Neural network-based aneurysm rupture prediction method and device and storage medium

Abstract

The invention provides an aneurysm rupture prediction method and device based on a neural network and a storage medium, and the method comprises the following steps: 1, obtaining a case of segmented CTA image data, carrying out the patch processing of the segmented CTA image data through three-dimensional modeling software, and constructing a 3D blood vessel model; according to the method, the exclusive preoperative neural network is trained for each patient by using a specific neural network structure, the positions of the blood vessel models of the patients can be accurately predicted, and the blood vessel models are adjusted to realize registration by inputting DSA images and rapidly outputting a camera parameter set during the operation, so that the accuracy and the efficiency of the operation are greatly improved; for an inexperienced doctor, the dependence on experience is reduced, and the vascular interventional operation can be more accurately performed; meanwhile, by evaluating and optimizing the registration result, the neural network structure is continuously adjusted, the number of training samples is increased, and the performance and reliability of the system are further improved.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a method, device and storage medium for predicting aneurysm rupture based on a neural network. Background Art

[0002] In the current medical field, vascular interventional surgery, as an advanced and vital treatment method, is playing an increasingly important role. With the continuous development of modern medical technology, vascular interventional surgery has brought new hope to many patients with its unique advantages. This surgical method has the characteristics of small trauma, fast recovery, and significant efficacy, and can effectively treat a variety of vascular diseases. For example, in the treatment of coronary heart disease, cerebrovascular disease, peripheral vascular disease, etc., vascular interventional surgery has become one of the main treatment methods. Compared with traditional surgical operations, vascular interventional surgery does not require large-scale surgery, but instead performs diagnosis and treatment by inserting catheters and other instruments into the patient's blood vessels. This minimally invasive surgical method greatly reduces the patient's pain and postoperative recovery time, while also reducing the risk of surgery and the incidence of complications. However, existing technologies have many defects and shortcomings in vascular interventional surgery;

[0003] First, there is currently a lack of an effective method to perform real-time registration of DSA (digital subtraction angiography) and preoperative CTA (computed tomography angiography) blood vessels in vascular interventional surgery. DSA is a commonly used vascular imaging technology that can provide high-resolution vascular images, but it can only display two-dimensional images and cannot provide three-dimensional vascular information. Preoperative CTA can provide a three-dimensional vascular model, but during surgery, how to accurately align the preoperative CTA vascular model with the intraoperative DSA image has always been a difficult problem;

[0004] Secondly, under the existing technical conditions, doctors mainly rely on experience to judge the real-time position of the guidewire in the human body when performing vascular interventional surgery. This method has great limitations. It is difficult for inexperienced doctors to independently perform vascular interventional surgery. The accumulation of experience requires long-term practice and a large number of surgical cases, and the experience levels of different doctors vary greatly, which makes it difficult to guarantee the quality and effect of the surgery;

[0005] In addition, existing technologies are inefficient in processing image information in vascular interventional surgery. Due to the lack of effective image registration methods, doctors need to spend a lot of time analyzing and judging images during surgery, which not only increases the time of surgery, but also increases the pain and risk of patients.

[0006] At the same time, existing technologies are also insufficient in terms of accuracy. Relying on experience to determine the position of the guidewire is prone to errors, and these errors may have a serious impact on the outcome of the operation. For example, when treating diseases such as aneurysms, the accuracy requirements are very high. If the position of the guidewire is inaccurate, it may cause the aneurysm to rupture and endanger the patient's life.

[0007] To this end, a method, device and storage medium for predicting aneurysm rupture based on a neural network are proposed. Summary of the invention

[0008] In view of this, the embodiments of the present invention hope to provide a method, device and storage medium for predicting aneurysm rupture based on a neural network to solve or alleviate the technical problems existing in the prior art and at least provide a beneficial option.

