CN114723937A - Method and system for classifying blood vessel surrounding gaps based on nuclear magnetic resonance image - Google Patents
Method and system for classifying blood vessel surrounding gaps based on nuclear magnetic resonance image Download PDFInfo
- Publication number
- CN114723937A CN114723937A CN202210305047.3A CN202210305047A CN114723937A CN 114723937 A CN114723937 A CN 114723937A CN 202210305047 A CN202210305047 A CN 202210305047A CN 114723937 A CN114723937 A CN 114723937A
- Authority
- CN
- China
- Prior art keywords
- region
- perivascular
- magnetic resonance
- nuclear magnetic
- interest
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention provides a perivascular clearance classification method and a perivascular clearance classification system based on nuclear magnetic resonance images, which belong to the field of image processing and deep learning, and comprise the following steps: acquiring a nuclear magnetic resonance image to be classified; the nuclear magnetic resonance images to be classified comprise T1weighted nuclear magnetic resonance images and T2weighted nuclear magnetic resonance images; determining a first classification result and a second classification result corresponding to the nuclear magnetic resonance image to be classified based on the gap classification model according to the nuclear magnetic resonance image to be classified; the first classification result is no perivascular space or perivascular space; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces or greater than 40 perivascular spaces. The method and the device realize the automatic determination of the perivascular gaps corresponding to the second classification and the fifth classification by using the T1 and T2weighted nuclear magnetic resonance images, and improve the accuracy of perivascular gap classification.
Description
Technical Field
The invention relates to the field of image processing and deep learning, in particular to a method and a system for classifying blood vessel surrounding gaps based on nuclear magnetic resonance images.
Background
The cerebrovascular-vascular disease (CSVD) is a common disease seriously harming the health of people in China, and is easily ignored by patients and even clinicians due to the hidden diseases. The enlarged perivascular space (EPVS) is one of the important imaging characteristics of the cerebrovascular and small vessel diseases, is an index reflecting the existence and severity of the cerebrovascular and small vessel diseases, is related to aging and cognitive decline, and is also one of the markers of neuroinflammation.
The evaluation of the peripheral vascular clearance can objectively reflect the severity of the peripheral vascular clearance, but the evaluation is difficult and time-consuming, and the currently widely used peripheral vascular clearance evaluation and evaluation method is mainly that a doctor manually interprets the peripheral vascular clearance through a magnetic resonance image, but the manual interpretation method has inter-marker and internal differences, so that the judgment result is not accurate enough and the efficiency is low.
Therefore, a method for classifying the grade of the perivascular space which can be automatically and accurately is needed.
Disclosure of Invention
The invention aims to provide a method and a system for classifying the perivascular gaps based on nuclear magnetic resonance images, which can improve the accuracy of the classification of the perivascular gaps.
In order to achieve the purpose, the invention provides the following scheme:
a perivascular space classification method based on nuclear magnetic resonance images comprises the following steps:
acquiring a nuclear magnetic resonance image to be classified; the nuclear magnetic resonance images to be classified comprise T1weighted nuclear magnetic resonance images and T2weighted nuclear magnetic resonance images;
determining a first classification result and a second classification result corresponding to the nuclear magnetic resonance image to be classified based on a gap classification model according to the nuclear magnetic resonance image to be classified; the first classification result is no perivascular space or perivascular space; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces or more than 40 perivascular spaces;
the gap classification model is obtained by adopting a training sample set to train the three-dimensional convolution neural network in advance; the training sample set comprises a plurality of groups of sample nuclear magnetic resonance images and a category corresponding to each group of sample nuclear magnetic resonance images; the sample NMR image comprises a T1weighted NMR image sample and a T2weighted NMR image sample; the categories represent the grades of the perivascular gaps and comprise a second category and a fifth category; the two classifications include avascular and perivascular gaps; the five classifications include avascular spaces, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces and greater than 40 perivascular spaces.
Optionally, the method for establishing the gap classification model includes:
extracting an interested area of the T1weighted nuclear magnetic resonance image sample to obtain a first interested area;
registering the first region of interest and the T1weighted nuclear magnetic resonance image sample to a T2weighted nuclear magnetic resonance image sample space to obtain a registered first region of interest and a registered T1weighted image;
extracting the interested region of the T2weighted nuclear magnetic resonance image according to the registered first interested region to obtain a second interested region;
performing iterative training on the three-dimensional convolutional neural network by adopting a back propagation and gradient descent algorithm based on a loss function according to the registered first region of interest, the registered second region of interest and the corresponding category until the loss function is converged to obtain an optimal three-dimensional convolutional neural network; the optimal three-dimensional convolutional neural network is a gap classification model.
