System for simulating stroke process in combination with virtual reality

Abstract

The invention discloses a system for simulating a stroke process in combination with virtual reality. The system comprises a pathology demonstration unit, a function evaluation unit, a treatment planning unit, a decision collaboration unit and a process tracing unit. The pathology demonstration unit adopts a multi-scale hemodynamic model to realize virtual reality simulation of a stroke lesion process; the function evaluation unit performs neural function quantitative evaluation based on deep learning and computer vision technologies; the treatment planning unit makes a scheme through a rule engine and multi-objective optimization; the decision cooperation unit realizes multi-party real-time consultation and opinion fusion; and the process tracing unit adopts a block chain technology to ensure the traceability of the decision-making process. The method provides systematic technical support for diagnosis and treatment decision of acute stroke, and has important clinical application value.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical informatization, and particularly relates to a system for simulating the stroke process in combination with virtual reality. Background Art

[0002] Ischemic stroke is an acute cerebrovascular disease caused by cerebrovascular occlusion, resulting in ischemia and hypoxia of brain tissue. Among them, extracranial large artery atherosclerosis is one of the important etiologies. In clinical practice, due to the acute onset characteristics of the disease and the time window limitation of treatment, it is often necessary for the patient's family members to act as proxy decision-makers to make treatment decisions within a very short time, especially for the selection of traumatic treatment options such as endovascular treatment.

[0003] In the prior art, relatively mature technical solutions have been formed for the clinical diagnosis and treatment of ischemic stroke. For example, intravenous thrombolysis has a treatment time window of 3 - 4.5 hours, and mechanical thrombectomy can extend the treatment time window to 6 - 24 hours. However, in actual clinical applications, due to the following problems, the treatment effect and prognosis are seriously affected:

[0004] 1. It is difficult to understand complex pathological mechanisms and treatment plans in a short time, and the existing text and picture information is not intuitive enough.

[0005] 2. It is difficult to intuitively quantify the treatment risks and benefits, and there is a lack of tools for integrating and displaying multi-dimensional data (clinical indicators, imaging features, etc.).

[0006] 3. The communication time between emergency doctors and patients is limited, and traditional oral and written explanations are prone to misunderstandings and cannot meet the psychological needs of decision-makers.

[0007] 4. There is a lack of standardized decision support tools, which cannot effectively reduce the pressure on proxy decision-makers and provide scientific references.

[0008] 5. The decision-making process lacks a recording and traceability mechanism, which is not conducive to post-event liability determination.

[0009] In summary, the prior art has obvious deficiencies in supporting proxy decision-makers of ischemic stroke endovascular treatment to participate in medical decision-making. It is urgent to develop new technical solutions to improve the efficiency and quality of medical decision-making, improve the prognosis of patients, and reduce the risk of doctor-patient disputes. Summary of the Invention

[0010] The present invention aims to provide a system for simulating the stroke process in combination with virtual reality. By integrating virtual reality technology, multi-scale modeling, deep learning, rule engine and blockchain technology, an intelligent auxiliary system integrating pathological display, function evaluation, treatment planning, decision-making collaboration and process traceability is constructed to solve the problems of low diagnosis and treatment efficiency, high decision-making risk and difficult data traceability in the prior art.

[0011] To achieve the above object, the system proposed by the present invention includes the following modules:

[0012] 1. Pathology Demonstration Unit

[0013] The Pathology Demonstration Unit simulates the stroke lesion process by constructing a three-dimensional medical virtual scene and combining a multi-scale hemodynamic model. At the macroscopic scale, the continuous medium equation is used to describe the characteristics of cerebral blood flow distribution; at the microscopic scale, platelets are modeled as discrete particles with adhesion characteristics, and the dynamic process of thrombus formation is simulated through the force function between platelets and endothelial cells. The blood vessel wall adopts a nonlinear anisotropic constitutive model, and the real-time visualization of stroke lesions is realized by combining the dynamic display function, providing an intuitive display of the abnormal area of cerebral hemodynamics and the ischemic range.

[0014] 2. Functional Evaluation Unit

[0015] The Functional Evaluation Unit performs brain region segmentation on the patient's cranial images through a deep learning model, and registers with a pre-set functional area template to generate an individualized functional area mapping. Combining computer vision technology to collect the patient's physical sign data, quantifying the neurological deficit conditions, and the evaluation contents include indicators such as consciousness level, gaze, visual field, facial paralysis, upper and lower limb movement, ataxia, sensation, language, dysarthria, and neglect. At the same time, a prognosis prediction model is established through the random forest algorithm, and the treatment effect and recovery possibility are predicted according to the patient's current evaluation data.

[0016] 3. Treatment Planning Unit

[0017] Based on the functional evaluation results, the Treatment Planning Unit calls the stroke treatment guideline rule base through a rule engine, and infers applicable treatment plans (such as intravenous thrombolysis, mechanical thrombectomy, or conservative treatment). Combining the complication prediction model, comprehensively analyzing the patient's characteristics and examination indicators, and using a multi-objective optimization algorithm to perform benefit and risk balance analysis, and output an individualized treatment plan, including treatment path, drug dosage, and complication risk prompts, etc.

[0018] 4. Decision-making Collaboration Unit

[0019] The Decision-making Collaboration Unit standardizes the medical term expressions in the treatment planning scheme through a term standardization processing mechanism, establishes a multi-party collaboration platform based on a distributed architecture, and supports real-time audio and video consultations and medical image sharing. Through an opinion fusion model based on evidence level, integrates the treatment suggestions of the multi-disciplinary team, and uses a structured discussion template to guide the rapid achievement of a diagnosis and treatment consensus.

