An artificial intelligence-based neurology intervention surgery risk assessment method

Abstract

The application discloses a kind of neurology department intervention surgery risk assessment methods based on artificial intelligence, belong to medical information technology field, specifically include: obtain the cerebral vascular multi-phase image of target patient and vascular wall biomechanics data;Based on image reconstruction three-dimensional blood vessel model, extract each segment lumen area change curve, and construct blood flow disharmonic network according to the transmission delay between curve and morphological difference;Based on biomechanics data, each blood vessel section is assigned to contain individualized blood vessel wall compliance curve fluid-solid coupling simulation parameter;Planning instrument path, and simulating instrument and blood vessel wall interaction generates local pressure pulse sequence;The sequence is input into blood flow disharmonic network, and the blood flow resonance intensity and phase shift cumulative amount caused in specific network node are calculated;Accordingly, high-risk blood vessel area existing delayed blood flow redistribution disorder is identified, and structured list is output.The application realizes the quantification evaluation and early warning of networked dynamics risk caused by intervention surgery.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, specifically to an artificial intelligence-based method for risk assessment in interventional neurological surgery. Background Technology

[0002] Interventional neurosurgery has become an important means of treating cerebrovascular diseases, and preoperative risk assessment is crucial for surgical planning and patient prognosis. Currently, risk assessment based on medical imaging and computational models is the main direction of technological development, aiming to non-invasively and qualitatively predict the hemodynamic changes and related complications that may be caused by surgery.

[0003] Currently, existing technologies in this field mainly focus on two main directions. First, they involve constructing static or quasi-static three-dimensional vascular geometric models based on high-resolution vascular imaging, and using computational fluid dynamics simulations to analyze local hemodynamic parameters at specific lesions (such as aneurysms and stenosis), such as wall shear force and pressure distribution. Second, they involve further introducing simplified fluid-structure interaction models to attempt to simulate the elastic deformation of the vessel wall under blood flow pressure and its reaction to the local flow field. These methods provide valuable information for understanding local blood flow states and have become auxiliary tools for preoperative assessment.

[0004] However, when dealing with the dynamic and systemic perturbation process of neurointerventional surgery, the aforementioned existing technologies mainly rely on the assessment of static vascular anatomy and blood flow parameters at single time points, or on simplified computational fluid dynamics simulations to estimate local hemodynamic changes, generally treating the cerebrovascular system as a passive network of pipes or isolated analytical units. This ignores the inherent dynamic characteristics of blood flow pulse waves propagating in complex vascular networks, especially the amplification and resonance effects that may occur when there are individual abnormalities in vascular wall compliance or detuning between different vascular segments. Therefore, existing technologies struggle to predict how local pressure disturbances generated by interventional device manipulation propagate and evolve in such abnormal networks, potentially generating delayed, nonlinear hemodynamic risks in certain areas, leading to blind spots in the assessment of potential complications. Summary of the Invention

[0005] The purpose of this invention is to provide an artificial intelligence-based method for risk assessment in interventional neurological surgery, addressing the following technical problems:

[0006] Current neurointerventional risk assessment methods rely on static anatomy or simplified flow field simulations, treating the vascular network as a passive unit. This neglects the individualized network dynamics of pulse wave propagation, particularly the amplification and resonance effects that may arise from abnormal vessel wall compliance and intersegmental motion mismatch. Therefore, they cannot effectively predict the propagation and evolution of device manipulation pressure disturbances within this abnormal network, creating a blind spot in assessing the resulting delayed, nonlinear hemodynamic risks.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] An artificial intelligence-based method for risk assessment in interventional neurological surgery includes the following steps:

[0009] S1. Acquire multi-phase images of cerebral blood vessels and biomechanical data of the vessel wall of the target patient. The multi-phase images of cerebral blood vessels are acquired at different times during the cardiac cycle, and the biomechanical data of the vessel wall is acquired based on magnetic resonance elastography.

[0010] S2. Based on the multi-phase images, reconstruct a three-dimensional vascular model, extract the lumen area change curves of each vascular segment, and construct a blood flow harmonic network based on the transmission delay and morphological differences between the curves.

[0011] S3. Based on the biomechanical data of the blood vessel wall, dynamic fluid-structure interaction simulation parameters are assigned to each blood vessel segment at the corresponding spatial location of the three-dimensional blood vessel model. The parameters include the blood vessel wall compliance curve that varies with the pressure inside the vessel.

[0012] S4. Based on the size and shape attributes of the instruments and the target points of the proposed interventional surgery, plan the instrument travel path in the three-dimensional vascular model, and generate a local pressure pulse sequence by simulating the interaction between the instrument and the vascular wall based on the fluid-structure interaction simulation parameters.

[0013] S5. Input the pressure pulse sequence into the blood flow harmonic network and calculate the cumulative amount of blood flow resonance intensity and phase shift caused by it at a specific network node during network propagation.

[0014] S6. Based on the resonance intensity and phase shift accumulation, identify high-risk vascular regions with delayed blood flow redistribution disorders, and output a structured list containing the region location, disorder type and risk level.

[0015] As a further aspect of the present invention: in S2, the process of constructing the blood flow harmonic network is as follows:

[0016] The luminal area change curve of each vascular segment in the three-dimensional vascular model is analyzed, and the period, amplitude and waveform envelope feature parameters of the curve are extracted. The feature transmission delay between the curves corresponding to anatomically connected vascular segments is calculated. The feature transmission delay is the lag of the time point of the main peak of the downstream vascular segment curve waveform relative to the time point of the main peak of the upstream vascular segment curve waveform. The dynamic time warping distance between the curves of connected vascular segments is calculated, and the waveform morphology difference is quantified by the dynamic time warping distance.

[0017] Using the vascular segments in the three-dimensional vascular model as nodes and the anatomical connections between vascular segments as edges, an initial network topology is constructed. The feature propagation delay is assigned to the corresponding edge as a direction weight, and the waveform morphology difference is assigned to the corresponding edge as an intensity weight, thereby generating the blood vessel harmonic network.

[0018] As a further aspect of the present invention: in S3, the process of assigning dynamic fluid-structure interaction simulation parameters to each blood vessel segment is as follows:

[0019] The biomechanical data of the blood vessel wall are spatially registered with the three-dimensional blood vessel model. The elastic modulus parameter and viscoelastic parameter are assigned to the surface mesh nodes of each blood vessel segment in the three-dimensional blood vessel model through three-dimensional interpolation. For each blood vessel segment, the change in lumen volume of the blood vessel segment under multiple arithmetically increasing internal pressure values ​​is calculated based on the assigned elastic modulus parameter, viscoelastic parameter, length of the blood vessel segment and baseline radius.

