CN111986101A - Cerebrovascular map construction method - Google Patents
Cerebrovascular map construction method Download PDFInfo
- Publication number
- CN111986101A CN111986101A CN202010656252.5A CN202010656252A CN111986101A CN 111986101 A CN111986101 A CN 111986101A CN 202010656252 A CN202010656252 A CN 202010656252A CN 111986101 A CN111986101 A CN 111986101A
- Authority
- CN
- China
- Prior art keywords
- data
- cerebrovascular
- atlas
- patch
- blood vessel
- 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.)
- Granted
Links
- 238000010276 construction Methods 0.000 title abstract description 5
- 210000004204 blood vessel Anatomy 0.000 claims abstract description 67
- 238000000034 method Methods 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 31
- 230000011218 segmentation Effects 0.000 claims abstract description 28
- 230000002490 cerebral effect Effects 0.000 claims abstract description 17
- 230000008569 process Effects 0.000 claims abstract description 12
- 238000002372 labelling Methods 0.000 claims abstract description 7
- 238000012216 screening Methods 0.000 claims abstract description 6
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 12
- 238000013507 mapping Methods 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 6
- 210000003484 anatomy Anatomy 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 5
- 230000008602 contraction Effects 0.000 claims description 4
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 210000000988 bone and bone Anatomy 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 230000002792 vascular Effects 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 3
- 238000003709 image segmentation Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 208000026106 cerebrovascular disease Diseases 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 208000024172 Cardiovascular disease Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000002526 effect on cardiovascular system Effects 0.000 description 1
- 238000011985 exploratory data analysis Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000007917 intracranial administration Methods 0.000 description 1
- 231100000518 lethal Toxicity 0.000 description 1
- 230000001665 lethal effect Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30101—Blood vessel; Artery; Vein; Vascular
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
A cerebrovascular atlas construction method includes preprocessing and labeling MRA data, screening required patch through an algorithm in a data segmentation process, inputting the patch into a modified U-Net network for training and prediction, converting a data set of a prediction result into the same coordinate system through registration, calculating similarity, distributing the similarity to a corresponding characteristic space, and constructing a cerebrovascular atlas. The invention greatly reduces the calculated amount in data segmentation, provides a method for making the blood vessel atlas, solves the subjectivity of manual operation in the traditional method, and realizes rapid, simple, accurate and automatic segmentation of the cerebral vessels.
Description
Technical Field
The invention relates to medical image segmentation, in particular to a method for constructing a cerebrovascular map.
Background
Research data indicate that cerebrovascular disease has become the most lethal disease. Cerebrovascular segmentation is an important topic in medical image analysis, as cerebrovascular is crucial for diagnosis, treatment planning and execution and assessment of clinical outcome in different fields like neurosurgery and cardiovascular and cerebrovascular diseases. The importance of the brain vessels and the specificity of their location pose challenges to the accuracy of their segmentation and the reliability of the results.
Automatic or semi-automatic segmentation methods also require at least one clinician to initially segment or evaluate the segmentation results, even if a large number of manual operations are omitted. With the rise of deep learning research enthusiasm, many researchers use convolutional neural networks for vessel segmentation. With the excellent performance of the U-Net network structure in medical image segmentation in recent years, more and more researches are focused on realizing intracranial vessel segmentation by using the U-shaped structure. However, due to the limitations of the segmentation technology and the characteristics of the medical image, the analysis and recognition of the image are still bottlenecks that restrict the development of the medical image segmentation technology. Both the traditional cerebrovascular segmentation and the current deep learning-based cerebrovascular segmentation have self-deficiency, and the real full-automatic accurate segmentation is difficult to realize. In the current deep learning-based method, the MRA data is generally large, and the direct input into the network seriously increases the amount of calculation, so researchers will cut each MRA data in the data set into small image blocks (patch). Since the volume occupied by the vessel in the voxels is small but widely distributed, a large number of patches containing no or only a small number of vessel voxels will appear in the direct traversal of the segmentation. The input of the above data into the network for training can greatly slow down the convergence speed of the model. Researchers tend to manually screen out these data, but the effort is enormous.