[0020] 5. Process Tracing Unit

[0021] The process tracing unit uses blockchain technology to record all key nodes of the diagnosis and treatment process throughout, generates encrypted hash values to ensure data integrity, and supports medical liability division and data traceability analysis through a compliance audit mechanism.

[0022] The advantages of the present invention are as follows:

[0023] 1. Provide three-dimensional pathological display: Use a multi-scale hemodynamic model to simulate the processes of platelet aggregation and plaque formation, and display the processes of carotid artery stenosis, thrombosis, and cerebral tissue ischemia in a virtual reality three-dimensional scene to achieve multi-dimensional observation.

[0024] 2. Achieve quantitative functional evaluation: Use deep learning brain segmentation technology to construct a functional area mapping, combine computer vision to collect physical sign data, and establish a standardized scoring system for neurological deficits.

[0025] 3. Establish an information conversion mechanism: Use a medical term conversion algorithm to standardize professional terms, and integrate medical image transmission and multi-party real-time communication functions through a remote consultation platform.

[0026] 4. Construct an auxiliary decision-making framework: Based on a clinical guideline rule base and a complication prediction model, use a multi-objective optimization algorithm to conduct a benefit-risk balance analysis to form a standardized decision support system.

[0027] 5. Establish a data traceability system: Use blockchain technology to encrypt and store key decision-making information, and achieve synchronous recording of information through distributed data collection to form a complete data traceability system. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Shows a schematic diagram of the system architecture;

[0029] Figure 2 Shows the system working flow chart. DETAILED DESCRIPTION OF THE INVENTION

[0030] Combined with Figure 1 , the present invention proposes a system for combining virtual reality to simulate the stroke process, and the system composition is as follows:

[0031] 1. Pathology demonstration unit

[0032] The pathology demonstration unit constructs a three-dimensional medical virtual scene through the Unity / Unreal engine, constructs a multi-scale hemodynamic model, describes the blood flow characteristics using the continuous medium equation at the macroscopic scale, models platelets as discrete particles with adhesion characteristics at the microscopic scale, and simulates the aggregation process through the platelet-endothelial cell force function; the blood vessel wall adopts a non-linear anisotropic constitutive model to achieve real-time simulation of the lesion process. The specific implementation method is as follows:

[0033] A. Construction of 3D Medical Virtual Scenes

[0034] A 3D medical scene is constructed using a computer graphics engine, which can be a Unity engine or an Unreal engine. The scene construction process includes the following steps:

[0035] First, 3D reconstruction of the patient's angiography data is required. Since medical images generally have noise interference, anisotropic diffusion filtering is used for preprocessing, and its mathematical expression is:

[0036]

[0037] where: I represents the image gray value matrix, and the domain is the entire image region Ω; t represents the diffusion time parameter, t ∈ [0, T], and T is the maximum diffusion time; c(·) represents the diffusion coefficient function, and its expression is: c(x) = exp(-(x / κ) 2 ), where κ is an adjustable diffusion parameter determined by optimizing the validation set; div is the divergence operator used to calculate the divergence of the vector field; is the gradient operator used to calculate the gradient of the function.

[0038] Next, the level set method is used for blood vessel contour extraction, and its evolution equation is:

[0039]

[0040] where: φ represents the level set function, defined on the computational domain Ω; g(I) represents the stopping function based on the image gradient; κ represents the curvature term; ν represents the dilation coefficient; α represents the weight coefficient; the parameters are determined based on the training data through an optimization algorithm; · represents the dot product operation.

[0041] This step aims to provide doctors with an accurate cerebrovascular anatomical structure model, and by clearly showing the stenosis and thrombus sites, it assists doctors in quickly judging the scope and severity of the lesion.

[0042] B. Construction of Hemodynamic Model

[0043] At the macroscopic scale, the incompressible Navier-Stokes equation is used to describe the blood flow characteristics:

[0044]

[0045] where: v represents the velocity vector field, defined on the fluid domain Ω f ; p represents the pressure field; ρ represents the blood density; μ represents the blood dynamic viscosity, and its value is calibrated according to experimental data.

[0046] The boundary conditions are set as follows: the inlet boundary Γ inApply a time-varying velocity profile: v = v 0 (t); on the wall boundary Γ w Adopt the no-slip condition: v = 0; on the outlet boundary Γ out Adopt the free outflow condition: At the microscale, platelets are modeled as discrete particles with adhesion properties, and their motion equation is:

[0047]

[0048] where: F h represents the fluid force, which is calculated from the local flow field; F c represents the intercellular force; F b represents the Brownian motion force, which is simulated using a stochastic process.

[0049] The intercellular force is described using the Morse potential function:

[0050]

[0051] U(r) = D e [exp(-2β(r - r e )) - 2exp(-β(r - r e ))]

[0052] where: D e represents the potential well depth parameter; β represents the potential function shape parameter; r e represents the equilibrium distance; r represents the actual distance between cells; the parameters are calibrated through molecular dynamics simulation results.

[0053] Generally speaking, the macroscopic scale simulates the cerebral blood flow distribution, providing a quantitative basis for doctors to judge the location of the ischemic core area and the penumbra; the microscopic scale reveals the dynamic process of thrombus formation, providing support for the use of thrombolytic drugs and the feasibility analysis of mechanical thrombectomy.

[0054] C. Vascular wall mechanical model

[0055] Considering the non-linear deformation characteristics of the vascular wall, a non-linear anisotropic hyperelastic model is adopted to describe its mechanical properties:

[0056]

[0057] where: W represents the strain energy density function; c 1 represents the matrix material parameter; k 1 、k 2 represent the fiber material parameters; I 1 represents the first invariant of the right Cauchy-Green deformation tensor; I 4Denote the stretch invariant in the fiber direction; the material parameters are calibrated through vascular wall mechanical tests.

