Three-dimensional operation risk model reconstruction method based on brain anatomy and functional atlas

Abstract

The invention discloses a three-dimensional surgical risk model reconstruction method based on brain anatomy and a functional atlas, and the method comprises the steps: S1, carrying out the image processing and fine segmentation of a brain MRI image or a DICOM sequence image of a patient, and recognizing a plurality of anatomical structures; s2, establishing a multi-dimensional risk assessment model, and performing quantitative scoring on each brain region from four dimensions of functional importance, structural vulnerability, recovery potential and clinical significance; s3, constructing a continuous three-dimensional risk field based on the risk score, and realizing the propagation of the risk value in the brain tissue based on a Gaussian diffusion model; and S4, based on an improved path planning algorithm, searching an optimal operation path in the three-dimensional risk field to realize risk minimization. The invention further discloses a corresponding system, electronic equipment and a computer readable storage medium.

Description

Technical Field

[0001] This invention relates to the fields of medical image processing and surgical planning technology, specifically to a method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas. Background Technology

[0002] Neurosurgery requires precise planning to minimize damage to critical brain structures. The human brain is one of the most complex organs, with a dense network of neural pathways and functional areas within the skull. Even minor surgical deviations can lead to severe neurological deficits.

[0003] Currently, preoperative disease diagnosis and surgical planning primarily rely on physicians' expertise and experience, based on three-dimensional medical images of the patient, to determine the appropriate treatment plan. However, traditional neurosurgical planning heavily depends on the surgeon's experience and qualitative assessment of preoperative images. While valuable, this method introduces variability and subjectivity into the surgical decision-making process. The subjectivity and variability of the surgeon's clinical experience make it difficult for them to make accurate judgments and plans, thus increasing surgical risks. Existing systems mostly employ coarse brain region divisions, typically identifying only major anatomical structures. Current risk assessments often use binary classification (critical / safe), failing to reflect the continuity of risk. Risk propagation models are overly simplistic, often using Euclidean distance and neglecting tissue anisotropy. Specifically:

[0004] In recent years, advances in neuroimaging technology have greatly improved our ability to visualize brain anatomy and function, primarily through: high-resolution magnetic resonance imaging (MRI) providing unprecedented detail in brain structures; diffusion tensor imaging (DTI) revealing white matter tract organization; and functional MRI (fMRI) mapping key functional regions. However, integrating this multimodal information into a coherent and quantitative framework for surgical risk assessment remains challenging.

[0005] 1. In terms of brain dissection techniques, traditional methods include:

[0006] (1) FreeSurfer system: an automated tool that can segment subcortical structures and cortical regions, but with limited granularity;

[0007] (2) Desikan map: using a probabilistic labeling algorithm, 34 cortical regions in each hemisphere were identified;

[0008] (3) Atlas-based method: Registers labeled atlases to patient images, but faces challenges due to anatomical variations and pathological conditions;

[0009] (4) Deep learning methods, including:

[0010] 3D CNN architecture: used for whole brain segmentation, outperforming traditional methods;

[0011] U-Net and its 3D variants: particularly influential in medical image segmentation.

[0012] 2. Surgical risk assessment techniques; existing risk assessment methods include:

[0013] (1) Functional MRI method: fMRI is used for preoperative risk assessment to identify areas of the eloquent cortex that need to be avoided;

[0014] (2) DTI integration method: Integrating diffusion tensor imaging to map white matter tracts, and preventing postoperative motor defects by preserving key pathways;

[0015] (3) Multi-parameter risk model: Create a risk map for pediatric epilepsy surgery by combining anatomical, functional and vascular information;

[0016] Limitations of existing risk assessment methods:

[0017] (1) It usually focuses on specific areas or functions of the brain, rather than providing a comprehensive assessment of the risk to the whole brain;

[0018] (2) Most of them use a binary classification ("critical" or "safe"), which fails to capture the range of potential consequences associated with surgical procedures on different structures.

[0019] 3. Risk field modeling techniques; existing risk field methods include:

[0020] (1) Cost function graph: A cost function graph is introduced for electrode placement in deep brain stimulation, and the proximity of blood vessels and ventricles is modeled as a risk factor;

[0021] (2) Safety Map: Create a "safety map" for neurosurgical planning, using distance transformation to propagate risk from critical structures;

[0022] (3) Simple Euclidean distance method: Many existing systems use a simple Euclidean distance metric to propagate risk, failing to take into account the anisotropy of brain tissue and the complex spatial relationships between adjacent but functionally different regions.

[0023] 4. Neurosurgical pathway planning techniques; existing pathway planning methods include:

[0024] (1) Graph segmentation-based framework: Considering the value of all brain regions, Dijkstra, A* and their variants are used to identify the optimal trajectory;

[0025] (2) DBS preoperative planning method: using a coarse-to-fine segmentation strategy to automatically and accurately depict anatomical structures;

[0026] (3) 3D trajectory planning method: developed for flexible needle guidance in neurosurgery;

[0027] (4) Deep reinforcement learning method: A heuristic algorithm for accelerating deep Q-network (DQN) is proposed;

[0028] Limitations of existing path planning methods**:

[0029] (1) Many methods still rely on simplified risk models and fail to fully capture the anatomical and functional complexity of the human brain;

[0030] (2) Lack of a whole-brain, refined, and continuous risk field model;

[0031] (3) The path planning failed to fully integrate multi-dimensional risk factors.

[0032] The aforementioned existing technologies have the following main drawbacks:

[0033] 1. Insufficient precision in brain segmentation: Most existing systems use coarse brain region segmentation, usually only recognizing major anatomical structures (such as FreeSurfer, which only recognizes limited subcortical structures). This limits the ability to provide a detailed risk assessment of the entire brain volume and makes it impossible to identify and assess subtle anatomical features that may affect surgical outcomes.

[0034] 2. Risk assessment lacks quantification and multidimensionality. Existing risk assessments usually use binary classification (critical / safe) or simple ordinal risk scales, which fail to reflect the continuity and multidimensionality of risk. This makes it impossible to provide a scientific and rigorous method for quantifying risk and to adjust the weights of different dimensions according to patient-specific factors and surgical goals.

[0035] 3. Regarding the risk propagation model, the model is too simple, mostly using Euclidean distance, and does not consider the anisotropy of the tissue and the complex spatial relationships between adjacent but functionally different regions. Therefore, it cannot provide a realistic risk distribution to guide surgical decisions and cannot identify the optimal surgical route to avoid critical structures.

[0036] 4. Surgical planning is highly subjective and highly variable: Current neurosurgical planning mainly relies on the surgeon's experience and qualitative assessment of preoperative images, which introduces variability and subjectivity into the surgical decision-making process, making it difficult to ensure the consistency and optimality of surgical planning.

[0037] 5. Lack of a comprehensive optimization framework: Existing path planning methods usually focus on a single optimization objective (such as the shortest path) and lack an optimization framework that comprehensively considers tissue safety and surgical feasibility. This may result in a path that is geometrically short but passes through high-risk anatomical areas, increasing surgical risks.

[0038] 6. There is a contradiction between computational efficiency and clinical applicability: Some advanced planning methods have high computational complexity and long planning time (e.g., some methods require 1247 seconds), which makes it difficult to meet the needs of real-time clinical planning, thus limiting the application of these methods in actual clinical environments.

[0039] 7. Lack of visualization and clinical decision support: Existing systems lack an intuitive 3D visualization interface, making it difficult for doctors to understand complex risk distributions and path planning results. This affects doctors' acceptance of the planning results and their willingness to use them, thus limiting the clinical promotion of computer-aided planning systems. Summary of the Invention

[0040] The purpose of this invention is to provide a method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas, thereby achieving refined and quantitative surgical risk assessment.

[0041] The first aspect of this invention is to provide a method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas, comprising:

[0042] S1 performs image processing and fine segmentation on the patient's brain MRI images or DICOM sequence images to identify multiple anatomical structures;

[0043] S2, establish a multi-dimensional risk assessment model to quantitatively score each brain region from four dimensions: functional importance, structural vulnerability, recovery potential and clinical significance;

[0044] S3 constructs a continuous three-dimensional risk field based on risk scoring and realizes the propagation of risk values ​​in brain tissue based on a Gaussian diffusion model;

[0045] S4, based on an improved path planning algorithm, searches for the optimal surgical path in the three-dimensional risk field to minimize risk.

[0046] Preferably, for cases with 280 brain regions in anatomical structure, S1 includes performing the fine segmentation based on six processing stages, including:

[0047] (1) In the preprocessing stage, N4 bias field correction is applied to remove intensity inhomogeneity, and the image is resampled to a resolution of 1×1×1mm and normalized to 256×256×256 voxels; including:

[0048] Identify and exclude anomalous pixels with intensity values ​​below 0 or above μ+2σ, where μ is the mean and σ is the standard deviation;

[0049] Linearly normalize the pixel intensity to the range [-1, 1];

[0050] Replace the excluded pixels with the minimum and maximum values ​​within the normalization range;

[0051] (2) In the cropping stage, the cropping network CNet constructed by 2D U-Net is used to remove the neck tissue below the skull base;

[0052] (3) During the skull dissection stage, the skull dissection network SSNet was used to extract brain tissue;

[0053] (4) During the segmentation stage, the segmentation network PNet, constructed using 2.5D U-Net, segmented the whole brain into 141 anatomical regions, including:

[0054] The target slice is combined with its adjacent slices above and below it along the channel direction to form an input dimension of 256×256×3.

[0055] Segmentation was performed using three models: coronal, sagittal, and axial planes.

[0056] The final prediction result is generated based on the highest average prediction probability of the three models.

[0057] (5) In the hemisphere recognition stage, the hemisphere network HNet is used to identify the left and right hemispheres;

[0058] (6) During the merging phase, 139 bilateral structures were separated based on hemispherical information, resulting in 280 brain regions.