Clustering is the most widely used technique in exploratory data analysis. When dealing with data without class labels in various scientific fields today, researchers have always gained a visual impression of the data by determining the classification of different samples in the data. Spectral clustering is an algorithm evolved from graph theory, and is widely applied to clustering. The main idea is to treat all data as points in space, which can be connected by edges. The edge weight value between two points farther away is lower, while the edge weight value between two points closer is higher. The graph formed by all data points is cut, so that the edge weight sum between different subgraphs after the graph is cut is as low as possible, and the edge weight sum in the subgraph is as high as possible, and the clustering purpose is achieved. And each new data after the atlas is obtained can be used for obtaining a corresponding clustering label by using the atlas, so that automatic segmentation is realized. Methods of atlas clustering have been widely used in knowledge and fiber atlases.
Disclosure of Invention
In order to overcome the defects that the existing cerebral vessel segmentation technology depends on the operation of a medical researcher too much and has strong subjectivity, the deep learning-based vessel segmentation has large workload in the data segmentation patch process, and poor segmentation easily causes great influence on network model training. In order to overcome the defects, the invention provides a method for constructing a cerebrovascular map.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a method of constructing a cerebrovascular map, the method comprising the steps of:
the method comprises the following steps: data preprocessing and label labeling, the process is as follows:
firstly, denoising MRA data, then performing a head bone removing operation by using a BET tool in FSL, labeling label to the processed data, and finally dividing a training set and a test set according to the number of data;
step two: the image data segmentation image block patch comprises the following steps:
selecting patch size and stride to segment the data due to overlarge three-dimensional MRA data; then selecting the patch with more vessel voxel ratios in the patch as main training data to be input into the network;
step three: network model, training and prediction, the process is as follows:
inputting the obtained data set into a network model for training, obtaining the weight after the model is stabilized according to a statistical result, and finally inputting the data set to obtain a prediction result;
step four: and (3) carrying out registration mapping on the cerebrovascular atlas cluster and the new sample, wherein the process is as follows:
based on the obtained cerebrovascular prediction result, a cerebrovascular atlas is constructed, a new MRA vascular data can obtain a segmentation result only through coordinate conversion with the atlas, and the creation of the atlas and the registration of a new sample both need to map a plurality of data sets to a common coordinate system through registration to carry out initial alignment or standardization of the cerebrovascular.
Further, in the second step, the segmenting the image block of the image data includes the following steps:
2.1 selecting proper patch size and stride to obtain the patch
Selecting the size of the patch size according to the size of the original data set, wherein the size can be 32 multiplied by 32, 64 multiplied by 64, 96 multiplied by 96, 128 multiplied by 128, and stride can be 8, 16, 24, 48, and the like, performing heuristic test, and selecting proper patch size and stride according to the upper limit of hardware computing power and the network convergence rate through the evaluation of a model;
2.2 finding the blood vessel density in patch
Performing statistical calculation on all the patches obtained in the last step, calculating the density of blood vessels in each patch, and searching the patches with the highest density as the centers of blood vessel regions;
2.3 Density-based gradient Screen Patch
After obtaining a plurality of centers of the blood vessel region, taking the central point as a starting point, and taking the gradient descending direction of the blood vessel density as a path screening patch, and stopping calculating until meeting a data boundary or reaching a lower threshold of the density or meeting a screening path from other central points; the lower threshold can be selected by calculation according to the resolution capability of the network and the accuracy requirement of data, and the condition that a large number of patches contain no blood vessels or are few in proportion is eliminated as much as possible.