[0020] Based on the data pairs of the change in lumen volume and the corresponding internal pressure value, the least squares method is used to fit and generate the vascular wall compliance curve of the vascular segment. The vascular wall compliance curve takes the pressure difference as input and the change rate of the lumen cross-sectional area as output. Based on the baseline radius and length of each vascular segment and the blood viscosity constant, the baseline blood flow resistance coefficient of each vascular segment is calculated.

[0021] As a further aspect of the present invention: the process of calculating the change in lumen volume of the blood vessel segment under multiple arithmetically increasing internal pressure values ​​is as follows:

[0022] A sequence of arithmetically increasing internal pressure values ​​is defined, the sequence covering a numerical range from below diastolic pressure to above systolic pressure. For each internal pressure value in the sequence, a three-dimensional finite element mechanical model of the blood vessel segment is established. The model input is the pressure value, as well as the elastic modulus and viscoelastic parameters assigned to the blood vessel segment.

[0023] Solve the static equilibrium equations of the three-dimensional finite element mechanical model to obtain the displacement vectors of each node of the vessel wall. Update the three-dimensional geometry of the vessel segment based on the displacement vectors, calculate the updated lumen volume, and subtract it from the baseline lumen volume under zero pressure to obtain the change in lumen volume corresponding to the pressure value.

[0024] As a further aspect of the present invention: in step S4, the specific process of planning the instrument's travel path in the three-dimensional vascular model is as follows:

[0025] Within the three-dimensional space of the lumen of the three-dimensional blood vessel model, the diameter of the circumscribed sphere of the instrument head is used as the collision detection threshold. A three-dimensional spatial traversal search is performed from the starting point of the path to the target point to generate an initial discrete passable coordinate set.

[0026] The discrete coordinate set is processed by a spline curve-based spatial path smoothing algorithm to obtain a three-dimensional spatial curve with continuous first and second derivatives as a smoothing path. On the smoothing path, samples are taken at fixed arc length intervals to generate a series of ordered critical path points. The blood vessel segment number, the equivalent diameter of the lumen at the point, and the tangent direction vector of the smoothing path at the point are recorded for each critical path point.

[0027] As a further aspect of the present invention: in S4, the specific process of generating a local pressure pulse sequence based on the interaction between the simulation device and the blood vessel wall according to the fluid-structure interaction simulation parameters is as follows:

[0028] Based on the average advancement speed of the device, the time-space relationship of the device's travel path is converted into a time-position sequence of the device head center point. At each simulation time step, the shortest distance field between the surface geometry of the device head and the inner wall of the three-dimensional blood vessel model is established. The normal contact force distribution is calculated based on the shortest distance field and the contact stiffness coefficient.

[0029] The system queries the vessel wall compliance curve corresponding to the current position of the device head center point, calculates the radial deformation of the local vessel wall based on the normal contact force distribution, uses the radial deformation as a moving boundary condition, substitutes it into the transient Navier-Stokes equation coupled with the dynamic flow resistance parameter for numerical solution, and outputs the three-dimensional pressure field of the entire computational domain at this time step.

[0030] Assemble the three-dimensional pressure field for all time steps in the simulated time sequence, and extract the pressure-time variation signal along the centerline of the three-dimensional vascular model to form the local pressure pulse sequence.

[0031] As a further aspect of the present invention: in step S5, the process of calculating the cumulative amount of blood flow resonance intensity and phase shift is as follows:

[0032] From the local pressure pulse sequence, the pressure change signal at each node of the blood flow harmonic network is extracted over time. The pressure signal of each node is subjected to spectrum analysis to identify its dominant oscillation frequency component and corresponding amplitude.

[0033] In the blood flow resonant network, the equivalent transfer function of all possible paths propagating from the pressure disturbance source node to the target node is calculated based on the direction weight and intensity weight of each connection. For each target node, the phase shift of the dominant frequency component is calculated based on the equivalent transfer function of all incoming paths. The phase shifts of all incoming paths are superimposed to obtain the cumulative phase shift. The blood flow resonance intensity is calculated by combining the amplitude of the incoming pressure signal.

[0034] As a further aspect of the present invention: in step S6, the process of outputting the structured list is as follows:

[0035] The blood flow resonance intensity and the cumulative phase shift are compared using a set threshold. Vascular regions that exceed the threshold are identified as high-risk vascular regions. A data record is created for each high-risk vascular region. The data record includes a region identifier, anatomical location description, a list of included vascular segments, a barrier type code, a risk level value, an average resonance intensity, and a maximum cumulative phase shift.

[0036] All data records of high-risk vascular areas are sorted in descending order of risk level value, and a statistical summary is generated. The statistical summary includes the risk level distribution and the lesion type distribution. The sorted data records and the statistical summary are integrated and output as a structured list in a predefined field format.

[0037] The beneficial effects of this invention are:

[0038] This invention constructs a blood flow harmonic network reflecting the motion coordination between different vascular segments by acquiring individualized multi-phase cerebral vascular imaging and vascular wall biomechanical data. Based on the biomechanical data, fluid-structure interaction simulation parameters containing individualized vascular wall compliance curves are assigned to each vascular segment. Furthermore, by simulating the interaction between interventional devices and the vascular wall to generate a pressure pulse sequence, and inputting it into the blood flow harmonic network, the hemodynamic resonance intensity and cumulative phase shift caused by this disturbance at specific nodes during propagation within the vascular network are calculated. This identifies high-risk areas with delayed blood flow redistribution disorders due to abnormal network dynamics. This method, for the first time, shifts the evaluation focus from static local anatomy and flow field parameters to dynamic networked system responses. It can identify fluctuation amplification and resonance risks caused by individual differences in vascular wall compliance and motion mismatch between vascular segments, risks that are undetectable by traditional methods. This provides preoperative planning with risk warnings based on system dynamics, transcending morphology, and helps avoid potential surgical complications caused by abnormal network dynamics effects. Attached Figure Description

[0039] The invention will now be further described with reference to the accompanying drawings.

[0040] Figure 1 This is a flowchart illustrating the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Please see Figure 1 As shown, this invention is an artificial intelligence-based method for risk assessment in interventional neurological surgery, comprising the following steps:

[0043] S1. Acquire multi-phase images of cerebral blood vessels and biomechanical data of the vessel wall of the target patient. The multi-phase images of cerebral blood vessels are acquired at different times during the cardiac cycle, and the biomechanical data of the vessel wall is acquired based on magnetic resonance elastography.