Optionally, the extracting the region of interest of the T1weighted nuclear magnetic resonance image sample to obtain a first region of interest specifically includes:
extracting a basal ganglia region and a hemioval central region of the T1weighted nuclear magnetic resonance image sample;
performing morphological binary expansion on the basal ganglia region to obtain an expanded basal ganglia region;
and respectively carrying out normalization treatment on the expansion basal ganglia region and the semi-oval central region to obtain a first region of interest.
Optionally, the three-dimensional convolutional neural network includes an input layer, a hidden layer, and an output layer, which are connected in sequence; the output layer comprises a global average pooling layer, a first linear layer and a second linear layer; the global average pooling layer is connected with the hidden layer, the first linear layer and the second linear layer respectively;
performing iterative training on the three-dimensional convolutional neural network by adopting a back propagation and gradient descent algorithm based on a loss function according to the categories corresponding to the registered first region of interest and the registered second region of interest, specifically comprising:
transmitting the registered first region of interest and the second region of interest to a hidden layer through an input layer;
respectively extracting the characteristics of the registered first interested region and the registered second interested region through a hidden layer to obtain initial characteristic information;
performing feature selection on the initial feature information through the global average pooling layer to obtain optimal feature information;
classifying the optimal characteristic information through the first linear layer to obtain a first classification result; the first classification result is no perivascular space or perivascular space;
classifying the optimal characteristic information through the second linear layer to obtain a second classification result; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces or more than 40 perivascular spaces;
determining a loss value of a loss function according to the first classification result, the second classification result, the class corresponding to the registered first region of interest and the class corresponding to the second region of interest;
and performing iterative training on the hidden layer, the global average pooling layer, the first linear layer and the second linear layer by adopting a back propagation and gradient descent algorithm based on the loss value of the loss function.
Optionally, the loss function is:
LPVS＝0.5*LCE+0.5*LICC；
wherein L isPVSTo be a loss function value, LCEFor cross entropy loss, LICCFor intra-group correlation coefficient loss, MSR is the mean square of the same grader rating different samples, MSC is the mean square of different graders rating the same sample, MSE is the mean square error of all graders rating all samples, and n is the number of samples.
In order to achieve the above purpose, the invention also provides the following scheme:
a magnetic resonance image based perivascular gap classification system, comprising:
the image acquisition unit is used for acquiring a nuclear magnetic resonance image to be classified; the nuclear magnetic resonance images to be classified comprise T1weighted nuclear magnetic resonance images and T2weighted nuclear magnetic resonance images;
the classification unit is connected with the image acquisition unit and used for determining a first classification result and a second classification result corresponding to the nuclear magnetic resonance image to be classified based on a gap classification model according to the nuclear magnetic resonance image to be classified; the first classification result is no perivascular space or perivascular space; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces or more than 40 perivascular spaces;
the gap classification model is obtained by adopting a training sample set to train the three-dimensional convolution neural network in advance; the training sample set comprises a plurality of groups of sample nuclear magnetic resonance images and a category corresponding to each group of sample nuclear magnetic resonance images; the sample NMR image comprises a T1weighted NMR image sample and a T2weighted NMR image sample; the categories represent the grades of the perivascular gaps and comprise a second category and a fifth category; the two classifications include avascular and perivascular gaps; the five classifications include avascular perivascular spaces, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces and greater than 40 perivascular spaces.
Optionally, the magnetic resonance image-based perivascular space classification system further includes:
the first extraction unit is used for extracting the interested region of the T1weighted nuclear magnetic resonance image sample to obtain a first interested region;
the registration unit is connected with the first extraction unit and is used for registering the first region of interest and the T1weighted nuclear magnetic resonance image sample to a T2weighted nuclear magnetic resonance image sample space to obtain a registered first region of interest and a registered T1weighted image;
the second extraction unit is connected with the registration unit and is used for extracting the interested area of the T2weighted nuclear magnetic resonance image according to the registered first interested area to obtain a second interested area;
the training unit is respectively connected with the registration unit, the second extraction unit and the classification unit and is used for performing iterative training on the three-dimensional convolutional neural network by adopting a back propagation and gradient descent algorithm based on a loss function according to the registered first region of interest, the registered second region of interest and the corresponding category until the loss function is converged to obtain an optimal three-dimensional convolutional neural network; the optimal three-dimensional convolutional neural network is a gap classification model.