[0059] Preferably, S2 includes:

[0060] S21. Establish a multi-dimensional risk assessment model, which includes four dimensions: functional importance, structural vulnerability, recovery potential, and clinical significance. Functional importance assesses the criticality of a structure in supporting basic life functions and cognitive processes; structural vulnerability assesses the physical properties of a tissue and its sensitivity to surgical trauma, including vascular density and tissue fragility; recovery potential assesses the ability to restore or compensate for functional impairment after injury; and clinical significance is scored based on evidence from clinical experience and surgical outcomes.

[0061] S22 quantifies each brain region based on four dimensions: functional importance, structural fragility, recovery potential, and clinical significance.

[0062] The functional importance dimension was set to a score range of 0-40, the structural vulnerability dimension to a score range of 0-30, the recovery potential dimension to a score range of 0-20, and the clinical significance dimension to a score range of 0-10. Among the structural vulnerability dimensions, the vascular density dimension was set to a score range of 0-15, and the tissue vulnerability dimension was set to a score range of 0-15.

[0063] The formula for calculating the total risk score for each brain region is shown in Equation (1), as follows:

[0064] (1);

[0065] in, Score the importance of the function. Score the structural vulnerability. To assess recovery potential, The clinical significance is scored.

[0066] Preferably, S3 includes:

[0067] S31, The basic framework for constructing a continuous three-dimensional risk field based on risk scoring is shown in equation (2):

[0068] (2);

[0069] in, For spatial points; For the first Each brain region; The normalized risk score, also known as the regional baseline risk value, represents the brain region. Its inherent, pre-defined risk level is a scalar between 0 and 1; V( ) is the volume compensation factor; For point to brain region The minimum distance; It is a decay function; For region-specific attenuation parameters;

[0070] S32, calculate the parameters of the basic architecture of the three-dimensional risk field, thereby realizing the propagation of risk values ​​in brain tissue based on a Gaussian diffusion model, wherein the Gaussian diffusion model includes a volume compensation factor. Attenuation function and region-specific attenuation parameters The volume compensation factor ensures that the small structure maintains appropriate influence in the three-dimensional risk field; the attenuation function The Gaussian diffusion form is adopted, and the volume compensation factor is... As shown in equation (3):

[0071] (3);

[0072] in, As an index variable, it usually represents the first index. A discrete brain region or location, i.e. a target brain region, refers to a specific anatomical region or voxel in a three-dimensional brain tissue model that has risk source attributes. The volume represents the brain region. The actual physical volume or number of voxels occupied in three-dimensional space; The maximum region volume represents the volume of the largest region among all considered risk source brain regions, used as the normalization benchmark; the exponent -0.5 is a power operation, equivalent to... As a compensation mechanism, its function is inversely proportional to the square root of the volume;

[0073] The attenuation function As shown in equation (4):

[0074] (4);

[0075] in, Euclidean distance refers to the distance from the location of the risk source in three-dimensional space. The straight-line distance between a point on the field and a destination; This is the attenuation coefficient, used to determine the rate at which risk decreases with distance. The larger the value, the faster the decay and the more concentrated the scope of the risk impact; conversely, the smaller the value, the more diffuse the impact. As the core part of the Gaussian function, and as a Gaussian diffusion model, it conforms to a normal distribution;

[0076] The region-specific attenuation parameter As shown in equation (5):

[0077] (5);

[0078] in, For each region-specific attenuation parameter, it represents the brain region of each risk source. Calculate a unique attenuation coefficient to make risk diffusion regionally specific; The baseline attenuation parameter is a globally set base attenuation rate that defines the overall level of risk diffusion. This is a linear transformation, with output values ​​in the interval [0.5, 1.0]. For high-risk regions, the calculated... The smaller the size, the slower the risk decays and the wider the impact, thus highlighting these areas that need to be avoided more prominently in the risk field.

[0079] Preferably, S4 includes: determining the risk corresponding to the optimal surgical path searched in the three-dimensional risk field based on formula (6), wherein formula (6) is as follows:

[0080] (6);

[0081] in, The overall risk score is used to quantitatively evaluate a candidate surgical approach. The optimal path is the path that minimizes this score, taking into account the overall cost or risk. and For two weight coefficients between 0 and 1, satisfying It is used for multi-objective trade-offs, weighing the two optimization objectives of safety and operability. High-risk surgical techniques, with a focus on avoiding high-risk tissues, allow for more complex surgical approaches. High optimization tends to choose short, straight, and smooth paths, which allows for a certain degree of organizational risk; The surgical path refers to a possible spatial trajectory from the surgical entrance to the target lesion. To organize the risk integral, the risk values ​​in all three-dimensional risk fields traversed by the path are accumulated, which is the path's risk integral. Line integrals or discrete summations of the risk values ​​at each point; This is used to normalize path complexity and quantify the geometric properties of the path.

[0082] Preferably, step S4 further includes: synchronously displaying the three-dimensional risk field and the optimal surgical path.

[0083] A second aspect of the present invention provides a three-dimensional surgical risk model reconstruction system based on brain anatomy and functional atlas for implementing the method of the first aspect, comprising:

[0084] The image segmentation module is used to perform image processing and fine segmentation on MRI images or DICOM sequence images of the patient's brain, and to identify multiple anatomical structures;

[0085] Risk assessment module: used to establish a multi-dimensional risk assessment model, which quantifies and scores each brain region from four dimensions: functional importance, structural vulnerability, recovery potential, and clinical significance;

[0086] Risk field construction module: used to construct a continuous three-dimensional risk field based on risk scores, and to realize the propagation of risk values ​​in brain tissue based on a Gaussian diffusion model;

[0087] Path planning module: Used to search for the optimal surgical path in the three-dimensional risk field based on an improved path planning algorithm, thereby minimizing risk.

[0088] Preferably, the system further includes:

[0089] A visualization module is used to simultaneously display the three-dimensional risk field and the optimal surgical path.

[0090] A third aspect of the present invention provides an electronic device including a processor and a memory, the memory storing a plurality of instructions, the processor being configured to read the instructions and execute the method as described in the first aspect.

[0091] A fourth aspect of the present invention provides a computer-readable storage medium storing a plurality of instructions which can be read by a processor and executed as described in the first aspect.

[0092] The beneficial effects of the method and system of the present invention are as follows:

[0093] This invention identifies 280 anatomical structures by finely segmenting MRI images of the patient's brain; establishes a multi-dimensional risk assessment model to quantify and score each brain region from four dimensions: functional importance, structural vulnerability, recovery potential, and clinical significance; constructs a continuous three-dimensional risk field based on the risk score and uses a Gaussian diffusion model to realize the propagation of risk values ​​in brain tissue; and utilizes an improved path planning algorithm to search for the optimal surgical path in the risk field to minimize risk, thereby improving the accuracy of surgeons' planning of the surgical procedure, reducing surgical risks, and facilitating the achievement of ideal surgical results that are scientific, precise, minimally invasive, and safe.

[0094] Compared to existing technologies, it has produced many significant beneficial effects, specifically reflected in:

[0095] First, this invention achieves refined segmentation of up to 280 brain regions in brain MRI images by introducing a six-stage processing flow based on the OpenMAP-T1 framework and a 2.5D U-Net architecture. The beneficial effects of this technical solution are: its multi-view fusion strategy and intelligent separation method for the left and right hemispheres not only preserve crucial spatial three-dimensional information but also significantly improve segmentation accuracy and robustness. Compared to traditional methods that typically identify only 34-141 brain regions, this invention improves the level of detail in basic anatomical information by approximately 2-8 times. This provides an unprecedentedly detailed anatomical base map for subsequent risk assessment, enabling the system to identify and evaluate subtle structures that are overlooked in traditional coarse segmentation but may pose a threat to surgical safety, fundamentally reducing the risk of intraoperative injury due to insufficient or incorrect anatomical information.

[0096] Secondly, this invention creatively constructs a multi-dimensional risk assessment model encompassing four dimensions: functional importance, structural vulnerability, recovery potential, and clinical significance. Its beneficial effect lies in transforming traditionally experience-based, qualitative, and even binary (critical / safety) risk assessments into a standardized, quantitative scoring system ranging from 0 to 100 points. This balance between quantification and multi-dimensionality makes risk assessment more comprehensive, objective, and repeatable. Physicians can flexibly adjust the weights of each dimension based on the patient's specific condition and surgical goals (for example, surgeries with high functional preservation requirements can emphasize functional importance scores), thereby generating more personalized risk maps and providing a powerful quantitative decision support tool for precision medicine.

[0097] Third, this invention employs a three-dimensional risk field construction method based on a Gaussian diffusion model (innovation point 3 and protection point 3), generating a continuous, smooth risk distribution that conforms to anatomical principles. Its beneficial effects are specifically achieved through innovations in the mathematical model: 1) Volume compensation factor. This ensures that small, critical structures (such as important nerve fiber bundles) are not marginalized in the risk field due to their small physical size, maintaining their due risk warning influence; 2) Region-specific attenuation parameters This allows the risk impact range of high-risk areas (such as major functional areas) to automatically expand. This adaptive mechanism better aligns with the clinical logic that "the more important the structure, the further away it should be." Compared to a simple Euclidean distance risk model, the risk field of this invention can more realistically simulate the propagation of risk in complex brain tissue, providing a more reliable basis for safe obstacle avoidance in path planning.

[0098] Fourth, the comprehensive optimized surgical path planning algorithm of this invention achieves an optimal balance between safety and feasibility by balancing tissue risk integral and path complexity. Its beneficial effects are reflected in the fact that the algorithm does not simply seek the geometric shortest path, but rather uses a formula... Multi-objective optimization was performed. Practice showed that the algorithm was willing to accept a moderate increase in path length of approximately 19.2% on average in exchange for a reduction in tissue risk exposure of up to 37.2%, a trade-off that is clinically valuable. The improved A* search algorithm incorporates a step cost function with real-time risk and angle penalties, combined with path smoothing and local optimization techniques, ultimately outputting a safe, smooth, and operable surgical trajectory, significantly improving the practicality and reliability of the planning results.