[0044] S2. Based on the multi-phase images, reconstruct a three-dimensional vascular model, extract the lumen area change curves of each vascular segment, and construct a blood flow harmonic network based on the transmission delay and morphological differences between the curves.

[0045] S3. Based on the biomechanical data of the blood vessel wall, dynamic fluid-structure interaction simulation parameters are assigned to each blood vessel segment at the corresponding spatial location of the three-dimensional blood vessel model. The parameters include the blood vessel wall compliance curve that varies with the pressure inside the vessel.

[0046] S4. Based on the size and shape attributes of the instruments and the target points of the proposed interventional surgery, plan the instrument travel path in the three-dimensional vascular model, and generate a local pressure pulse sequence by simulating the interaction between the instrument and the vascular wall based on the fluid-structure interaction simulation parameters.

[0047] S5. Input the pressure pulse sequence into the blood flow harmonic network and calculate the cumulative amount of blood flow resonance intensity and phase shift caused by it at a specific network node during network propagation.

[0048] S6. Based on the resonance intensity and phase shift accumulation, identify high-risk vascular regions with delayed blood flow redistribution disorders, and output a structured list containing the region location, disorder type and risk level.

[0049] In a preferred embodiment of the present invention, in step S1, when acquiring multi-phase images of the cerebral blood vessels of the target patient, electrocardiogram gating technology is used to trigger magnetic resonance scanning at specific phase points of the cardiac cycle. These phase points typically include 5 to 8 time points, such as peak systole, early diastole, and end diastole, thereby capturing the periodic expansion and contraction sequence of the vascular lumen as the heart beats. After three-dimensional reconstruction, the images acquired at each time point can yield a series of temporally continuous three-dimensional vascular models.

[0050] Biomechanical data of the vascular wall were acquired using magnetic resonance elastography (MRI). This technique applies a low-frequency mechanical vibration of 50 Hz to the patient's skull, which propagates through the brain tissue as a shear wave. The scanning sequence captures the propagation of the shear wave, and an inversion algorithm calculates the elastic modulus value at various points on the vascular wall, expressed in kilopascals, to quantify the stiffness of the vascular wall. This results in a vascular wall elastic modulus distribution map aligned with a three-dimensional vascular model space. The combination of multi-phase images and the elastic modulus distribution map provides a data foundation for subsequent analysis of individual dynamic characteristics.

[0051] In another preferred embodiment of the present invention, the process of constructing the blood flow harmonic network in S2 is as follows:

[0052] First, the cross-sectional area of ​​the lumen of each vascular segment is extracted from a sequence of three-dimensional vascular models covering multiple moments in the cardiac cycle. Specifically, for a 5-millimeter-long vascular segment, its cross-sectional area is calculated at eight time points uniformly collected during the cardiac cycle. These eight area values ​​are connected in chronological order to form a curve showing the change in the lumen area of ​​the vascular segment. This curve visually reflects the periodic contraction and relaxation movements of the vascular segment as the heart beats.

[0053] Next, feature parameter analysis is performed on each luminal area change curve. The extracted features mainly include three parts. The first is the curve period, which is the time interval between two adjacent area peaks or troughs, usually corresponding to heart rate; for example, at a heart rate of 75 beats / min, the period is approximately 0.8 seconds. The second is the curve amplitude, defined as the difference between the maximum and minimum area values ​​within a single cycle, used to quantify the amplitude of vasoconstriction and vasodilation. The third is the curve waveform envelope, formed by connecting the peak and trough points within each cycle to form upper and lower envelope lines, used to describe the trend of amplitude change over time, such as whether the amplitude gradually increases, decreases, or remains stable over several consecutive cycles. These features together constitute a digital fingerprint describing the motion pattern of this vascular segment.

[0054] The key technical step lies in calculating the coordination of motion patterns between connected vessel segments, which introduces two core quantitative indicators. The first indicator is the "feature transmission delay." During calculation, the waveform peak (i.e., the point of maximum area) of the upstream vessel segment curve within a complete cycle is first located, and its occurrence time stamp is recorded. Next, the timestamp of the waveform peak of the directly downstream vessel segment curve within the same cycle is located. The time difference between the two, for example, if the downstream peak is delayed by 0.05 seconds compared to the upstream peak, is the feature transmission delay in that connection direction. This delay reflects the time required for a pressure pulse wave or blood flow pulse to propagate from upstream to downstream. In a healthy, elastic vascular network, this delay is typically within a small, stable range.

[0055] The second metric is "waveform morphology difference," calculated using a dynamic time warping algorithm. The core of this algorithm lies in solving the optimal matching problem between two curves that may not be exactly the same length. Even if two vessel segments have similar curve periods and amplitudes, subtle differences in waveform shape, such as the steepness of the ascending limb, the smoothness of the descending limb, or the presence of additional small fluctuations, can exist. The dynamic time warping algorithm finds the optimal corresponding point for each data point of the two curves by non-linearly stretching or compressing the time axis and calculates the sum of the distances between all matching point pairs. This summed distance, the dynamic time warping distance, is used to precisely quantify the overall morphological dissimilarity between the two curves. The larger the distance value, the more significant the difference in shape between the motion waveforms of the two vessel segments.

[0056] After completing the above quantitative calculations, the "blood vessel harmonic network" is constructed. This network is a weighted directed graph model. The network nodes consist of all the segmented blood vessel segments in the 3D blood vessel model. For example, a middle cerebral artery and its main branches can be divided into dozens of segment nodes. The network edges are established based on the actual anatomical connections of the blood vessels, with the direction consistent with the blood flow direction, i.e., from upstream nodes to downstream nodes. Each connection edge is assigned two key weight values. The first weight value is the "feature propagation delay" calculated earlier, which acts as the directional weight of the edge and describes the time cost of a signal or disturbance propagating along this edge. The second weight value is the "waveform morphology difference," i.e., the dynamic time warping distance, which acts as the strength weight of the edge and describes the degree of waveform distortion that may occur when the signal passes through this edge.