Optionally, the first extraction unit includes:
the region extraction module is used for extracting a basal ganglia region and a half oval center region of the T1weighted nuclear magnetic resonance image sample;
the expansion module is connected with the region extraction module and used for performing morphological binary expansion on the basal ganglia region to obtain an expanded basal ganglia region;
and the normalization module is respectively connected with the expansion module and the region extraction module and is used for respectively carrying out normalization processing on the expansion basal ganglia region and the semi-oval central region to obtain a first region of interest.
Optionally, the three-dimensional convolutional neural network includes an input layer, a hidden layer, and an output layer, which are connected in sequence; the output layer comprises a global average pooling layer, a first linear layer and a second linear layer; the global average pooling layer is connected with the hidden layer, the first linear layer and the second linear layer respectively;
the input layer is used for transmitting the registered first region of interest and the registered second region of interest to a hidden layer;
the hidden layer is used for extracting the characteristics of the registered first region of interest and the registered second region of interest to obtain initial characteristic information;
the global average pooling layer is connected with the hidden layer and is used for carrying out feature selection on the initial feature information to obtain optimal feature information;
the first linear layer is connected with the global average pooling layer and is used for classifying the optimal feature information to obtain a first classification result; the first classification result is no perivascular space or perivascular space;
the second linear layer is connected with the global average pooling layer and is used for classifying the optimal feature information to obtain a second classification result; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces, or greater than 40 perivascular spaces.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
based on the T1 and T2weighted nuclear magnetic resonance images, a gap classification model is established through a deep learning method, so that the peripheral vascular gaps corresponding to the second classification and the fifth classification are automatically determined by using the T1 and T2weighted nuclear magnetic resonance images, two grading tasks of the second classification and the fifth classification of the peripheral vascular gaps are combined into one model, and the accuracy of the peripheral vascular gap classification is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flowchart of a method for classifying perivascular spaces based on MRI images according to the present invention;
FIG. 2 is a schematic diagram of a gap classification model;
fig. 3 is a schematic block diagram of a perivascular space classification system based on mri according to the present invention.
Description of the symbols:
the image acquisition unit-1, the classification unit-2, the first extraction unit-3, the registration unit-4, the second extraction unit-5 and the training unit-6.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a method and a system for classifying the surrounding gaps of blood vessels based on a nuclear magnetic resonance image, wherein a gap classification model is established by a depth learning method based on T1 and T2weighted nuclear magnetic resonance images, so that the surrounding gaps of the blood vessels corresponding to two classifications and five classifications are automatically determined by using the T1 and T2weighted nuclear magnetic resonance images, and meanwhile, two grading tasks of the second classification and the five classifications of the surrounding gaps of the blood vessels are combined into one model, and the accuracy of the classification of the surrounding gaps of the blood vessels is improved.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
In magnetic resonance imaging, contrast depends on the magnetic properties and the number of hydrogen nuclei in the imaged region. By running different sequences with different weights, different contrasts in the region to be imaged can be selected. If the magnetic resonance image mainly reflects the T1 value difference between tissues, the magnetic resonance image is a T1weighted image (T1weighted image, T1WI), and the T1 image has higher resolution in the direction vertical to the cross section and can well reflect the brain structure information; if the difference between T2 values is mainly reflected, the weighted image T2 is the T2weighted image (T2 WI).
As shown in fig. 1, the method for classifying the perivascular space based on mri of the present invention includes:
s1: acquiring a nuclear magnetic resonance image to be classified; the to-be-classified MRI images include T1-weighted MRI images and T2-weighted MRI images.
S2: and determining a first classification result and a second classification result corresponding to the nuclear magnetic resonance image to be classified based on a gap classification model according to the nuclear magnetic resonance image to be classified. The first classification result is no perivascular space or perivascular space. The second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces, or greater than 40 perivascular spaces.
The gap classification model is obtained by adopting a training sample set to train the three-dimensional convolution neural network in advance. The training sample set comprises a plurality of groups of sample nuclear magnetic resonance images and the corresponding category of each group of sample nuclear magnetic resonance images. The sample nmr images include T1weighted nmr image samples and T2weighted nmr image samples. The categories represent the grades of the perivascular gaps and comprise a second category and a fifth category; the two classifications include avascular and perivascular gaps. The five classifications include avascular perivascular spaces, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces and greater than 40 perivascular spaces.