[0058] The dynamic response of the vascular wall satisfies the following motion equation:

[0059]

[0060] where: ρ s denotes the vascular wall density; u denotes the displacement field, defined on the domain Ω s ; σ denotes the stress tensor; f denotes the external force term.

[0061] This model predicts the stress concentration in the stenotic part of the vascular wall, helping doctors evaluate the risk of arterial rupture and the potential damage of mechanical thrombectomy operation to the blood vessels.

[0062] D. Fluid-Structure Interaction Solving Strategy

[0063] The finite element method is used for spatial discretization. The streamline upwind / Petrov-Galerkin (SUPG) method is adopted for the fluid field discretization to improve the computational stability, and the total Lagrangian description is used for the structural field to handle large deformation problems.

[0064] The following strong coupling iterative strategy is adopted:

[0065] S1. Predict the fluid-structure interface position based on the results of the previous time step;

[0066] S2. Solve the fluid field control equations to obtain the fluid field distribution;

[0067] S3. Calculate the coupling force on the fluid-structure interface;

[0068] S4. Solve the structural field control equations to obtain the deformation response;

[0069] S5. Update the computational grid;

[0070] S6. Check the convergence. If not converged, return to the second step. The convergence criterion uses the displacement increment norm: ||x (k+1) -x (k) || ≤ ε, where: x represents the displacement or force at the coupling interface; k represents the number of iteration steps; ε represents the given convergence tolerance; the specific tolerance value is determined according to the calculation accuracy requirements.

[0071] The above fluid-structure interaction provides dynamic data for doctors, simulates the blood flow recovery effect after the relief of vascular stenosis, and helps evaluate the degree of improvement in cerebral tissue perfusion after treatment.

[0072] E. Visualization Implementation

[0073] The vascular surface rendering uses a physically based rendering method, and its material parameters include diffuse reflectance, specular coefficient, and roughness, which are adjusted according to the display effect. The lighting calculation uses the Phong model to provide a more realistic lighting effect.

[0074] The blood flow field visualization solution includes: the streamline density adopts an adaptive control mechanism to ensure display clarity; the velocity field adopts a gradient color scale mapping to improve the data expression effect; the transparency changes with the velocity gradient to highlight key flow features.

[0075] The interaction design solution includes: the perspective control system supports free viewing of the scene; the time control system supports dynamic adjustment of the simulation process; the control parameters can be configured according to specific application requirements.

[0076] The visualization technology enables doctors to intuitively understand the core lesion area of stroke and adjust treatment decisions in real time; it supports doctor-patient communication and helps patient representatives quickly understand the condition through interactive displays.

[0077] 2. Functional Evaluation Unit

[0078] The functional evaluation unit uses a deep learning model for brain segmentation, and registers with a pre-set brain functional area template to generate an individualized functional area map; combines computer vision technology to collect patient sign data for quantitative evaluation of neurological deficits, and the evaluation items include level of consciousness, gaze, visual field, facial palsy, upper and lower limb movement, ataxia, sensation, language, dysarthria, neglect; establishes a prognosis prediction model through the random forest algorithm. The specific implementation is as follows:

[0079] A. Construction of Brain Segmentation Model

[0080] Deep convolutional neural network is used for brain segmentation, and the network structure adopts an encoder-decoder architecture, and its mathematical expression is:

[0081] F(x) = D(E(x))

[0082] Where: represents the input brain image data, H represents the image height, W represents the image width, C represents the number of channels; E represents the encoder mapping function that maps the input to the feature space; D represents the decoder mapping function that maps the features back to the image space, and M represents the number of target segmentation categories.

[0083] Encoder feature extraction process:

[0084] h l = f l (W l * h l-1 + b l )

[0085] Wherein: represents the feature map of the l-th layer; represents the convolutional kernel weight, and k represents the convolutional kernel size; represents the bias term; represents the activation function, and the ReLU function f l (x) = max(0, x); * represents the convolution operation; l ∈ {1, 2,..., L} represents the network layer index.

[0086] Decoder upsampling process:

[0087]

[0088] Wherein: represents the reconstructed feature map of the l-th layer; represents the transposed convolutional kernel weight; represents the bias term; g l represents the activation function, and the Softmax function is used for the last layer, and the ReLU function is used for other layers; represents the transposed convolution operation.

[0089] Brain region segmentation and functional mapping can enable doctors to accurately identify damaged areas (such as functional areas responsible for movement and language), and guide subsequent treatments (such as the adaptive evaluation of mechanical thrombectomy).

[0090] Provide a quantitative description of stroke-related brain regions, providing an anatomical basis for prognosis evaluation.

[0091] B. Registration mapping generation

[0092] The spatial correspondence between the individual image and the standard template is achieved by using a non-rigid registration algorithm, and the registration process optimizes the following energy functional:

[0093] E(T) = E D (I 1 , T(I 2 )) + λE R (T)

[0094] Wherein: T represents the spatial transformation operator; I 1 , I 2 represent the target image and the floating image respectively; E D represents the data term; E R represents the regularization term; λ > 0 represents the weight coefficient.

[0095] The data term is measured by mutual information:

[0096] E D = -∑ i,j p(i, j) log p(i, j) / p 1 (i) p 2(j)

[0097] where: p(i, j) represents the joint probability density, p(i, j) ≥ 0 and ∑ i,j p(i, j) = 1; p 1 (i) and p 2 (j) represent the marginal probability densities; i, j ∈ {1, 2,..., N b} represents the gray value quantization level, and N b represents the number of histogram bins.