[0057] The resulting "distuned blood flow network" is not a static anatomical diagram, but a model rich in individualized dynamic information. It characterizes the time delays and waveform distortions experienced by pressure pulse waves during propagation within the target patient's cerebral vascular network, caused by the spatial heterogeneity of the vascular wall's mechanical properties. The concept of "distuning" describes how, when the delays and morphological differences of multiple connections in the network significantly deviate from normal physiological ranges, the entire network's response to internal disturbances (such as interventional procedures) may exhibit abnormal amplification, resonance, or attenuation effects. This network provides a crucial computational framework and quantitative foundation for subsequent simulations of how local pressure disturbances caused by surgical instruments propagate and evolve within the network.

[0058] In another preferred embodiment of the present invention, the process of assigning dynamic fluid-structure interaction simulation parameters to each blood vessel segment in step S3 is as follows:

[0059] The first step is spatial registration and parameter assignment. Since the biomechanical data of the blood vessel wall acquired from magnetic resonance elastography (MRI) initially exists in its own three-dimensional imaging coordinate space, while the three-dimensional blood vessel model is reconstructed from multi-phase blood vessel images, the two are not automatically aligned spatially. Therefore, strict spatial registration is required. This is typically achieved by identifying common anatomical landmarks in both datasets, such as the internal carotid siphon and the bifurcation of the middle cerebral artery. Using these landmarks, an optimal spatial transformation matrix is ​​calculated, which accurately maps the coordinate system of the biomechanical data to the coordinate system of the three-dimensional blood vessel model. After registration, three-dimensional interpolation can be performed. The biomechanical data mainly consists of elastic modulus and viscoelastic parameters, existing in discrete voxel form. Using a trilinear interpolation algorithm, the coordinate position of each surface mesh node of the blood vessel segment in the three-dimensional blood vessel model is calculated in the registered space, and the elastic modulus and viscoelastic parameter values ​​at that node are extracted from the biomechanical data voxels accordingly. The elastic modulus, measured in kilopascals (kPa), describes the ability of the blood vessel wall material to resist elastic deformation; a higher value indicates a "harder" blood vessel wall. Viscoelastic parameters describe the properties of the blood vessel wall material, which combines the characteristics of a viscous liquid and an elastic solid, determining its deformation rate dependence and energy dissipation characteristics under stress. Through this process, the originally abstract overall biomechanical data is concretized into a spatially resolved distribution of mechanical properties attached to the surface of the geometric model of each blood vessel segment.

[0060] After parameter assignment, the calculation process for the personalized vascular wall compliance curve is initiated for each individual vascular segment, such as a branch of the middle cerebral artery approximately 10 mm long with a baseline radius of 1.5 mm. Vascular wall compliance is a key biomechanical concept that quantitatively describes the ability of a blood vessel's lumen volume or cross-sectional area to change with variations in the pressure difference between the inside and outside of the vessel. The calculation process employs a physics-based numerical simulation method.

[0061] First, define an arithmetic progression sequence of internal pressure values. This sequence needs to cover the entire range from below physiological diastolic pressure to above physiological systolic pressure to ensure the curve is effective within both physiological and potential pathological pressure fluctuations. For example, for a patient with blood pressure of 120 / 80 mmHg, the pressure sequence might be set to start at 50 mmHg, increment in 10 mmHg steps, and increase to 150 mmHg, for a total of 11 pressure sampling points.

[0062] For each specific internal pressure value in the sequence, such as 100 mmHg, a three-dimensional finite element mechanical model corresponding to that blood vessel segment needs to be established. The finite element method is a numerical technique that discretizes a complex continuum into a large number of simple small elements for mechanical analysis. In this embodiment, the three-dimensional geometric surface mesh of the blood vessel segment is further filled with a solid mesh, such as tetrahedral elements. The material properties of each element, namely its elastic modulus and viscoelastic parameters, are obtained by interpolation from the parameters previously assigned to the surface nodes according to its location. The boundary conditions of this finite element model are set as follows: a uniform, perpendicular force distributed to the vessel wall is applied to the inner wall surface of the blood vessel segment, the magnitude of which is equal to the current internal pressure value; the outer wall surface of the blood vessel segment is usually set as a free boundary or a small tissue constraint pressure is applied. The input to the model is the current pressure value and the personalized material parameters distributed throughout the model.

[0063] Subsequently, the static equilibrium equations of the three-dimensional finite element mechanical model are solved. This step is accomplished using a finite element solver, with the goal of calculating the displacement of each node in the vessel segment mesh when mechanical equilibrium is reached under given pressure loads and material properties. The solution is a displacement vector field that details the direction and distance of movement of each point on the vessel wall under internal pressure; for example, a node may move outward by 0.02 millimeters.

[0064] Next, the three-dimensional geometry of the vessel segment is updated based on the calculated displacement vectors of all nodes. That is, the original coordinates of each node are added to its displacement vector to obtain the new coordinates of that node under the current internal pressure. Using these new coordinates, the deformed three-dimensional model of the vessel segment can be reconstructed. Then, the luminal volume of this deformed model is calculated. The specific volume calculation method can be to fill the deformed luminal space with voxels and count the volume, or it can be achieved through integration methods in computational geometry.

[0065] Finally, the deformed lumen volume is subtracted from the original lumen volume under the zero pressure difference reference state, i.e., the baseline state. The difference is the change in lumen volume of this vessel segment under the specific internal pressure of 100 mmHg. Repeating the complete process of model building, displacement calculation, geometry update, and volume change calculation for each point in the pressure sequence will ultimately yield a series of one-to-one data pairs: pressure value 1 corresponds to volume change 1, pressure value 2 corresponds to volume change 2, and so on.

[0066] After obtaining this set of data pairs, the next step is to use the least squares method to fit and generate a vessel wall compliance curve. Since the relationship between volume change and pressure is usually non-linear, the compliance curve itself describes instantaneous compliance, i.e., the rate of volume change caused by a unit pressure change. In practice, it is more common to fit a curve relating pressure to the rate of change of the vessel's cross-sectional area. Therefore, the calculated volume change needs to be converted into an average cross-sectional area change based on the length of the vessel segment, and then the rate of change relative to the baseline area needs to be calculated. Using these processed data pairs as input, the least squares fitting algorithm will find a continuous mathematical curve that best represents the trend of all data points. This curve is the personalized vessel wall compliance curve for that vessel segment. In subsequent fluid-structure interaction simulations, when it is necessary to know the geometric state of a vessel segment under a certain instantaneous pressure, simply input this pressure value into its specific compliance curve to directly query the predicted rate of change of the vessel segment's cross-sectional area at that moment, thereby quickly updating its geometric dimensions and avoiding time-consuming three-dimensional finite element calculations each time.