The invention uses a deep learning method, finishes the automatic classification of the grading of the peripheral gaps of the expanded blood vessels by weighting the nuclear magnetic resonance image data T1 and T2, has complete and objective classification results, avoids the evaluation difference caused in the observation process of medical personnel, can quickly and accurately predict the future disability condition of a patient by means of the strong calculation capability of a computer, saves the fussy evaluation time and effectively reduces the marking pressure of a doctor. In addition, a multi-task learning method is utilized, the two-classification and five-classification grading tasks are combined, overfitting of the gap classification model is relieved, and generalization capability of the gap classification model is improved.
Further, the method for establishing the gap classification model specifically comprises the following steps:
s101: and extracting the interested region of the T1weighted nuclear magnetic resonance image sample to obtain a first interested region. The first region of interest includes a basal ganglia region and a hemioval central region. Since the enlarged perivascular gap mainly occurs in the two regions of the basal ganglia and the center of the hemioval, in order to reduce the redundant information input model, the two ROIs are extracted by a region of interest (ROI) extraction method. Since the T1weighted nmr image contains abundant structural information, the masks of the basal ganglia and the central region of the hemioval were extracted using the T1weighted nmr image using the freeschurr software.
S102: and registering the first region of interest and the T1weighted nuclear magnetic resonance image sample to a T2weighted nuclear magnetic resonance image sample space to obtain a registered first region of interest and a registered T1weighted image.
S103: and extracting the interested region of the T2weighted nuclear magnetic resonance image according to the registered first interested region to obtain a second interested region. Specifically, the fundamentals, the central hemioval region and the T1weighted nmr image samples are first registered to the T2weighted nmr image sample space, and then the registered mask is applied to the registered T1weighted nmr image and the T2weighted nmr image of the corresponding samples, so as to obtain the ROIs of the T1 and T2weighted nmr images.
S104: performing iterative training on the three-dimensional convolutional neural network by adopting a back propagation and gradient descent algorithm based on a loss function according to the registered first region of interest, the registered second region of interest and the corresponding category until the loss function is converged to obtain an optimal three-dimensional convolutional neural network; the optimal three-dimensional convolutional neural network is a gap classification model.
Specifically, step S101 specifically includes:
and extracting a basal ganglia region and a hemioval center region of the T1weighted nuclear magnetic resonance image sample.
And performing morphological binary expansion on the basal zone to obtain an expanded basal zone. Since the expanded perivascular space is generally distributed on the border of the basal ganglia, the invention performs morphological binary expansion on the basal ganglia region, so that the perivascular space is clearer.
And respectively carrying out normalization treatment on the expansion basal ganglia region and the semi-oval central region to obtain a first region of interest. In this embodiment, the edges are calculated first, and the maximum value of all samples is taken for clipping, so that the sizes of the nuclear magnetic resonance image samples in the sample set are uniform, batch processing is facilitated, the calculation amount is reduced, and the calculation resources are saved. Normalization processing is carried out through the expansion basal ganglia region and the semi-oval center region, interference of background voxel information on data distribution is prevented, and accuracy of the gap classification model is improved.
Furthermore, the three-dimensional convolution neural network comprises an input layer, a hidden layer and an output layer which are connected in sequence; the output layer comprises a global average pooling layer, a first linear layer and a second linear layer; the global average pooling layer is connected with the hidden layer, the first linear layer and the second linear layer respectively. In this embodiment, the first linear layer and the second linear layer are independent of each other to realize the second classification and the fifth classification, respectively. The first linear layer includes 1 neuron and the second linear layer includes 5 neurons.
Step S104 specifically includes:
and transmitting the registered first region of interest and the second region of interest to a hidden layer through an input layer.
And respectively carrying out feature extraction on the registered first region of interest and the registered second region of interest through a hidden layer to obtain initial feature information.
And performing feature selection on the initial feature information through the global average pooling layer to obtain optimal feature information.
Classifying the optimal characteristic information through the first linear layer to obtain a first classification result; the first classification result is no perivascular space or perivascular space.
Classifying the optimal characteristic information through the second linear layer to obtain a second classification result; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces, or greater than 40 perivascular spaces.
And determining a loss value of a loss function according to the first classification result, the second classification result, the category corresponding to the first region of interest and the category corresponding to the second region of interest.
And performing iterative training on the hidden layer, the global average pooling layer, the first linear layer and the second linear layer by adopting a back propagation and gradient descent algorithm based on the loss value of the loss function.
The enlarged perivascular space is generally tubular in the direction perpendicular to the cross-section, so its visual rating requires a comparison between multiple cross-sections, taking the most severe layer as a rating for the entire area, taking into account contextual information. The method constructs a convolutional neural network model PVS-DenseNet by means of a 3D convolutional neural network based on the DenseNet thought. The 3D convolutional neural network can effectively extract the context information, and a neural network model is built by using a Pythrch deep learning framework. And adjusting hyper-parameters of the neural network in the training process, wherein the hyper-parameters comprise learning rate, batch training size and training turns. The parameters of each convolution are optimized each iteration during training.
