and distortion cell removal, correct distorted meshes, reduce mesh distortion rate, and ensure the regularity and computational convergence of mesh cells at each scale; at the same time, carry out the self-developed high-precision mesh mapping algorithm to establish the spatial mapping relationship and data interaction interface of macroscopic, microscopic and subcellular three-layer meshes, realize the seamless connection of meshes of different scales, real-time parameter transmission and scale transition coupling, completely solve the problem of multi-scale mesh data fragmentation and poor simulation synchronization, and ensure that the multi-scale coupled simulation process is smooth and the data interaction is accurate and distortion-free.
[0093] In one possible implementation, S200 includes: S201, LBM fluid domain modeling: The brain parenchyma is modeled as a porous medium. The D3Q19 lattice model is used, and the unsteady Navier-Stokes equations of the interstitial fluid and cerebrospinal fluid in the brain tissue are numerically solved through the BGK collision model to obtain high-precision flow field information of the entire domain.
[0094] S202, DEM Discrete Phase Modeling: The discrete transport entity is modeled as DEM particles, and the particles are given realistic physical property parameters such as mechanics, adhesion, and size. The particle translation, collision, adhesion and boundary interaction behavior are simulated based on Newton's dynamic equations.
[0095] S203. Two-way coupling mechanism setting: The two-way force transmission between LBM and DEM is realized through the momentum exchange formula; the trilinear interpolation algorithm is used to realize the global continuous data interaction between the Euler LBM fixed mesh and the Lagrange DEM free particles, ensuring the coupling stability.
[0096] S204. Deep Learning Proxy Model Association: A proxy model is constructed using a CNN-LSTM hybrid architecture. Local microenvironment parameters are used as input to predict the transmembrane transport probability, rate, and transport mode of particles in real time. The transport modes include endocytosis, carrier-mediated transport, passive diffusion, and tight junction pathways.
[0097] Furthermore, in S201, the expression for the D3Q19 lattice model is:
[0098] In the formula: These are the discrete velocity vectors in each direction corresponding to the D3Q19 model. i The index number (number) represents the direction of discrete velocity, covering the stationary direction in three-dimensional space, the forward and reverse axial directions, and the combined oblique directions, which can fully meet the three-dimensional flow solution requirements of low-speed viscous fluid in the brain.
[0099] Furthermore, in S201, the expression for the BGK collision model is the evolution equation of the single relaxed particle distribution function, as follows:
[0100] In the formula: Spatial position at time t The particle distribution function at the i-th direction; This represents the equilibrium distribution function in the corresponding direction; The dimensionless relaxation time determines the viscous relaxation characteristics of the fluid; To calculate the time step, the model achieves a stable evolution of fluid particle collisions and relaxation through a single relaxation rate.
[0101] Furthermore, in S201, the unsteady Navier-Stokes equations include a continuity equation and a momentum equation, adapted to the transient flow scenario of porous media in the brain, and the specific expressions are as follows: Continuity equation:
[0102] Momentum equation:
[0103] In the formula: Interstitial fluid density; For fluid velocity vector; For time; For flow field pressure; For fluid dynamic viscosity; It is the vector of gravitational acceleration; This is a resistance source term for porous media, used to characterize the resistance effect of brain tissue and tumor tissue on fluids, and its dynamic changes with spatial porosity and permeability.
[0104] Furthermore, in S201, the physical quantities of the flow field include three categories: fluid kinematic parameters, dynamic parameters, and medium property parameters, specifically: fluid velocity vector, fluid pressure, fluid density, fluid kinematic viscosity, local porosity, local permeability, flow field shear stress, flow flux, pressure gradient, and velocity gradient. These physical quantities comprehensively characterize the transient flow characteristics and spatial heterogeneity distribution of the porous fluid in the brain.
[0105] Furthermore, in S202, the discrete transport entity includes drug particles.
[0106] Furthermore, S203 includes: S203a, Bidirectional force transfer mechanism based on momentum exchange formula This step relies on the strict principle of momentum conservation to construct a closed-loop two-way force transmission system of "fluid driving particles and particles counteracting the flow field". It abandons the simplified approximation mode of traditional simulation that only considers unidirectional driving and fully conforms to the real physical mechanism of dynamic mutual disturbance between fluid and particles in the body.