Still further, in the fourth step, the registration mapping of the cerebrovascular atlas cluster and the new sample comprises the following steps:
4.1 rigid and non-rigid registration
Rigid and non-rigid registration is needed when a cerebrovascular atlas is generated, and based on a complex cerebrovascular tree structure, entropy-based registration is performed in a multi-scale mode: performing affine transformation, performing non-rigid b-spline transformation, and aligning blood vessels from a plurality of samples; converting each input blood vessel into a representation with a fixed length, and representing each branch of each blood vessel through an end point, a middle point and two middle points;
4.2 cerebrovascular clustering
In the feature space, all blood vessels calculate feature vectors through the similarity matrix, and represent the feature vectors as points in the feature space, after vessel registration, each vessel is converted to a point in a feature space, clustering is performed to automatically generate a cerebrovascular atlas, the feature space represents different vessels according to the similarity between vessels, classifying and naming the blood vessels according to the corresponding anatomical structures of different blood vessels, assigning each blood vessel from each sample to the nearest cerebral blood vessel map in the feature space, if the similarity of the blood vessels exceeds 2 standard deviations relative to the average similarity of the blood vessels, judging the blood vessel as an abnormal blood vessel, removing the abnormal blood vessel from the map, comparing the result of the first clustering with the original data MRA and the anatomical structure of the cerebral vessels in order to more accurately ensure the accuracy of the cerebral vessel map, and carrying out the second clustering to finally obtain the cerebral vessel map;
4.3 New sample registration mapping
After the cerebrovascular atlas is established, the cerebrovascular atlas is used for automatically segmenting new sample blood vessels, each new blood vessel sample is represented in a characteristic space which is initially clustered, and then a clustering label is distributed according to the nearest clustering center of mass; firstly, the coordinate transformation of a new sample is carried out and the new sample is registered on a blood vessel atlas, and then atlas mapping is carried out according to the characteristic vector of the new sample to obtain the cerebral vessels of the new sample.
Further, in the third step, the network model, training and prediction comprises the following steps:
3.1 network model
The method adopts a hot U-Net structure in the field of medical image processing, and consists of two parts of encoding and decoding. Two 3 x 3 convolutional layers are applied repeatedly on the punctured path, i.e., the left coding path, and a Batch Normalization layer follows the convolutional layer to accelerate network convergence. Then, a ReLU activation function is added, and finally, a down-sampling operation is performed with a 2 × 2 × 2 maximum pooling layer with a step size of 2. After the downsampling is finished, the convolution layer is used for doubling the number of channels, and the expression capability of the model is enhanced. The upsampling result is connected with the corresponding feature map of the slave contraction path, and image integrity information is reserved through padding operation. Two further convolutions of 3 x 3, each convolution followed by a ReLU active layer; finally, each 64-component feature vector is mapped to the required number of classes using a 1 × 1 × 1 convolution and then normalized using the softmax activation function, with 0 and 1 representing the probability of each pixel being background or cerebrovascular;
3.2 training and prediction
The preprocessed data needs to be input into the network for training. The training optimizer selects Adam. And setting the batch size according to the GPU performance and the video memory size. Setting different learning rates for testing, selecting a proper learning rate according to loss reduction of each epoch, training, obtaining the weight after the model is stable according to a statistical result, and using the weight to input a data set to obtain a prediction result.
The invention has the beneficial effects that: the influence of poor segmentation of deep learning data on the network is avoided, and the construction and automatic segmentation of the cerebrovascular atlas are realized.
Drawings
FIG. 1 is a flow chart of the implementation steps of the method.
Fig. 2 is a flow of slicing data blocks.
Fig. 3 is a diagram of the network architecture used.
FIG. 4 shows the process of constructing a cerebrovascular map
Detailed description of the preferred embodiments
The present invention is further described below.