Because the relevance of the two-classification task and the five-classification task is high, the multi-task learning method is applied to the interval classification model, the multi-task learning adopts hard parameter sharing, and the task parameters are updated independently only in the last linear layer, so that the interval classification model can simultaneously and accurately classify and expand two grades of the second grade and the fifth grade of the interval around the blood vessel.
The dilated perivascular gap is low signal on the T1weighted nmr image and high signal on the T2weighted nmr image, and the gap classification model of the invention uses a combination of fixed-size three-dimensional brain T1 and T2weighted nmr image data as input. In order to relieve the problem of gradient disappearance and improve the utilization efficiency of features, inspired by DenseNet, feature maps learned by different convolution blocks are connected in series, so that the input variables of subsequent layers are increased, the feature utilization efficiency and the learning effect of the network are improved, and each convolution block integrates the information of all the output feature maps of the previous convolution blocks to be input.
As shown in fig. 2, the hidden layer has 5 repeated Convolutional blocks (Convolutional blocks), each of which contains two identical Convolutional units (Convolutional units) and one 2 × 2 × 1 max-pooling layer with step size of 2. The convolution unit contains a 3 × 3 × 3 convolution layer with step size 1, a ReLU activation and a 3D bulk normalization layer. In the first volume block, the number of feature channels is set to 8 and doubled after entering the next volume block to deduce a sufficiently rich representation of the brain information.
Specifically, the loss function is:
LPVS＝0.5*LCE+0.5*LICC；
wherein L isPVSTo the value of the loss function, LCEFor cross entropy loss, LICCFor intra-group correlation coefficient loss, MSR is the mean square of the same grader rating different samples, MSC is the mean square of different graders rating the same sample, MSE is the mean square error of all graders rating all samples, and n is the number of samples. In this embodiment, LCEConventional cross-entropy losses are employed.
ICC (intragroup correlation coefficient) is one of the confidence coefficient indexes for measuring and evaluating the confidence and retest confidence among observers, and is defined as follows:
introduction of LICCThe loss can effectively improve the classification effect of the model and obviously improve the classification evaluation index ICC (extraction)About 37% higher).
In addition, after the model training is finished, the model is verified by using a cross-validation method, and the model is verified by using AUC and ICC indexes.
As shown in fig. 3, the perivascular space classification system based on mri of the present invention includes: an image acquisition unit 1 and a classification unit 2.
The image acquisition unit 1 is used for acquiring a nuclear magnetic resonance image to be classified. The to-be-classified MRI images include T1-weighted MRI images and T2-weighted MRI images.
The classifying unit 2 is connected with the image acquiring unit 1, and the classifying unit 2 is configured to determine a first classifying result and a second classifying result corresponding to the nuclear magnetic resonance image to be classified based on a gap classification model according to the nuclear magnetic resonance image to be classified. The first classification result is no perivascular space or perivascular space. The second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces or greater than 40 perivascular spaces.
The gap classification model is obtained by adopting a training sample set to train a three-dimensional convolution neural network in advance; the training sample set comprises a plurality of groups of sample nuclear magnetic resonance images and a category corresponding to each group of sample nuclear magnetic resonance images; the sample NMR image comprises a T1weighted NMR image sample and a T2weighted NMR image sample; the categories represent the grades of the perivascular gaps and comprise a second category and a fifth category; the two classifications include avascular and perivascular gaps; the five classifications include avascular perivascular spaces, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces and greater than 40 perivascular spaces.
Further, the perivascular space classification system based on magnetic resonance imaging further comprises: a first extraction unit 3, a registration unit 4, a second extraction unit 5 and a training unit 6.
The first extraction unit 3 is configured to extract a region of interest of the T1weighted mri sample, so as to obtain a first region of interest. The region of interest includes a basal ganglia region and a hemioval central region.
The registration unit 4 is connected to the first extraction unit 3, and the registration unit 4 is configured to register the first region of interest and the T1weighted nmr image sample to a T2weighted nmr image sample space, so as to obtain a registered first region of interest and a registered T1weighted image.
The second extracting unit 5 is connected to the registering unit 4, and the second extracting unit 5 is configured to extract the region of interest of the T2weighted nuclear magnetic resonance image according to the registered first region of interest, so as to obtain a second region of interest.