[0107] (1) Fluid → Particle positive force transmission: After obtaining the global flow field parameters by solving the LBM, an accurate fluid drag force formula is constructed based on the difference of the distribution function of the lattice model. The dynamic force of the fluid on the DEM particles at any position is calculated in real time and directly coupled to the DEM particle dynamic equation to drive the particle motion evolution.
[0108] The specific formula is as follows:
[0109] In the formula: The driving force for fluid coupling to DEM particles; This is the volume weighting coefficient of the particles within the LBM mesh element; This represents the three-dimensional spatial coordinate vector of the DEM particle centroid within the LBM Euler computational domain. It is used to locate the mesh cell to which the particle belongs and serves as the spatial reference for calculating the volume weight coefficient and flow field interpolation. The remaining parameters correspond one-to-one with the parameters of the D3Q19 lattice model. This method can dynamically correct the force according to the spatial heterogeneity of the tumor region porosity, the degree of BBB damage, and the vascular shear force, accurately adapting to the non-uniform flow field characteristics of the pathological microenvironment.
[0110] (2) Particle → Fluid reverse force feedback: Based on the law of conservation of momentum, the moving, colliding and adhering DEM particles generate an equivalent reverse force on the local fluid, which is transformed into the dynamic momentum source term of the LBM evolution equation, and the local flow field velocity and pressure distribution are corrected in real time to achieve closed-loop coupling control.
[0111] The revised BGK evolution equation and source term formula are as follows:
[0112]
[0113]
[0114] In the formula: For LBM mesh cell volume, This mechanism represents the reverse perturbation force exerted by particles on the fluid. It can accurately reproduce the flow field disturbance phenomena in tumor vascular malformations and particle-rich regions, eliminating the false results of traditional simulation flow field homogenization.
[0115] Understandably, the force transmission mechanism in S203a plays a key role and has the following advantages: it achieves physical closed-loop coupling between fluid and particles, without empirical parameter fitting, resulting in extremely high physical realism; it can differentiate and adapt the force transmission patterns of tumor core, infiltrative margin, and normal brain tissue, restoring the heterogeneous transport characteristics of pathology; compared with traditional unidirectional coupling models, it significantly reduces multi-scale simulation errors and significantly improves simulation fidelity.
[0116] S203b, Cross-scale data interaction mechanism based on trilinear interpolation algorithm: Using the three-dimensional spatial coordinates of any moving DEM particle as a reference, an interpolation unit is constructed by selecting eight adjacent LBM grid nodes of the cubic cell containing the particle. Based on the known flow field physical quantities of the grid nodes, the accurate and continuous microenvironment parameters of any non-grid location of the particle are solved through three-dimensional weighted interpolation. At the same time, reverse interpolation is supported, which maps the microscopic parameters such as the motion state, adhesion state, and drug loading of the DEM particle to the corresponding LBM grid cell, realizing bidirectional real-time synchronization between the macroscopic flow field and the microscopic particle state.
[0117] The core formula is as follows:
[0118] In the formula: For interpolating particle positions; The original discrete calculated values of the same flow field physical quantity are located on the 8 neighboring grid nodes of the cubic grid cell to which the particle belongs. The normalized relative coordinates of the particles within the grid cell.
[0119] Understandably, S203b addresses the inherent contradiction between the spatial location mismatch and scale fragmentation of LBM fixed Euler mesh and DEM free Lagrange particles by employing a three-dimensional trilinear interpolation algorithm to achieve global, bidirectional, and continuous data interaction, thus resolving the problems of parameter jumps, numerical oscillations, and loss of microscopic details caused by traditional nearest-neighbor value taking and mean mapping.
[0120] The core function of data interaction is to break down the data barriers between the macroscopic fluid scale of LBM and the microscopic particle scale of DEM, and to achieve continuous discrete grid data; to provide real-time and accurate local microenvironment parameters for dynamically moving particles, supporting high-precision cross-membrane prediction by deep learning surrogate models; and to ensure that multi-scale coupled calculations are continuous, without abrupt changes or oscillations throughout the entire process.
