Method and device for detecting intracranial aneurysm based on MRA image

Abstract

The application provides an intracranial aneurysm detection method and device based on MRA images, which comprises the following steps: pre-processing MRA image samples; extracting target image blocks based on the pre-processed MRA image samples; wherein the target image blocks are image blocks containing arteries; generating MIP images based on the target image blocks; generating a training data set based on the MIP images and the target image blocks corresponding to the MIP images; obtaining a trained aneurysm detection model based on the constructed aneurysm detection model and the training data set; and predicting whether the to-be-detected MRA image contains an aneurysm based on the trained aneurysm detection model and the to-be-detected MRA image. The MRA images are fully decomposed, so that the aneurysm detection model can more accurately identify and extract the features of the aneurysm. The deep learning algorithm is combined with the logic of manual reading to detect the aneurysm in the MRA images, and the sensitivity and accuracy of the detection are high, and the occurrence of false positive results is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, and in particular to an intracranial aneurysm detection method and device based on MRA images. BACKGROUND

[0002] Magnetic resonance angiography (TOF-MRA) is an imaging technique sensitive to blood flow in the brain, unlike digital subtraction angiography (DSA) and CT angiography (CTA) which require radiation exposure and injection of contrast agents, MRA can allow patients to be performed without radiation exposure and contrast agents, therefore, it is widely used in clinical for the identification of unruptured intracranial aneurysms (UIAs) in asymptomatic patients. In healthy adults, the prevalence of unruptured intracranial aneurysms is about 3.2%, but if not treated in time, UIA is likely to rupture and then cause subarachnoid hemorrhage, with a mortality rate of up to 40%, and a poor prognosis, and the disability rate of survivors is also high.

[0003] Screening aneurysms is a challenging and time-consuming task, so medical staff need to combine machine automated analysis to improve efficiency and accuracy of disease judgment. Most of the existing traditional machine learning algorithms are difficult to process input images in original form; in order to obtain good performance, it is usually necessary to extract calculated "image features". For example, in the aneurysm detection problem, curvature or Hessian matrix derived features are often used. But CNN can take the image itself as input and automatically learn good features through a processing method similar to human visual recognition, thereby achieving better performance. Therefore, people naturally use CNN for CAD and use simple images as input, and most aneurysm detection models usually take the original image as input, use convolutional neural network as a detection model to determine which area on the image contains an aneurysm. But in the existing method, relying on pure machine recognition, it lacks effective combination with the accumulated clinical experience and the logic of manual reading in the development of medicine, and taking the original MRA image as input, the sensitivity of aneurysm judgment is low, and false positive results are prone to occur. SUMMARY

[0004] The purpose of the present application is to provide an intracranial aneurysm detection method and device based on MRA images, to solve the problem in the prior art that relying on pure machine recognition lacks effective combination with the accumulated clinical experience and the logic of manual reading in the development of medicine, and the sensitivity of aneurysm judgment in MRA images is low, and false positive results are prone to occur.

[0005] To solve the above technical problems, according to some embodiments, the present application provides, in one aspect, a method for detecting intracranial aneurysm based on MRA image, comprising: preprocessing MRA image samples; extracting target image blocks based on the preprocessed MRA image samples; wherein the target image blocks are image blocks containing arteries; generating MIP images based on the target image blocks; generating a training data set based on the MIP images and the target image blocks corresponding to the MIP images; obtaining a trained aneurysm detection model based on the constructed aneurysm detection model and the training data set; and predicting whether the to-be-detected MRA image contains an aneurysm based on the trained aneurysm detection model and the to-be-detected MRA image.

[0006] Further, the preprocessing of the MRA image samples comprises: manually labeling aneurysm images in the MRA image samples; generating mask information of the aneurysm based on the aneurysm images; and converting the manually labeled MRA image samples and the mask information of the aneurysm into three-dimensional matrices respectively to obtain MRA matrix images and mask matrices of the aneurysm.

[0007] Further, the extraction of the target image blocks based on the preprocessed MRA image samples comprises: dividing the MRA matrix images into a plurality of preset image blocks; performing blood vessel segmentation on the MRA matrix images to obtain target regions; wherein the target regions are regions containing arteries; screening a plurality of preset image blocks of the target regions to obtain target image blocks; and screening image blocks containing aneurysms based on the target image blocks and the mask matrices of the aneurysm.

[0008] Further, after the extraction of the target image blocks based on the preprocessed MRA image samples, the method comprises: performing data enhancement on the image blocks containing aneurysms; and labeling all target image blocks to mark whether the target image blocks contain aneurysms; wherein the probability of marking the image blocks containing aneurysms is 1, and the probability of marking the image blocks not containing aneurysms is 0.

[0009] Further, the generation of the MIP images based on the target image blocks comprises: projecting the target image blocks from multiple directions to generate a plurality of MIP images corresponding to the multiple directions.

[0010] Further, the constructed aneurysm detection model comprises a plurality of image classification models; and the obtaining of the trained aneurysm detection model based on the constructed aneurysm detection model and the training data set comprises: the plurality of image classification models learning the plurality of MIP images and the target image blocks corresponding to the MIP images in the training data set from different directions one by one to obtain the trained aneurysm detection model.

[0011] Further, the image classification model comprises a convolution layer, a max pooling layer, a full connection layer and an output layer; a ReLU function except the output layer is selected as an activation function, and the activation function is normalized; the output of the full connection layer is converted into a probability of containing an aneurysm through a Sigmoid function; wherein, network parameters are initialized, calculation is performed according to an input MIP image, and a prediction probability of containing an aneurysm is obtained; the prediction probability of containing an aneurysm is compared with a probability of whether containing an aneurysm which has been marked, and back propagation is performed to update network parameters, and a trained aneurysm detection model is obtained through multiple iteration calculations.

[0012] Further, based on the trained aneurysm detection model and the to-be-detected MRA image, whether the to-be-detected MRA image contains an aneurysm is predicted, comprising: converting the to-be-detected MRA image into a three-dimensional matrix image to obtain a to-be-detected MRA matrix image; dividing the to-be-detected MRA matrix image into a plurality of to-be-detected preset image blocks; performing blood vessel segmentation on the to-be-detected MRA matrix image to obtain a to-be-detected target region; wherein, the to-be-detected target region is a region containing an artery; screening a plurality of to-be-detected preset image blocks of the to-be-detected target region; respectively projecting each to-be-detected preset image block from multiple directions to generate a plurality of to-be-detected MIP images corresponding to the multiple directions; inputting the plurality of to-be-detected MIP images according to different directions in the multiple directions into a plurality of image classification models one by one to obtain a plurality of probabilities of containing an aneurysm; if an average value of the plurality of probabilities of containing an aneurysm is greater than a preset threshold value, it is determined that the to-be-detected MRA image contains an aneurysm, and the to-be-detected preset image block containing an aneurysm is marked as a to-be-detected target image block.

[0013] Further, the intracranial aneurysm detection method based on the MRA image further comprises: if the plurality of to-be-detected target image blocks are adjacent, splicing the plurality of adjacent to-be-detected target image blocks.

[0014] Another aspect of the present application proposes an intracranial aneurysm detection device based on an MRA image, comprising: a preprocessing module for preprocessing MRA image samples; an extraction module for extracting target image blocks based on the preprocessed MRA image samples; wherein, the target image blocks are image blocks containing arteries; a MIP image generation module for generating MIP images based on the target image blocks; a data set generation module for generating a training data set based on the MIP images and the target image blocks corresponding to the MIP images; a training module for obtaining a trained aneurysm detection model based on the constructed aneurysm detection model and the training data set; and a prediction module for predicting whether a to-be-detected MRA image contains an aneurysm based on the trained aneurysm detection model and the to-be-detected MRA image.

[0015] The above technical solutions of the present application have at least the following beneficial technical effects:

[0016] 1、The embodiment of the present application can effectively combine the accumulated clinical experience and the logic of manual reading in medical development into automatic analysis by preprocessing the MRA image sample, and extract the target image block based on the preprocessed MRA image sample, so as to refine the arterial region in the MRA image sample, filter out the region without arteries in the MRA image sample, and reduce the analysis cost without affecting the judgment.

[0017] 2、The embodiment of the present application can more accurately identify and extract the features of aneurysm by fully decomposing the MRA image, so that the aneurysm detection model has high sensitivity and accuracy in detecting aneurysm in the MRA image by using a deep learning algorithm combined with the logic of manual reading, and reduces the occurrence of false positive results.

[0018] 3、The embodiment of the present application combines the convolutional neural network, the maximum density projection and the information of multiple aneurysm positions to simulate the diagnosis logic of radiologists, creates a strong learning model composed of multiple weak learning models, and can reduce the occurrence of false positive results while maintaining the high sensitivity of the neural network detection result. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0020] Figure 1 is a flowchart of an intracranial aneurysm detection method based on MRA image provided by an embodiment of the present application;

[0021] Figure 2 is an MRA image sample in an embodiment of the present application;

[0022] Figure 3 is a mask information diagram of an aneurysm in an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of an aneurysm detection model in an embodiment of the present application;

[0024] Figure 5 is a schematic diagram of an image classification model structure in an embodiment of the present application;

[0025] Figure 6It is a structure schematic view of an intracranial aneurysm detection device based on MRA images provided by an embodiment of the present application. DETAILED DESCRIPTION

[0026] At present, in the existing method or device, pure machine recognition is relied on, and there is a lack of effective combination of clinical experience accumulated in medical development and logical artificial reading, and the sensitivity of aneurysm judgment in the MRA image is low, and false positive results are prone to occur.

[0027] To solve the above problems, an embodiment of the present application provides an intracranial aneurysm detection method based on MRA images, as shown in the following formula (I), which specifically comprises the following steps: Figure 1

[0028] Step S100, pre-processing the MRA image sample;

[0029] Step S200, extracting a target image block based on the pre-processed MRA image sample; wherein the target image block is an image block containing an artery;

[0030] Step S300, generating a MIP image based on the target image block;

[0031] Step S400, generating a training data set based on the MIP image and the target image block corresponding to the MIP image;

[0032] Step S500, obtaining a trained aneurysm detection model based on the constructed aneurysm detection model and the training data set;

[0033] Step S600, predicting whether the to-be-detected MRA image contains an aneurysm based on the trained aneurysm detection model and the to-be-detected MRA image.

[0034] In an embodiment of the present application, as shown in the following formula (II), the method specifically comprises the following steps: Figure 1 ​As shown, the present application can effectively combine the clinical experience accumulated in medical development and the logic of manual reading into automatic analysis by preprocessing the MRA image sample; and based on the preprocessed MRA image sample, the target image block is extracted, so that the arterial region in the MRA image sample is refined, the region without arteries in the MRA image sample is screened out, and the cost of analysis is reduced without affecting the judgment result; the MIP image is generated based on the target image block; the training data set is generated based on the MIP image and the target image block corresponding to the MIP image; the image block containing the artery is extracted into a two-dimensional MIP image and a training data set is generated, a trained aneurysm detection model is obtained by learning and training, and whether the to-be-detected MRA image contains an aneurysm is predicted based on the trained aneurysm detection model and the to-be-detected MRA image. Through sufficient decomposition of the MRA image, the aneurysm detection model can more accurately identify and extract the features of the aneurysm. By using a deep learning algorithm and combining the logic of manual reading, the detection of the aneurysm in the MRA image has high sensitivity and accuracy, and the occurrence of false positive results is reduced.

[0035] In an embodiment of the present application, step S100, the preprocessing of the MRA image sample comprises:

[0036] Step S110, manually annotating the aneurysm image in the MRA image sample;

[0037] Step S120, generating mask information of the aneurysm based on the aneurysm image;

[0038] Step S130, converting the MRA image sample and the mask information of the aneurysm after manual annotation into three-dimensional matrices respectively to obtain an MRA matrix image and a mask matrix of the aneurysm.

[0039] In this technical solution, first, the Mimics software can be used to read the Dicom format MRA image, as shown in Figure 2 Then, the obtained MRA image sample is preprocessed, the MRA image sample can be annotated by a group of clinical doctors and experts, the position and contour of the determined aneurysm image in the MRA image sample are annotated, and the key area, transition area, growth direction and other features of the aneurysm can be distinguished and annotated; then the mask information (Mask) of the aneurysm is generated based on the aneurysm image, as shown in Figure 3 for subsequent model learning and training, so as to accurately identify the features of the aneurysm in the form of machine language.

[0040] Convert all artificially annotated MRA image samples and aneurysm mask information (Mask) into Nifti format, and then use the Python-based Nibabel library to read the Nifti format file into a three-dimensional matrix respectively for subsequent image processing. Since the MRA images collected by different devices may have different resolutions, it is necessary to unify the resolution of all obtained MRA matrix images and aneurysm mask matrix. The MRA matrix image and the aneurysm mask matrix are resampled to a first preset size interval by trilinear interpolation (Trilinear Interpolation), and the first preset size interval can be 0.4*0.4*0.4mm.

[0041] In an embodiment of the present application, step S200, the target image block is extracted based on the preprocessed MRA image sample, comprising:

[0042] Step S210, the MRA matrix image is divided into a plurality of preset image blocks;

[0043] Step S220, blood vessel segmentation is performed on the MRA matrix image to obtain a target region; wherein the target region is a region containing an artery;

[0044] Step S230, a plurality of preset image blocks of the target region are screened to obtain a target image block;

[0045] Step S240, based on the target image block and the aneurysm mask matrix, an image block containing an aneurysm is screened out.

[0046] In this technical solution, the MRA matrix image is divided into a plurality of preset image blocks; since aneurysms only exist on arteries, only the region containing the artery, i.e. the target region, needs to be studied and analyzed, and then a plurality of preset image blocks of the target region are screened to obtain a target image block. Finally, based on the target image block and the aneurysm mask matrix, an image block containing an aneurysm is screened out.

[0047] Wherein, the blood vessel segmentation of the MRA matrix image can use the adaptive threshold segmentation method, such as Otsu method; traverse all HU values in the MRA matrix image, calculate all corresponding variances, and take the threshold value corresponding to the maximum variance; since the HU value of the artery is obviously different from that of other tissues and is higher, the Otsu method can be used for preliminary blood vessel segmentation.

[0048] Screening a plurality of preset image blocks of the target region to obtain a target image block; optionally, performing a sliding window operation on the entire MRA matrix image in a manner that the window size is half of the image block size, and dividing the entire MRA image into a plurality of image blocks of a second preset size; optionally, the second preset size is 16*16*16 mm.

[0049] In an embodiment of the present application, after the target image block is extracted based on the preprocessed MRA image sample, step S200 includes:

[0050] Step S250, according to the mask information (Mask) of the aneurysm corresponding to the MRA image sample, screening out all image blocks containing aneurysms, and performing data enhancement on the image blocks containing aneurysms. In an exemplary embodiment, the data enhancement method includes: the image blocks containing aneurysms can be rotated, flipped, etc. along the x, y and z axes, so that more angles and directions can be extracted and learned in the subsequent steps of extracting MIP images and learning and training the features of aneurysms.

[0051] Step S260, labeling all target image blocks to mark whether the target image blocks contain aneurysms; wherein the probability of marking the image blocks containing aneurysms is 1, and the probability of marking the image blocks not containing aneurysms is 0.

[0052] In an embodiment of the present application, step S300, generating MIP images based on the target image block includes: projecting and imaging the target image block from multiple directions to generate a plurality of MIP images corresponding to the multiple directions.

[0053] In an exemplary embodiment, the maximum intensity value encountered by the target image block along the ray is encoded and projected and imaged from nine directions: x-axis, y-axis, z-axis, diagonal of x-y plane, diagonal of y-z plane, and diagonal of z-x plane, thereby generating nine two-dimensional MIP images with a size of 16*16 mm.

[0054] In the above technical solution, even if the image blocks not containing arteries are filtered out, the proportion of the aneurysm region in the entire MRA image sample is still very small. In order to learn more about the feature information of aneurysms, in step S400, generating training data set based on the MIP image and the target image block corresponding to the MIP image specifically can include: generating training data set in a ratio of 1:1 of positive and negative samples from the MIP image projected in each of the nine directions and the image block corresponding thereto and having been marked whether containing aneurysms, for use in subsequent training of an aneurysm detection model.

[0055] In an embodiment of the present application, the constructed aneurysm detection model comprises a plurality of image classification models; and the step S500 of obtaining the trained aneurysm detection model based on the constructed aneurysm detection model and the training data set can specifically comprise: a plurality of image classification models correspondingly learning a plurality of MIP images and the target image blocks corresponding to the MIP images in the training data set from different directions to obtain the trained aneurysm detection model.

[0056] In the technical solution, KNN, logistic regression, SVM, neural network, etc. are used to predict whether the MIP image contains an aneurysm, and the predicted results are voted. Preferably, a plurality of image classification models based on convolutional neural network (CNN) are built in an ensemble learning manner, and each image classification model learns the MIP images generated from different directions and the labels corresponding to the MIP images. The input of the image classification model is a two-dimensional matrix of the MIP image, and the output value is the probability of whether the MIP image contains an aneurysm. As shown in Figure 4 Fig. 1, the image classification models 1-9 learn a plurality of MIP images and the target image blocks corresponding to the MIP images in the training data set from nine directions respectively, and each image classification model obtains a corresponding probability of whether the MIP image contains an aneurysm.

[0057] In an embodiment of the present application, as shown in Figure 5 Fig. 2, the image classification model comprises a convolutional layer, a max-pooling layer, a fully connected layer and an output layer; since the size of the generated MIP image is small, only a neural network comprising two convolutional layers, two max-pooling layers and one fully connected layer is built as the image classification model.

[0058] A ReLU function except the output layer is selected as an activation function, and the activation function is normalized;

[0059] The output of the fully connected layer is converted into a probability of containing an aneurysm by a Sigmoid function; the value of the probability is in the range of 0 to 1.

[0060] Wherein, the network parameters are initialized, the input MIP image is calculated to obtain a predicted probability of containing an aneurysm; the predicted probability of containing an aneurysm is compared with the labeled probability of containing an aneurysm, and back propagation is performed to update the network parameters, and after a plurality of iterations, a trained aneurysm detection model is obtained to realize output of an accurate predicted probability.

[0061] Specifically, as shown in Figure 5 Fig. 3, Figure 5is a schematic diagram of an image classification model in one embodiment of the present application, wherein Input MIP represents an input MIP image; Convolutions represents a convolution layer; Max-Pooling represents a maximum pooling layer; Fully-Connected represents a full connection layer; and Prediction represents an output layer. The training data set received by the aneurysm detection model is calculated layer by layer through the built neural network, and the calculation result is output, and then the loss function is used to compare the calculation result with the predicted output value (1 or 0). The key to training the neural network is to obtain the gradient of the loss function first, so as to iterate in the direction of gradient descent, and optimize the network parameters. In order to improve the calculation efficiency, the stochastic gradient descent (Stochastic Gradient Descent) is used, that is, only part of the random samples in the current training data set is used to calculate the gradient of the loss function at each training iteration. When building the model, the integrated learning is selected as the basis, a plurality of weak learning models are trained, and a certain combination strategy (Soft Voting) is added, and finally a strong learning model is formed.

[0062] In one embodiment of the present application, step S600, based on the trained aneurysm detection model and the to-be-detected MRA image, whether the to-be-detected MRA image contains an aneurysm is predicted, which can specifically include the following steps:

[0063] Step S610, the to-be-detected MRA image is converted into a three-dimensional matrix image to obtain a to-be-detected MRA matrix image;

[0064] Step S620, the to-be-detected MRA matrix image is divided into a plurality of to-be-detected preset image blocks;

[0065] Step S630, blood vessel segmentation is performed on the to-be-detected MRA matrix image to obtain a to-be-detected target region; wherein the to-be-detected target region is a region containing an artery;

[0066] Step S640, a plurality of to-be-detected preset image blocks of the to-be-detected target region are screened out;

[0067] Step S650, each to-be-detected preset image block is respectively projected and imaged from a plurality of directions to generate a plurality of to-be-detected MIP images corresponding to the plurality of directions;

[0068] Step S660, the plurality of to-be-detected MIP images are input into a plurality of image classification models one by one according to different directions in the plurality of directions, to obtain a plurality of probabilities of containing an aneurysm;

[0069] Step S670, if the average value of the plurality of probabilities of containing an aneurysm is greater than a preset threshold value, it is determined that the to-be-detected MRA image contains an aneurysm, and the to-be-detected preset image block containing the aneurysm is marked as a to-be-detected target image block.

[0070] In one embodiment of the present application, as shown in Figure 4 the obtained plurality of to-be-tested MIP images are input into the trained aneurysm detection model, that is, the plurality of to-be-tested MIP images are input into the nine image classification models one by one according to different directions in the nine directions, nine probabilities of containing aneurysms are obtained, and if the average value of the nine probabilities of containing aneurysms is greater than a preset threshold, the voter in the aneurysm detection model determines that the to-be-tested MRA image contains an aneurysm.

[0071] The preset threshold can also be set reversely with reference to the MRA image sample manually labeled and the mask information (Mask) of the aneurysm. Preferably, the preset threshold is 0.5, that is, if the average value of the plurality of probabilities of containing aneurysms is greater than 0.5, it is determined that the to-be-tested MRA image contains an aneurysm. The present application combines the convolutional neural network, the maximum density projection and the information of multiple aneurysm positions to simulate the diagnosis logic of radiologists. A strong learning model composed of a plurality of learning models is created, which can reduce the occurrence of false positive results while maintaining the high sensitivity of the detection results of the neural network.

[0072] In one embodiment of the present application, the intracranial aneurysm detection method based on the MRA image further comprises:

[0073] In step S680, if the plurality of to-be-tested target image blocks are adjacent, the plurality of adjacent to-be-tested target image blocks are spliced.

[0074] In the technical solution, in order to assist medical staff in judging the condition of the aneurysm in the to-be-tested MRA image, a reminder mark is generated on the to-be-tested target image block, that is, on the image block determined to contain an aneurysm. The reminder mark can be a red square to remind the medical staff. Further, if the plurality of to-be-tested target image blocks are adjacent, the plurality of red squares are spliced to mark the complete aneurysm image.

[0075] Based on the same inventive concept, another aspect of the present application proposes an intracranial aneurysm detection device based on an MRA image, which is used to execute the above-mentioned intracranial aneurysm detection method based on an MRA image, as shown in Figure 6 The device comprises:

[0076] A preprocessing module 10 for preprocessing the MRA image sample;

[0077] An extraction module 20 for extracting a target image block based on the preprocessed MRA image sample; wherein the target image block is an image block containing an artery;

[0078] A MIP image generation module 30 for generating a MIP image based on the target image block;

[0079] generating a training data set based on the MIP image and the target image block corresponding to the MIP image;

[0080] training a trained aneurysm detection model based on the constructed aneurysm detection model and the training data set;

[0081] predicting whether the to-be-detected MRA image contains an aneurysm based on the trained aneurysm detection model and the to-be-detected MRA image.

[0082] Wherein, the specific shape structure of the pre-processing module 10, the extraction module 20, the MIP image generating module 30, the data set generating module 40, the training module 50 and the prediction module 60 in the embodiment of the present application is not limited, and those skilled in the art can set them arbitrarily according to their functions, which will not be repeated here. In addition, the specific implementation process and implementation effect of the operation steps realized by each module in the embodiment of the present application are the same as the specific implementation process and implementation effect of steps S100-S600, which can be referred to the above description, and will not be repeated here.

[0083] It should be understood that the above specific embodiments of the present application are only used for illustrative or explanatory purposes of the principles of the present application, and do not constitute a limitation of the present application. Therefore, any modification, equivalent replacement, improvement, etc. made without departing from the spirit and scope of the present application shall be included in the protection scope of the present application. In addition, the appended claims of the present application are intended to cover all variations and modifications falling within the scope and boundary of the appended claims, or the equivalent forms of such scope and boundary.

Claims

1. A method for detecting intracranial aneurysms based on MRA images, characterized in that , comprising: Preprocessing the MRA image sample, specifically including: manually labeling the aneurysm image in the MRA image sample; generating mask information of the aneurysm based on the aneurysm image; converting the manually labeled MRA image sample and the mask information of the aneurysm into three-dimensional matrices respectively to obtain MRA matrix images and aneurysm mask matrices; Extracting target image blocks based on the preprocessed MRA image sample; wherein the target image block is an image block containing an artery; specifically including: dividing the MRA matrix image into a plurality of preset image blocks; performing blood vessel segmentation on the MRA matrix image to obtain a target region; wherein the target region is a region containing an artery; screening a plurality of preset image blocks of the target region to obtain target image blocks; screening image blocks containing aneurysms based on the target image blocks and the aneurysm mask matrix; Generating MIP images based on the target image blocks, specifically including: projecting the target image blocks from multiple directions to generate a plurality of MIP images corresponding to the multiple directions; Generating a training data set based on the MIP images and the target image blocks corresponding to the MIP images; Based on the constructed aneurysm detection model and the training data set, a trained aneurysm detection model is obtained; wherein the constructed aneurysm detection model includes a plurality of image classification models, and the plurality of image classification models correspond to the plurality of MIP images and the target image blocks corresponding to the MIP images in the training data set from different directions; Based on the trained aneurysm detection model and the MRA image to be measured, whether the MRA image to be measured contains an aneurysm is predicted.

2. The method of claim 1, wherein, After extracting the target image blocks based on the preprocessed MRA image sample, it includes: Data augmentation is performed on the image blocks containing aneurysms; All target image blocks are labeled to mark whether the target image blocks contain aneurysms; wherein the probability of marking the image blocks containing aneurysms is 1, and the probability of marking the image blocks not containing aneurysms is 0.

3. The method of claim 1, wherein, The image classification model includes a convolution layer, a max pooling layer, a full connection layer and an output layer; Selecting a ReLU function as an activation function except for the output layer, and normalizing the activation function; The output of the full connection layer is converted into a probability of containing an aneurysm through a Sigmoid function; wherein, Initialize the network parameters, calculate according to the input MIP image, and obtain the prediction probability of containing an aneurysm; Compare the prediction probability of containing an aneurysm with the probability of whether it contains an aneurysm that has been marked, and perform back propagation to update the network parameters. After multiple iterations of calculation, a trained aneurysm detection model is obtained.

4. The method of claim 3, wherein, Based on the trained aneurysm detection model and the MRA image to be measured, whether the MRA image to be measured contains an aneurysm is predicted, including: Converting the MRA image to be measured into a three-dimensional matrix image to obtain a MRA matrix image to be measured; Divide the MRA matrix image to be measured into a plurality of test preset image blocks; The MRA image to be tested is subjected to blood vessel segmentation to obtain a target region to be tested; the target region to be tested is a region containing arteries; A plurality of preset image blocks to be tested in the target region to be tested are screened out; Each preset image block to be tested is subjected to projection imaging from multiple directions to generate a plurality of MIP images to be tested corresponding to the multiple directions; The plurality of MIP images to be tested are input into a plurality of image classification models one by one according to different directions in the multiple directions to obtain a plurality of probabilities of containing an aneurysm; If an average value of the plurality of probabilities of containing an aneurysm is greater than a preset threshold value, it is determined that the MRA image to be tested contains an aneurysm, and the preset image block to be tested containing an aneurysm is marked as a target image block to be tested.

5. The method of claim 4, wherein, Further comprising: If the plurality of target image blocks to be tested are adjacent, the plurality of adjacent target image blocks to be tested are spliced.

6. A device for detecting intracranial aneurysms based on MRA imaging, characterized in that, Comprise: A preprocessing module is configured to preprocess MRA image samples, specifically configured to generate mask information of an aneurysm based on an aneurysm image in the MRA image samples after manually labeling the aneurysm image, and convert the MRA image samples after manual labeling and the mask information of the aneurysm into three-dimensional matrices respectively to obtain MRA matrix images and a mask matrix of the aneurysm; An extraction module is configured to extract target image blocks based on the MRA image samples after preprocessing; the target image blocks are image blocks containing arteries; the extraction module is specifically configured to divide the MRA matrix images into a plurality of preset image blocks, perform blood vessel segmentation on the MRA matrix images to obtain a target region, the target region is a region containing arteries, and screen out a plurality of preset image blocks of the target region to obtain target image blocks; and screen out image blocks containing aneurysms based on the target image blocks and the mask matrix of the aneurysm; A MIP image generation module is configured to generate MIP images based on the target image blocks, specifically configured to project and image the target image blocks from multiple directions to generate a plurality of MIP images corresponding to the multiple directions; A data set generation module is configured to generate a training data set based on the MIP images and the target image blocks corresponding to the MIP images; A training module is configured to obtain a trained aneurysm detection model based on the constructed aneurysm detection model and the training data set; the constructed aneurysm detection model comprises a plurality of image classification models, and the plurality of image classification models one by one correspond to learn a plurality of MIP images in the training data set and the target image blocks corresponding to the MIP images from different directions; A prediction module is configured to predict whether the MRA image to be tested contains an aneurysm based on the trained aneurysm detection model and the MRA image to be tested.

Images