The training unit 6 is connected to the registration unit 4 and the second extraction unit 5, respectively, and the training unit 6 is configured to perform iterative training on the three-dimensional convolutional neural network by using a back propagation and gradient descent algorithm based on a loss function according to the registered first region of interest, the registered second region of interest, and the corresponding category until the loss function converges to obtain an optimal three-dimensional convolutional neural network; the optimal three-dimensional convolution neural network is a gap classification model.
Specifically, the first extraction unit 3 includes: the device comprises an area extraction module, an expansion module and a normalization module.
The region extraction module is used for extracting a basal ganglia region and a hemioval central region of the T1weighted nuclear magnetic resonance image sample.
The expansion module is connected with the region extraction module and is used for performing morphological binary expansion on the basal ganglia region to obtain an expanded basal ganglia region.
The normalization module is respectively connected with the expansion module and the region extraction module, and is used for respectively carrying out normalization processing on the expansion basal ganglia region and the semi-oval central region to obtain a first region of interest.
Furthermore, the three-dimensional convolution neural network comprises an input layer, a hidden layer and an output layer which are connected in sequence; the output layer comprises a global average pooling layer, a first linear layer and a second linear layer; the global average pooling layer is connected with the hidden layer, the first linear layer and the second linear layer respectively;
the input layer is used for transmitting the registered first region of interest and the registered second region of interest to a hidden layer.
The hidden layer is used for extracting the characteristics of the registered first interested region and the registered second interested region to obtain initial characteristic information.
The global average pooling layer is connected with the hidden layer and used for carrying out feature selection on the initial feature information to obtain optimal feature information.
The first linear layer is connected with the global average pooling layer and is used for classifying the optimal feature information to obtain a first classification result; the first classification result is no perivascular space or perivascular space.
The second linear layer is connected with the global average pooling layer and is used for classifying the optimal feature information to obtain a second classification result; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces, or greater than 40 perivascular spaces.
Compared with the prior art, the perivascular clearance classification system based on the nuclear magnetic resonance image has the same beneficial effects as the perivascular clearance classification method based on the nuclear magnetic resonance image, and is not repeated herein.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (9)
1. A perivascular space classification method based on a nuclear magnetic resonance image is characterized by comprising the following steps:
acquiring a nuclear magnetic resonance image to be classified; the nuclear magnetic resonance images to be classified comprise T1weighted nuclear magnetic resonance images and T2weighted nuclear magnetic resonance images;
determining a first classification result and a second classification result corresponding to the nuclear magnetic resonance image to be classified based on a gap classification model according to the nuclear magnetic resonance image to be classified; the first classification result is no perivascular space or perivascular space; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces or more than 40 perivascular spaces;
the gap classification model is obtained by adopting a training sample set to train the three-dimensional convolution neural network in advance; the training sample set comprises a plurality of groups of sample nuclear magnetic resonance images and a category corresponding to each group of sample nuclear magnetic resonance images; the sample NMR image comprises a T1weighted NMR image sample and a T2weighted NMR image sample; the categories represent the grades of the perivascular gaps and comprise a second category and a fifth category; the two classifications include avascular and perivascular gaps; the five classifications include avascular perivascular spaces, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces and greater than 40 perivascular spaces.
2. The method for classifying intravascular space based on magnetic resonance images according to claim 1, wherein the method for establishing the space classification model comprises:
extracting an interested region of the T1weighted nuclear magnetic resonance image sample to obtain a first interested region;
registering the first region of interest and the T1weighted nuclear magnetic resonance image sample to a T2weighted nuclear magnetic resonance image sample space to obtain a registered first region of interest and a registered T1weighted image;
extracting an interested area of the T2weighted nuclear magnetic resonance image according to the registered first interested area to obtain a second interested area;
performing iterative training on the three-dimensional convolutional neural network by adopting a back propagation and gradient descent algorithm based on a loss function according to the registered first region of interest, the registered second region of interest and the corresponding category until the loss function is converged to obtain an optimal three-dimensional convolutional neural network; the optimal three-dimensional convolutional neural network is a gap classification model.
3. The method for classifying perivascular gaps based on mri as claimed in claim 2, wherein the extracting the region of interest of the T1weighted mri sample to obtain a first region of interest specifically includes:
extracting a basal ganglia region and a hemioval central region of the T1weighted nuclear magnetic resonance image sample;
performing morphological binary expansion on the basal ganglia region to obtain an expanded basal ganglia region;
and respectively carrying out normalization treatment on the expansion basal ganglia region and the semi-oval central region to obtain a first region of interest.