[0098] C. Sign data collection

[0099] Use a depth camera to obtain the patient's motion state data, and extract the feature point coordinates through a human key point detection algorithm: P = {p i = (x i , y i , z i ) | i = 1, 2,..., N}, where: represents the three-dimensional coordinates of the i-th key point; represent the position components in the spatial Cartesian coordinate system respectively; N represents the total number of key points, which is determined by the anatomical feature points of the human body.

[0100] Calculation of motion parameters:

[0101]

[0102] where: θ j ∈ [0, π] represents the j-th joint angle; represents the adjacent bone vector; j ∈ {1, 2,..., J} represents the joint index, and J represents the total number of joints.

[0103] D. Neurological function assessment

[0104] Establish a functional scoring model based on the obtained sign data:

[0105]

[0106] where: S k ∈ [0, S max represents the score of the k-th function, and S max represents the highest score; w i ∈ [0, 1] represents the weight coefficient, and f i represents the sub-item scoring function; represents the feature vector, d represents the feature dimension; k ∈ {1, 2,..., K} represents the functional assessment item index, K represents the total number of assessment items; M represents the number of sub-items for scoring each function.

[0107] Feature definitions for various function evaluations:

[0108] Consciousness level evaluation: Including pupil activity and expression changes, scoring function f 1 :

[0109]

[0110] Among them:

[0111] A pupil,i : Pupil activity, with the value range being the change range of pupil diameter, unit: millimeter (mm), measured by an eye tracker; A min ,A max : The minimum and maximum standard values of pupil activity, taking 2mm and 9mm respectively, based on medical reference standards; E expression,i : The amplitude of expression change, unit: pixel, calculated by the displacement of expression feature points; E max : The maximum standard value of expression change, with the value being 30 pixels (based on the experimental measurement of the normal expression change amplitude); w pupil,i ,w expression,i : Weight coefficient, and w pupil,i +w expression,i = 1.

[0112] Gaze evaluation: Including eye movement trajectory and speed, scoring function f 2 :

[0113]

[0114] Among them: v eye,i (t): Eye movement speed, unit: degree / second, measured by an eye movement tracker; T = t 1 -t 0 : Total test time, unit: second, taking 10 seconds; σ trajectory,i : Standard deviation of eye movement trajectory, unit: pixel, calculated by the variance of movement trajectory data;

[0115] γ: Adjustment coefficient, with the default value being 0.2, based on experimental optimization.

[0116] Visual field evaluation: Including visual stimulus position and reaction time, scoring function f 3 :

[0117]

[0118] Among them: R j : The accuracy of the jth visual stimulus response, with the value range [0,1], calculated by the correct answer rate; T j: The corresponding response time, unit: second, recorded by the key press reaction time; n: The total number of stimuli, set to 20 times.

[0119] Facial paralysis assessment: Including facial feature point displacement, symmetry, scoring function f 4 :

[0120]

[0121] Where: Δd symmetry,i : Facial symmetry deviation, unit: pixel, calculated by the distance difference of facial feature points on both sides; Δd movement,i : Amplitude of facial feature point displacement, unit: pixel; d max : Normalized maximum value, set to 50 pixels.

[0122] Upper and lower limb movement assessment: Including joint range of motion, movement speed, scoring function f 5 :

[0123]

[0124] Where: θ k : The movement angle of the k-th joint, unit: degree, calculated by the motion capture system; θ max : Maximum joint movement angle, referring to the normal physiological range of motion, taken as 90 degrees; v k : Movement speed, unit: degree / second; v max : Maximum value of normal movement speed, taken as 150 degrees / second; M: Total number of joints evaluated, taken as 8 (major joints of the upper and lower limbs).

[0125] Ataxia assessment: Including movement trajectory curvature, acceleration, scoring function f 6 :

[0126]

[0127] Where: κ j : Curvature deviation of the j-th movement trajectory, unit: dimensionless, calculated by the trajectory curvature; n: Number of tests, set to 10 times.

[0128] Sensory assessment: Including stimulus intensity, reaction delay, scoring function f 7 :

[0129]

[0130] Where: F j : Stimulus intensity, unit: Newton, measured by the force sensor; F max: Standardized maximum stimulus intensity, take 5 Newtons; t j : Reaction delay, unit: second; t 0 : Basic reaction time, take 0.2 seconds; λ: Parameter controlling reaction sensitivity, take 3.

[0131] Language assessment: Including voice features, semantic understanding, scoring function f 8 :

[0132]

[0133] Among them: P semantic,m : The accuracy rate of semantic understanding of the m-th item, value range [0,1]; C phonetic,m : Clarity of voice features, unit: decibel (dB), calculated through acoustic analysis; M: Total number of test sentences, set to 10 sentences.

[0134] Articulation assessment: Including pronunciation accuracy, fluency, scoring function f 9 :

[0135] f 9 = A accuracy / A max ·(1 - σ pause / T speech )

[0136] Among them: A accuracy : Pronunciation accuracy, unit: dimensionless, calculated through the speech recognition system; A max : Maximum accuracy of normal pronunciation, value 1; σ pause : Standard deviation of pause time in speech, unit: second; T speech : Total pronunciation time, unit: second.

[0137] Neglect assessment: Including spatial attention distribution characteristics, scoring function f 10 :

[0138]

[0139] Among them: A attention (x,y): Attention distribution, unit: probability density; A ideal (x,y): Ideal attention distribution; A ROI : Area of the region of interest, unit: square pixel.

[0140] Among them d 1 ,d 2 ,...,d 10 Respectively represent the feature dimensions of each functional assessment.

[0141] By collecting and analyzing multi-modal data, subjective evaluation errors are avoided, and the objectivity and accuracy of functional evaluation are improved; refined analysis of the patient's motor function and language ability is provided, laying a foundation for rehabilitation treatment planning.

[0142] D. Prognosis prediction model

[0143] A prognosis prediction model is constructed using the random forest algorithm:

[0144]

[0145] Where: represents the prediction result, indicating the probability of good prognosis; represents the number of decision trees; represents the prediction function of the k-th decision tree; represents the input feature vector, D represents the total dimension of features; each decision tree T k is constructed through the following recursive process:

[0146] S1. Node splitting criterion:

[0147] G(S,j,t)=|S L | / |S|H(S L )+|S R | / |S|H(S R )

[0148] Where: S represents the sample set of the current node; j∈{1,2,...,D} represents the feature index; represents the splitting threshold; S L ={(x,y)∈S|x j ≤t} represents the left child node sample set; S R ={(x,y)∈S|x j >t} represents the right child node sample set; p c represents the information entropy, p c represents the proportion of class c.

[0149] S2. Feature importance evaluation:

[0150]

[0151] Where: I j ≥0 represents the importance score of the j-th feature; Δi(j,k)≥0 represents the information gain of feature j in tree k; j∈{1,2,...,D} represents the feature index. Model performance evaluation uses multiple cross-validation:

[0152] Where: respectively represent the number of positive and negative samples; fi , f j ∈ [0, 1] represents the predicted score of the sample; I represents the indicator function, which takes 1 when the condition holds and 0 otherwise; AUC ∈ [0, 1] represents the area under the receiver operating characteristic curve.

[0153] Systematic prognosis prediction provides doctors with an intuitive reference for treatment risks and benefits, especially applicable for treatment decision-making support for high-risk patients; it provides understandable recovery probability data for patients and their families, helping surrogate decision-makers participate in the selection of treatment plans.

[0154] F. Medical scenario support

[0155] The core application scenarios of the function evaluation unit in clinical practice include:

[0156] a). Diagnosis and evaluation in the acute stage of stroke

[0157] Quickly identify the degree of brain function deficit in patients, providing data support for thrombolytic or thrombectomy treatment.

[0158] Provide real-time function scores, shortening the time required for traditional manual evaluation.

[0159] b). Precision treatment decision-making

[0160] The personalized scoring results generated by the system are directly connected to the treatment planning unit, guiding doctors to optimize the treatment path.

[0161] Help doctors evaluate the potential risk of complications (such as intracranial hemorrhage) through quantitative analysis results.

[0162] c). Patient prognosis evaluation and communication

[0163] The prognosis prediction model updated at any time provides scientific recovery predictions for patients and their families, helping surrogate decision-makers reasonably select treatment plans.

[0164] 3. Treatment planning unit

[0165] The treatment planning unit formulates treatment plans based on the function evaluation results, performs plan reasoning by calling the clinical guideline rule base through a rule engine; establishes a complication prediction model, comprehensively analyzes patient characteristics and examination indicators, and conducts benefit-risk balance analysis through a multi-objective optimization algorithm to form treatment plans. The specific implementation methods are as follows:

[0166] A. Rule base construction

[0167] The system builds a rule base of stroke treatment guidelines (such as the AHA / ASA guidelines), covering the decision-making paths of intravenous thrombolysis, mechanical thrombectomy, and conservative treatment. The rule engine infers the applicable treatment plan through the following condition matching:

[0168] Time window determination: Infer the adaptability of intravenous thrombolysis (3 - 4.5 hours) or mechanical thrombectomy (6 - 24 hours) based on the patient's arrival time at the hospital and the onset time.

[0169] Lesion feature matching: Determine the treatment feasibility by combining imaging data (such as thrombus location, cerebral blood flow parameters).

[0170] Patient's basic status: Evaluate the applicability of the treatment plan comprehensively considering age, medical history, and functional score.

[0171] The rules are represented in production form: R = {r i |i = 1, 2,..., N R}

[0172] Where: r i represents the i-th rule; represents the total number of rules; i represents the rule index; Each rule r i is in the form: If C i then A i (CF i ), where, C i represents the condition set, c ij represents the j-th condition of the i-th rule, j ∈ {1, 2,..., m i}; A i represents the action set, a ik represents the k-th action of the i-th rule, k ∈ {1, 2,..., n i}; CF i ∈ [0, 1] represents the certainty factor of the rule; represents the number of conditions of the i-th rule; represents the number of actions of the i-th rule;

[0173] Condition matching function:

[0174]

[0175] Where: F = {f 1 , f 2 ,..., f M} represents the fact set, represents the j-th feature value; μ ij represents the membership function of the j-th condition of the i-th rule; represents the feature dimension.

[0176] B. Inference mechanism

[0177] Adopt the forward inference mechanism:

[0178]

[0179] Among them: Indicates the rule subset with the conclusion of A k ; Conf(A k ) ∈ [0, 1] indicates the confidence of conclusion A k ; k represents the conclusion index, k ∈ {1, 2,..., N A}, Indicates the total number of possible conclusions;

[0180] The conflict resolution strategy formula is Among them, Utility represents the utility function; Indicates the set of all possible actions; A * Indicates the finally selected action.

[0181] C. Complications prediction model

[0182] Construct a risk prediction model based on patient characteristics:

[0183]

[0184] Among them: Indicates the patient characteristic vector; y ∈ {0, 1} indicates the complication occurrence indicator variable, 1 indicates occurrence, and 0 indicates non-occurrence; P(y|x) ∈ [0, 1] indicates the conditional probability of complication occurrence given the characteristic x; Indicates the feature dimension; g represents the risk scoring function, and the risk scoring function adopts an additive model:

[0185]

[0186] Among them: Indicates the intercept term; Indicates the weight coefficient of the jth feature; h j Indicates the transformation function of the jth feature; x j Indicates the jth feature component.

[0187] D. Multi-objective optimization

[0188] Establish an optimization model for treatment plans, and the objective function is:

[0189] min θ {f 1 (θ), f 2 (θ),..., f K (θ)}

[0190] Among them: Indicates the treatment plan parameter vector; Θ indicates the set of feasible plans; f kdenotes the k-th optimization objective function; denotes the number of objective functions; denotes the dimension of the solution parameters.

[0191] The constraint conditions are:

[0192] g i (θ) ≤ 0, i = 1, 2, ..., m

[0193] h j (θ) = 0, j = 1, 2, ..., n

[0194] where: g i denotes the i-th inequality constraint function; h j denotes the j-th equality constraint function; denotes the number of inequality constraints; denotes the number of equality constraints.

[0195] Population update: P t+1 = Select(P t ∪ Q t )

[0196] where: denotes the population at the t-th generation, denotes the generation number of evolution; denotes the offspring generated by crossover and mutation; Select denotes the selection operator based on non-dominated sorting E. Solution generation

[0197] The solution scoring formula is:

[0198]

[0199] where: w k ∈ [0, 1] denotes the weight of the k-th objective, and denotes the k-th objective function after normalization; Score denotes the comprehensive scoring function of the solution.

[0200] Final solution selection:

[0201] where: denotes the Pareto optimal solution set; θ opt denotes the parameters of the finally selected treatment plan.

[0202] 4. Decision collaboration unit

[0203] The decision-making collaboration unit standardizes the terminology of treatment planning schemes; establishes a multi-party collaboration platform based on a distributed architecture, ensures the security of consultations through a two-way authentication mechanism, and realizes real-time sharing of audio and video transmission and medical images; establishes an opinion fusion model based on evidence levels, and uses a structured discussion template to guide consensus. The specific implementation methods are as follows:

[0204] A. Terminology Standardization

[0205] a). Build a standard terminology knowledge graph based on the medical terminology mapping model:

[0206] G=(V,E,Λ)

[0207] Where: V = {v i |i=1,2,...,N V} represents a term node set; E = {e k =(v i ,v j ,r)|i,j∈{1,...,N V},r∈R} represents the edge set; Λ:V∪E→A represents the attribute mapping function; Represents the total number of nodes; R represents the set of relationship types; A represents the attribute value domain.

[0208] b). Node attribute vector:

[0209] P(v)=[p 1 (v),p 2 (v),...,p L (v)]

[0210] Where: v∈V represents a term node; p l represents the lth attribute function; Indicates the value range of the lth attribute; Represents an attribute dimension.

[0211] c).Term similarity calculation:

[0212] sim(t,s)=α·lexical(t,s)+β·semantic(t,s)+γ·context(t,s)

[0213] in: denotes the pair of terms to be compared, represents the term space; lexical represents the string similarity function; semantic represents the semantic similarity function; context represents the context similarity function; α, β, γ∈[0,1] represent the weight coefficients, and α+β+γ=1.

[0214] B. Multi-party collaboration mechanism

[0215] a). Identity authentication protocol:

[0216] Auth(i,j) = {Token ij , Sign(H(m ij ), SK i )}

[0217] Where: i, j ∈ {1, 2,..., N P} represents the participant index; represents the number of participants; Token ij represents the session token; m ij represents the authentication message; H represents the cryptographic hash function; SK i represents the private key of participant i; Sign represents the digital signature function.

[0218] b). Real-time data transmission configuration:

[0219] StreamConfig = {ω a , ω v , τ, δ}

[0220] Where: ω a ∈ Ω a represents the audio coding parameter set; ω v ∈ Ω v represents the video coding parameter set; represents the maximum allowable delay; represents the minimum bandwidth requirement.

[0221] C. Opinion fusion model

[0222] a). Evidence level assessment:

[0223] E(o) = w 1 L(o) + w 2 C(o) + w 3 R(o)

[0224] Where: represents the expert opinion, represents the opinion space; L represents the evidence level scoring function; C represents the clinical relevance scoring function; w 1 , w 2 , w 3 ∈ [0, 1] represents the weight coefficient, and w 1 + w 2 + w 3 = 1; R represents the research quality scoring function:

[0225] R = w impact · S impact + wrigor ·S rigor +w validity ·S validity

[0226] Among them: the research quality score R reflects the overall quality of the cited research for collaborative decision support, and its value range is [0, 1]; w impact , w rigor , w validity : weight coefficients, which measure the importance of clinical influence, research rigor, and evidence validity respectively, and w impact +w rigor +w validity = 1; S impact , S rigor , S validity : corresponding sub-scoring functions, and their value ranges are all [0, 1]. The definitions are as follows:

[0227] 1. Clinical influence score S impact

[0228] Definition: Measure the clinical applicability and influence of the cited research in stroke diagnosis and treatment, including the coverage of the target population and the direct clinical application value of the research results.

[0229] Expression: S impact = N relevant / N total

[0230] Among them: N relevant : the number of patient samples related to stroke; N total : the total number of patient samples in the research.

[0231] 2. Research rigor score S rigor

[0232] Definition: Quantify the scientific design and data analysis quality of the cited research, including the randomization method, sample size, and correctness of statistical methods, etc.

[0233] Expression: S rigor = w design ·S design +w analysis ·S analysis

[0234] Among them: S design : research design score, which measures whether randomization and reasonable control groups are adopted; S analysis : data analysis score, which measures whether the statistical method is rigorous and the conclusion is reliable; w design , w analysis : weight coefficients.

[0235] 3. Evidence validity score S validity

[0236] Definition: Evaluate the replication of the cited study and its compatibility with existing evidence, reflecting the robustness of the research conclusion.

[0237] Expression: S validity = N verified / N cited

[0238] Where: N verified : The number of replicative studies of the cited study, indicating the number of times its conclusion is repeatedly supported in other independent studies; N cited : The total number of citations of the study.

[0239] b). Opinion synthesis strategy:

[0240]

[0241] Where: Represents the solution to be evaluated, Represents the solution space; Represents the number of opinions collected; I represents the support indication function; Vote represents the weighted voting score of the solution.

[0242] D. Structured discussion process

[0243] a). Discussion state transition:

[0244] S t+1 = T(S t , A t , C t )

[0245] Where: Represents the discussion state at time t, Represents the state space; Represents the discussion behavior at time t, Represents the behavior space; Represents the constraint condition at time t, Represents the constraint space; T represents the state transition function; Represents the discussion time step.

[0246] b). Discussion progress evaluation:

[0247]

[0248] Where: φ k Represents the k-th progress index function; Represents the index weight; Represents the number of progress indicators; P represents the overall progress evaluation function.

[0249] c). Consensus determination criterion:

[0250]

[0251] Where: η ∈ [0, 1] represents the progress threshold; θ ∈ [0, 1] represents the consistency threshold; Consensus represents the consensus achievement indicator function.

[0252] 5. Process Traceability Unit

[0253] The process traceability unit records the entire process of stroke diagnosis and treatment decisions through blockchain technology, achieving data integrity verification and process compliance auditing. The specific implementation method is as follows:

[0254] A. Decision Data Storage Structure

[0255] Organize the diagnosis and treatment process data into a time-series event sequence:

[0256] E = {e i =(d i , t i , s i )|i = 1, 2,..., n}

[0257] Where: e i represents the i-th event; d i ∈ {0, 1} * represents the event data; represents the timestamp, accurate to milliseconds; s i ∈ {0, 1} 256 represents the digital signature; represents the total number of events. The data block d i contains three parts:

[0258] Functional evaluation data: Among them, nihss i ∈ [0, 42] represents the NIHSS score result; represents the physical sign data vector, m is the feature dimension; predict i ∈ [0, 1] represents the prognosis prediction probability;

[0259] Treatment planning data: Among them, rule i ∈ {0, 1} k represents the rule matching result vector, k is the number of rules; risk i ∈ [0, 1] p represents the risk assessment vector, p is the number of risk types; represents the treatment plan code, q is the number of plan parameters.

[0260] Decision collaboration data: Among them, represents the expert opinion vector, and r is the number of experts; evidence i ∈ [1, 5] r represents the evidence level vector; consensus i ∈ {0, 1} indicates whether a consensus is reached.

[0261] B. Integrity verification mechanism

[0262] Define the hash function: h: {0, 1} * → {0, 1} 256 , implemented based on the SHA-256 algorithm, with the input being a bit string of any length and the output being a 256-bit hash value;

[0263] Data block verification function:

[0264] V d (d i ) = [h(d i ) = H i ∧ [t i > t i-1 ∧ [s i = sign(d i )]

[0265] Among them: H i ∈ {0, 1} 256 represents the stored hash value; sign: {0, 1} * → {0, 1} 256 represents the signature function.

[0266] Evidence chain verification function:

[0267]

[0268] Among them, L(C) is the chain integrity verification function: B i represents the i-th block; B i .prev represents the previous hash value stored in the block.

[0269] C. Compliance audit algorithm

[0270] Audit rule set definition:

[0271] R = {r j = (item j , threshold j , weight j )|j = 1, 2,..., m}

[0272] Among them: item j represents the inspection item evaluation function; represents the compliance threshold; weight j ∈[0,1] represents the rule weight; represents the total number of rules.

[0273] Rule compliance calculation function:

[0274]

[0275] Among them, eval represents the evaluation function.

[0276] Overall compliance score:

[0277]

[0278] Among them, Score represents the compliance scoring function.

[0279] D. Process traceability algorithm

[0280] Decision dependence graph: G=(V,E)

[0281] Among them: V={v i |i = 1,2,...,|V|} represents the decision event set; represents the event dependence relationship set.

[0282] Critical path identification function:

[0283] Among them: pred(v i ) represents the set of predecessor nodes of node v i ; P represents the node importance evaluation function; θ∈[0,1] represents the importance threshold.

[0284] Path integrity verification function:

[0285] E. Performance metric calculation

[0286] Time efficiency metric: E t =T standard / T actual ×100%

[0287] Among them: represents the standard processing time; represents the actual processing time; E t ∈[0,∞) represents the time efficiency percentage.

[0288] Data quality metric:

[0289]

[0290] where: q i ∈ [0, 1] represents the quality score of the i-th item; w i ∈ [0, 1] represents the corresponding weight; Q d ∈ [0, 100] represents the data quality percentage.

[0291] Combined with Figure 2 , the workflow of this system is as follows:

[0292] S1. Pathological process simulation and visualization

[0293] The pathological demonstration unit receives imaging data such as the patient's head CT and CTA, as well as clinical data, constructs a three-dimensional medical virtual scene through the Unity / Unreal engine, constructs a multi-scale hemodynamic model, describes the blood flow characteristics using the continuous medium equation at the macroscopic scale, models platelets as discrete particles with adhesion characteristics at the microscopic scale, and simulates the aggregation process through the platelet-endothelial cell force function; the blood vessel wall adopts a nonlinear anisotropic constitutive model to realize real-time simulation of the lesion process. The simulation results are used to assist in the assessment of functional impairment.