[0009] In order to solve the above technical problems, a technical solution adopted in the present application is: a method for predicting aneurysm rupture based on a neural network, comprising the following steps:

[0010] Step 1: Obtain a segmented CTA image data, and use 3D modeling software to process the segmented CTA image data into slices to construct a 3D vascular model;

[0011] Step 2: Based on the simulated X-ray projection algorithm, the 3D blood vessel model is projected by adjusting various camera parameters to generate multiple simulated DSA images, and the camera parameter set corresponding to each simulated DSA image is recorded;

[0012] Step 3: Preprocess the simulated DSA images and the intraoperative DSA images obtained during the actual surgery;

[0013] Step 4: Use the deep learning framework CNN or U-net to build the neural network structure, construct the neural network model, and determine the loss function and optimization algorithm;

[0014] Step 5: Use the simulated DSA images as training data and the camera parameter sets corresponding to the simulated DSA images as supervision labels to train the neural network model and learn to restore the corresponding camera parameter sets from the intraoperative DSA images of the patients;

[0015] Step 6: In actual surgery, the intraoperative DSA image is input, and the neural network automatically outputs the corresponding camera parameter set. According to this camera parameter set, the 3D vascular model is adjusted to be registered with the intraoperative DSA image to generate the registration result;

[0016] Step 7: Evaluate the registration results according to the evaluation indicators, and take corresponding optimization measures based on the evaluation results.

[0017] As a further preferred embodiment of the present technical solution, in step 1, the method for constructing a 3D blood vessel model comprises the following steps:

[0018] Step 1, obtaining segmented CTA image data;

[0019] Step 2: Import the segmented CTA images into the 3D modeling software, and use the adaptive patch segmentation strategy to adjust the patch size according to the curvature and diameter changes of the blood vessels;

[0020] Step 3: Based on the 3D modeling software, the panels are gradually constructed to form a continuous 3D structure.

[0021] Step 4: Smooth the constructed 3D vascular model, remove noise and discontinuities, and optimize and adjust the model based on clinical information.

[0022] As a further preferred embodiment of the present technical solution, in step 2, the camera parameter set includes focal length, viewing angle, exposure, rotation angle, translation distance, and magnification / reduction ratio.

[0023] As a further preferred embodiment of the present technical solution, in step three, the image preprocessing includes image denoising processing, image enhancement processing, image registration processing and image correction processing.

[0024] As a further preferred embodiment of the present technical solution, in step four, the loss function is a mean square error function or a cross entropy function, and the optimization algorithm is a stochastic gradient descent algorithm or an Adam optimizer.

[0025] As a further preferred embodiment of the present technical solution, in step five, when training the neural network, a small batch stochastic gradient descent method is used, a quantitative simulated DSA image and its corresponding camera parameter set are randomly selected in each batch for training, and a dynamic learning rate adjustment strategy is set during the training process.

[0026] As a further preferred embodiment of the present technical solution, in step six, when adjusting the 3D vascular model, the adjustment is performed using a stepwise approximation method, and the alignment result is displayed in real time. After each adjustment, the similarity between the 3D vascular model and the intraoperative DSA image is calculated. If the similarity reaches a preset threshold, the adjustment is stopped.

[0027] As a further preferred embodiment of the present technical solution, in step seven, the evaluation indicators include alignment accuracy, registration error and registration efficiency; and the optimization measures include adjusting neural network parameters, improving image preprocessing algorithms and optimizing camera parameters.

[0028] To solve the above technical problems, another technical solution adopted in the present application is: an aneurysm rupture prediction device based on a neural network, comprising: an image acquisition module, a three-dimensional modeling module, a projection module, an image preprocessing module, a neural network construction module, a training module, a registration module and an evaluation optimization module;

[0029] The image acquisition module is used to acquire the segmented CTA image data;

[0030] The three-dimensional modeling module is used to process the segmented CTA image data into slices using three-dimensional modeling software to construct a 3D blood vessel model;

[0031] The projection module is used to project the 3D blood vessel model based on a simulated X-ray projection algorithm by adjusting a variety of camera parameters to generate a plurality of simulated DSA images, and record a camera parameter set corresponding to each simulated DSA image, wherein the camera parameter set includes focal length, viewing angle, exposure, rotation angle, translation distance, and magnification ratio;

[0032] The image preprocessing module is used to perform denoising, enhancement and image standardization processing on the simulated DSA images and the intraoperative DSA images obtained in the actual surgery;

[0033] The neural network building module is used to build a neural network structure using a deep learning framework CNN or U-net, determine a mean square error function or a cross entropy function as a loss function, and a stochastic gradient descent algorithm or an Adam optimizer as an optimization algorithm;

[0034] The training module is used to use the simulated DSA images as training data and the camera parameter sets corresponding to the simulated DSA images as supervision labels to train the neural network and learn to restore the corresponding camera parameter sets from the intraoperative DSA images of the patient;

[0035] The registration module is used to input intraoperative DSA images in actual surgery, and the neural network automatically outputs the corresponding camera parameter set, adjusts the 3D blood vessel model according to the camera parameter set, and displays the registration result in real time;

[0036] The evaluation and optimization module is used to evaluate the registration result according to the evaluation index and optimize according to the evaluation result.