4. The method for classifying perivascular gaps based on magnetic resonance images according to claim 2, wherein the three-dimensional convolutional neural network comprises an input layer, a hidden layer and an output layer which are connected in sequence; the output layer comprises a global average pooling layer, a first linear layer and a second linear layer; the global average pooling layer is connected with the hidden layer, the first linear layer and the second linear layer respectively;
performing iterative training on the three-dimensional convolutional neural network by adopting a back propagation and gradient descent algorithm based on a loss function according to the categories corresponding to the registered first region of interest and the registered second region of interest, specifically comprising:
transmitting the registered first region of interest and the second region of interest to a hidden layer through an input layer;
respectively extracting the characteristics of the registered first interested region and the registered second interested region through a hidden layer to obtain initial characteristic information;
performing feature selection on the initial feature information through the global average pooling layer to obtain optimal feature information;
classifying the optimal characteristic information through the first linear layer to obtain a first classification result; the first classification result is no perivascular space or perivascular space;
classifying the optimal characteristic information through the second linear layer to obtain a second classification result; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces or more than 40 perivascular spaces;
determining a loss value of a loss function according to the first classification result, the second classification result, the category corresponding to the first region of interest and the category corresponding to the second region of interest;
and performing iterative training on the hidden layer, the global average pooling layer, the first linear layer and the second linear layer by adopting a back propagation and gradient descent algorithm based on the loss value of the loss function.
5. The method of claim 2, wherein the loss function is:
LPVS＝0.5*LCE+0.5*LICC；
wherein L isPVSTo be a loss function value, LCEFor cross entropy loss, LICCFor intra-group correlation coefficient loss, MSR is the mean square of the same grader rating different samples, MSC is the mean square of different graders rating the same sample, MSE is the mean square error of all graders rating all samples, and n is the number of samples.
6. A magnetic resonance image-based perivascular gap classification system, the magnetic resonance image-based perivascular gap classification system comprising:
the image acquisition unit is used for acquiring a nuclear magnetic resonance image to be classified; the nuclear magnetic resonance images to be classified comprise T1weighted nuclear magnetic resonance images and T2weighted nuclear magnetic resonance images;
the classification unit is connected with the image acquisition unit and used for determining a first classification result and a second classification result corresponding to the nuclear magnetic resonance image to be classified based on a gap classification model according to the nuclear magnetic resonance image to be classified; the first classification result is no perivascular space or perivascular space; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces or more than 40 perivascular spaces;
the gap classification model is obtained by adopting a training sample set to train the three-dimensional convolution neural network in advance; the training sample set comprises a plurality of groups of sample nuclear magnetic resonance images and a category corresponding to each group of sample nuclear magnetic resonance images; the sample NMR image comprises a T1weighted NMR image sample and a T2weighted NMR image sample; the categories represent the grades of the perivascular gaps and comprise a second category and a fifth category; the two classifications include avascular and perivascular gaps; the five classifications include avascular perivascular spaces, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces and greater than 40 perivascular spaces.
7. The mri image-based perivascular gap classification system according to claim 6, further comprising:
the first extraction unit is used for extracting the interested region of the T1weighted nuclear magnetic resonance image sample to obtain a first interested region;
the registration unit is connected with the first extraction unit and is used for registering the first region of interest and the T1weighted nuclear magnetic resonance image sample to a T2weighted nuclear magnetic resonance image sample space to obtain a registered first region of interest and a registered T1weighted image;
the second extraction unit is connected with the registration unit and used for extracting the interested region of the T2weighted nuclear magnetic resonance image according to the registered first interested region to obtain a second interested region;
the training unit is respectively connected with the registration unit, the second extraction unit and the classification unit and is used for performing iterative training on the three-dimensional convolutional neural network by adopting a back propagation and gradient descent algorithm based on a loss function according to the registered first region of interest, the registered second region of interest and the corresponding category until the loss function is converged to obtain an optimal three-dimensional convolutional neural network; the optimal three-dimensional convolutional neural network is a gap classification model.
8. The system according to claim 7, wherein the first extraction unit comprises:
the region extraction module is used for extracting a basal ganglia region and a half oval center region of the T1weighted nuclear magnetic resonance image sample;
the expansion module is connected with the region extraction module and used for performing morphological binary expansion on the basal ganglia region to obtain an expanded basal ganglia region;
and the normalization module is respectively connected with the expansion module and the region extraction module and is used for respectively carrying out normalization processing on the expansion basal ganglia region and the semi-oval central region to obtain a first region of interest.