[0294] S2. Quantitative evaluation of neurological function

[0295] Based on the simulation results of the pathological demonstration unit, the functional evaluation unit uses a deep learning model to perform brain region segmentation, and registers with a pre-set brain functional area template to generate an individualized functional area mapping; combines computer vision technology to collect the patient's physical sign data, and conducts a quantitative evaluation of neurological deficits. The evaluation items include the level of consciousness, gaze, visual field, facial palsy, upper and lower limb movement, ataxia, sensation, language, dysarthria, and neglect; establishes a prognosis prediction model through the random forest algorithm. The evaluation results are transmitted to the treatment planning unit.

[0296] S3. Personalized treatment planning

[0297] The treatment planning unit obtains the evaluation results of the functional evaluation unit, formulates a treatment plan based on the evaluation results, and performs plan reasoning by calling the clinical guideline rule base through a rule engine; establishes a complication prediction model, comprehensively analyzes the patient's characteristics and examination indicators, and conducts a benefit-risk balance analysis through a multi-objective optimization algorithm to form a treatment plan. The planned treatment plan is transmitted to the decision-making collaboration unit for multi-party demonstration.

[0298] S4. Multi-party collaborative decision-making

[0299] The decision-making collaboration unit receives the treatment plan from the treatment planning unit and standardizes the terms of the treatment plan; it establishes a multi-party collaboration platform based on a distributed architecture, ensures the security of the consultation through a two-way authentication mechanism, and realizes real-time audio and video transmission and real-time sharing of medical images; it establishes an opinion fusion model based on the evidence level and uses a structured discussion template to guide the achievement of consensus. The decision-making process is transmitted to the process tracing unit in real time.

[0300] S5. Full-process tracing

[0301] The process tracing unit records the working processes of the above-mentioned units throughout the process, realizes the full-process tracing of the decision-making process, marks the key decision-making nodes using blockchain technology, ensures the integrity of the evidence chain through hash verification, and establishes a compliance audit mechanism to conduct traceability analysis on the decision-making process. The system finally outputs the visualized results of lesion simulation, the report on neurological function assessment, the personalized treatment plan, the multi-party consensus decision-making result, and the full-process traceability analysis report.

[0302] Adopting the above technical solutions, the present invention realizes the precise virtual reality simulation of the stroke occurrence process through the pathological demonstration unit, the quantitative assessment of neurological function impairment is completed by the function assessment unit, the personalized treatment plan is formulated by the treatment planning unit, the decision-making collaboration unit supports multi-party real-time collaborative decision-making, and the process tracing unit ensures the full-process data traceability, thereby providing systematic technical support for the rapid and accurate diagnosis and treatment decision-making in the acute stage of stroke.

[0303] In terms of the scope of rights protection, the present invention is not limited to the specific implementation manners described in the text. On the contrary, it should be clear that the scope of rights protection of the present invention covers all reasonable changes, modifications, and equivalent implementation manners made based on the content of this document. The scope of rights protection we apply for is intended to encompass all innovative points of the present invention and its potential applications, so as to ensure that our intellectual property rights can be fully protected and respected.

Claims

1. A system for simulating stroke process in combination with virtual reality, characterized in that: include: A) A pathology demonstration unit, used to construct a three-dimensional medical virtual scene, wherein the pathology demonstration unit includes a multi-scale hemodynamic model, which is used to describe blood flow characteristics using a continuous medium equation at a macroscopic scale, and to model platelets as discrete particles with adhesion characteristics at a microscopic scale; B). Functional evaluation unit, used to receive the simulation results of the pathological demonstration unit, use the deep learning model to segment the brain area and generate individualized functional area mapping, and combine computer vision technology to collect patient vital sign data for quantitative evaluation of neurological function deficit; C) A treatment planning unit, which is used to call the clinical guideline rule base through a rule engine to perform scheme reasoning based on the evaluation results of the functional evaluation unit, and to perform benefit-risk balance analysis through a multi-objective optimization algorithm; D) Decision-making collaboration unit, used to standardize the terminology of the treatment plan generated by the treatment planning unit, establish a multi-party collaboration platform through a distributed architecture, and realize real-time sharing of medical images and fusion of opinions based on evidence level; E). Process tracing unit, which uses blockchain technology to record the entire decision-making process and ensure the integrity of the evidence chain through hash verification.

2. The system according to claim 1, characterized in that In the pathological demonstration unit: the vascular wall adopts a nonlinear anisotropic constitutive model, including: Wherein: W represents the strain energy density function; c1 represents the matrix material parameters; k1 and k2 represent the fiber material parameters; I1 represents the first invariant of the right Cauchy-Green deformation tensor; I4 represents the fiber direction tensile invariant; the material parameters are calibrated by vascular wall mechanics tests.

3. The system according to claim 1, characterized in that The quantitative assessment of neurological deficits in the functional assessment unit includes assessment of consciousness level, gaze, visual field, facial paralysis, upper and lower limb movement, ataxia, sensation, language, articulation, and neglect, and a prognosis prediction model is established using a random forest algorithm.

4. The system according to claim 1, characterized in that The terminology standardization process in the decision-making collaborative unit builds a standard terminology knowledge graph based on the medical terminology mapping model, and the knowledge graph includes: Node set: conceptual nodes used to represent different medical terms, each node corresponds to an independent medical term; Edge set: used to represent the semantic relationship between terms, including synonymy, subordination, and causal relationships; Attribute mapping function: used to describe the context information of a term node, wherein the context information includes the term's indication range, associated treatment plans, and clinical applicability restrictions.

5. The system according to claim 1, characterized in that The process tracing unit uses blockchain technology to realize the tamper-proof record of the decision-making process, and its storage structure includes: Decision event data: records include timestamps of treatment plan generation, modification, and collaborative decision consensus; Hash check field: used to verify the integrity of decision data and generate a unique identifier through a hash function; Signature verification module: used to record and trace the identity information of decision-making participants, and use digital signature technology to ensure the non-repudiation of the identities of the participants.

Images