[0037] In order to solve the above technical problems, another technical solution adopted by the present application is: a computer-readable storage medium storing program instructions capable of implementing the aneurysm rupture prediction method based on a neural network as described above.

[0038] The embodiment of the present invention has the following advantages due to the adoption of the above technical solution:

[0039] The present invention uses artificial intelligence image processing technology to achieve an intuitive display of the position information of surgical instruments during the operation. Specifically, a specific neural network structure is used to train a unique preoperative neural network for each patient, which can accurately predict the patient's vascular model position. During the operation, the DSA image is input and the camera parameter set is quickly output, and the vascular model is adjusted to achieve registration, which greatly improves the accuracy and efficiency of the operation. For inexperienced doctors, it reduces dependence on experience and can perform vascular interventional surgery more accurately. At the same time, by evaluating and optimizing the registration results, the neural network structure is continuously adjusted and the number of training samples is increased, thereby further improving the performance and reliability of the system.

[0040] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 It is a flowchart of a method for predicting aneurysm rupture based on a neural network according to the present invention;

[0043] Figure 2 A schematic diagram of the process of constructing a 3D blood vessel model according to the present invention;

[0044] Figure 3 This is a schematic diagram of the functional modules of an aneurysm rupture prediction device based on a neural network according to the present invention. DETAILED DESCRIPTION

[0045] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0046] It should be clear that the following embodiments of the present disclosure are described by specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other in the absence of conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0047] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0048] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0049] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0050] Figure 1 FIG. 1 is a flowchart of a method for predicting aneurysm rupture based on a neural network according to an embodiment of the present invention. It should be noted that if there are substantially the same results, the method of the present application is not limited to the method of predicting aneurysm rupture based on a neural network. Figure 1 The process sequence shown is limited. Figure 1-Figure 2 As shown: A method for predicting aneurysm rupture based on a neural network, comprising the following steps:

[0051] Step 1: Obtain a segmented CTA image data, and use 3D modeling software to process the segmented CTA image data into pieces to construct a 3D vascular model; CTA image data is usually acquired by medical imaging equipment (such as CT scanners), and these data are usually stored in DICOM format, containing detailed 3D information of the patient's blood vessels; image segmentation methods include threshold segmentation, region growing, level set method, machine learning algorithm, etc., and the segmented image data should only contain the blood vessel part, removing other non-vascular tissues such as bones and muscles; 3D modeling software can be selected from BlueDer, Maya, 3dsMax or specialized medical image processing software (such as Mimics, 3D-Doctor, etc.);

[0052] Step 2: Based on the simulated X-ray projection algorithm, the 3D vascular model is projected by adjusting various camera parameters to generate multiple simulated DSA images, and the camera parameter set corresponding to each simulated DSA image is recorded; wherein the simulated X-ray projection algorithm simulates the attenuation process of X-rays passing through an object (in this case, a blood vessel) to generate a two-dimensional projection image, which usually involves voxelizing the 3D model (i.e., dividing the 3D space into small cubic units) and then applying the X-ray attenuation formula to each voxel; by changing the camera parameters, the same 3D vascular model is projected multiple times to generate multiple simulated DSA images;

[0053] Step 3: Perform image preprocessing on the simulated DSA images and intraoperative DSA images obtained during actual surgery. By selecting appropriate preprocessing algorithms, evaluating preprocessing effects, and selecting professional preprocessing software, high-quality simulated DSA images and intraoperative DSA images can be obtained, providing strong support for the success of the surgery.