[0067] In addition to the mechanical behavior of the blood vessel wall, the simulation also requires resistance parameters of blood flow. Therefore, a baseline blood flow resistance coefficient needs to be calculated for each blood vessel segment. This calculation is based on the classical fluid dynamics principle of Poiseuille's law. The formula is: the resistance coefficient equals 8 times the blood viscosity multiplied by the blood vessel segment length, then divided by the product of pi and the fourth power of the blood vessel segment's baseline radius. Here, blood viscosity is a physiological constant, typically between 3.5 and 4.0 centipoise. The length of the blood vessel segment and the baseline radius can be directly measured from its three-dimensional geometric model. The calculation here is the theoretical flow resistance under the baseline geometry. In the dynamic simulation, when the cross-sectional area of ​​the blood vessel segment changes according to the compliance curve due to pressure variations, its instantaneous effective radius also changes. At this time, the baseline resistance coefficient needs to be adjusted in real time according to the fourth power relationship of the radius change, thus obtaining the dynamically changing flow resistance parameter.

[0068] In summary, through the aforementioned series of complex yet orderly technical steps, this invention generates two key core simulation parameters for each vascular segment in the patient's cerebral vascular network: an individualized, nonlinear vascular wall compliance curve, and a baseline blood flow resistance coefficient that can be dynamically adjusted with geometric changes. These two parameters together form the cornerstone of subsequent high-fidelity, individualized fluid-structure interaction simulations, enabling the simulations to realistically reflect the impact of the patient's unique vascular wall mechanical properties on hemodynamics.

[0069] In another preferred embodiment of the present invention, the specific process of S4, which involves planning the instrument's travel path in the three-dimensional vascular model based on the instrument's size and shape attributes and target points for the proposed interventional surgery, and simulating the interaction between the instrument and the vascular wall to generate a local pressure pulse sequence based on the fluid-structure interaction simulation parameters, is as follows:

[0070] First, the instrument's path is planned. This process takes place within the luminal space of a reconstructed, personalized 3D vascular model of the patient. The starting point is typically the puncture site in the femoral or radial artery, while the endpoint is the target site for surgical intervention, such as an aneurysm in the middle cerebral artery. The core challenge of planning lies in finding a collision-free and efficient route for an instrument with specific physical dimensions to reach its target within a winding, variable-diameter vascular network.

[0071] To achieve this goal, a geometric threshold for collision detection must first be defined. Considering that the device head (such as the tip of a microcatheter) is typically not a perfectly regular geometry, a practical simplification is to approximate the space it occupies with the smallest circumscribed sphere that just completely encloses the device head. The diameter of this sphere serves as the collision detection threshold. For example, for a microcatheter with a maximum head cross-sectional diameter of approximately 2.0 mm, the diameter of its circumscribed sphere might be set to 2.3 mm, allowing for a small safety clearance between the device and the catheter wall.

[0072] Next, a three-dimensional path search algorithm, such as the improved A, will be used. The algorithm, or fast traversal method, searches the lumen voxel data of the vascular model from the starting point to the ending point. At each step, the algorithm checks whether the spherical space centered on the current candidate point and bounded by the diameter of its circumscribed sphere is entirely within the lumen space defined by the vessel wall. Only when the entire spherical space is not occupied by vessel wall voxels is the point considered "passable." In this way, the algorithm can automatically avoid narrow regions with lumen diameters smaller than the safe passage threshold for the device, ultimately generating an initial path composed of a large number of discrete 3D coordinate points. This path is geometrically safe, but due to the discrete nature of the search algorithm, its trajectory often exhibits a harsh, broken line shape, lacking smoothness and failing to reflect the kinematic characteristics of continuous device movement in flexible blood vessels.

[0073] Therefore, this initial discrete path needs to be smoothed. A spline-based spatial path smoothing algorithm is applied here. Specifically, the discrete coordinates of the initial path can be used as control points to fit a three-dimensional B-spline curve or a non-uniform rational B-spline curve. These curves have continuous parametric equations, and their key advantage is that they ensure the curve itself, as well as its first derivative (tangent direction) and second derivative (curvature), change continuously throughout the entire path. This characteristic is crucial because the tangent direction represents the instantaneous direction of the instrument's movement, and the continuity of curvature means that the change in direction is gradual, without abrupt changes. The smoothed curve is the final smoothed path for the instrument's movement; it is a mathematically smooth and physically feasible spatial trajectory.

[0074] To further drive subsequent mechanical simulations, key physical parameters need to be extracted from this smooth path. Along the smooth path, sampling is performed at fixed arc-length intervals, for example, taking a point every 0.5 mm arc length, generating a series of ordered critical path points. For each critical path point, three sets of information need to be recorded. The first set is the vessel segment number, which is determined through spatial querying to identify the specific vessel segment unit in the 3D vessel model where the point is located. The second set is the equivalent lumen diameter. At this point, a plane perpendicular to the tangent direction of the path is drawn, intersecting the vessel wall to form a cross-section. The area of ​​this cross-section is calculated and represented as the area of ​​a circle; the diameter of this circle is the equivalent lumen diameter at that point. This value reflects the local spatial constraints faced by the device at that point. The third set is the tangent direction vector, i.e., the unit vector of the first derivative direction of the smooth path curve at this point, indicating the theoretical direction of movement of the device head center at that point. All critical path point information is encapsulated into a structured path description file, providing a spatiotemporal framework for subsequent dynamic simulations.

[0075] Next, we move on to the second core process: the interaction between the simulated device and the blood vessel wall generates a sequence of local pressure pulses. This is a dynamic, time-discrete fluid-structure interaction simulation process.

[0076] The simulation first requires establishing a timeline. Based on common clinical operating speeds, an average advancement speed for the instrument is set, for example, 1.5 millimeters per second. Combined with the previously planned smooth path, the path length information can be easily converted into time information. Assuming the total path length is 150 millimeters, at the average speed, it would take approximately 100 seconds for the instrument head's center point to travel the entire distance. Discretizing the entire travel time into small time steps, such as 0.01 seconds, generates a sequence showing the position of the instrument head's center point changing over time: time point 0 seconds corresponds to the starting coordinates of the path, time point 0.01 seconds corresponds to the coordinates after advancing 0.015 millimeters along the path, and so on. This time-position sequence is the "script" driving the entire dynamic simulation.