[0121] Algorithm advantages and effects: Global continuous interpolation completely eliminates grid discretization errors and significantly improves the stability of coupled calculations; it accurately preserves the details of local parameter mutations in the BBB interface and tumor microvessels, adapting to subcellular level fine simulation; it achieves true multi-scale linkage of "macro-flow field controlling micro-particles and micro-behavior feeding back to the macro-flow field", and the data matching accuracy is significantly improved compared with traditional coupling methods, solving the core defects of existing technologies such as scale fragmentation, simulation distortion, and poor stability from the bottom up.
[0122] Further, see Figure 4 In S204, the CNN-LSTM hybrid architecture comprises three core layers connected in series: a multi-layer CNN feature extraction layer, a CNN spatial pattern recognition layer, and an LSTM temporal analysis layer. This allows for the simultaneous mining of spatial features and analysis of temporal dynamic patterns of brain microenvironment parameters, precisely adapting to the spatiotemporal coupling changes in particle transmembrane transport. The specific detailed structure and functions of each layer are as follows: (1) Multi-layer CNN feature extraction layer This layer serves as the foundational pre-model layer, consisting of three stacked 2D convolutional layers, two pooling layers, and a normalized activation layer. Its core function is to automatically extract shallow and mid-level features and optimize noise reduction for the multi-dimensional microenvironment parameters. Specifically, the first convolutional layer uses a 3×3 kernel and 32 channels to perform preliminary feature mapping on four input parameters: pH, blood vessel wall shear force, endothelial cell receptor density, and BBB permeability, preserving effective feature information of the original microenvironment parameters. The second convolutional layer uses a 3×3 kernel and 64 channels to further refine the initial features, strengthening the fundamental correlations between parameters. The third convolutional layer uses a 1×1 kernel and 128 channels to achieve feature dimensionality reduction and feature fusion, eliminating redundant and noisy features. Max pooling layers are used between adjacent convolutional layers to perform feature compression, reducing model computation. Simultaneously, ReLU activation layers perform nonlinear transformations, addressing the issue of insufficient linear fitting ability and providing a high-quality feature dataset for subsequent spatial pattern recognition.
[0123] (2) CNN spatial pattern recognition layer This layer is the core layer for spatial feature analysis of the model. It inherits the fusion features from the feature extraction layer and consists of a feature fusion convolutional layer, a spatial attention layer, and a global pooling layer, specifically designed for spatial pattern recognition. Specifically, spatial pattern recognition refers to identifying the distribution patterns, parameter coupling correlations, and spatial differences of microenvironmental parameters at different spatial locations, based on the spatial heterogeneity of the brain's BBB interface. It accurately captures the differentiated combinations of microenvironmental parameters in different regions of the tumor core, tumor infiltration margin, and normal brain tissue, as well as the corresponding mapping relationship between these spatial distribution patterns and particle transmembrane transport methods and efficiency. This layer automatically weights and strengthens the feature weights of key sensitive regions of the BBB interface through a spatial attention mechanism, weakens invalid background features, accurately identifies the core microenvironmental factors that dominate transmembrane transport at different spatial locations, and outputs a standardized spatial feature vector, achieving precise matching of "spatial location - microenvironmental parameters - transmembrane transport features."
[0124] (3) LSTM timing analysis layer This layer is the core layer for dynamic temporal analysis of the model. It consists of two stacked LSTM memory units, a dropout regularization layer, and a temporally fully connected layer. It is used to receive the static spatial features output by the CNN and to mine the temporal patterns of the dynamic evolution of microenvironment parameters over simulation time. Each of the two LSTM units has 128 hidden neurons. Through built-in input gate, forget gate, and output gate structures, it forgets the invalid temporal features of the past and remembers the long-term dynamic changes of microenvironment parameters. This accurately captures the continuous impact of temporal dynamic features such as BBB permeability fluctuations and periodic changes in vascular shear force on the transmembrane transport process of particles. Combined with the dropout layer to suppress model overfitting, the temporally fully connected layer completes the spatiotemporal feature fusion and outputs the final prediction results, namely the transmembrane transport probability of particles, the real-time transport rate, and the corresponding four transport modes.
[0125] Furthermore, in S204, the input parameters include pH value, blood vessel wall shear force, and endothelial cell receptor density, and the output parameters include transmembrane transport probability, transport rate, and transport mode.
[0126] Furthermore, the deep learning agent model is trained using molecular dynamics simulation data and in vitro BBB-on-a-chip experimental data.
[0127] It is understood that in S204, the proxy model is associated with DEM particles to calculate the transmembrane behavior of the particles in real time.
[0128] In one possible implementation, in S300, the spatially differentiated pathological parameters include: In the glioma simulation process, permeability is set according to the center-edge gradient of BBB disruption; In the Alzheimer's disease simulation, the driving force of the glial lymphatic system was weakened, reducing the efficiency of cerebrospinal fluid and tissue fluid exchange.
[0129] In one possible implementation, in S400, at each time step, LBM flow field calculation, fluid force transmission, DEM particle motion and collision calculation, particle reaction force feedback, surrogate model transmembrane transport calculation, and particle state update are performed.
[0130] Furthermore, the particle state includes: particle spatial position, motion velocity, motion acceleration, remaining particle size, spatial orientation, collision state, adhesion state, transmembrane transport state, local microenvironment parameters, particle survival state, and drug loading.
[0131] In one possible implementation, S500 includes: S501. The calculated data is dynamically rendered in real time to obtain visualization results including flow field cloud map, particle motion trajectory, transmembrane transport heat map, and multi-scale structure fusion visualization map; S502. Extract the biophysical parameters of the visualization results, perform quantitative analysis, and obtain quantitative results including spatial permeability, solute clearance efficiency, vascular wall shear force, drug concentration change curve and transmembrane transport efficiency, and generate a standardized simulation report. S503. Compare the quantification results with the experimental data to verify and calibrate them.
[0132] The following detailed embodiments will provide further explanation.
[0133] This embodiment uses a brain-targeted nanomedicine delivery simulation in a high-grade glioma patient as an example to explain in detail the specific implementation process of the present invention and verify the effectiveness and practicality of the present invention.
[0134] 1. Experimental Preparation and Data Acquisition 1.1 Multi-source data acquisition: Preoperative multimodal medical imaging data of a patient with high-grade glioma (WHO grade IV) was acquired, including T1-enhanced MRI, T2-FLAIR MRI, and DSA data; high-resolution pathological section images (100 nm resolution) and tissue clarity imaging data of a mouse glioma model were acquired; the patient's clinicopathological report was acquired, including information such as tumor location, size, and grade; experimental data were acquired, including the mechanical parameters (Young's modulus 1.2 GPa, surface adhesion energy) of brain-targeting peptide-modified PLGA nanoparticles (PLGA-A2). (Diameter 100nm), BBB permeability parameters, reference values for vascular wall shear stress, etc.
[0135] 1.2 Computing Platform Setup: A GPU cluster is used as the computing platform, consisting of 8 NVIDIA A100 GPU nodes, each equipped with 64GB of video memory; the MPI parallel communication library, VTK visualization engine, and Python deep learning framework (TensorFlow) are installed; the LBM solver and DEM solver are configured to build a parallel computing environment.
[0136] 2. Multi-source data integration and 3D reconstruction 2.1 Data Preprocessing: This invention classifies and performs comprehensive preprocessing on image-based structural data and numerical pathological data, fully utilizing all multi-source data. Specific steps are as follows: First, for MRI and DSA macroscopic medical imaging data, a Gaussian filtering algorithm is used for denoising, effectively eliminating equipment acquisition noise and motion artifacts, optimizing the clarity of blood vessel and brain region boundaries, and ensuring the accuracy of macroscopic vascular network segmentation. Second, for high-resolution pathological slide images and tissue transparency imaging data—two types of microstructural image data—dedicated preprocessing operations are performed separately: For pathological slide images, adaptive grayscale enhancement, noise filtering, and edge sharpening are applied to weaken background interference and enhance the contrast and recognizability of microstructures such as tumor cell clusters, perivascular cells, astrocyte terminale, and microvascular walls; for tissue transparency imaging data, three-dimensional tomographic calibration and inter-layer artifact removal are performed. The system employs several techniques: First, image processing corrects inter-layer shifts and light scattering artifacts generated during transparent imaging, preserving the fine three-dimensional spatial structure of the BBB interface and co-selected vascular regions. Second, a rigid registration algorithm is used to align the preprocessed MRI and DSA macroscopic images with pathological slides and tissue transparent microscopic imaging data across scales, normalizing them to the same three-dimensional coordinate system to eliminate spatial misalignment and scale bias in macroscopic and microscopic structural data. Third, numerical pathological data (tumor location, tumor size, WHO pathological grade, BBB theoretical rupture rate) undergoes data cleaning, outlier removal, and quantification normalization, transforming discrete clinical pathological indicators into standardized numerical parameters that can be directly used for model parameter assignment. This achieves precise matching between image structural information and pathological numerical information, providing complete and accurate data source support for subsequent multi-scale three-dimensional reconstruction and differentiated mesh modeling.
[0137] 2.2 Image Segmentation: The image segmentation submodule in the data integration and 3D reconstruction module of this embodiment is run. An improved geometric active contour model based on fully parallelized LBM is used to automatically segment the preprocessed MRI and DSA data, extracting the 3D geometric structure of the whole brain vascular network, brain region contours and tumor regions (approximately 3.5 cm in diameter). The segmentation results are manually corrected to eliminate segmentation errors and ensure the segmentation accuracy of the vascular network (especially the malformed blood vessels in the tumor region).
[0138] 2.3 3D Reconstruction and Mesh Generation: By integrating high-resolution 3D pathological imaging data of mouse gliomas with clinical image segmentation results, refined 3D reconstruction was performed within a 5mm radius of the tumor region, generating submicron-precision cellular-level geometric structures, including large-diameter malformed blood vessels, vascular co-selection regions, tumor cell clusters, astrocyte terminale, pericytes, etc.; a multi-scale computational mesh was constructed: a macroscopic tissue-scale mesh (100μm resolution, covering the entire brain), a microscopic cellular-scale mesh (1μm resolution, covering the tumor and surrounding areas), and a subcellular-scale mesh (100nm resolution, covering the BBB interface); mesh mapping was used to connect the meshes of different scales, ensuring smooth data interaction.
[0139] 3. Construction and parameter setting of multi-scale coupled model 3.1 LBM fluid domain modeling: The brain parenchyma was modeled as a porous medium, with the porosity of the tumor core region set to 0.45, the porosity of the tumor infiltration edge set to 0.35, and the porosity of normal brain tissue set to 0.25. A D3Q19 lattice model was used, with a relaxation time set to 0.8. The unsteady Navier-Stokes equations were solved using the BGK collision model to simulate the flow of interstitial fluid and cerebrospinal fluid.
[0140] 3.2 DEM Discrete Phase Modeling: PLGA-A2 (65:35 polylactic acid-glycolic acid copolymer) nanoparticles were modeled as DEM particles, with a particle diameter of 100 nm, density of 1200 kg / m³, Young's modulus of 1.2 GPa, and surface adhesion energy set as follows. J / m², friction coefficient 0.3; simulated particle number is There are 1, evenly distributed at the entrance of the artery.
[0141] 3.3 Two-way coupling mechanism setting: Two-way force transmission between LBM and DEM is realized through momentum exchange formula. The fluid force calculation coefficient is set to 0.01 and the particle reaction force feedback coefficient is set to 0.01. The trilinear interpolation algorithm is used to realize data interaction between LBM mesh and DEM particles to ensure coupling stability.
[0142] 3.4 Deep Learning Proxy Model Integration: A Proxy Model was constructed using a CNN-LSTM hybrid architecture. The input parameters were pH value (7.2-7.4), blood vessel wall shear force (0.1-1.0 Pa), and endothelial cell receptor density (1×10^5-1×10^6 receptors / μm²). The output parameters were transmembrane transport probability, transport rate, and transport mode. The model was trained using molecular dynamics simulation data (1000 sets) and in vitro BBB-on-a-chip experimental data (500 sets). The prediction accuracy of the trained model reached 92%. The Proxy Model was associated with DEM particles to calculate the transmembrane behavior of particles in real time.
[0143] 3.5 Boundary Conditions and Pathological Parameter Settings: A pulsatile pressure waveform based on the patient's pulse pressure data was applied to the arterial inlet (systolic blood pressure 120 mmHg, diastolic blood pressure 80 mmHg, heart rate 60 bpm); the venous outlet was set as a pressure outlet (10 mmHg); the brain tissue surface was set as an impermeable boundary; according to the glioma "center-perimeter" gradient model, the baseline permeability of the tumor core region's BBB was set to... The permeability of the tumor infiltration margin was set to 0.3 × 10⁻³ cm / s, and the permeability of normal brain tissue was set to [missing value]. cm / s.
[0144] 4. High-performance coupled computing 4.1 Parallel Task Allocation: LBM computation is allocated to 4 GPU cores, DEM computation to 2 GPU cores, and deep learning proxy model computation to 2 GPU cores; MPI is used to realize communication between GPU nodes and synchronize computation progress.
[0145] 4.2 Coupled Calculation Execution: The total calculation time step is set to 86,400 steps (corresponding to 24 hours of actual time, with each step corresponding to 1 second of actual time). Within each time step, LBM flow field calculation, fluid force transmission, DEM particle motion and collision calculation, particle reaction force feedback, surrogate model transmembrane transport calculation, and particle state update are executed sequentially. During the calculation, the flow field and particle state are monitored in real time, and the relaxation time and collision coefficient are adjusted to avoid numerical oscillations. The calculation data (flow field data and particle data) are stored in real time on the server, and the LZ4 compression algorithm is used to save storage space.
[0146] 5. Dynamic visualization and quantitative analysis 5.1 Dynamic Visualization: Run the dynamic visualization submodule and use the VTK engine to render the following results in real time: a) Interstitial flow velocity and pressure field cloud map clearly shows that the flow velocity in the tumor area is higher than that in normal brain tissue, and the pressure exhibits a distribution characteristic of "high in the center and low at the edge"; b) The trajectory and spatiotemporal distribution of PLGA-A2 nanoparticles showed that the particles exhibited a "side-in" effect at the malformed blood vessels, were enriched in the tumor core area, and were less distributed at the infiltration edge. c) BBB transmembrane transport dynamics heatmap, showing that there are more transmembrane transport events in the tumor core area than in the peripheral area; d) Multi-scale fusion visualization to show the spatial relationships between macroscopic vascular networks, microscopic tumor cells, and nanoparticles. Export the visualization results as an MP4 video for presentation.
[0147] 5.2 Quantitative Analysis: The quantitative analysis submodule automatically extracted key indicators, including: the peak drug concentration in the tumor core area was 8.2 μg / mL, occurring 6 hours after administration; the peak drug concentration at the tumor invasion edge was 2.1 μg / mL, occurring 8 hours after administration; the glial lymphatic system clearance efficiency was 45%; the vascular wall shear force was highest at malformed vessels, reaching 0.9 Pa; and the transmembrane transport efficiency was 32%, of which receptor-mediated endocytosis accounted for 75%.
[0148] 5.3 Model Validation and Calibration: An in vitro 3D BBB-on-a-chip model was constructed, and the same PLGA-A2 nanoparticles were used for drug delivery experiments. OCT was used to observe the distribution of the drug in the microvascular network. The in vitro experimental data were compared with the simulation results, and the correlation coefficient was calculated to be 0.89, indicating good consistency between the simulation and experimental results. Based on the comparison results, the BBB permeability parameters were fine-tuned, improving the model prediction accuracy to 94%.
[0149] 5.4 Report Generation: Generate a standardized simulation report, including basic patient information, data sources, model parameters, visualization results, and quantitative analysis conclusions. The report clearly indicates that the concentration of PLGA-A2 nanoparticles at the tumor invasion edge is low, and recommends optimizing the nanoparticle surface ligands to enhance their targeting of tumor cells at the invasion edge, providing data support for optimizing clinical dosing regimens.
[0150] 6. Results and Analysis This application's embodiments validate the effectiveness of the method provided by simulating the delivery process of targeted nanomedicines in high-grade gliomas. The results show that this application's embodiments can accurately simulate the distribution, movement, and cross-BBB transport of drugs in a heterogeneous pathological microenvironment. The generated visualization results can be directly compared with in vitro experimental data, and the quantitative analysis results can provide precise support for drug optimization and clinical dosing regimen formulation. Compared to traditional homogeneous models, the heterogeneous pathological model of this application's embodiments predicts a 40% lower drug concentration at the tumor invasion margin, which better matches clinically observed tumor recurrence patterns, fully demonstrating the advanced nature and practicality of this application's embodiments.
[0151] See Figure 5 This is a visualization example of the simulation results for high-grade gliomas in this application embodiment. Through the heatmap of the distribution difference between the two, it is clearly shown that the model provided by this application can more accurately simulate the heterogeneous distribution of drugs in the tumor region.
[0152] See Figure 6This is a comparative diagram showing the accumulation of metabolic waste particles in the brain parenchyma before and after a simulated weakening of the glial lymphatic system function, as described in this application embodiment. The diagram clearly shows the distribution of waste and intuitively demonstrates the impact of glial lymphatic system function on waste clearance.
[0153] Those skilled in the art should understand that this embodiment is merely exemplary and does not constitute a limitation on the scope of protection of the present invention; the present invention can also be applied to the simulation of material transport in other brain diseases (such as Alzheimer's disease and cerebrovascular diseases), or the simulation of delivery of other types of drugs (such as antibody drugs and gene drugs). The corresponding simulation function can be achieved simply by adjusting the model parameters and data input according to the specific scenario.
[0154] The following describes the multi-scale brain material transport visualization simulation device provided in this application. The multi-scale brain material transport visualization simulation device described below can be referred to in correspondence with the multi-scale brain material transport visualization simulation method described above.
[0155] Figure 7 This is a schematic diagram of the structure of the multi-scale brain material transport visualization simulation device provided in the embodiments of this application, such as... Figure 7 As shown, it includes: Module 71 is used to acquire multi-source datasets and construct a multi-scale computational grid that covers the three scales of subcellular, cellular, and tissue heterogeneity. The coupling module 72 is used to simulate the flow of interstitial fluid and cerebrospinal fluid in porous media using LBM and to simulate the motion and interaction of discrete phase particles at the BBB interface using DEM. The transmembrane coupling of LBM and DEM is realized through bidirectional force transmission, and a deep learning surrogate model is associated with each DEM particle to calculate the transmembrane transport behavior of the particles in real time. Setting module 73 is used to set the pulsatile pressure boundary at the artery inlet, the pressure boundary at the venous outlet, and the impermeable boundary of brain tissue; and to set pathological parameters spatially differentiated according to pathological characteristics. Execution module 74 is used to perform LBM-DEM coupled calculations and transmembrane transport behavior in parallel to obtain calculated data including real-time flow field data and particle data; The output module 75 is used to perform real-time dynamic rendering of the calculated data, visualize the material transport process, extract key biophysical parameters, perform quantitative analysis, generate a standardized simulation report, and verify and calibrate the data through experimental data.
[0156] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communications bus 840, wherein the processor 810, communications interface 820, and memory 830 communicate with each other via the communications bus 840. The processor 810 can call logical instructions in the memory 830 to execute a multi-scale brain material transport visualization simulation method.
[0157] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0158] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the multi-scale brain material transport visualization simulation method provided by the above methods.
[0159] In another aspect, this application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the multi-scale brain material transport visualization simulation method provided by the above methods.
[0160] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0161] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A multi-scale visualization simulation method for brain material transport, characterized in that, include: Acquire multi-source datasets and construct a multi-scale computational grid that covers three scales: subcellular, cellular, and tissue, with heterogeneity. The flow of interstitial fluid and cerebrospinal fluid in porous media was simulated using LBM, and the motion and interaction of discrete phase particles at the BBB interface were simulated using DEM. Transmembrane coupling between LBM and DEM was achieved through bidirectional force transmission, and a deep learning surrogate model was associated with each DEM particle to calculate the transmembrane transport behavior of the particles in real time. Define the pulsatile pressure boundary at the arterial inlet, the pressure boundary at the venous outlet, and the impermeable boundary of the brain tissue; set pathological parameters spatially differentiated according to pathological characteristics. Parallel LBM-DEM coupled calculations and transmembrane transport behavior were performed to obtain computational data including real-time flow field data and particle data. The calculated data is rendered dynamically in real time to visualize the material transport process; key biophysical parameters are extracted, quantitatively analyzed, and a standardized simulation report is generated, which is then verified and calibrated using experimental data.
2. The method according to claim 1, characterized in that, The multi-source dataset includes MRI data, DSA data, pathological slide images, tissue transparency imaging data, pathological data, and experimental data.
3. The method according to claim 1, characterized in that, The method of using LBM to simulate the flow of interstitial fluid and cerebrospinal fluid in porous media includes: modeling brain parenchyma as a porous medium, using the D3Q19 lattice model, and numerically solving the unsteady Navier-Stokes equations for interstitial fluid and cerebrospinal fluid through the BGK collision model to obtain high-precision flow field information over the entire domain. The mechanical parameters of the DEM particles include elasticity, adhesion, and coefficient of friction, which are set according to the actual characteristics of the discrete entity. The bidirectional force transfer is achieved through momentum exchange. Fluid forces drive the movement of DEM particles, and the particle reaction forces are fed back to the LBM flow field as momentum sources. A trilinear interpolation algorithm is used to achieve continuous global data interaction between the Euler LBM fixed grid and the Lagrange DEM free particles.
4. The method according to claim 1, characterized in that, The deep learning proxy model adopts a CNN-LSTM hybrid architecture, using local microenvironment parameters as input, to predict the transmembrane transport probability, rate, and transport mode of particles in real time; the transport modes include endocytosis, carrier-mediated transport, passive diffusion, and tight junction pathways.
5. The method according to claim 1, characterized in that, The spatially differentiated pathological parameters include: In the glioma simulation process, permeability is set according to the center-edge gradient of BBB disruption; In the Alzheimer's disease simulation, the driving force of the glial lymphatic system was weakened, reducing the efficiency of cerebrospinal fluid and tissue fluid exchange.
6. The method according to claim 1, characterized in that, The process involves real-time dynamic rendering of the calculated data to visualize the material transport process; extraction of key biophysical parameters for quantitative analysis; generation of a standardized simulation report; and verification and calibration using experimental data, including: The calculated data is rendered dynamically in real time to obtain visualization results including flow field cloud map, particle motion trajectory, transmembrane transport heat map, and multi-scale structure fusion visualization map; The biophysical parameters of the visualization results were extracted and quantitatively analyzed to obtain quantitative results including spatial permeability, solute clearance efficiency, vascular wall shear force, drug concentration change curve, and transmembrane transport efficiency. The quantification results are compared with experimental data for verification and calibration.
7. A multi-scale visualization simulation device for brain material transport, characterized in that, include: The module is used to acquire multi-source datasets and construct a multi-scale computational grid that covers the three scales of subcellular, cellular, and tissue heterogeneity. The coupling module is used to simulate the flow of interstitial fluid and cerebrospinal fluid in porous media using LBM and to simulate the motion and interaction of discrete phase particles at the BBB interface using DEM. The transmembrane coupling of LBM and DEM is achieved through bidirectional force transmission, and a deep learning surrogate model is associated with each DEM particle to calculate the transmembrane transport behavior of the particles in real time. The settings module is used to set the pulsatile pressure boundary at the arterial inlet, the pressure boundary at the venous outlet, and the impermeable boundary of brain tissue; and to set pathological parameters spatially differentiated according to pathological characteristics. The execution module is used to perform LBM-DEM coupled calculations and transmembrane transport behavior in parallel, and obtain computational data including real-time flow field data and particle data. The output module is used to perform real-time dynamic rendering of the calculated data, visualize the material transport process, extract key biophysical parameters, perform quantitative analysis, generate standardized simulation reports, and verify and calibrate them through experimental data.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the multi-scale brain material transport visualization simulation method as described in any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-scale brain material transport visualization simulation method as described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-scale brain material transport visualization simulation method as described in any one of claims 1 to 6.