[0054] Step 4: Use the deep learning framework CNN or U-net to build the neural network structure, construct the neural network model, and determine the loss function and optimization algorithm; the loss function is used to measure the difference between the predicted results of the neural network model and the actual results; the role of the optimization algorithm is to minimize the loss function by adjusting the model parameters;

[0055] Step 5: Use the simulated DSA images as training data and the camera parameter sets corresponding to the simulated DSA images as supervision labels to train the neural network model and learn to restore the corresponding camera parameter sets from the intraoperative DSA images of the patients;

[0056] Specifically: First, the simulated DSA images and their corresponding camera parameter sets are divided into training set, validation set and test set; usually, the training set accounts for most of the data and is used to train the neural network model; the validation set is used to evaluate the performance of the neural network model and perform hyperparameter tuning during the training process; the test set is used to finally evaluate the generalization ability of the neural network model;

[0057] Then, write a training loop, including steps such as forward propagation, loss calculation, backpropagation, and parameter update. In each training iteration, use a batch of simulated DSA images and their corresponding camera parameter sets to train the neural network model.

[0058] During the training process, monitor the changes in the loss function value to evaluate the training progress and performance of the neural network model; at the end of each training cycle, use the validation set to evaluate the performance of the neural network model, and adjust hyperparameters such as learning rate and batch size based on the validation results; if the validation performance does not improve significantly after multiple consecutive training cycles, stop training to avoid overfitting;

[0059] After training is completed, use the test set to evaluate the generalization ability of the neural network model, calculate the error between the predicted camera parameters and the actual camera parameters, and evaluate the accuracy, robustness and other indicators of the neural network model; if the model performance is poor, hyperparameters can be tuned through grid search, random search or Bayesian optimization; based on the evaluation and tuning results, the neural network model structure, loss function or optimization algorithm can be improved;

[0060] Finally, the trained neural network model was applied to intraoperative DSA images obtained during actual surgery to restore the corresponding camera parameter set; the accuracy and reliability of the neural network model were verified by comparing the restored camera parameters with the camera parameters actually used during surgery;

[0061] Step 6: In actual surgery, the intraoperative DSA image is input, and the neural network automatically outputs the corresponding camera parameter set. According to this camera parameter set, the 3D vascular model is adjusted to be registered with the intraoperative DSA image to generate the registration result;

[0062] Specifically: First, during the operation, use the DSA device to obtain the patient's intraoperative DSA images to ensure that the image quality is good and has undergone necessary preprocessing steps;

[0063] Then, the intraoperative DSA image is input into the previously trained neural network model, which will automatically output the corresponding camera parameter set, which describes the position and posture of the intraoperative DSA image relative to a certain reference coordinate system;

[0064] Next, based on the camera parameter set output by the neural network, the 3D vascular model is transformed to make it consistent with the intraoperative DSA image in space and posture; this may involve operations such as translation, rotation, and scaling to ensure accurate alignment between the 3D model and the 2D image;

[0065] Finally, a registration algorithm is used to further refine the alignment between the 3D vascular model and the intraoperative DSA image. When the alignment between the 3D vascular model and the intraoperative DSA image reaches a satisfactory level, a registration result is generated.

[0066] Step 7. Evaluate the registration results according to the evaluation indicators, and take corresponding optimization measures based on the evaluation results; By effectively evaluating the registration results and taking corresponding optimization measures based on the evaluation results, the accuracy and reliability of the registration can be improved, which will provide more accurate and reliable information support for surgical navigation, intraoperative monitoring and other related applications.

[0067] In one embodiment, specifically, in step 1, the method for constructing a 3D blood vessel model includes the following steps:

[0068] Step 1, obtaining segmented CTA image data; wherein, the CTA image data after professional image segmentation processing is ensured to contain only blood vessels, and remove interference from other non-vascular tissues such as bones and muscles;

[0069] Step 2: Import the segmented CTA images into the 3D modeling software, and use the adaptive patch segmentation strategy to adjust the patch size according to the curvature and diameter change of the blood vessel. The adaptive patch segmentation strategy ensures that the patch can fit the blood vessel contour more closely in areas where the blood vessel is curved or has a large diameter change, thereby improving the accuracy of the model.

[0070] Step 3: Based on the 3D modeling software, the meshes are gradually constructed to form a continuous 3D structure. Specifically, in the 3D modeling software, mesh construction tools or plug-ins are used to gradually add meshes to cover the entire vascular structure, ensuring that the connection between the meshes is smooth and continuous to avoid gaps or overlaps. The preview function of the software can be used to view the 3D structure in real time during the construction process so as to adjust and optimize it in time.

[0071] Step 4: Smooth the constructed 3D vascular model to remove noise and discontinuities, and optimize and adjust the model based on clinical information; the combined clinical information includes the anatomical structure of the blood vessels, the condition of the lesions, etc., and further optimize and adjust the model, which may include adjusting parameters such as the diameter, length, and curvature of the blood vessels to ensure that the model is consistent with the actual conditions of the patient.

[0072] In one embodiment, specifically, in step 2, the camera parameter set includes focal length, viewing angle, exposure, rotation angle, translation distance, and zoom ratio;

[0073] Among them, the adjustment of the focal length can affect the projection size of the 3D model in the virtual view, thereby affecting the matching degree with the DSA image. By adjusting the focal length, the projection of the 3D model in the virtual view can be made closer to the vascular structure in the DSA image;

[0074] The adjustment of the viewing angle can change the visible range of the 3D model in the virtual view. By adjusting the viewing angle, it can ensure that the projection of the 3D model matches the field of view in the DSA image, thereby improving the accuracy of the registration;

[0075] Exposure adjustment can ensure that the vascular structure in the DSA image is clearly visible, which helps to improve the accuracy of registration. However, it should be noted that the exposure adjustment should be based on the image quality requirements to avoid image distortion caused by overexposure or underexposure.

[0076] Adjusting the rotation angle can change the direction of the 3D model in the virtual view. By adjusting the rotation angle, the projection of the 3D model and the vascular structure in the DSA image can be kept consistent in direction, thereby improving the accuracy of registration.

[0077] Adjusting the translation distance can change the position of the 3D model in the virtual view. By adjusting the translation distance, the projection of the 3D model can be aligned with the vascular structure in the DSA image, thereby further improving the accuracy of the registration.

[0078] Adjusting the zoom ratio can change the size of the 3D model in the virtual view. By adjusting the zoom ratio, the projection of the 3D model can be matched with the vascular structure in the DSA image in size, thereby ensuring the accuracy of the registration.

[0079] In one embodiment, specifically, in step three, the image preprocessing includes image denoising processing, image enhancement processing, image registration processing, and image correction processing;

[0080] Specifically: Image denoising can use spatial domain filtering methods, such as Gaussian filtering, arithmetic mean filtering and median filtering, to remove or reduce noise in the image. According to the type and distribution characteristics of the noise, select the appropriate filtering algorithm and parameters;

[0081] Image enhancement processing can use methods such as histogram equalization and contrast stretching to enhance the contrast of the image. Other image enhancement techniques such as edge enhancement and sharpening can also be selected according to the image characteristics and processing purpose.

[0082] Image registration is the process of aligning images at different times, different perspectives or different modalities to ensure their spatial consistency. It can use feature point-based registration methods, mutual information-based registration methods or other advanced registration algorithms.

[0083] Image correction processing uses an image correction algorithm to eliminate the geometric distortion or distortion of intraoperative DSA images that may be affected by factors such as the equipment and patient position, so as to ensure the accuracy and reliability of the image. Correction methods include morphological control point detection and polynomial-based correction coefficient solution.

[0084] In one embodiment, specifically, in step 4, the loss function is a mean square error function or a cross entropy function, and the optimization algorithm is a stochastic gradient descent algorithm or an Adam optimizer;

[0085] The mean square error function (MSE) is the average value of the sum of squares of the difference between the predicted value and the true value. In the registration task, MSE is often used to measure the difference in pixel values ​​between the projection of the 3D vascular model and the intraoperative DSA image. By minimizing MSE, the projection of the 3D model can be made closer to the vascular structure in the DSA image.

[0086] The cross entropy function is a measure of the difference between two probability distributions and is often used in classification tasks. Although the cross entropy function is mainly used for classification tasks, in some registration tasks, if the registration problem can be transformed into a classification problem (for example, by segmenting or detecting the key points of blood vessels), the cross entropy can also be used as an effective loss function. However, in direct pixel-level registration tasks, MSE is usually more commonly used.

[0087] Stochastic Gradient Descent (SGD) is an iterative optimization algorithm used to minimize the objective function. It updates the model parameters by randomly selecting a small batch of data, thereby speeding up training and reducing memory usage. In the registration task, SGD can be used to update the weights of the neural network to minimize the loss function. The randomness of SGD helps to escape from the local optimal solution, but it may also cause instability and oscillation during training.

[0088] The Adam optimizer is an adaptive learning rate optimization algorithm based on first-order moment estimation and second-order moment estimation. It combines the advantages of SGD and introduces momentum terms and adaptive learning rate adjustment mechanisms to speed up training and improve training stability. In registration tasks, the Adam optimizer usually has better performance and stability than SGD. It can automatically adjust the learning rate and reduce the risk of oscillation and overfitting during training. Therefore, the Adam optimizer is one of the commonly used optimization algorithms in registration tasks.

[0089] In one embodiment, specifically, in step five, when training the neural network, a small batch stochastic gradient descent method is used, a quantitative simulated DSA image and its corresponding camera parameter set are randomly selected in each batch for training, and a dynamic learning rate adjustment strategy is set during the training process;

[0090] Among them, Mini-batch Stochastic Gradient Descent (Mini-batch SGD) is an optimization algorithm that combines the advantages of Batch Gradient Descent (BGD) and Stochastic Gradient Descent (SGD). In each iteration, it randomly selects a mini-batch of data (i.e., a subset) to calculate the gradient and update the model parameters instead of using the entire data set or a single data point. In the registration task, Mini-batch SGD is used to train the neural network. Specifically, a certain number of simulated DSA images and their corresponding camera parameter sets are randomly selected from the training set as a batch each time, and the gradient of the loss function on the batch is calculated, and the gradient is used to update the weights of the neural network. Mini-batch SGD can speed up training and reduce memory usage by reducing the amount of data for calculating gradients in each iteration. At the same time, since each iteration uses a different data subset, it helps to increase the randomness and generalization ability of training.

[0091] In one embodiment, specifically, in step six, when adjusting the 3D vascular model, the adjustment is performed by using a stepwise approximation method, and the registration result is displayed in real time. After each adjustment, the similarity between the 3D vascular model and the intraoperative DSA image is calculated. If the similarity reaches a preset threshold, the adjustment is stopped.

[0092] Among them, stepwise approximation is an iterative optimization method that approaches the target value or optimal solution by gradually adjusting parameters. In the registration task, the stepwise approximation method is used to adjust the parameters of the 3D vascular model so that it gradually matches the intraoperative DSA image. In each iteration, the parameters of the 3D vascular model (such as focal length, viewing angle, exposure, rotation angle, translation distance, and zoom ratio in the camera parameter set) are fine-tuned according to the current registration result and similarity measurement. These adjustments can be optimization steps based on gradient descent or iterative steps based on heuristic search algorithms (such as genetic algorithms, particle swarm optimization, etc.). The stepwise approximation adjustment method can ensure that each adjustment is made in the direction of improving the registration result, thereby gradually approaching the optimal solution. At the same time, since each adjustment is small, instability or incorrect registration caused by excessive adjustments can be avoided.

[0093] In one embodiment, specifically, in step seven, the evaluation indicators include alignment accuracy, registration error, and registration efficiency; the optimization measures include adjusting neural network parameters, improving image preprocessing algorithms, and optimizing camera parameters;

[0094] Among them, the accuracy of alignment is measured by comparing the position and morphological consistency of the blood vessels in the 3D vascular model and the intraoperative DSA images;

[0095] The registration error was determined by calculating the distance or difference between the corresponding points in the 3D vascular model and the intraoperative DSA image;

[0096] Registration efficiency was evaluated by calculating the time taken for the registration process.

[0097] In summary, the neural network-based aneurysm rupture prediction method provided in the embodiment of the present invention provides an efficient and accurate method for predicting aneurysm rupture by combining neural networks and image processing technology, realizes the intuitive display of surgical instrument position information, greatly improves the accuracy and efficiency of the operation, and reduces the dependence on experience for inexperienced doctors, so that they can also perform vascular interventional surgery more accurately.

[0098] Figure 3 is a functional module diagram of a neural network-based aneurysm rupture prediction device according to an embodiment of the present application, such as Figure 3 As shown, a device for predicting aneurysm rupture based on a neural network includes: an image acquisition module, a three-dimensional modeling module, a projection module, an image preprocessing module, a neural network construction module, a training module, a registration module and an evaluation and optimization module;

[0099] An image acquisition module, used for acquiring segmented CTA image data;

[0100] A three-dimensional modeling module is used to process the segmented CTA image data into slices using three-dimensional modeling software to construct a 3D vascular model;

[0101] A projection module is used to project the 3D blood vessel model based on a simulated X-ray projection algorithm by adjusting various camera parameters to generate multiple simulated DSA images, and record the camera parameter set corresponding to each simulated DSA image, the camera parameter set including focal length, viewing angle, exposure, rotation angle, translation distance, and magnification ratio;

[0102] An image preprocessing module is used to perform denoising, enhancement and image standardization on simulated DSA images and intraoperative DSA images obtained during actual surgery;

[0103] Neural network building module, which is used to build the neural network structure using the deep learning framework CNN or U-net, determine the mean square error function or cross entropy function as the loss function, and the stochastic gradient descent algorithm or Adam optimizer as the optimization algorithm;

[0104] A training module is used to train a neural network using simulated DSA images as training data and camera parameter sets corresponding to the simulated DSA images as supervision labels, so as to learn to restore the corresponding camera parameter sets from the intraoperative DSA images of patients;

[0105] The registration module is used to input intraoperative DSA images during actual surgery. The neural network automatically outputs the corresponding camera parameter set. According to this camera parameter set, the 3D vascular model is adjusted and the registration result is displayed in real time.

[0106] The evaluation and optimization module is used to evaluate the registration results according to the evaluation indicators and optimize according to the evaluation results.

[0107] In summary, the neural network-based aneurysm rupture prediction device provided in the embodiment of the present invention realizes accurate registration from CTA images to intraoperative DSA images by integrating multiple modules such as image acquisition, three-dimensional modeling, projection, image preprocessing, neural network construction, training, registration and evaluation optimization. This not only provides strong technical support for the prediction of aneurysm rupture, but also provides new ideas and methods for the development of fields such as medical image analysis and surgical navigation.

[0108] For other details about the technical solutions for implementing each module in the aneurysm rupture prediction device based on a neural network in the above embodiment, please refer to the description of an aneurysm rupture prediction method based on a neural network in the above embodiment, which will not be repeated here.

[0109] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0110] According to the computer-readable storage medium of the embodiment of the present disclosure, non-transitory computer-readable instructions are stored thereon. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the aneurysm rupture prediction method based on a neural network in the above-mentioned embodiments of the present disclosure are executed.

[0111] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0112] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0113] The basic principles of the present disclosure are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present disclosure are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present disclosure. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, and are not limitations. The above details do not limit the present disclosure to the necessity of adopting the above specific details to be implemented.

[0114] In the present disclosure, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. The block diagrams of the devices, devices, equipment, and systems involved in the present disclosure are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.

[0115] Additionally, as used herein, "or" used in a list of items beginning with "at least one" indicates a separate list, so that, for example, a list of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not mean that the example described is preferred or better than other examples.

[0116] It should also be noted that in the system and method of the present disclosure, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present disclosure.

[0117] Various changes, substitutions, and modifications of the techniques described herein may be made without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of the present disclosure is not limited to the specific aspects of the processes, machines, manufactures, compositions of events, means, methods, and actions described above. Currently existing or later to be developed processes, machines, manufactures, compositions of events, means, methods, or actions that perform substantially the same functions or achieve substantially the same results as the corresponding aspects described herein may be utilized. Thus, the appended claims include such processes, machines, manufactures, compositions of events, means, methods, or actions within their scope.

[0118] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0119] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A method for predicting aneurysm rupture based on a neural network, characterized in that: The following steps are involved: Step 1: Obtain a segmented CTA image data, and use 3D modeling software to process the segmented CTA image data into slices to construct a 3D vascular model; Step 2: Based on the simulated X-ray projection algorithm, the 3D blood vessel model is projected by adjusting various camera parameters to generate multiple simulated DSA images, and the camera parameter set corresponding to each simulated DSA image is recorded; Step 3: Preprocess the simulated DSA images and the intraoperative DSA images obtained during the actual surgery; Step 4: Use the deep learning framework CNN or U-net to build the neural network structure, construct the neural network model, and determine the loss function and optimization algorithm; Step 5: Use the simulated DSA images as training data and the camera parameter sets corresponding to the simulated DSA images as supervision labels to train the neural network model and learn to restore the corresponding camera parameter sets from the intraoperative DSA images of the patients; Step 6: In actual surgery, the intraoperative DSA image is input, and the neural network automatically outputs the corresponding camera parameter set. According to this camera parameter set, the 3D vascular model is adjusted to be registered with the intraoperative DSA image to generate the registration result; Step 7: Evaluate the registration results according to the evaluation indicators, and take corresponding optimization measures based on the evaluation results.

2. The method for predicting aneurysm rupture based on a neural network according to claim 1, characterized in that: In step 1, the method for constructing a 3D blood vessel model comprises the following steps: Step 1, obtaining segmented CTA image data; Step 2: Import the segmented CTA images into the 3D modeling software, and use the adaptive patch segmentation strategy to adjust the patch size according to the curvature and diameter changes of the blood vessels; Step 3: Based on the 3D modeling software, the panels are gradually constructed to form a continuous 3D structure. Step 4: Smooth the constructed 3D vascular model, remove noise and discontinuities, and optimize and adjust the model based on clinical information.

3. The method for predicting aneurysm rupture based on a neural network according to claim 1, characterized in that: In step 2, the camera parameter set includes focal length, viewing angle, exposure, rotation angle, translation distance, and zoom ratio.

4. The method for predicting aneurysm rupture based on a neural network according to claim 1, characterized in that: In step three, the image preprocessing includes image denoising, image enhancement, image registration and image correction.

5. The method for predicting aneurysm rupture based on a neural network according to claim 1, characterized in that: In step 4, the loss function is a mean square error function or a cross entropy function, and the optimization algorithm is a stochastic gradient descent algorithm or an Adam optimizer.

6. The method for predicting aneurysm rupture based on a neural network according to claim 1, characterized in that: In step five, when training the neural network, a small batch stochastic gradient descent method is used. A certain amount of simulated DSA images and their corresponding camera parameter sets are randomly selected in each batch for training, and a dynamic learning rate adjustment strategy is set during the training process.

7. The method for predicting aneurysm rupture based on a neural network according to claim 1, characterized in that: In step six, when adjusting the 3D vascular model, the adjustment is performed using a stepwise approximation method, and the registration result is displayed in real time. After each adjustment, the similarity between the 3D vascular model and the intraoperative DSA image is calculated. If the similarity reaches a preset threshold, the adjustment is stopped.

8. The method for predicting aneurysm rupture based on a neural network according to claim 1, characterized in that: In step seven, the evaluation indicators include alignment accuracy, registration error and registration efficiency; the optimization measures include adjusting neural network parameters, improving image preprocessing algorithms and optimizing camera parameters.

9. A neural network-based aneurysm rupture prediction device, applied to a neural network-based aneurysm rupture prediction method according to any one of claims 1 to 8, characterized in that: include: Image acquisition module, 3D modeling module, projection module, image preprocessing module, neural network building module, training module, registration module and evaluation and optimization module; The image acquisition module is used to acquire the segmented CTA image data; The three-dimensional modeling module is used to process the segmented CTA image data into slices using three-dimensional modeling software to construct a 3D blood vessel model; The projection module is used to project the 3D blood vessel model based on a simulated X-ray projection algorithm by adjusting a variety of camera parameters to generate a plurality of simulated DSA images, and record a camera parameter set corresponding to each simulated DSA image, wherein the camera parameter set includes focal length, viewing angle, exposure, rotation angle, translation distance, and magnification ratio; The image preprocessing module is used to perform denoising, enhancement and image standardization processing on the simulated DSA images and the intraoperative DSA images obtained in the actual surgery; The neural network building module is used to build a neural network structure using a deep learning framework CNN or U-net, determine a mean square error function or a cross entropy function as a loss function, and a stochastic gradient descent algorithm or an Adam optimizer as an optimization algorithm; The training module is used to use the simulated DSA images as training data and the camera parameter sets corresponding to the simulated DSA images as supervision labels to train the neural network and learn to restore the corresponding camera parameter sets from the intraoperative DSA images of the patient; The registration module is used to input intraoperative DSA images in actual surgery, and the neural network automatically outputs the corresponding camera parameter set, adjusts the 3D blood vessel model according to the camera parameter set, and displays the registration result in real time; The evaluation and optimization module is used to evaluate the registration result according to the evaluation index and optimize according to the evaluation result.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the neural network-based aneurysm rupture prediction method described in any one of claims 1-8.

Images