[0099] Fifth, the complete system architecture and advanced visualization solution provided by this invention bring excellent clinical practicality and user experience. Its beneficial effects include: 1) High efficiency: It reduces the average surgical planning time from approximately 1200 seconds of manual planning to 41.2 seconds, improving efficiency by more than 96%, meeting the needs of real-time or near-real-time replanning during surgery, and greatly reducing the burden on doctors; 2) Intuitiveness: Through multi-plane synchronous views, color-coded 3D risk heat maps, and interactive path comparison interfaces, complex risk data and planning results are presented in an intuitive and easy-to-understand way, helping doctors quickly understand the spatial relationship between lesions and key structures, enhancing doctors' trust in and willingness to adopt intelligent planning results; 3) Integrability: The modular design and standardized interfaces of the system enable it to be easily integrated into existing neuronavigation or surgical robot systems, with good scalability and wide applicability, and can be used for various neurosurgical procedures such as tumor resection and deep brain stimulation electrode implantation. Attached Figure Description

[0100] To more clearly illustrate the technical solutions in the specific embodiments or related technologies of the present invention, the drawings used in the description of the specific embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0101] Figure 1A A flowchart illustrating a method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas, according to an embodiment of the present invention;

[0102] Figure 1B This is a flowchart illustrating an embodiment of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas provided by an embodiment of the present invention;

[0103] Figure 2 This is a flowchart illustrating another embodiment of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas provided by an embodiment of the present invention;

[0104] Figure 3 This is a flowchart illustrating another embodiment of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas provided by the present invention.

[0105] Figure 4 This is a flowchart illustrating another embodiment of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas provided by the present invention.

[0106] Figure 5 This is a flowchart illustrating yet another embodiment of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas provided by an embodiment of the present invention;

[0107] Figure 6 This is a schematic diagram of a structure of an embodiment of a three-dimensional surgical risk model reconstruction system based on brain anatomy and functional atlas provided by the present invention.

[0108] Figure 7 A schematic diagram of the structure of an embodiment of an electronic device provided according to an embodiment of the present invention. Detailed Implementation

[0109] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0110] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0111] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" 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 invention based on the specific circumstances.

[0112] The first aspect of this invention is to provide a method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas, comprising:

[0113] S1, Image processing and fine segmentation are performed on the patient's brain MRI image or DICOM sequence image to identify multiple anatomical structures; of course, the number of anatomical structures can be set by those skilled in the art as needed, and all are within the protection scope of this invention;

[0114] S2, establish a multi-dimensional risk assessment model to quantitatively score each brain region from four dimensions: functional importance, structural vulnerability, recovery potential and clinical significance;

[0115] S3 constructs a continuous three-dimensional risk field based on risk scoring and realizes the propagation of risk values ​​in brain tissue based on a Gaussian diffusion model;

[0116] S4, based on an improved path planning algorithm, searches for the optimal surgical path in the three-dimensional risk field to minimize risk.

[0117] Preferably, for cases with 280 brain regions in anatomical structure, S1 includes performing the fine segmentation based on six processing stages, including:

[0118] (1) In the preprocessing stage, N4 bias field correction is applied to remove intensity inhomogeneity, and the image is resampled to a resolution of 1×1×1mm and normalized to 256×256×256 voxels; including:

[0119] Identify and exclude anomalous pixels with intensity values ​​below 0 or above μ+2σ, where μ is the mean and σ is the standard deviation;

[0120] Linearly normalize the pixel intensity to the range [-1, 1];

[0121] Replace the excluded pixels with the minimum and maximum values ​​within the normalization range;

[0122] (2) In the cropping stage, the cropping network CNet constructed by 2D U-Net is used to remove the neck tissue below the skull base;

[0123] (3) During the skull dissection stage, the skull dissection network SSNet was used to extract brain tissue;

[0124] (4) During the segmentation stage, the segmentation network PNet, constructed using 2.5D U-Net, segmented the whole brain into 141 anatomical regions, including:

[0125] The target slice is combined with its adjacent slices above and below it along the channel direction to form an input dimension of 256×256×3.

[0126] Segmentation was performed using three models: coronal, sagittal, and axial planes.

[0127] The final prediction result is generated based on the highest average prediction probability of the three models.

[0128] (5) In the hemisphere recognition stage, the hemisphere network HNet is used to identify the left and right hemispheres;

[0129] (6) During the merging phase, 139 bilateral structures were separated based on hemispherical information, resulting in 280 brain regions.

[0130] Preferably, S2 includes:

[0131] S21. Establish a multi-dimensional risk assessment model, which includes four dimensions: functional importance, structural vulnerability, recovery potential, and clinical significance. Functional importance assesses the criticality of a structure in supporting basic life functions and cognitive processes; structural vulnerability assesses the physical properties of a tissue and its sensitivity to surgical trauma, including vascular density and tissue fragility; recovery potential assesses the ability to restore or compensate for functional impairment after injury; and clinical significance is scored based on evidence from clinical experience and surgical outcomes.

[0132] S22 quantifies each brain region based on four dimensions: functional importance, structural fragility, recovery potential, and clinical significance.

[0133] The functional importance dimension was set to a score range of 0-40, the structural vulnerability dimension to a score range of 0-30, the recovery potential dimension to a score range of 0-20, and the clinical significance dimension to a score range of 0-10. Among the structural vulnerability dimensions, the vascular density dimension was set to a score range of 0-15, and the tissue vulnerability dimension was set to a score range of 0-15.

[0134] The formula for calculating the total risk score for each brain region is shown in Equation (1), as follows:

[0135] (1);

[0136] in, Score the importance of the function. Score the structural vulnerability. To assess recovery potential, The clinical significance is scored.

[0137] Preferably, S3 includes:

[0138] S31, The basic framework for constructing a continuous three-dimensional risk field based on risk scoring is shown in equation (2):

[0139] (2);

[0140] in, For spatial points; For the first Each brain region; The normalized risk score, also known as the regional baseline risk value, represents the brain region. The inherent, pre-defined risk level (e.g., higher risk values ​​in functional areas and areas with high vascularity) is a scalar value between 0 and 1 (assuming risk has been normalized); V( ) is the volume compensation factor; For point to brain region The minimum distance; It is a decay function; For region-specific attenuation parameters;

[0141] S32, calculate the parameters of the basic architecture of the three-dimensional risk field, thereby realizing the propagation of risk values ​​in brain tissue based on a Gaussian diffusion model, wherein the Gaussian diffusion model includes a volume compensation factor. Attenuation function and region-specific attenuation parameters The volume compensation factor ensures that the small structure maintains appropriate influence in the three-dimensional risk field; the attenuation function The Gaussian diffusion form is adopted, and the volume compensation factor is... As shown in equation (3):

[0142] (3);

[0143] in, As an index variable, it usually represents the first index. A discrete brain region or location, i.e. a target brain region, refers to a specific anatomical region or voxel in a three-dimensional brain tissue model that has risk source attributes. The volume represents the brain region. The actual physical volume or number of voxels occupied in three-dimensional space; The maximum region volume represents the volume of the largest region among all considered risk source brain regions, used as the normalization benchmark; the exponent -0.5 is a power operation, equivalent to... As a compensation mechanism, its function is inversely proportional to the square root of the volume, which is used to prevent small critical structures (such as small nerves and blood vessels) from being "submerged" by the risk signals of nearby large structures in the entire three-dimensional risk field due to their small physical volume, and to ensure that small structures can still maintain an influence commensurate with their clinical importance.

[0144] The attenuation function As shown in equation (4):

[0145] (4);

[0146] in, Euclidean distance refers to the distance from the location of the risk source in three-dimensional space. The straight-line distance between a point on the field and a destination; This is the attenuation coefficient, used to determine the rate at which risk decreases with distance. The larger the value, the faster the decay and the more concentrated the scope of the risk impact; conversely, the smaller the value, the more diffuse the impact. As the core part of the Gaussian function (normal distribution), it is used as a Gaussian diffusion model to simulate the natural diffusion process of risk in an organization. Its characteristics are smooth and continuous decay, and the squared distance term makes the risk value decrease rapidly in places far away from the risk source, which is consistent with the characteristics of many biophysical diffusion processes.

[0147] The region-specific attenuation parameter As shown in equation (5):

[0148] (5);

[0149] in, For each region-specific attenuation parameter, it represents the brain region of each risk source. Calculate a unique attenuation coefficient to make risk diffusion regionally specific; The baseline attenuation parameter is a globally set base attenuation rate that defines the overall level of risk diffusion. Assuming it's a linear transformation, the output value is in the interval [0.5, 1.0]. In [0,1]), for high-risk areas ( (High), calculated The smaller the size, the slower the risk decays and the wider the impact, thus highlighting these areas that need to be avoided more prominently in the risk field, thereby realizing the clinical logic of "the higher the risk, the wider the warning range".

[0150] Preferably, S4 includes: determining the risk corresponding to the optimal surgical path searched in the three-dimensional risk field based on formula (6), wherein formula (6) is as follows:

[0151] (6);

[0152] in, The overall risk score is used to quantitatively evaluate a candidate surgical approach. The optimal path is the path that minimizes this score, taking into account the overall "cost" or "risk". and For two weight coefficients between 0 and 1, satisfying Used for multi-objective trade-offs, in terms of safety ( ) and operability ( A trade-off is made between the two optimization objectives. High-risk surgical procedures tend to avoid high-risk tissues and may involve more complex surgical approaches. High optimization tends to choose short, straight, and smooth paths, which may allow for some organizational risk and constraints. This ensured the standardization of the relative importance of the two; The surgical path refers to a possible spatial trajectory from the surgical entrance to the target lesion. To organize the risk integral, the risk values ​​in all three-dimensional risk fields traversed by the path are accumulated, representing the "safety" cost of the path. The more high-risk areas traversed, the larger this value becomes. It is a scalar, typically representing the path's risk score. Line integrals or discrete summations of the risk values ​​at each point; Normalized path complexity is used to quantify the geometric characteristics of a path, such as length, curvature, steepness, and proximity to critical structures. It represents the "operability" or "invasiveness" cost of the path. The longer and more tortuous the path, the greater the operational difficulty and trauma it may cause. It is a scalar and is usually normalized to a certain range (e.g., [0,1]).

[0153] Step S4 defines how to comprehensively evaluate any candidate path within this three-dimensional risk field, taking into account the geometric characteristics of the surgical path. This is achieved by adjusting the weights. and It can adapt to different surgical strategies (such as radical or conservative) to find the optimal comprehensive path.

[0154] In a preferred embodiment, the method further includes:

[0155] S5, Simultaneously display the three-dimensional risk field and the optimal surgical path.

[0156] A second aspect of the present invention provides a three-dimensional surgical risk model reconstruction system based on brain anatomy and functional atlas for implementing the method of the first aspect, comprising:

[0157] The image segmentation module is used to perform image processing and fine segmentation on the patient's brain MRI images or DICOM sequence images, and to identify multiple (280 in this embodiment) anatomical structures;

[0158] Risk assessment module: used to establish a multi-dimensional risk assessment model, which quantifies and scores each brain region from four dimensions: functional importance, structural vulnerability, recovery potential, and clinical significance;

[0159] Risk field construction module: used to construct a continuous three-dimensional risk field based on risk scores, and to realize the propagation of risk values ​​in brain tissue based on a Gaussian diffusion model;

[0160] Path planning module: Used to search for the optimal surgical path in the three-dimensional risk field based on an improved path planning algorithm, thereby minimizing risk.

[0161] In a preferred embodiment, the system further includes:

[0162] A visualization module is used to simultaneously display the three-dimensional risk field and the optimal surgical path.

[0163] See Figure 1A This figure is a flowchart illustrating the inventive concept of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas, as provided in this application. Figure 1A As shown, the core idea of ​​this method is as follows: First, the patient's preoperative medical images (DICOM data) are processed, including whole-brain segmentation and tumor segmentation; then, risk calculation is performed, and a risk atlas is established based on a multi-dimensional assessment of 280 brain regions; next, the risk map is visualized to generate a three-dimensional risk heatmap; finally, puncture path planning is performed, searching for the optimal surgical path under the constraints of the risk field. The entire process is completed within a unified framework, providing a quantitative and visualized risk assessment and path optimization tool for neurosurgery.

[0164] See Figure 1B This figure is a flowchart illustrating an embodiment of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas provided in this application. The method includes:

[0165] S101: Fine segmentation of the patient's brain MRI images to identify 280 anatomical structures.

[0166] In this embodiment, a whole-brain segmentation framework based on OpenMAP-T1 was used to preprocess T1-weighted brain MRI images, remove redundant tissue, and perform other operations such as whole-brain frontal spectroscopy to identify 280 neuroanatomically defined brain regions. This fine segmentation provides an anatomical basis for subsequent risk assessment.

[0167] S102: Establish a multi-dimensional risk assessment model to quantitatively score each brain region from four dimensions: functional importance, structural vulnerability, recovery potential, and clinical significance.

[0168] In this embodiment, the 280 identified brain regions were risk-scored from four dimensions:

[0169] Functional Importance (0-40 points): Assess the criticality of the structure in supporting basic life functions and cognitive processes.

[0170] Structural vulnerability (0-30 points): Assessing the physical properties of tissues and their sensitivity to surgical trauma.

[0171] Recovery potential (0-20 points): Assessing the ability to recover or compensate for function after injury.

[0172] Clinical significance (0-10 points): Evidence score based on clinical experience and surgical outcomes.

[0173] The total risk score for each brain region is calculated as the sum of scores across four dimensions, ranging from 0 to 100.

[0174] S103: Construct a continuous three-dimensional risk field based on risk scoring, and use a Gaussian diffusion model to realize the propagation of risk values ​​in brain tissue.

[0175] In this embodiment, a three-dimensional risk field is constructed based on a multi-dimensional risk score using the following mathematical model:

[0176] For any point in the brain volume Risk value The calculation formula is shown in equation (2):

[0177] (2);

[0178] This approach ensures that the risk impact of high-risk structures is more widespread, while avoiding large structures dominating the risk field through volume compensation.

[0179] S104: Using an improved path planning algorithm, the optimal surgical path is searched in the risk field to minimize the risk.

[0180] In this embodiment, surgical path planning is modeled as an optimization problem. The discretized trajectory from the skull surface to the tumor target is solved, and the path risk assessment adopts the following comprehensive model, as shown in Equation (6):

[0181] (6);

[0182] By minimizing This allows for the discovery of surgical pathways that achieve the optimal balance between tissue safety and surgical feasibility.

[0183] In this embodiment, the integration of refined whole-brain segmentation, multi-dimensional risk assessment, continuous risk field construction, and optimized path planning provides neurosurgeons with a quantitative and visualized surgical planning tool, which helps reduce surgical risks and improve patient prognosis.

[0184] See Figure 2 This figure is a flowchart illustrating another embodiment of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas provided in this application; as shown... Figure 2 As shown, based on the above embodiments, a specific implementation of "performing fine segmentation of the patient's brain MRI image and identifying 280 anatomical structures" in S101 includes:

[0185] For each T1-weighted brain MRI image, perform the following operations:

[0186] S1011: Preprocess MRI images.

[0187] In this embodiment, N4 bias field correction is applied to remove intensity non-uniformity, and the image is resampled to a resolution of 1×1×1 mm and normalized to 256×256×256 voxels. Specifically, abnormal pixels with intensity values ​​below 0 or above μ+2σ (where μ is the average value and σ is the standard deviation) are identified and excluded. The pixel intensity is linearly normalized to the range of [-1, 1], and the excluded pixels are replaced with the minimum and maximum values ​​within the normalized range.

[0188] S1012: Cropping head regions and extracting brain tissue using a cropping network and a cranial dissection network.

[0189] In this embodiment, non-brain soft tissues, such as skin, fat, and muscle, can obfuscate whole-brain segmentation. In particular, variations in neck tissue intensity can interfere with image segmentation. To reduce the influence of neck tissue in image segmentation and partitioning, a cropping stage is incorporated to consistently eliminate regions below the brain using a CNet cropping network. CNet performs 2D segmentation on individual cross-sections of a 3D brain MRI and then stacks them vertically for high-speed processing. By leveraging a 2D U-Net architecture, the model can be trained using a large number of images from a single MRI. To prevent small gaps or missing regions in the output mask, a closure process is applied using a 3×3×3 filter. Dilation and erosion are each performed three times.

[0190] In this embodiment, given the varying intensity ranges caused by different scanner types, scanning sequences, and parameters, the image contrast between gray matter and white matter, as well as cerebrospinal fluid, must be adjusted. However, the signal variations from extracranial soft tissues (such as fat, bone, and muscle) are significant and can disrupt the stable signal intensity curves of intracranial structures. This variation can adversely affect the performance of deep learning models. To further remove residual extracranial soft tissue signals after the initial cropping stage, cranial dissection was applied to the images. By using the cranial dissection network SSNet, irrelevant regions were effectively removed for later stages of block segmentation and hemispherical analysis.

[0191] S1013: Whole brain segmentation and left and right hemisphere segmentation of the whole brain are performed by segmentation network, and the resulting labeled map of 280 brain regions is generated by merging.

[0192] In this embodiment, the segmentation network PNet, constructed using 2.5D U-Net, segments the whole brain into 141 anatomical regions. PNet is specifically designed for 2D segmentation of 3D brain MRI cross-sections, vertically stacking these regions. PNet combines the target slice with its adjacent slices along the channel direction, forming an input dimension of 256×256×3. Segmentation is performed using three models: coronal, sagittal, and axial planes, with the final prediction result generated based on the highest average prediction probability of the three models. According to the JHU atlas, excluding the 3rd and 4th ventricles, there are a total of 278 regions in the left and right hemispheres. During the cloning stage, this merging reduces the target regions of PNet to 141, effectively achieving 142 classes for the segmentation task, including the background.

[0193] In this embodiment, a hemispherical network HNet is used to identify the left and right hemispheres. HNet performs three-class classification: background, right hemisphere, and left hemisphere, which aims to determine the left and right boundaries of 139 out of 141 region labels generated in the tiling stage. HNet performs 2D segmentation on any cross section, stacking these sections vertically. It specifically uses axial and coronal portions to output hemispherical labels. A post-processing step involving expansion is implemented to separate each region generated in the tiling stage to the left and right sides. The process is as follows: (1) Expand only the left hemisphere label. (2) Modify the overlap between the expanded left hemisphere label and the right hemisphere label to the right side. (3) Expand only the right hemisphere label. (4) Modify the overlap between the expanded right hemisphere label and the left hemisphere label to the left side. This method allows hemispherical labels to expand while minimizing the impact on boundaries.

[0194] In this embodiment, 139 bilateral structures were separated based on hemispheric information, resulting in 280 brain regions. In the 141 region partitioning map, two regions without left-right distinction (the 3rd and 4th ventricles) were prioritized over hemispheric labels.

[0195] Through these six stages of processing, high-resolution whole-brain segmentation based on T1-weighted MRI was achieved, providing a precise anatomical basis for subsequent risk assessment.

[0196] See Figure 3 This figure is a flowchart illustrating another embodiment of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas provided in this application; as shown... Figure 2 As shown, based on the above embodiments, a specific implementation of S102, "establishing a multi-dimensional risk assessment model to quantitatively score each brain region from four dimensions: functional importance, structural vulnerability, recovery potential, and clinical significance," includes:

[0197] S1021: Establish a comprehensive, multidimensional brain risk assessment model and risk scoring mechanism:

[0198] In this embodiment, the risk assessment framework evaluates each brain structure from four key dimensions: functional score (0-40 points): assessing the functional importance of the brain region and its criticality to life / cognitive function; structural score (0-30 points): assessing the vascular richness and tissue vulnerability of the brain region; recovery score (0-20 points): assessing the recovery capacity and neural plasticity of the brain region after injury; and clinical score (0-10 points): additional risk adjustment based on clinical practice.

[0199] S1022: Refine the internal risk scoring rules for the four-dimensional risk scoring mechanism.

[0200] In this embodiment, the risk scoring rules and specific explanations for each dimension of the four-dimensional comprehensive multidimensional brain risk assessment model in S1021 are established.

[0201] For brain tissue functional scores (0-40 points): organization definition (30-40 points) indicates that it is a structure essential to basic life functions or influences key cognitive processes; organization definition (20-30 points) indicates that it is a structure that supports important cognitive, sensory, or motor functions; organization definition (10-20 points) indicates that it is a structure that participates in general cognitive processes and has some redundancy; organization definition (0-10 points) indicates that it is a structure with unknown function or high redundancy.

[0202] For the structural score of brain tissue (0-30 points): the tissue definition includes two parts, vascular density and hemorrhage risk (0-15 points) and tissue fragility and mechanical properties (0-15 points). The higher the vascular density and the greater the hemorrhage risk of the tissue, the more fragile the tissue, the worse the mechanical properties, and the higher the corresponding risk score.

[0203] For brain tissue recovery scores (0-20 points): Tissue definition (15-20 points) indicates a structure with very low recovery potential (e.g., brainstem nuclei); Tissue definition (10-15 points) indicates a structure with limited recovery potential; Tissue definition (5-10 points) indicates a structure with moderate recovery potential; Tissue definition (0-5 points) indicates a structure that can be well recovered through neuroplasticity.

[0204] For clinical scoring of brain tissue (0-10 points): Tissue definition (8-10 points) indicates that damage consistently leads to severe defects; Tissue definition (6-8 points) indicates that damage frequently leads to moderate defects; Tissue definition (4-6 points) indicates that damage sometimes leads to minor defects; Tissue definition (0-4 points) indicates that damage rarely leads to significant defects.

[0205] The formula for calculating the total risk score for each brain region, as described above, is: in Score the importance of the function. Score the structural vulnerability. To assess recovery potential, The clinical significance is scored.

[0206] S1023: Explainable individual differences adjustment under quantitative risk assessment scoring.

[0207] In this embodiment, the highest risk scores were observed for critical functional areas of the brain, such as the primary motor cortex (precentral gyrus) and brainstem regions, reflecting their crucial role in basic motor function and life maintenance. Damage to these areas can lead to severe motor impairment or life-threatening conditions. High risk scores were also observed for important subcortical structures, including the thalamus, hippocampus, and basal ganglia, indicating their key roles in neural circuits. These areas are associated with memory, motor regulation, and sensory transmission, and have limited potential for recovery.

[0208] The risk distribution in language-related areas is asymmetrical, with structures in the left hemisphere (right-handed individuals) such as the left angular gyrus and the left inferior frontal triangle typically receiving higher risk scores than their counterparts in the right hemisphere. Cerebrospinal fluid spaces such as the ventricles have the lowest risk scores, reflecting their limited direct functional role; however, surrounding structures must still be considered when performing surgery through these spaces. White matter pathways typically receive moderate scores, reflecting their crucial role in neural connectivity and their potential for partial recovery through neural plasticity.

[0209] This multidimensional scoring system surpasses traditional binary or simple ordinal risk classification, providing a more detailed and scientific method for risk quantification, and can adjust the weights of different dimensions according to patient-specific factors and surgical goals.

[0210] See Figure 4 This figure is a flowchart illustrating another embodiment of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas provided in this application; as shown... Figure 4 As shown, based on the above embodiments, a specific implementation of S103, "constructing a continuous three-dimensional risk field based on risk scores and using a Gaussian diffusion model to realize the propagation of risk values ​​in brain tissue," includes:

[0211] After establishing a multi-dimensional risk assessment model in S102, and quantifying each brain region based on four dimensions—functional importance, structural vulnerability, recovery potential, and clinical significance—the following operations were performed:

[0212] S1031: Constructing a mathematical model of a Gaussian decay function for a continuous three-dimensional risk field spanning the entire brain volume:

[0213] In this embodiment, the formula for constructing the three-dimensional risk field in the step is Equation (2):

[0214]

[0215] in For spatial points, For the first Each brain region, To normalize the risk score, V( ) is the volume compensation factor. For point to brain region The minimum distance, The decay function, For region-specific attenuation parameters.

[0216] S1032: Set the Gaussian attenuation series and structure-specific attenuation parameters based on Gaussian diffusion:

[0217] In this embodiment, in order to create a realistic representation of risk propagation that conforms to anatomical principles, an advanced decay function based on Gaussian diffusion is implemented: the decay function is shown in equation (4): (4) The Gaussian diffusion form is adopted; this function ensures a smooth transition of risk between regions, while allowing different risk decay rates between regions. The parameter α controls the rate at which risk decays with distance; the higher the value of α, the faster the decay rate.

[0218] The specific attenuation parameters for the region are shown in equation (5): (5); in The baseline attenuation parameter is used. This formula allows the risk from high-risk structures to propagate further than that from low-risk structures, reflecting the larger "radius of influence" of key brain regions.

[0219] In this embodiment, to prevent large structures from dominating the risk area simply because of their size, a volume compensation factor is introduced as shown in equation (3): (3); in, Indicates the region Transaction volume The trading volume in the largest area indicates that the smaller structure maintains an appropriate influence in the risk field.

[0220] See Figure 5 This figure is a flowchart illustrating another embodiment of a three-dimensional surgical risk model reconstruction method based on brain anatomy and functional atlas provided in this application; as shown... Figure 5 As shown, based on the above embodiments, a specific implementation of S104, "using an improved path planning algorithm to search for the optimal surgical path in the risk field to minimize risk," includes:

[0221] S1041: Definition of Trajectory Optimization Problem under Whole-Brain Risk Field and Construction of Minimum Comprehensive Risk Trajectory Evaluation Model.

[0222] In this embodiment, trajectory planning is treated as an optimization problem, where each potential path from the skull surface to the tumor target must be evaluated according to multiple clinical criteria. Represents a discretized trajectory, where Indicates the entry point on the surface of the skull. The target location corresponds to the tumor. This represents the spatial coordinates of the planned trajectory in the patient's brain coordinate system.

[0223] The corresponding formula (6) for the minimum comprehensive risk trajectory assessment model is shown below:

[0224] (6);

[0225] in, The overall risk score is used to quantitatively evaluate a candidate surgical approach. The optimal path is the path that minimizes this score, taking into account the overall "cost" or "risk". and For two weight coefficients between 0 and 1, satisfying Used for multi-objective trade-offs, in terms of safety ( ) and operability ( A trade-off is made between the two optimization objectives. High-risk surgical procedures tend to avoid high-risk tissues and may involve more complex surgical approaches. High optimization tends to choose short, straight, and smooth paths, which may allow for some organizational risk and constraints. This ensured the standardization of the relative importance of the two; The surgical path refers to a possible spatial trajectory from the surgical entrance to the target lesion. To organize the risk integral, the risk values ​​in all three-dimensional risk fields traversed by the path are accumulated, representing the "safety" cost of the path. The more high-risk areas traversed, the larger this value becomes. It is a scalar, typically representing the path's risk score. Line integrals or discrete summations of the risk values ​​at each point; Normalized path complexity is used to quantify the geometric characteristics of a path, such as length, curvature, steepness, and proximity to critical structures. It represents the "operability" or "invasiveness" cost of the path. The longer and more tortuous the path, the greater the operational difficulty and trauma it may cause. It is a scalar and is usually normalized to a certain range (e.g., [0,1]).

[0226] S1042: A detailed explanation of the components and scientific principles of the minimum comprehensive risk trajectory assessment model.

[0227] In this embodiment, the minimum comprehensive risk trajectory assessment model described in S1041 is divided into two parts. The first part is the organizational risk score. The second part is path complexity. .

[0228] Organizational risk score : Utilizing pre-calculated whole-brain risk field This quantifies the cumulative anatomical risk exposure over the entire trajectory. The mathematical expression for this component is shown in Equation (7):

[0229] (7);

[0230] in, Indicates the total trajectory length. The differential arc length element along the path is normalized by the trajectory length to ensure that the organizational risk component reflects the average risk density rather than the cumulative risk, thereby avoiding a systematic bias towards shorter trajectories that may affect target accessibility. Representation function At point The function value at that point, It is a scalar field function whose domain is three-dimensional space and whose range is a closed interval [0,1].

[0231] For practical computational implementation, the continuous integral is discretized using uniform sampling at 0.5 mm intervals along the trajectory, resulting in equation (8):

[0232] (8);

[0233] in, The risk value at the midpoint of the path segment is obtained by querying a pre-calculated three-dimensional risk field, and is selected from adjacent sampling points. and Sampling risk values ​​at the midpoint of the connection is a numerical integration method (similar to the midpoint rule). It samples the risk field at the midpoint of each trajectory segment to minimize discretization error while maintaining computational efficiency suitable for real-time surgical planning applications. It can effectively reduce discretization error and is more accurate than sampling at the endpoints. The number of nodes after discretization of the path represents the continuous surgical path. Discretized Sampling points arranged in sequence (generally It's the entrance. (The target point) For the first on the path Several sampling points, which are typically placed at fixed intervals (e.g., 0.5 mm) along the path. Uniform distribution, used for numerical calculations;

[0234] Path complexity This multifaceted indicator addresses the surgical difficulty and technical risks associated with trajectory execution, encompassing three different factors that influence surgical success, as shown in equation (9):

[0235] (9);

[0236] in, , and The weights for each complexity component are used to adjust the relative importance of length, curvature, and depth penalties in the overall complexity evaluation. These weights are determined by expert experience or optimization algorithms, reflecting the emphasis placed on each indicator in different surgical scenarios. , and These are non-negative weighting coefficients. They are typically... Or it can be achieved through normalization; The curvature of the path can be mathematically defined as the average, integral, or maximum curvature of the curvature at each point on the path. High curvature means that the path turns sharply and is not smooth, which will increase the difficulty of instrument operation, the risk of tissue damage, and navigation errors. This is a depth penalty term used to penalize the path that reaches deep targets. The deeper the surgical path, the worse the operative field of vision, the greater the challenge to instrument stability, and the higher the potential risks such as bleeding. Its specific calculation is based on the maximum depth, average depth, or length of the path into deep tissue.

[0237] Normalized trajectory length The relationship between puncture distance and positioning accuracy was considered, because a longer trajectory inherently carries an increased risk of cumulative positioning error and greater tissue damage. The calculation formula is Equation (10):

[0238] (10);

[0239] in, ; Representing a path The total physical length, that is, from the surgical entry point To the target point The arc length along the path curve.

[0240] Curvature components Recognizing that stereo orientation systems perform best on linear trajectories, any deviation from a straight path introduces additional complexity and potential errors. The trajectory curvature is quantified as the average angular deviation along the path, calculated as shown in Equation (11):

[0241] (11);

[0242] in, Point The angular deviation at a point is calculated by the dot product of continuous trajectory segments; It is a three-dimensional vector. That is, from point Point of view A directed line segment; It is a three-dimensional vector, representing a vector defined by adjacent sampling points. That is, from point Point of view A directed line segment.

[0243] The depth penalty component addresses the clinical reality that deeper penetration introduces increased uncertainty due to accumulated localization errors and greater distance from surface landmarks. The factor is expressed as Equation (12):

[0244] (12);

[0245] in, Point The depth from the skull surface to the target point is the reference depth, a constant used for normalization, and corresponds to the typical maximum penetration depth in stereotactic surgery. Provides standardized references corresponding to the typical maximum penetration depth in stereotactic surgery; Indicates to Take the average value, that is This reflects the overall depth level of the path;

[0246] To ensure numerical stability and enable meaningful comparisons across diverse trajectory configurations, a comprehensive normalization was applied to the path complexity components, as shown in Equation (13):

[0247] (13);

[0248] in, The complexity of the reference straight-line trajectory representing the farthest reachable target location within the brain volume provides a meaningful baseline for comparison; The original score for path complexity is calculated using formula (9); To normalize path complexity, the complexity scores of different paths are scaled to a relative scale by dividing by a reference complexity. The technical objectives are: numerical stability (avoiding excessively large or small original values ​​and improving the stability of the optimization algorithm); and comparability (providing a meaningful baseline (the complexity of the reference straight line trajectory is 1) so that the complexity of any path can be compared with it. For example, if... If the complexity of the path is 1.5 times that of the reference straight line trajectory, then it means that the complexity of the path is 1.5 times that of the reference straight line trajectory.

[0249] These three formulas together quantify the "operational difficulty" of the surgical path: Formula (11) converts the curvature of the path into an average angular deviation. Formula (12) converts the depth of the path into a normalized average depth penalty. Formula (13) normalizes the original complexity after weighted aggregation to ensure numerical stability and comparability across paths. These indicators are finally substituted into Formula (6) and weighed against the "safety" indicator (tissue risk score) to complete the comprehensive evaluation of the surgical path.

[0250] S1043: Visualization of the whole-brain integrated risk field and the minimum integrated risk trajectory.

[0251] In this embodiment, the continuous risk field generated by S103 is visualized using a heatmap, supporting multi-planar views (axial, sagittal, and coronal) and 3D volume rendering. High-risk areas are displayed in warm colors (red and orange), while low-risk areas are displayed in cool colors (blue and green). This method generates a risk field that respects anatomical boundaries while providing a smooth risk gradient, which can guide surgical path planning and identify the optimal surgical route that avoids critical structures.

[0252] In this embodiment, for the lowest comprehensive risk trajectory calculated by S1041, the colored path of the lowest comprehensive risk trajectory is visualized in the whole brain comprehensive risk field according to the path depth, and the paths of other brain planning algorithms are visualized for comparison.

[0253] See Figure 6 The figure is a schematic diagram of the structure of a three-dimensional surgical risk model reconstruction system based on brain anatomy and functional atlas provided in this application. Figure 6 As shown, the system 60 includes an image segmentation module 61, a risk assessment module 62, a wind site construction module 63, a path planning module 64, and a visualization module 65.

[0254] Furthermore, based on the above embodiments, the image segmentation module 61 is specifically used to perform image processing on the acquired DICOM sequence images to achieve fine segmentation of 280 brain regions. This module integrates the OpenMAP-T1 framework to complete the entire process from preprocessing to the final 280-region partition map.

[0255] The method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0256] Furthermore, based on the above embodiments, the risk assessment module 62 is specifically used for quantitative scoring of a multi-dimensional risk assessment model. This module stores four-dimensional scoring standards for 280 brain regions, automatically calculates the comprehensive risk score for each brain region, and can dynamically adjust to provide personalized risk scores for patients.

[0257] The method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0258] Furthermore, based on the above embodiments, the risk field construction module 63 is specifically used for generating a three-dimensional continuous risk field; implementing a Gaussian diffusion model to convert discrete brain region risk scores into a continuous three-dimensional risk field, and providing multi-plane and three-dimensional preoperative and intraoperative visualization solutions.

[0259] The method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0260] Furthermore, based on the above embodiments, the path planning module 64 is specifically used to perform the optimal surgical path search; implement the comprehensive risk optimization algorithm, and generate multiple candidate paths for clinical selection.

[0261] The method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas provided in this application embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0262] Furthermore, based on the above embodiments, the visualization module 65 is specifically used to synchronously display the three-dimensional risk field and surgical path; it provides an interactive interface to support doctors in viewing the risk distribution and planning the path from different perspectives.

[0263] The method and system for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas provided in this application can execute the technical solutions shown in the above method embodiments. The implementation principle and beneficial effects are similar, and will not be repeated here.

[0264] See Figure 7 This figure is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 70 may include a processor 71 and a memory 72.

[0265] The processor 71 is communicatively connected to the memory 72, which is used to store computer execution instructions. The processor 71 is configured to execute the technical solutions in any of the foregoing method embodiments via the computer execution instructions stored in the memory 72.

[0266] Optionally, the memory 72 can be either standalone or integrated with the processor 71. Optionally, when the memory 72 is a device independent of the processor 71, the electronic device 60 may further include a bus for connecting the aforementioned devices.

[0267] The electronic device can be a server, workstation, or embedded computing system for executing the risk model reconstruction and path planning algorithms of this application. In practical deployment, the device can be integrated into a hospital's imaging workstation or navigation system to provide real-time surgical planning support for clinical practice.

[0268] This application also provides a computer-readable storage medium storing computer-executable instructions. When these instructions are executed by a processor, they are used to implement the above-described method. The implementation principle and technical effects are similar and will not be repeated here.

[0269] The computer-readable storage medium may be a non-volatile storage medium such as ROM, RAM, disk or optical disk, storing program code that implements the method of this application, and can be deployed on different computing platforms.

[0270] I. Alternatives to the fine segmentation step (S1) of brain images

[0271] 1. Alternative Solution 1: Use a full 3D U-Net architecture for segmentation.

[0272] Specific details: A fully 3D convolutional neural network (3D U-Net) is used instead of 2.5D U-Net to directly extract and segment features from 3D volumetric data. The network input is a complete T1-weighted MRI 3D volume, and isotropic spatial features are learned through 3D convolutional kernels.

[0273] Advantages: It can capture more complete and isotropic 3D context information, and theoretically can obtain more accurate spatial continuity segmentation results.

[0274] Disadvantages: The computational complexity and GPU memory consumption are significantly higher than those of the 2.5D method, the model training and inference time are longer, and the hardware requirements are high.

[0275] Application steps: This scheme is directly applied to step S1 as the core network architecture for achieving fine segmentation of brain regions.

[0276] 2. Alternative Solution 2: Use a Transformer-based architecture (such as UNETR) for segmentation.

[0277] Specifically, the Vision Transformer (ViT) or a hybrid model combining it with a CNN (such as UNETR) is used as the segmentation network. This architecture uses a self-attention mechanism to model long-range dependencies between image patches.

[0278] Advantages: Powerful global context modeling capabilities, which may result in better segmentation performance in areas with complex anatomical structures or blurred boundaries.

[0279] Disadvantages: It usually requires a large amount of training data to avoid overfitting, has a large number of parameters, requires extremely high computational resources, and the results are relatively poorly interpretable.

[0280] Application steps: This scheme is directly applied to step S1 as another implementation of the segmentation network.

[0281] 3. Alternative Solution 3: Use multi-map registration and label fusion method for segmentation.

[0282] Specific instructions: Instead of using deep learning, a set of finely labeled brain atlases (templates) is prepared. Multiple atlases are registered to the MRI space of the target patient, and then the labels of each atlas are fused through voting (such as majority voting, locally weighted voting) to obtain the segmentation result of the target image.

[0283] Advantages: It does not rely on a large amount of labeled training data, can utilize existing high-quality maps, and the method is mature and stable.

[0284] Disadvantages: computation is time-consuming (especially when there are many maps), it is poorly adapted to brains with large anatomical variations or pathological changes, and the segmentation accuracy is severely limited by the registration accuracy.

[0285] Application steps: This scheme is directly applied to step S1 as another technical route to achieve brain region segmentation.

[0286] II. Alternatives to the Multidimensional Risk Assessment Step (S2)

[0287] 1. Alternative Solution 1: Risk Weight Adaptive Learning Based on Machine Learning

[0288] Specific explanation: Instead of pre-defining a fixed four-dimensional scoring standard, historical data including preoperative imaging and postoperative prognosis (such as neurological deficits) are collected. Using regression models (such as linear regression, random forest) or shallow neural networks, with the imaging features and location features of the brain region as input, the system automatically learns and outputs the comprehensive risk value or the weights of each dimension for that brain region.

[0289] Advantages: It can uncover potential risk patterns from real-world data that are not explicitly recognized by experts, enabling data-driven risk assessment.

[0290] Disadvantages: It heavily relies on a large amount of high-quality, consistently labeled historical data; the model becomes a "black box," with poor interpretability, which is not conducive to clinical trust and adjustment.

[0291] Application steps: This scheme is applied to step S2 to generate or adjust the BaseRisk(ri) scores or weights of each dimension for each brain region.

[0292] 2. Alternative Option 2: Dynamic score adjustment incorporating patient-specific factors

[0293] Specific explanation: Based on the basic four-dimensional scoring system, individualized parameter adjustment factors for patients are introduced. For example, the functional importance or recovery potential scores of relevant brain regions are dynamically weighted or shifted according to the patient's age, dominant hand, specific medical history (such as previous stroke), tumor pathology type, etc.

[0294] Advantages: It enables risk assessment to move from "group average" to "individual precision", improving the personalization and clinical relevance of planning.

[0295] Disadvantages: It requires the collection and integration of multimodal clinical data, increasing system complexity and data requirements; the determination of adjustment factors requires solid clinical research evidence.

[0296] Application steps: This scheme is applied to step S2, which is the step of making personalized adjustments to BaseRisk(ri) after the basic score is completed.

[0297] 3. Alternative Solution 3: Real-time functional area calibration using fused functional magnetic resonance imaging (fMRI) data

[0298] Specific instructions: Obtain preoperative task-oriented or resting-state fMRI data from patients, and identify patient-specific areas of brain activity (such as the motor cortex and language areas) through analysis. Overlay the location information of these functional areas onto the anatomical segmentation, and significantly increase their functional importance score, or mark them as "no-go zones" that must be avoided.

[0299] Advantages: Provides the most direct location of the patient's actual functional areas, greatly improving the accuracy of risk assessment for key functional areas.

[0300] Disadvantages: It adds extra imaging scans and costs, fMRI data processing takes time, and it is not suitable for patients who cannot cooperate to complete the task.

[0301] Application steps: This scheme mainly affects step S2, which is used to set the “functional importance” dimension score of a specific brain region (determined by fMRI) with extremely high precision, thereby affecting its BaseRisk(ri).

[0302] III. Alternatives to the 3D Risk Field Construction Step (S3)

[0303] 1. Alternative Solution 1: Anisotropic Diffusion Model Based on Diffusion Tensor Imaging (DTI)

[0304] Specific explanation: Information on the direction of white matter fiber bundles provided by DTI data is incorporated into risk propagation. The Gaussian attenuation function is modified so that risk attenuates more slowly (propagates further) along the main direction of white matter fibers and attenuates more quickly perpendicular to the fiber direction. In other words, the attenuation coefficient α is direction-dependent.

[0305] Advantages: It is more in line with the biological and physical nature of white matter as a risk transmission pathway (such as edema spread and electrical signal transmission), and the model is more realistic.

[0306] Disadvantages: DTI data must be acquired, which increases the cost and complexity of data collection; model calculations are more complex; DTI images themselves contain noise and estimation errors.

[0307] Application steps: This approach is applied to step S3 and is a fundamental modification of the core Decay(d, α) function model to construct a risk field that is more consistent with neuroanatomy.

[0308] 2. Alternative Solution 2: Risk field modeling based on probabilistic graphical models (such as Markov random fields, MRF)

[0309] In detail: The three-dimensional risk field is considered as an MRF, where the risk value of each voxel is a random variable. By defining the pairwise potential function between voxels (encouraging smoothing of risk values ​​of adjacent voxels) and the single-point potential function (using prior risk scores as observational evidence), the risk field construction problem is transformed into solving the maximum a posteriori probability estimation problem.

[0310] Advantages: It provides a rigorous probabilistic framework and can flexibly encode complex spatial constraints and prior knowledge.

[0311] Disadvantages: Model inference (such as using belief propagation and graph cut algorithms) involves a large amount of computation; the design of the potential function and parameter estimation are relatively complex.

[0312] Application steps: This scheme is applied to step S3 and is an alternative mathematical model to the Gaussian diffusion model for generating a continuous risk field R(p) from the discrete BaseRisk(ri).

[0313] 3. Alternative Solution 3: Construction of Multi-Scale Hierarchical Risk Field

[0314] Specific instructions: First, a coarse risk field is constructed on the low-resolution image to quickly determine the approximate distribution range of the risk; then, upsampling is performed on the region of interest or the entire image, and the risk field is refined at a higher resolution. This can be achieved using image pyramid techniques.

[0315] Advantages: High computational efficiency, especially suitable for high-resolution images; coarse scale can grasp the overall picture, while fine scale can depict local details.

[0316] Disadvantages: Effective cross-scale information transmission and fusion strategies need to be designed to avoid loss of detail or introduction of artifacts.

[0317] Application steps: This scheme is applied to step S3 and is an engineering optimization strategy for constructing the final risk field R(p). It can be used in conjunction with any of the risk propagation models mentioned above.

[0318] IV. Alternatives to the Path Planning Search Step (S4)

[0319] 1. Alternative Solution 1: Use the Fast Exploratory Random Tree (RRT / RRT*) algorithm

[0320] Detailed explanation: Starting from the path's origin, the system randomly samples within a 3D space and attempts to expand the nearest node in the tree towards the sampling point, generating new nodes and gradually constructing a path that covers the spatial tree. RRT* is its asymptotically optimal version. During expansion, it is necessary to check whether the new path segment collides with obstacles (high-risk areas) and guide the growth direction with path cost (such as PathRisk).

[0321] Advantages: It has high search efficiency in high-dimensional configuration space and can quickly find feasible paths, especially suitable for complex constraint and non-convex obstacle environments.

[0322] Disadvantages: The pathways are usually not optimal, and randomness leads to non-repeatable results; post-processing (such as pruning and smoothing) is required to obtain usable clinical pathways.

[0323] Application steps: This scheme is applied to step S4 as an alternative to the improved A* algorithm, a global planner for searching the initial path in the three-dimensional risk field.

[0324] 2. Alternative Solution 2: Use Deep Reinforcement Learning (DRL) for path planning

[0325] Specific explanation: Path planning is modeled as a sequential decision-making process. The agent (the tip of a surgical instrument) starts from the origin, observes the current local risk field state, selects a direction of movement (action), and is trained with the goal of minimizing cumulative risk (negative reward). A neural network policy is trained using algorithms such as PPO and SAC.

[0326] Advantages: It has strong end-to-end learning capabilities and may discover navigation strategies that exceed the design scope of traditional algorithms; it has fast inference speed.

[0327] Disadvantages: It requires a massive amount of interactive simulation environment data for training; safety and reliability are difficult to guarantee, and the "black box" nature makes clinical validation difficult; the model's generalization ability may be insufficient.

[0328] Application steps: This scheme is applied to step S4, completely replacing the graph-based search algorithm and directly outputting the path sequence from the starting point to the target point.

[0329] 3. Alternative Solution 3: Use evolutionary algorithms (such as genetic algorithms) for path optimization.

[0330] Specific explanation: A path is encoded as a set of control points (such as control points of a B-spline curve) as genes. A path population is initialized, and the population is iteratively evolved through operations such as selection (preserving paths with low PathRisk), crossover (exchanging path segments), and mutation (fine-tuning control points) until it converges to an approximately optimal path.

[0331] Advantages: It does not depend on gradient information and is suitable for non-convex, nonlinear, and multi-peak cost functions; it is easy to parallelize.

[0332] Disadvantages: computation is time-consuming, convergence is slow and may not converge to the global optimum; evolutionary parameters (population size, mutation rate, etc.) need to be carefully tuned.

[0333] Application steps: This scheme is applied to step S4, and is typically used for local optimization based on the initial path generated by traditional planners (such as A*, RRT), or for direct optimization of parameterized paths.

[0334] V. System Implementation and Visualization Scheme (covering the entire system)

[0335] 1. Alternative Solution 1: Visualization and Interaction Based on Virtual Reality (VR) / Augmented Reality (AR)

[0336] Specific instructions: Use VR headsets to provide doctors with a fully immersive experience of 3D brain models, risk heat maps, and planning pathways; or use AR devices (such as HoloLens) to overlay virtual risk fields and pathways onto real surgical scenarios or patient models.

[0337] Advantages: Provides unparalleled depth spatial perception and a sense of presence, greatly facilitating the understanding of complex spatial relationships and preoperative simulation.

[0338] Disadvantages: Requires dedicated hardware, which is costly; prolonged use may cause dizziness; software development is complex.

[0339] Specific application: This solution is applied to the human-computer interaction and result display stages of the entire system, and is a revolutionary alternative to the final planning result presentation method.

[0340] 2. Alternative Solution 2: Deployment of a cloud-based microservice architecture

[0341] In detail: Computationally intensive modules such as segmentation, risk field construction, and path planning are encapsulated as Docker containers and deployed on a cloud server. Hospitals upload image data via a webpage or a lightweight client and receive the results returned from the cloud.

[0342] Advantages: No high-performance computer required locally, facilitating maintenance and upgrades; supports multi-center collaboration and remote consultation; computing resources can be scaled elastically.

[0343] Disadvantages: It relies on a stable and high-speed network connection; the security and privacy protection of sensitive medical data involved in cloud migration are major challenges.

[0344] Specific application: This solution is applied to the deployment architecture of the entire system, changing the way the software is used.

[0345] 3. Alternative Solution 3: Embed as a standalone module in existing neuronavigation or surgical robot systems

[0346] Specific description: The algorithm modules of this invention (especially segmentation and path planning) are integrated into the software framework of commercial neural navigation systems (such as Brainlab, StealthStation) or surgical robots (such as ROSA, Neuromate) in the form of SDK or plug-in.

[0347] Advantages: Seamlessly integrates into existing clinical workflows, directly utilizing the tracking, registration, and visualization capabilities of navigation systems to achieve integration from planning to navigation.

[0348] Disadvantages: Development is limited by the host platform's interfaces and specifications; requires in-depth cooperation with device manufacturers.

[0349] Specific application: This approach is applied to the integration of the entire system with the clinical environment, and is one of the key ways to realize its clinical value.

[0350] Application Examples: Clinical Application Scenarios

[0351] (a) Applicable surgical types:

[0352] 1. Stereotactic biopsy of brain tumors: the most direct application, providing the optimal puncture path for preoperative planning;

[0353] 2. Deep brain stimulation (DBS) electrode implantation: Optimize electrode trajectory to avoid blood vessels and important structures;

[0354] 3. Brain tumor resection: to help determine the surgical approach and assess the risks of different approach options;

[0355] 4. External ventricular drainage: Planning the catheter placement path;

[0356] 5. Epilepsy lesion resection: Plan a safe resection area near the functional area.

[0357] (II) Integration of Clinical Workflows:

[0358] 1. Preoperative: Automatic segmentation and risk assessment are performed immediately after obtaining MRI images;

[0359] 2. Preoperative planning meeting: Use the system's visualization function to display risk distribution and recommended pathways;

[0360] 3. Operating Room: Import the planning results into the neuronavigation system for intraoperative reference;

[0361] 4. Postoperative: Compare the actual surgical path with the planned path, and assess the deviation and its impact.

[0362] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the system and device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments.

[0363] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention 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 or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas, characterized in that, include: S1 performs image processing and fine segmentation on the patient's brain MRI images or DICOM sequence images to identify multiple anatomical structures; S2, establish a multi-dimensional risk assessment model to quantitatively score each brain region from four dimensions: functional importance, structural vulnerability, recovery potential and clinical significance; S3 constructs a continuous three-dimensional risk field based on risk scoring and realizes the propagation of risk values ​​in brain tissue based on a Gaussian diffusion model; S4, based on an improved path planning algorithm, searches for the optimal surgical path in the three-dimensional risk field to minimize risk.

2. The method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas according to claim 1, characterized in that, For a brain region with 280 anatomical regions, S1 includes fine segmentation based on six processing stages, including: (1) In the preprocessing stage, N4 bias field correction is applied to remove intensity inhomogeneity, and the image is resampled to a resolution of 1×1×1mm and normalized to 256×256×256 voxels; including: Identify and exclude anomalous pixels with intensity values ​​below 0 or above μ+2σ, where μ is the mean and σ is the standard deviation; Linearly normalize the pixel intensity to the range [-1, 1]; Replace the excluded pixels with the minimum and maximum values ​​within the normalization range; (2) In the cropping stage, the cropping network CNet constructed by 2D U-Net is used to remove the neck tissue below the skull base; (3) During the skull dissection stage, the skull dissection network SSNet was used to extract brain tissue; (4) During the segmentation stage, the segmentation network PNet, constructed using 2.5D U-Net, segmented the whole brain into 141 anatomical regions, including: The target slice is combined with its adjacent slices above and below it along the channel direction to form an input dimension of 256×256×3. Segmentation was performed using three models: coronal, sagittal, and axial planes. The final prediction result is generated based on the highest average prediction probability of the three models. (5) In the hemisphere recognition stage, the hemisphere network HNet is used to identify the left and right hemispheres; (6) During the merging phase, 139 bilateral structures were separated based on hemispherical information, resulting in 280 brain regions.

3. The method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas according to claim 2, characterized in that, S2 includes: S21. Establish a multi-dimensional risk assessment model, which includes four dimensions: functional importance, structural vulnerability, recovery potential, and clinical significance. Functional importance assesses the criticality of a structure in supporting basic life functions and cognitive processes; structural vulnerability assesses the physical properties of a tissue and its sensitivity to surgical trauma, including vascular density and tissue fragility; recovery potential assesses the ability to restore or compensate for functional impairment after injury; and clinical significance is scored based on evidence from clinical experience and surgical outcomes. S22 quantifies each brain region based on four dimensions: functional importance, structural fragility, recovery potential, and clinical significance. The functional importance dimension was set to a score range of 0-40, the structural vulnerability dimension to a score range of 0-30, the recovery potential dimension to a score range of 0-20, and the clinical significance dimension to a score range of 0-10. Among the structural vulnerability dimensions, the vascular density dimension was set to a score range of 0-15, and the tissue vulnerability dimension was set to a score range of 0-15. The formula for calculating the total risk score for each brain region is shown in Equation (1), as follows: （1）； in, Score the importance of the function. Score the structural vulnerability. To restore potential score, The clinical significance is scored.

4. The method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas according to claim 3, characterized in that, S3 includes: S31, The basic framework for constructing a continuous three-dimensional risk field based on risk scoring is shown in equation (2): （2）； in, For spatial points; For the first Each brain region; The normalized risk score, also known as the regional baseline risk value, represents the brain region. Its inherent, pre-defined risk level is a scalar between 0 and 1; V( ) is the volume compensation factor; For point to brain region The minimum distance; It is a decay function; For region-specific attenuation parameters; S32, calculate the parameters of the basic architecture of the three-dimensional risk field, thereby realizing the propagation of risk values ​​in brain tissue based on a Gaussian diffusion model, wherein the Gaussian diffusion model includes a volume compensation factor. Attenuation function and region-specific attenuation parameters The volume compensation factor ensures that the small structure maintains appropriate influence in the three-dimensional risk field; the attenuation function The Gaussian diffusion form is adopted, and the volume compensation factor is... As shown in equation (3): （3）； in, As an index variable, it usually represents the first index. A discrete brain region or location, i.e. a target brain region, refers to a specific anatomical region or voxel in a three-dimensional brain tissue model that has risk source attributes. The volume represents the brain region. The actual physical volume or number of voxels occupied in three-dimensional space; The maximum region volume represents the volume of the largest region among all considered risk source brain regions, used as the normalization benchmark; the exponent -0.5 is a power operation, equivalent to... As a compensation mechanism, its function is inversely proportional to the square root of the volume; The attenuation function As shown in equation (4): （4）； in, Euclidean distance refers to the distance from the location of the risk source in three-dimensional space. The straight-line distance between a point on the field and a destination; This is the attenuation coefficient, used to determine the rate at which risk decreases with distance. The larger the value, the faster the decay and the more concentrated the scope of the risk impact; conversely, the smaller the value, the more diffuse the impact. As the core part of the Gaussian function, and as a Gaussian diffusion model, it conforms to a normal distribution; The region-specific attenuation parameter As shown in equation (5): （5）； in, For each region-specific attenuation parameter, it represents the brain region of each risk source. Calculate a unique attenuation coefficient to make risk diffusion regionally specific; The baseline attenuation parameter is a globally set base attenuation rate that defines the overall level of risk diffusion. This is a linear transformation, with output values ​​in the interval [0.5, 1.0]. For high-risk regions, the calculated... The smaller the size, the slower the risk decays and the wider the impact, thus highlighting these areas that need to be avoided more prominently in the risk field.

5. The method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas according to claim 4, characterized in that, S4 includes: determining the risk corresponding to the optimal surgical path searched in the three-dimensional risk field based on formula (6), wherein formula (6) is as follows: （6）； in, The overall risk score is used to quantitatively evaluate a candidate surgical approach. The optimal path is the path that minimizes this score, taking into account the overall cost or risk. and For two weight coefficients between 0 and 1, satisfying It is used for multi-objective trade-offs, weighing the two optimization objectives of safety and operability. High-risk surgical techniques, with a focus on avoiding high-risk tissues, allow for more complex surgical approaches. High optimization tends to choose short, straight, and smooth paths, which allows for a certain degree of organizational risk; The surgical path refers to a possible spatial trajectory from the surgical entrance to the target lesion. To organize the risk integral, the risk values ​​in all three-dimensional risk fields traversed by the path are accumulated, which is the path's risk integral. Line integrals or discrete summations of the risk values ​​at each point; This is used to normalize path complexity and quantify the geometric properties of the path.

6. The method for reconstructing a three-dimensional surgical risk model based on brain anatomy and functional atlas according to claim 5, characterized in that, S4 further includes: synchronously displaying the three-dimensional risk field and the optimal surgical path.

7. A three-dimensional surgical risk model reconstruction system based on brain anatomy and functional atlas, used to implement the method described in any one of claims 1-6, characterized in that, include: The image segmentation module is used to perform image processing and fine segmentation on MRI images or DICOM sequence images of the patient's brain, and to identify multiple anatomical structures; Risk assessment module: used to establish a multi-dimensional risk assessment model, which quantifies and scores each brain region from four dimensions: functional importance, structural vulnerability, recovery potential, and clinical significance; Risk field construction module: used to construct a continuous three-dimensional risk field based on risk scores, and to realize the propagation of risk values ​​in brain tissue based on a Gaussian diffusion model; Path planning module: Used to search for the optimal surgical path in the three-dimensional risk field based on an improved path planning algorithm, thereby minimizing risk.

8. The method according to claim 7, characterized in that, The system also includes: A visualization module is used to simultaneously display the three-dimensional risk field and the optimal surgical path.

9. An electronic device comprising a processor and a memory, the memory storing a plurality of instructions, the processor being configured to read the instructions and execute the method as claimed in any one of claims 1-6.

10. A computer-readable storage medium storing a plurality of instructions, the plurality of instructions being readable by a processor and executing the method of any one of claims 1-6.

Images