[0077] At each simulation time step, such as the 50th second (when the device may have already entered the intracranial blood vessel), it is necessary to calculate the interaction force between the device and the vessel wall. For this purpose, the concept of a "shortest distance field" is introduced for calculation. The specific steps are as follows: at the current center point of the device head, based on the actual geometric model of the device head (which could be a more refined aspherical model), determine the three-dimensional coordinates of a large number of sampling points on its surface. For each sampling point on the device surface, calculate the shortest Euclidean distance from it to all triangles in the triangular mesh of the inner wall of the three-dimensional blood vessel model. This series of distance values ​​constitutes the "distance field" from the device surface to the blood vessel wall. A positive distance indicates that the sampling point has not contacted the vessel wall, while a distance of zero or negative indicates that penetration has occurred (which is physically impossible and needs to be resolved through mechanical calculations).

[0078] Based on this shortest distance field, combined with a preset "contact stiffness coefficient," the normal contact force distribution can be calculated. The contact stiffness coefficient is a physical parameter, measured in force per unit area per unit distance, which comprehensively reflects the contact mechanical properties of the device material and the vessel wall material. Its basic physical principle is similar to a spring: the amount of compression between the two (i.e., the theoretical penetration depth or negative distance) multiplied by the stiffness coefficient yields the contact pressure per unit area. By integrating over all areas on the device surface where "contact" or "potential contact" occurs (distance less than a small threshold), the total normal contact force exerted by the device on the vessel wall at the current moment and its distribution on the contact surface are obtained.

[0079] Subsequently, the mechanical calculations proceed to the fluid-structure interaction stage. The vessel segment corresponding to the current location of the instrument head's center point is queried, and the personalized vessel wall compliance curve generated in step S3 is obtained for that segment. Based on the normal contact pressure acting on the vessel wall in that local area, calculated in the previous step, this compliance curve can be used to inversely calculate the radial deformation of the vessel wall at that location to balance this contact pressure. This deformation is a scalar value, representing the outward displacement of the vessel wall due to the instrument's pressure.

[0080] This calculated local radial deformation is crucial for bridging solid mechanics and fluid dynamics. It is defined as the "moving boundary condition" of the vessel wall geometry at the current time step. This local deformation condition, superimposed on the deformation of other parts of the vascular network due to their internal blood flow pressure (determined by the compliance curves of each segment), collectively defines the instantaneous geometry of the entire vascular system at this moment.

[0081] Next, within this updated instantaneous vascular geometry, the dynamic equations for blood flow, the transient Navier-Stokes equations, are solved. These equations describe the motion of a viscous, incompressible fluid. The equations include dynamic flow resistance parameters (reflected in the source terms or boundary conditions of the equations) that are updated in real-time with changes in the vessel cross-section, calculated in step S3. Numerical solutions to this equation, which couples complex boundaries and variable parameters, typically employ computational fluid dynamics methods, such as the finite volume method. The goal is to obtain the blood velocity and pressure values ​​at every spatial point within the vascular system under the current geometry, current boundary motion (vessel wall deformation), and current upstream blood flow input. The key output of this step is the three-dimensional pressure field of the entire computational domain at this time step, i.e., a three-dimensional data array containing pressure values ​​at hundreds of thousands or even millions of spatial points.

[0082] Finally, the three-dimensional pressure fields output at each discrete time step (from 0 seconds to 100 seconds, totaling 10,000 time steps) on the simulation timeline are assembled in chronological order to form a four-dimensional pressure data volume (three-dimensional space + one-dimensional time). To extract information relevant to network propagation analysis, a series of feature points (usually coinciding with the node positions of the blood vessel detuned network) are selected along the centerline of the three-dimensional vascular model, and the pressure values ​​at these points are extracted from the four-dimensional data volume as a function of time. The pressure-time signal at each point records the pressure fluctuations experienced at that point throughout its entire process, from the advancement of the device from a distance, through its vicinity, to its departure. All these signals, organized spatially, constitute the final local pressure pulse sequence. This sequence contains not only the intensity information of the pressure disturbance but also its temporal phase information during spatial propagation, serving as direct input data for subsequent analysis of how the disturbance induces resonance phenomena in the detuned network.

[0083] In another preferred embodiment of the present invention, the process of calculating the cumulative amount of blood flow resonance intensity and phase shift in step S5 is as follows:

[0084] The local pressure pulse sequence is a four-dimensional dataset containing spatial location and temporal information. The goal is to extract the pressure-time variation signal at the spatial coordinates corresponding to each node in the blood flow harmonic network. For example, given the coordinates of a node representing the M1 segment of the middle cerebral artery in the network, the corresponding spatial voxel is located based on these coordinates within the four-dimensional data volume of the pressure pulse sequence. The pressure values ​​of this voxel at all simulated time points are then read, resulting in a pressure-time series of the same length as the number of simulation time steps.

[0085] Next, spectral analysis is performed on the pressure time series acquired at each node to identify its periodic oscillation components. A commonly used method is the Fast Fourier Transform (FFT). This algorithm converts the time-domain signal into a frequency-domain representation, resulting in a spectrum describing the energy strength of different frequency components in the signal. In the spectrum, the frequencies corresponding to peaks with energy significantly higher than the background noise level are considered the dominant oscillation frequency components of the pressure signal at that node. For example, the pressure signal spectrum at a node might show a sharp peak at 1.2 Hz; this frequency is the dominant oscillation frequency, and its corresponding amplitude, i.e., the height of the peak, is recorded as the amplitude of that frequency component. This frequency is usually related to heart rate or its harmonics, but may also contain other frequency components excited by instrument disturbances.

[0086] The core of network propagation analysis then follows: calculating the propagation effect of pressure disturbances from the source to other nodes in the network. Here, the concept of the equivalent transfer function is introduced. In a blood-harmonic network, multiple propagation paths connected by directed edges may exist between any two nodes. The equivalent transfer function quantitatively describes how the amplitude and phase of a signal of a specific frequency change as it travels from the source node to the target node along a specific path. The construction of this function depends entirely on the direction and intensity weights of all connecting edges along that path.

[0087] Directional weight, or characteristic propagation delay, directly affects the phase of the signal. During propagation, the signal's phase lags by an amount after each connecting edge, and this lag is proportional to the signal frequency and the propagation delay of that edge. Specifically, for a signal with frequency *f*, passing through an edge with a delay of *τ* results in a phase lag of 2πfτ. Intensity weight, or waveform morphology difference, can be extended here to be considered a signal amplitude attenuation coefficient. A larger difference indicates more severe waveform distortion when the signal passes through that edge, which can be equivalent to a greater attenuation of the signal amplitude. Therefore, for a path composed of several edges connected sequentially, its total phase lag is the sum of the phase lags of all edges in the path, and its total amplitude attenuation is the product of the attenuation coefficients of all edges in the path. This mathematical description, which integrates the relationship between phase change and amplitude attenuation, is the equivalent transfer function of the path for a specific frequency signal.

[0088] Calculating the equivalent transfer function of all possible paths from the pressure disturbance source node to each target node requires path search using graph theory algorithms. The pressure disturbance source node refers to the network node corresponding to those vessel segments that the device directly contacts during the simulation with increased force. The search will limit the maximum path length, for example, to paths with no more than 5 edges, to focus on the main propagation effects.

[0089] For each target node, such as a branch node of a lenticule-like artery far from the source, after obtaining the equivalent transfer functions of all paths originating from the source, the phase shift of the dominant frequency component at that node can be calculated. Specifically, the dominant oscillation frequency component identified at the source node is substituted into the equivalent transfer function of each incoming path to calculate the theoretical phase value of the signal reaching the target node via each path. Since the lengths and weights of each path differ, these theoretical phase values ​​vary. The phase contributions of all incoming paths are then superimposed. This superposition is a vector-like superposition, as each phase contribution is accompanied by an amplitude weight, i.e., the attenuation coefficient of that path. Through complex vector summation, a composite phase value is obtained. The difference between this composite phase and the original phase of the source signal is the phase shift at the target node caused by network propagation. This total phase shift, taking into account the contributions of all incoming paths, is called the cumulative phase shift. It reflects the degree of signal phase disturbance caused by the network topology and dynamics.

[0090] Finally, the blood flow resonance intensity is calculated by combining the amplitude of the incoming signal. Here, the amplitude of the incoming signal refers to the actual composite amplitude of the pressure oscillation at that frequency reaching the target node after propagation and superposition through all paths. The blood flow resonance intensity is defined as a comprehensive index, whose value is positively correlated with the magnitude of the composite amplitude and also positively correlated with the absolute value of the accumulated phase shift. A simplified calculation method is to multiply the composite amplitude by the absolute value of the accumulated phase shift and then normalize the result. This intensity index quantifies the aberration amplification and synchronization effect of pressure oscillations that may occur at the target node due to the network propagation characteristics, i.e., the level of "resonance" risk. The higher the intensity value, the greater the risk of hemodynamic instability at that node under device disturbance.

[0091] In another preferred embodiment of the present invention, the process of outputting the structured list in step S6 is as follows:

[0092] Two thresholds are pre-defined: one is the blood flow resonance intensity threshold, for example, 0.75; the other is the cumulative phase shift threshold, for example, 1.0 radians. These two thresholds are derived from historical clinical data and a fluid dynamics simulation database. The resonance intensity value and cumulative phase shift of all nodes in the network are compared with their respective thresholds. Any node is marked as a risk node if either its resonance intensity or cumulative phase shift exceeds the threshold. Subsequently, spatially continuous or adjacent risk nodes are clustered and grouped into independent high-risk vascular regions. For example, three adjacent risk nodes on the middle cerebral artery may be grouped into the same high-risk region.

[0093] A detailed data record is created for each identified high-risk vascular region. Fields in the record are generated automatically: the region identifier is an automatically generated unique code; the anatomical location description is obtained by mapping the region's center coordinates to a standard brain atlas, such as "upper trunk of the M2 segment of the left middle cerebral artery"; the list of included vascular segments is extracted directly from the node information constituting the region; the obstacle type code is automatically determined based on the typical characteristics of the cumulative phase shift of nodes within the region: if the average phase shift is positive and large, it is coded as "delayed"; if it is negative and has a large absolute value, it is coded as "erosion"; if the phase shift is close to zero but the resonance intensity is extremely high, it is coded as "fluctuating"; the risk level value is calculated using a weighted scoring model based on the average resonance intensity and the maximum cumulative phase shift within the region, with the result quantified as an integer from 1 to 5, where 5 represents the highest risk; the average resonance intensity and the maximum cumulative phase shift are statistical calculations of the corresponding values ​​of all nodes within the region.

[0094] Next, the data records for all high-risk vascular areas are processed. They are sorted in descending order of risk level, so that the most dangerous areas appear at the top of the report. At the same time, a statistical summary is generated, which includes two parts: one is the risk level distribution, such as "Level 5: 1 area, Level 4: 2 areas"; the other is the barrier type distribution, such as "Delayed type: 2 areas, Fluctuating type: 1 area".

[0095] Finally, the sorted detailed data records and statistical summaries are integrated and arranged according to a predefined field format. This format uses either plain text comma-separated value format or a structured Extensible Markup Language (EXPLAIN) format, with each field having a clear label. For example, a record might be presented as "ID:R01,Location:Left MCA M2,Risk Level:5,Type:Delay,AvgStrength:0.88,MaxPhaseShift:1.2". This final structured list constitutes the core risk assessment report output by this method. It provides quantified, localized, and clearly categorized risk information that can be directly used to support clinical decision-making.

[0096] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for risk assessment in interventional neurological surgery based on artificial intelligence, characterized in that, Includes the following steps: S1. Acquire multi-phase images of cerebral blood vessels and biomechanical data of the vessel wall of the target patient. The multi-phase images of cerebral blood vessels are acquired at different times during the cardiac cycle, and the biomechanical data of the vessel wall is acquired based on magnetic resonance elastography. S2. Based on the multi-phase images, reconstruct a three-dimensional vascular model, extract the lumen area change curves of each vascular segment, and construct a blood flow harmonic network based on the transmission delay and morphological differences between the curves. S3. Based on the biomechanical data of the blood vessel wall, dynamic fluid-structure interaction simulation parameters are assigned to each blood vessel segment at the corresponding spatial location of the three-dimensional blood vessel model. The parameters include the blood vessel wall compliance curve that varies with the pressure inside the vessel. S4. Based on the size and shape attributes of the instruments and the target points of the proposed interventional surgery, plan the instrument travel path in the three-dimensional vascular model, and generate a local pressure pulse sequence by simulating the interaction between the instrument and the vascular wall based on the fluid-structure interaction simulation parameters. S5. Input the pressure pulse sequence into the blood flow harmonic network and calculate the cumulative amount of blood flow resonance intensity and phase shift caused by it at a specific network node during network propagation. S6. Based on the resonance intensity and phase shift accumulation, identify high-risk vascular regions with delayed blood flow redistribution disorders, and output a structured list containing region location, disorder type and risk level. In S2, the process of constructing the blood flow harmonic network is as follows: The luminal area change curve of each vascular segment in the three-dimensional vascular model is analyzed, and the period, amplitude and waveform envelope feature parameters of the curve are extracted. The feature transmission delay between the curves corresponding to anatomically connected vascular segments is calculated. The feature transmission delay is the lag of the time point of the main peak of the downstream vascular segment curve waveform relative to the time point of the main peak of the upstream vascular segment curve waveform. The dynamic time warping distance between the curves of connected vascular segments is calculated, and the waveform morphology difference is quantified by the dynamic time warping distance. Using the vascular segments in the three-dimensional vascular model as nodes and the anatomical connections between vascular segments as edges, an initial network topology is constructed. The feature transmission delay is assigned to the corresponding edge as a direction weight, and the waveform morphology difference is assigned to the corresponding edge as an intensity weight, thereby generating the blood vessel harmonic network. In S3, the process of assigning dynamic fluid-structure interaction simulation parameters to each blood vessel segment is as follows: The biomechanical data of the blood vessel wall are spatially registered with the three-dimensional blood vessel model. The elastic modulus parameter and viscoelastic parameter are assigned to the surface mesh nodes of each blood vessel segment in the three-dimensional blood vessel model through three-dimensional interpolation. For each blood vessel segment, the change in lumen volume of the blood vessel segment under multiple arithmetically increasing internal pressure values ​​is calculated based on the assigned elastic modulus parameter, viscoelastic parameter, length of the blood vessel segment and baseline radius. Based on the data pairs of the change in lumen volume and the corresponding internal pressure value, the least squares method is used to fit and generate the vascular wall compliance curve of the vascular segment. The vascular wall compliance curve takes the pressure difference as input and the change rate of the lumen cross-sectional area as output. Based on the baseline radius and length of each vascular segment and the blood viscosity constant, the baseline blood flow resistance coefficient of each vascular segment is calculated. In step S5, the process of calculating the cumulative amount of blood flow resonance intensity and phase shift is as follows: From the local pressure pulse sequence, the pressure change signal at each node of the blood flow harmonic network is extracted over time. The pressure signal of each node is subjected to spectrum analysis to identify its dominant oscillation frequency component and corresponding amplitude. In the blood flow resonant network, the equivalent transfer function of all incoming paths propagating from the pressure disturbance source node to the target node is calculated based on the direction weight and intensity weight of each connection. For each target node, the phase shift of the dominant frequency component is calculated based on the equivalent transfer function of all incoming paths. The phase shifts of all incoming paths are superimposed to obtain the cumulative phase shift. The blood flow resonance intensity is calculated in combination with the amplitude of the incoming pressure signal.

2. The method for risk assessment of interventional neurological surgery based on artificial intelligence according to claim 1, characterized in that, The process of calculating the change in lumen volume of the blood vessel segment under multiple arithmetically increasing internal pressure values ​​is as follows: A sequence of arithmetically increasing internal pressure values ​​is defined, the sequence covering a numerical range from below diastolic pressure to above systolic pressure. For each internal pressure value in the sequence, a three-dimensional finite element mechanical model of the blood vessel segment is established. The model input is the pressure value, as well as the elastic modulus and viscoelastic parameters assigned to the blood vessel segment. Solve the static equilibrium equations of the three-dimensional finite element mechanical model to obtain the displacement vectors of each node of the vessel wall. Update the three-dimensional geometry of the vessel segment based on the displacement vectors, calculate the updated lumen volume, and subtract it from the baseline lumen volume under zero pressure to obtain the change in lumen volume corresponding to the pressure value.

3. The method for risk assessment of interventional neurological surgery based on artificial intelligence according to claim 1, characterized in that, In step S4, the specific process of planning the instrument's travel path in the three-dimensional vascular model is as follows: Within the three-dimensional space of the lumen of the three-dimensional blood vessel model, the diameter of the circumscribed sphere of the instrument head is used as the collision detection threshold. A three-dimensional spatial traversal search is performed from the starting point of the path to the target point to generate an initial discrete passable coordinate set. The discrete coordinate set is processed by a spline curve-based spatial path smoothing algorithm to obtain a three-dimensional spatial curve with continuous first and second derivatives as a smoothing path. On the smoothing path, samples are taken at fixed arc length intervals to generate a series of ordered critical path points. The vessel segment number, the equivalent diameter of the lumen at the critical path point, and the tangent direction vector of the smoothing path at the critical path point are recorded.

4. The method for risk assessment of interventional neurological surgery based on artificial intelligence according to claim 3, characterized in that, In step S4, the specific process of generating a local pressure pulse sequence based on the interaction between the simulation device and the blood vessel wall using the fluid-structure interaction simulation parameters is as follows: Based on the average advancement speed of the device, the time-space relationship of the device's travel path is converted into a time-position sequence of the device head center point. At each simulation time step, the shortest distance field between the surface geometry of the device head and the inner wall of the three-dimensional blood vessel model is established. The normal contact force distribution is calculated based on the shortest distance field and the contact stiffness coefficient. The system queries the vessel wall compliance curve corresponding to the current position of the device head center point, calculates the radial deformation of the local vessel wall based on the normal contact force distribution, uses the radial deformation as a moving boundary condition, substitutes it into the transient Navier-Stokes equation coupled with the dynamic flow resistance parameter for numerical solution, and outputs the three-dimensional pressure field of the entire computational domain at this time step. Assemble the three-dimensional pressure field for all time steps in the simulated time sequence, and extract the pressure-time variation signal along the centerline of the three-dimensional vascular model to form the local pressure pulse sequence.

5. The method for risk assessment of interventional neurological surgery based on artificial intelligence according to claim 1, characterized in that, In step S6, the process of outputting the structured list is as follows: The blood flow resonance intensity and the cumulative phase shift are compared using a set threshold. Vascular regions that exceed the threshold are identified as high-risk vascular regions. A data record is created for each high-risk vascular region. The data record includes a region identifier, anatomical location description, a list of included vascular segments, a barrier type code, a risk level value, an average resonance intensity, and a maximum cumulative phase shift. All data records of high-risk vascular areas are sorted in descending order of risk level value, and a statistical summary is generated. The statistical summary includes the risk level distribution and the lesion type distribution. The sorted data records and the statistical summary are integrated and output as a structured list in a predefined field format.

Images