9. The system according to claim 7, wherein the three-dimensional convolutional neural network comprises an input layer, a hidden layer and an output layer which are connected in sequence; the output layer comprises a global average pooling layer, a first linear layer and a second linear layer; the global average pooling layer is connected with the hidden layer, the first linear layer and the second linear layer respectively;
the input layer is used for transmitting the registered first region of interest and the second region of interest to a hidden layer;
the hidden layer is used for extracting the characteristics of the registered first region of interest and the registered second region of interest to obtain initial characteristic information;
the global average pooling layer is connected with the hidden layer and is used for carrying out feature selection on the initial feature information to obtain optimal feature information;
the first linear layer is connected with the global average pooling layer and is used for classifying the optimal feature information to obtain a first classification result; the first classification result is no perivascular space or perivascular space;
the second linear layer is connected with the global average pooling layer and is used for classifying the optimal feature information to obtain a second classification result; the second classification result is no perivascular space, 1-10 perivascular spaces, 11-20 perivascular spaces, 21-40 perivascular spaces, or greater than 40 perivascular spaces.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210305047.3A CN114723937A (en) | 2022-03-25 | 2022-03-25 | Method and system for classifying blood vessel surrounding gaps based on nuclear magnetic resonance image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210305047.3A CN114723937A (en) | 2022-03-25 | 2022-03-25 | Method and system for classifying blood vessel surrounding gaps based on nuclear magnetic resonance image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114723937A true CN114723937A (en) | 2022-07-08 |
Family
ID=82239288
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210305047.3A Pending CN114723937A (en) | 2022-03-25 | 2022-03-25 | Method and system for classifying blood vessel surrounding gaps based on nuclear magnetic resonance image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114723937A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116309355A (en) * | 2023-01-31 | 2023-06-23 | 优脑银河(浙江)科技有限公司 | Redundant information processing method and device for brain image and storage medium |
-
2022
- 2022-03-25 CN CN202210305047.3A patent/CN114723937A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116309355A (en) * | 2023-01-31 | 2023-06-23 | 优脑银河(浙江)科技有限公司 | Redundant information processing method and device for brain image and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109035263B (en) | Automatic brain tumor image segmentation method based on convolutional neural network | |
| CN109522894B (en) | Method for detecting dynamic covariation of fMRI brain network | |
| CN113616184B (en) | Brain network modeling and individual prediction method based on multi-mode magnetic resonance image | |
| CN107016395B (en) | Identification system for sparsely expressed primary brain lymphomas and glioblastomas | |
| CN111192245A (en) | Brain tumor segmentation network and method based on U-Net network | |
| CN112446891B (en) | Medical image segmentation method based on U-Net network brain glioma | |
| CN112767417B (en) | Multi-modal image segmentation method based on cascaded U-Net network | |
| CN107766874B (en) | Measuring method and measuring system for ultrasonic volume biological parameters | |
| CN107644420A (en) | Blood-vessel image dividing method, MRI system based on central line pick-up | |
| CN113420826B (en) | Liver focus image processing system and image processing method | |
| CN115496771A (en) | Brain tumor segmentation method based on brain three-dimensional MRI image design | |
| CN112508953B (en) | Meningioma rapid segmentation qualitative method based on deep neural network | |
| CN114926477A (en) | Brain tumor multi-modal MRI (magnetic resonance imaging) image segmentation method based on deep learning | |
| CN111523617B (en) | Epilepsy detection system based on white matter fusion characteristic diagram and residual error attention network | |
| CN115147600A (en) | GBM multi-mode MR image segmentation method based on classifier weight converter | |
| CN114581701A (en) | Method for generating dynamic enhanced image features by weighting image features through T2 | |
| CN114266939A (en) | Brain extraction method based on ResTLU-Net model | |
| CN113989551A (en) | Alzheimer disease classification method based on improved ResNet network | |
| CN109191422A (en) | Cerebral arterial thrombosis early detection system and method based on conventional CT image | |
| CN114723937A (en) | Method and system for classifying blood vessel surrounding gaps based on nuclear magnetic resonance image | |
| CN110458842A (en) | Brain tumor dividing method based on the three-dimensional intensive connection network of binary channels | |
| Suwalska et al. | CMB-HUNT: automatic detection of cerebral microbleeds using a deep neural network | |
| CN114119626A (en) | Cerebrovascular image segmentation method based on statistical model and multi-scale filtering | |
| CN116503593A (en) | Retina OCT image hydrops segmentation method based on deep learning | |
| Wang et al. | Adaptive Weights Integrated Convolutional Neural Network for Alzheimer's Disease Diagnosis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |