Postoperative cerebral apoplexy analysis method, system and device based on multi-modal image analysis

Abstract

The invention discloses a postoperative cerebral apoplexy analysis method based on multi-modal image analysis, and the method comprises the following steps: inputting medical data, including neurological function evaluation of a patient, intraoperative parameters, and preoperative and postoperative DWI images; labeling the medical data according to whether the cerebral apoplexy after the transvascular interventional operation occurs or not, and forming a data set by the medical data and labels; constructing a deep learning model; training a deep learning model by using the data set to obtain a risk prediction model for predicting whether the cerebral apoplexy after the transvascular interventional operation occurs or not; and inputting the medical data of the patient into the risk prediction model to output a prediction result of the cerebral apoplexy risk after the transvascular interventional operation. The invention further provides a postoperative cerebral apoplexy analysis system and device. According to the method provided by the invention, super-early risk prediction, brain region positioning and individualized rehabilitation guidance of dominant stroke after the transvascular intervention operation can be realized.

Description

Technical Field

[0001] The present invention belongs to the field of medical intelligent assistance, and in particular relates to a postoperative stroke analysis method, system and device based on multimodal image analysis. Background Art

[0002] Currently, clinical practice faces three core challenges: The lack of proactive early warning for silent stroke: Current diagnosis relies on manual postoperative DWI interpretation, requiring time-consuming analysis by radiology and neurology departments, and resulting in a localization error rate of 20%-35% (particularly in the basal ganglia and small subcortical infarcts); The mechanism of symptom conversion is unclear: a lack of multimodal (imaging + clinical) prediction models makes it impossible to distinguish the risk of conversion from silent infarction to symptomatic stroke; and Inadequate timeliness of rehabilitation intervention: Clinicians typically initiate intervention after the onset of overt symptoms (median delay >24 hours), missing the golden treatment window of 6 hours after stroke.

[0003] Patent CN118737436A discloses an intelligent decision-making system for stroke diagnosis and treatment. Although this system utilizes AI image analysis technology, it suffers from a single data type, relying primarily on imaging data. Furthermore, it targets stroke patients in a broad sense and fails to integrate the perioperative multimodal data specific to transaortic cardiovascular interventions, which is crucial for prognostication. Therefore, the system is unable to capture the unique risk factors for stroke after transaortic cardiovascular interventions, and naturally cannot address the aforementioned challenge of "symptom conversion prediction."

[0004] Patent CN119400394A discloses a thrombolysis prediction method based on deep learning and multimodal fusion. Although this method integrates imaging and clinical features, its application goal is to determine whether "thrombolytic therapy" is suitable. More importantly, its model construction relies on static, single-point imaging data, which is a "static assessment" model. It is unable to process and analyze the dynamic evolution characteristics of DWI images from preoperative to postoperative during the perioperative period of trans-aortic cardiovascular intervention, and also ignores the key variable of intraoperative operation. Therefore, this method cannot achieve "early and proactive warning" of overt stroke.

[0005] In summary, existing technologies are unable to effectively solve a series of interconnected clinical problems, such as risk prediction of latent cerebral infarction turning into overt stroke after trans-aortic cardiovascular intervention, localization of responsible lesions, and targeted rehabilitation guidance, either because the application scenarios are too generalized or because the data dimensions are too single and static. Summary of the Invention

[0006] The purpose of the present invention is to provide a method, system and device for postoperative overt stroke analysis based on multimodal image analysis. This method addresses the problems of delayed identification of latent infarction, inaccurate prediction of symptom conversion and lack of targeting of rehabilitation intervention. It can effectively reduce the time and cost of multi-department collaboration and shorten the time window from diagnosis to stroke intervention, thereby providing a comprehensive and reliable reference for ultra-early risk prediction of overt stroke, brain region localization and individualized rehabilitation guidance.

[0007] To achieve the first objective of the present invention, the following scheme is provided: a method for analyzing overt stroke after transaortic valve cardiovascular intervention based on multimodal image analysis, comprising the following steps: Input medical data, including the patient's neurological function assessment, intraoperative parameters, and preoperative and postoperative DWI images; Label the medical data based on whether a stroke occurred after a vascular interventional procedure, and combine the medical data and labels into a dataset. Build a deep learning model, including a multi-source data integration module, an image feature extraction module, a feature fusion module, and a prediction module; The multi-source data integration module includes an embedding layer and a fully connected layer. The embedding layer is used to map the categorical variables in the input neurological function assessment and intraoperative parameters into dense vectors. The fully connected layer is used to standardize the continuous variables in the input neurological function assessment and intraoperative parameters to output feature vectors. Multimodal feature weights are calculated based on the dense vectors and the feature vectors to obtain corresponding structured data feature vectors. The image feature extraction module is used to preprocess the DWI image using an offset correction formula and calculate the corresponding spatiotemporal differences. At the same time, a threshold segmentation algorithm is used to identify new infarcts in the DWI image and extract corresponding radiomic features. The corresponding spatiotemporal-image features are constructed based on the calculated spatiotemporal differences and the extracted radiomic features. The feature fusion module performs multimodal feature fusion on the structured data feature vector and the spatiotemporal-image feature through adaptive weight gating to output a corresponding fused feature vector; The prediction module performs prediction based on the fused feature vector to output a prediction result; Using the data set to train a deep learning model to obtain a risk prediction model for predicting whether postoperative stroke after transvascular interventional surgery will occur; The patient's medical data is input into a risk prediction model to output a prediction result of the risk of postoperative stroke after transvascular interventional surgery, wherein the prediction result includes whether postoperative stroke after transvascular interventional surgery occurs and the location of the lesion.

[0008] The present invention achieves ultra-early risk prediction of overt stroke, brain region localization and individualized rehabilitation guidance by integrating preoperative neurological function scores, preoperative and postoperative DWI images and intraoperative parameter factors.

[0009] Specifically, the transvascular interventional surgery refers to a transcatheter structural cardiac interventional surgery, which includes transcatheter aortic valve replacement (TAVR) and transcatheter mitral valve edge-to-edge suture surgery.

[0010] Specifically, the DWI image needs to be processed with a brain template before being input.

[0011] Specifically, the MNI152 template was used to perform image registration on the input DWI images, and the registration accuracy error was controlled within ±1 mm.

[0012] Specifically, the DWI images include skull DWI images before and within 24 hours after surgery, wherein the imaging parameters are: DICOM format, slice thickness ≤ 3 mm, b value = 1000 s / mm², matrix size ≥ 256×256.

[0013] Specifically, the neurological function assessment includes the modified Rankin Scale (mRS) score (0-6 points) used to reflect the preoperative neurological function status and the Mini-Mental State Examination (MMSE) score (0-30 points) used to assess cognitive function, among which the mRS score ≥2 points is input as a categorical variable for baseline functional impairment, and the MMSE score ≤24 points is significantly associated with the risk of postoperative stroke.

[0014] Specifically, the intraoperative parameters include anesthesia type, whether a second valve is implanted, prosthetic valve type, paravalvular leakage, sheath insertion time, intraoperative blood loss, left coronary sinus implantation depth, and non-coronary sinus implantation depth, among which paravalvular leakage ≥ grade 3 is an independent risk factor.

[0015] Specifically, the image feature extraction module adopts an improved 3D ResNet-50 network structure, including: (1) DWI image preprocessing algorithm: Offset correction formula: ; in, is the signal intensity without diffusion weighting, b is the diffusion sensitivity coefficient, ADC is the apparent diffusion coefficient, and S is the diffusion-weighted signal intensity; (2) Spatiotemporal feature extraction algorithm: Assume that the preoperative imaging is , postoperative imaging is , then the calculation formula for the time-space difference is: ; in 24 hours; (3) New infarct identification algorithm: Threshold segmentation combined with morphological processing: ; in ( is the background mean, is the standard deviation), =27mm³.

[0016] Specifically, the imaging features include quantitative parameters obtained by comparing preoperative and postoperative cranial DWI: the total number of new infarcts in each brain region (TLN), the volume of each new infarct (ILV), the total new infarct volume (TLV), where TLV>500mm³ is set as the high-risk threshold for stroke, and imaging omics features, including shape features such as sphericity. , texture feature contrast , i and j represent the grayscale values ​​(gray levels) of two adjacent pixels in the image.

[0017] For example, for an 8-bit grayscale image, the value range of i and j is 0-255 (corresponding to 256 gray levels). It is an element in the gray-level co-occurrence matrix (GLCM), which represents the probability that a pixel with gray value i appears adjacent to a pixel with gray value j in a specific direction and distance. GLCM is the core matrix of texture analysis, which quantifies image texture by counting the gray value relationship of pixel pairs.

[0018] Specifically, the brain regions are divided and constructed based on the Harvard-Oxford brain atlas, including the basal ganglia, cortical regions and cerebellum regions, and the corresponding symptom types are movement disorders, aphasia and ataxia, respectively.

[0019] Specifically, the multi-source data integration module adopts a layered coding strategy, including: (1) Categorical variable embedding algorithm: For categorical variables such as valve type, learnable embeddings are used: ; in is the embedding dimension, set to 64; (2) Standardization formula for continuous variables: ,in is the mean of the training set, is the standard deviation; (3) Multimodal feature weight calculation: ; in, It represents feature concatenation, MLP is a multi-layer perceptron, and Softmax is a key nonlinear normalization function. Its function is to convert the raw scores (logits) output by MLP into a probability distribution, ensuring that the sum of the weights of all modal features is 1, while highlighting the most important features. (4) Structured data feature vector generation: ; Specifically, the attention mechanism includes an improved multi-head cross attention mechanism and an asymmetric multimodal attention mechanism: (1) Multi-head attention calculation formula: ; in ; ; (2) Positional encoding fusion: For time series DWI data, position encoding is introduced: ; ; (3) Adaptive weight gating: ; in is the sigmoid function, is the learnable weight matrix, is the clinical feature vector; is the image feature vector; It is a bias term used to adjust the threshold of gate activation and enhance the flexibility of the model.

[0020] (4) Asymmetric attention score: ; Where Gate comes from the output of (3) adaptive weight gating; Element-wise multiplication (Hadamard product) for gating the dynamic modulation of attention weights; is a key nonlinear normalization function that converts the raw scores (logits) output by the MLP into a probability distribution, ensuring that the sum of the weights of all modal features is 1 while highlighting the most important features. Q is the query matrix (Query), which comes from the feature representation of the current modality; K is the key matrix (Key), which comes from the feature representation of other modalities. The symbol T represents the transpose operation of the matrix K. Represents the dimension of the key vector, used to scale the dot product result (to prevent gradient vanishing).

[0021] in is the inter-modal prior weight matrix, To adjust the intensity parameter.

[0022] Specifically, the prediction module adopts a hierarchical prediction strategy, including: (1) Risk probability calculation formula: ; in is the fusion feature vector, is the sigmoid activation function; is the learnable weight matrix; is the bias term; (2) Brain region localization algorithm: Probabilistic mapping based on Harvard-Oxford graph: ; where i {basal ganglia, cortex, cerebellum}; (3) Symptom prediction model: Using multi-label classification: ; where j {movement disorders, aphasia, ataxia}; is the fusion feature vector; is the learnable weight vector of the jth symptom; is the bias term of the jth symptom; (4) Confidence calculation: , where H is the entropy function: ; Specifically, the deep learning model is trained using a composite loss function: (1) Overall loss function: ; in , , , ; (2)Focal Loss specific formula: ; in ,set up , ; (3) Regional positioning loss: ; in , ; (4) Regularization term: ,in , .

[0023] Specifically, the model training adopts an adaptive learning strategy, including: (1) Learning rate scheduling algorithm: ; in , is the maximum number of training rounds; (2) Data enhancement strategy: Image Enhancement: Random Rotation , random scaling ; Noise injection: ,in ; (3) Early stopping judgment conditions: When continuous epochs satisfy Stop training when

[0024] Specifically, the quantitative standard of the model performance index is: Classification performance: , sensitivity , specificity , ; Positioning accuracy: Dice coefficient , Hausdorff distance ; Timeliness: Single-case reasoning time 5 seconds, system response time 60 seconds.

[0025] In order to achieve the second object of the present invention, the following technical solution is provided: a postoperative stroke analysis system, for executing the steps of the above-mentioned postoperative stroke analysis method based on multimodal image analysis, comprising an input unit, a data analysis unit, and an auxiliary unit; The input unit is used to collect medical data of the patient; The data analysis unit is used to analyze the collected medical data to output a prediction result, that is, to run the risk prediction model, extract structured data feature vectors and spatiotemporal-image features based on the collected medical data, perform feature fusion, and predict the risk of postoperative stroke after transvascular interventional surgery; The auxiliary unit is used to generate corresponding medication and rehabilitation strategy references based on the output prediction results.

[0026] Specifically, the analysis system adopts a distributed architecture.

[0027] In order to achieve the third object of the present invention, the following technical solution is provided: a postoperative stroke analysis device, used to execute the steps of the above-mentioned postoperative stroke analysis method based on multimodal image analysis.

[0028] Compared with the prior art, the present invention has the following beneficial effects: Novel integration dimension: For the first time, key intraoperative operating parameters of TAVR (such as artificial valve type and implantation depth) and dynamic pre- and post-operative DWI imaging features are jointly modeled, breaking through the limitation of traditional models that only rely on pre-operative or static data.

[0029] Accurate prediction and positioning: Through deep learning models and refined feature engineering, high-precision prediction of the risk of overt stroke is achieved (target AUC 0.85, sensitivity 0.80), and can perform anatomical localization of the responsible lesion based on the Harvard-Oxford atlas standard.

[0030] Closed clinical decision loop: The system not only provides risk values, but also maps the lesion location with expected neurological deficits (such as movement disorders and aphasia), and directly generates targeted rehabilitation recommendations (such as robot-assisted training and AI voice rehabilitation), realizing a closed clinical decision loop from prediction to intervention.

[0031] Efficient and practical deployment: The system adopts lightweight deployment optimization (such as INT8 quantization), which can be seamlessly connected to the hospital's existing information system to achieve automated and fast (single-case prediction < 5 seconds) analysis process, with strong clinical practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1A schematic diagram of the overall process of the postoperative stroke analysis method based on multimodal image analysis provided in this embodiment; Figure 2 A schematic diagram of the deep learning model architecture provided for this embodiment; Figure 3 Schematic diagram of the multi-head cross attention mechanism structure provided for this embodiment; Figure 4 A schematic diagram of the distributed architecture of the system provided in this embodiment; Figure 5 This is the personalized treatment recommendation system interface output by the post-operative stroke analysis system for transvascular interventional surgery provided in this embodiment. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0034] like Figure 1 As shown in FIG, a postoperative stroke analysis method based on multimodal image analysis provided by this embodiment, the specific steps are as follows: Step S1: Medical data input and preprocessing Input medical data includes: DWI imaging data: DWI images of the skull before and after surgery within 24 hours, DICOM format, layer thickness 3mm, b value = 1000 s / mm² Neurological function assessment: mRS score (0-6 points) and MMSE score (0-30 points) Intraoperative parameters: including 8 key parameters such as anesthesia type, valve type, and paravalvular leak grade.

[0035] Preprocessing of DWI images: Deskew: ; MNI152 template registration: The registration accuracy error is controlled within ±1mm.

[0036] Step S2: Data annotation and dataset construction The occurrence of symptomatic stroke within 30 days was used as the gold standard for labeling, and the labeling was confirmed by three neuroradiologists with a kappa value of >0.85.

[0037] Construct the data set: training set 70% (700 cases), validation set 15% (150 cases), and test set 15% (150 cases).

[0038] Step S3: Deep learning model construction like Figure 2 As shown, a deep learning model consisting of four core modules is constructed: (1) Multi-source data integration module: Categorical variable embedding: .

[0039] Standardization of continuous variables: .

[0040] Feature weight calculation: .

[0041] Integration vector generation ; (2) Image feature extraction module: Spatiotemporal difference calculation; ; Identification of new infarcts: ; Radiomics feature extraction: Sphericity ; Contrast ; (3) Feature fusion module: like Figure 3 As shown, an improved multi-head cross attention mechanism is adopted: Multi-head attention: ; .

[0042] Positional encoding: ; .

[0043] Adaptive Gating: .

[0044] Asymmetric Attention: .

[0045] (4) Prediction module: Risk probability prediction: .

[0046] Brain area localization ; Symptom prediction: .

[0047] Confidence calculation: .

[0048] Steps S4-S7: Each module implements corresponding functions according to the above algorithm, and specific parameter settings are: Learning rate: , batch size: , Number of training rounds: maximum 200 rounds, early stopping .

[0049] Step S8: Model training process. Using composite loss function: ; : .

[0050] Region localization loss: .

[0051] Regularization: .

[0052] Use adaptive learning rate scheduling: .

[0053] Step S9: Model application, that is, inputting new patient data into the trained model, and its output includes: the probability of symptomatic stroke within 30 days (0-1), the location and confidence of the responsible brain area, the corresponding symptom type prediction and individualized treatment recommendations.

[0054] This embodiment also provides a postoperative stroke analysis system for executing the steps of the above-mentioned postoperative stroke analysis method based on multimodal image analysis.

[0055] like Figure 4 As shown, the system architecture provided by this embodiment includes: (1) Input unit: DICOM image interface: supports multi-vendor devices, transmission rate ≥ 100MB / s; HIS system interface: HL7 FHIR standard, real-time data synchronization; Intraoperative parameter entry: Mobile APP, supports offline entry.

[0056] (2) Data analysis unit: A distributed processing architecture is used, as follows: Data Pipeline: DICOM -> preprocessing -> feature extraction -> model inference -> result output.

[0057] Load Balancing: ; Performance indicators: Single-case processing delay < 60 seconds; Concurrent processing capacity ≥ 50 cases / minute; GPU inference time is < 5 seconds.

[0058] (3) Security mechanism: Data transmission: AES-256 encryption; Access control: multi-level permissions based on RBAC; Audit log: MD5 verification, retention period ≥ 3 years.

[0059] (4) Auxiliary unit, generation of individualized treatment strategies: Medication Instructions: : : : return "It is recommended to replace with monoclonal antibody treatment (clopidogrel 75mg)"; else: return "It is recommended to start dual antiplatelet therapy (aspirin 100mg + clopidogrel 75mg)"; else: return "Conventional anticoagulation therapy".

[0060] Rehabilitation Program: : plans = {"Basal Ganglia - Movement Disorders": "Upper Limb Robotic Assisted Training 30 minutes / day × 6 weeks", "Language Area - Aphasia": "AI Speech Rehabilitation System 20 minutes / day", "Cerebellar Ataxia": "Balance training 45 minutes / day + VR technology"} return plans.get(f"{ }-{ }", "Standard Rehabilitation Program").

[0061] In order to better illustrate the effect of the technical solution provided in this embodiment, a specific description is given based on a clinical verification case.

[0062] Patient's basic information: Age: 78 years old, male, underlying disease: severe aortic valve stenosis.

[0063] The input data is as follows: Preoperative data: DWI imaging: no acute infarction; mRS score: 1 point; MMSE score: 26 points.

[0064] Intraoperative parameters: Anesthesia type: local anesthesia; Valve type: Venus A; Second valve: No; Paravalvular leak: Grade 3; Sheath time: 145 min; Blood loss: 200 ml; Left coronary sinus depth: 5.1 mm; No coronary sinus depth: 3.7 mm.

[0065] Model prediction results: System output {" ": 0.82,"confidence": 0.91, " ": {"basal ganglia": 0.75,"cortex": 0.20,"cerebellum": 0.05}, " ": {"dyskinesia": 0.78,"aphasia": 0.15,"ataxia":0.07}.

[0066] like Figure 5 As shown in the figure, the individualized treatment recommendations output by the post-vascular intervention stroke analysis system are: Medication instructions: It is recommended to start dual antiplatelet therapy (aspirin 100mg + clopidogrel 75mg).

[0067] Rehabilitation plan: Robot-assisted upper limb exercise training (30 minutes / day × 6 weeks) began the next day.

[0068] Monitoring strategy: Repeat DWI 24 hours after surgery and closely observe changes in upper limb muscle strength.

[0069] Clinical verification results: The patient developed mild left limb weakness 26 hours after the operation.

[0070] DWI reexamination showed a new small infarct (volume 580 mm³) in the right basal ganglia.

[0071] Prediction Accuracy: The model predictions are highly consistent with the actual results.

[0072] Treatment effect: After early intervention, the patient's mRS score was 1 at 3-month follow-up.

[0073] Performance evaluation: Model performance indicators: AUC: 0.87 (95% CI: 0.83-0.91).

[0074] Sensitivity: 0.83.

[0075] Specificity: 0.78.

[0076] F1-score: 0.80.

[0077] Dice coefficient: 0.89.

[0078] System performance indicators: Average response time: 45 seconds.

[0079] Concurrent processing capacity: 65 cases / minute.

[0080] System availability: 99.7%.

[0081] Prediction accuracy: 83.5%.

[0082] This embodiment further provides a postoperative stroke analysis device for executing the steps of the postoperative stroke analysis method based on multimodal image analysis provided in the above embodiment.

[0083] The difference between the solution provided by the present invention and the prior art is that: Integration of static intraoperative parameters: For the first time, this system systematically integrates key intraoperative parameters of TAVR (such as prosthetic valve type and sheath insertion duration) with radiomics modeling, breaking through the limitations of traditional reliance solely on preoperative data. Precise brain region-symptom mapping: Based on the Harvard-Oxford brain atlas, the responsible area is located to guide targeted rehabilitation; Strong clinical practicality: The risk prediction model adopts lightweight deployment (single-case prediction < 5 seconds) and is adapted to the hospital's existing information system.

[0084] In addition, the terms "upper", "lower", "inner", "outer", "front", and "back" are used for descriptive purposes only and should not be understood as indicating or implying relative importance. Unless otherwise specifically stated, the relative steps, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the present invention.

[0085] Of course, the above description is only a specific embodiment of the present invention and is not intended to limit the scope of implementation of the present invention. Any equivalent changes or modifications made based on the structure, features and principles described in the scope of the patent application of the present invention should be included in the scope of the patent application of the present invention.

[0086] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A postoperative stroke analysis method based on multimodal image analysis, characterized in that: The following steps are involved: Input medical data, including the patient's neurological function assessment, intraoperative parameters, and preoperative and postoperative DWI images; Label the medical data based on whether a stroke occurred after a vascular interventional procedure, and combine the medical data and labels into a dataset. Build a deep learning model, including a multi-source data integration module, an image feature extraction module, a feature fusion module, and a prediction module; The multi-source data integration module is used to perform multimodal feature extraction on the input neurological function assessment and intraoperative parameters to construct corresponding structured data feature vectors; The image feature extraction module is used to preprocess the DWI images and calculate the corresponding spatiotemporal differences, identify new infarcts in the DWI images and extract corresponding radiomic features, and construct corresponding spatiotemporal-image features based on the calculated spatiotemporal differences and the extracted radiomic features; The feature fusion module is used to perform multimodal feature fusion on the structured data feature vector and the spatiotemporal-image feature to output a corresponding fused feature vector; The prediction module performs prediction based on the fused feature vector to output a prediction result; The data set is used to train a deep learning model to obtain a risk prediction model for predicting whether postoperative stroke after transvascular interventional surgery will occur.

2. The postoperative stroke analysis method based on multimodal image analysis according to claim 1, characterized in that: The DWI image needs to be processed by brain template registration before input. The MNI152 template is used to perform image registration on the input DWI image, and the registration accuracy error is controlled within ±1mm.

3. The postoperative stroke analysis method based on multimodal image analysis according to claim 1, characterized in that: The DWI images include the head DWI images before and within 24 hours after the operation.

4. The postoperative stroke analysis method based on multimodal image analysis according to claim 1, characterized in that: The neurological function assessment included the modified Rankin Scale score for reflecting the preoperative neurological function status and the Mini-Mental State Examination score for evaluating cognitive function.

5. The postoperative stroke analysis method based on multimodal image analysis according to claim 1, characterized in that: The multi-source data integration module includes an embedding layer and a fully connected layer. The embedding layer is used to map the categorical variables in the input neurological function assessment and intraoperative parameters into dense vectors. The fully connected layer is used to standardize the continuous variables in the input neurological function assessment and intraoperative parameters to output feature vectors. The multimodal feature weights are calculated based on the dense vectors and the feature vectors to obtain the corresponding structured data feature vectors.

6. The postoperative stroke analysis method based on multimodal image analysis according to claim 1, characterized in that: The imaging features include quantitative parameters obtained by comparing preoperative and postoperative cranial DWI, as well as imaging omics features: The quantitative parameters include the total number of new infarct foci in each brain region, the volume of each new infarct foci, and the total volume of new infarct foci; The imaging omics features include shape feature sphericity and texture feature contrast.

7. The postoperative stroke analysis method based on multimodal image analysis according to claim 6, characterized in that: The brain regions are divided and constructed based on the Harvard-Oxford brain atlas, which includes the basal ganglia, cortical regions and cerebellum regions.

8. The postoperative stroke analysis method based on multimodal image analysis according to claim 1, characterized in that: The deep learning model training adopts an adaptive learning strategy, including: learning rate scheduling algorithm, data enhancement strategy including rotation and noise injection; early stopping judgment condition is continuous satisfy Stop training when 9. A postoperative stroke analysis system, characterized in that: The method for performing the postoperative stroke analysis method based on multimodal image analysis according to any one of claims 1 to 8 comprises an input unit, a data analysis unit and an auxiliary unit; The input unit is used to collect medical data of the patient; The data analysis unit is used to analyze the collected medical data to output a prediction result; The auxiliary unit is used to generate corresponding medication and rehabilitation strategy references based on the output prediction results.

10. A postoperative stroke analysis device, characterized in that: Steps for executing the postoperative stroke analysis method based on multimodal image analysis as described in any one of claims 1 to 8.

Images