Referring to fig. 1 to 4, a cerebrovascular map construction method includes the following steps:
the method comprises the following steps: data preprocessing and label labeling, the process is as follows:
firstly, denoising MRA data, then performing a head bone removing operation by using a BET tool in FSL, labeling label to the processed data, and finally dividing a training set and a test set according to the number of data;
step two: the image data segmentation image block patch comprises the following steps:
due to the fact that the three-dimensional MRA data is overlarge, proper patch size and stride need to be selected to segment the data; then, the patch with a large ratio of vessel voxels in the patch is selected as main training data to be input into a network, so that the reliability of the data is ensured, the influence of the patch with a low vessel voxel content on the network is eliminated, and the calculation amount of training is greatly reduced;
further, in the second step, the image data segmentation image block includes the following specific steps:
2.1 selecting proper patch size and stride to obtain the patch
Selecting the size of the patch size according to the size of the original data set, wherein the size can be 32 multiplied by 32, 64 multiplied by 64, 96 multiplied by 96, 128 multiplied by 128, and stride can be 8, 16, 24, 48, and the like, performing heuristic test, and selecting proper patch size and stride according to the upper limit of hardware computing power and the network convergence rate through the evaluation of a model;
2.2 finding the blood vessel density in patch
Performing statistical calculation on all the patches obtained in the last step, calculating the density of blood vessels in each patch, and searching the patches with the highest density as the centers of blood vessel regions;
2.3 Density-based gradient Screen Patch
After obtaining a plurality of centers of the blood vessel region, taking the central point as a starting point, taking the gradient descending direction of the blood vessel density as a path to screen the patch, and stopping computing until meeting a data boundary or reaching a lower threshold limit of the density or meeting screening paths from other central points, wherein the lower threshold limit can be selected by computing according to the resolution capability of the network and the precision requirement of data, so that the condition that a large number of patches contain no blood vessels or have a small proportion is eliminated as much as possible;
step three: network model, training and prediction, the process is as follows:
and inputting the obtained data set into a network model for training, obtaining the weight after the model is stabilized according to the statistical result, and finally inputting the data set to obtain a prediction result.
Further, in the third step, the network model, training and prediction include the following specific steps:
3.1 network model
The method adopts a hot U-Net structure in the field of medical image processing, and consists of two parts of encoding and decoding. Two 3 x 3 convolutional layers are applied repeatedly on the punctured path, i.e., the left coding path, and a Batch Normalization layer follows the convolutional layer to accelerate network convergence. Then, a ReLU activation function is added, and finally, a down-sampling operation is performed with a 2 × 2 × 2 maximum pooling layer with a step size of 2. After the downsampling is finished, the convolution layer is used for doubling the number of channels, and the expression capability of the model is enhanced. The upsampling result is connected with the corresponding feature map of the slave contraction path, and image integrity information is reserved through padding operation. Two further convolutions of 3 x 3, each convolution followed by a ReLU active layer; finally, each 64-component feature vector is mapped to the required number of classes using a 1 × 1 × 1 convolution and then normalized using the softmax activation function, with 0 and 1 representing the probability of each pixel being background or cerebrovascular;
3.2 training and prediction
The preprocessed data needs to be input into the network for training. The training optimizer selects Adam. And setting the batch size according to the GPU performance and the video memory size. Setting different learning rates for testing, selecting a proper learning rate according to loss reduction of each epoch, training, obtaining a weight after the model is stable according to a statistical result, and using the weight to input a data set to obtain a prediction result;
step four: and (3) carrying out registration mapping on the cerebrovascular atlas cluster and the new sample, wherein the process is as follows:
based on the obtained cerebrovascular prediction result, the invention creatively provides a cerebrovascular atlas, a new MRA vascular data can obtain a segmentation result only through coordinate conversion with the atlas, and the creation of the atlas and the registration of a new sample both need to map a plurality of data sets to a common coordinate system through registration to carry out initial alignment or standardization of the cerebrovascular.
In the fourth step, the registration of the cerebrovascular atlas cluster and the new sample comprises the following steps:
4.1 rigid and non-rigid registration
Rigid and non-rigid registration is needed when a cerebrovascular atlas is generated, and based on a complex cerebrovascular tree structure, entropy-based registration is performed in a multi-scale mode: performing affine transformation, performing non-rigid b-spline transformation, and aligning blood vessels from a plurality of samples; converting each input blood vessel into a representation with a fixed length, and representing each branch of each blood vessel through an end point, a middle point and two middle points;
4.2 cerebrovascular clustering
In the feature space, all blood vessels calculate feature vectors through the similarity matrix, and represent the feature vectors as points in the feature space, after vessel registration, each vessel is converted to a point in a feature space, clustering is performed to automatically generate a cerebrovascular atlas, the feature space represents different vessels according to the similarity between vessels, classifying and naming the blood vessels according to the corresponding anatomical structures of different blood vessels, assigning each blood vessel from each sample to the nearest cerebral blood vessel map in the feature space, if the similarity of the blood vessels exceeds 2 standard deviations relative to the average similarity of the blood vessels, judging the blood vessel as an abnormal blood vessel, removing the abnormal blood vessel from the map, comparing the result of the first clustering with the original data MRA and the anatomical structure of the cerebral vessels in order to more accurately ensure the accuracy of the cerebral vessel map, and carrying out the second clustering to finally obtain the cerebral vessel map;
4.3 New sample registration mapping
After the cerebrovascular atlas is built, the cerebrovascular atlas is used for automatically segmenting new sample blood vessels, each new blood vessel sample is represented in a feature space which is initially clustered, and then a clustering label is distributed according to the nearest clustering centroid. Firstly, the coordinate transformation of a new sample is carried out and the new sample is registered on a blood vessel atlas, and then atlas mapping is carried out according to the characteristic vector of the new sample to obtain the cerebral vessels of the new sample.
Claims (4)
1. A method for constructing a cerebrovascular map is characterized by comprising the following steps: the method comprises the following steps:
the method comprises the following steps: data preprocessing and label labeling, the process is as follows:
firstly, denoising MRA data, then performing a head bone removing operation by using a BET tool in FSL, labeling label to the processed data, and finally dividing a training set and a test set according to the number of data;
step two: the image data segmentation image block patch comprises the following steps:
selecting patch size and stride to segment the data due to overlarge three-dimensional MRA data; then selecting the patch with more vessel voxel ratios in the patch as main training data to be input into the network;
step three: network model, training and prediction, the process is as follows:
inputting the obtained data set into a network model for training, obtaining the weight after the model is stabilized according to a statistical result, and finally inputting the data set to obtain a prediction result;
step four: and (3) carrying out registration mapping on the cerebrovascular atlas cluster and the new sample, wherein the process is as follows:
based on the obtained cerebrovascular prediction result, a cerebrovascular atlas is constructed, a new MRA vascular data can obtain a segmentation result only through coordinate conversion with the atlas, and the creation of the atlas and the registration of a new sample both need to map a plurality of data sets to a common coordinate system through registration to carry out initial alignment or standardization of the cerebrovascular.
2. The method for constructing a cerebrovascular atlas as claimed in claim 1, wherein in the second step, the step of segmenting image blocks of the image data comprises the following steps:
2.1 selecting proper patch size and stride to obtain the patch
Selecting the size of the patch size according to the size of the original data set, wherein the size can be 32 multiplied by 32, 64 multiplied by 64, 96 multiplied by 96, 128 multiplied by 128, and stride can be 8, 16, 24, 48, and the like, performing heuristic test, and selecting proper patch size and stride according to the upper limit of hardware computing power and the network convergence rate through the evaluation of a model;
2.2 finding the blood vessel density in patch
Performing statistical calculation on all the patches obtained in the last step, calculating the density of blood vessels in each patch, and searching the patches with the highest density as the centers of blood vessel regions;
2.3 Density-based gradient Screen Patch
After obtaining a plurality of centers of the blood vessel region, taking the central point as a starting point, and taking the gradient descending direction of the blood vessel density as a path screening patch, and stopping calculating until meeting a data boundary or reaching a lower threshold of the density or meeting a screening path from other central points; the lower threshold can be selected by calculation according to the resolution capability of the network and the accuracy requirement of data, and the condition that a large number of patches contain no blood vessels or are few in proportion is eliminated as much as possible.
3. A method as claimed in claim 1 or 2, wherein in step four, the registration mapping of cerebrovascular atlas cluster and new sample comprises the following steps:
4.1 rigid and non-rigid registration
Rigid and non-rigid registration is needed when a cerebrovascular atlas is generated, and based on a complex cerebrovascular tree structure, entropy-based registration is performed in a multi-scale mode: performing affine transformation, performing non-rigid b-spline transformation, and aligning blood vessels from a plurality of samples; converting each input blood vessel into a representation with a fixed length, and representing each branch of each blood vessel through an end point, a middle point and two middle points;
4.2 cerebrovascular clustering
In the feature space, all blood vessels calculate feature vectors through the similarity matrix, and represent the feature vectors as points in the feature space, after vessel registration, each vessel is converted to a point in a feature space, clustering is performed to automatically generate a cerebrovascular atlas, the feature space represents different vessels according to the similarity between vessels, classifying and naming the blood vessels according to the corresponding anatomical structures of different blood vessels, assigning each blood vessel from each sample to the nearest cerebral blood vessel map in the feature space, if the similarity of the blood vessels exceeds 2 standard deviations relative to the average similarity of the blood vessels, judging the blood vessel as an abnormal blood vessel, removing the abnormal blood vessel from the map, comparing the result of the first clustering with the original data MRA and the anatomical structure of the cerebral vessels in order to more accurately ensure the accuracy of the cerebral vessel map, and carrying out the second clustering to finally obtain the cerebral vessel map;
4.3 New sample registration mapping
After the cerebrovascular atlas is established, the cerebrovascular atlas is used for automatically segmenting new sample blood vessels, each new blood vessel sample is represented in a characteristic space which is initially clustered, and then a clustering label is distributed according to the nearest clustering center of mass; firstly, the coordinate transformation of a new sample is carried out and the new sample is registered on a blood vessel atlas, and then atlas mapping is carried out according to the characteristic vector of the new sample to obtain the cerebral vessels of the new sample.
4. A method as claimed in claim 1 or 2, wherein in step three, the network model, training and prediction comprises the following steps:
3.1 network model
The U-Net structure which is popular in the field of medical image processing and consists of two parts of encoding and decoding is adopted, two convolution layers of 3 x 3 are repeatedly applied on a contraction path, namely a left encoding path, and a Batch Normalization layer is followed after the convolution layers for accelerating network convergence; then adding a ReLU activation function, and finally performing down-sampling operation with a 2 multiplied by 2 maximum pooling layer with the step length of 2; after the down-sampling is finished, the convolution layer is used for doubling the number of channels, so that the expression capability of the model is enhanced; connecting the up-sampling result with the corresponding feature map of the slave contraction path, and reserving complete information of the image through padding expansion operation; two further convolutions of 3 x 3, each convolution followed by a ReLU active layer; finally, each 64-component feature vector is mapped to the required number of classes using a 1 × 1 × 1 convolution and then normalized using the softmax activation function, with 0 and 1 representing the probability of each pixel being background or cerebrovascular;
3.2 training and prediction
The preprocessed data needs to be input into a network for training, Adam is selected by a training optimizer, batch size is set according to GPU performance and video memory size, different learning rates are set for testing, a proper learning rate is selected according to loss reduction of each epoch, after training, the weight after the model is stabilized is obtained according to a statistical result, and the weight is used for inputting a data set to obtain a prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010656252.5A CN111986101B (en) | 2020-07-09 | 2020-07-09 | Cerebrovascular map construction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010656252.5A CN111986101B (en) | 2020-07-09 | 2020-07-09 | Cerebrovascular map construction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111986101A true CN111986101A (en) | 2020-11-24 |
| CN111986101B CN111986101B (en) | 2024-06-18 |
Family
ID=73438582
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010656252.5A Active CN111986101B (en) | 2020-07-09 | 2020-07-09 | Cerebrovascular map construction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111986101B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112767350A (en) * | 2021-01-19 | 2021-05-07 | 深圳麦科田生物医疗技术股份有限公司 | Method, device, equipment and storage medium for predicting maximum interval of thromboelastogram |
| CN112907563A (en) * | 2021-03-17 | 2021-06-04 | 复旦大学附属华山医院 | CTA full-automatic collateral circulation scoring method and system based on deep learning |
| CN114170594A (en) * | 2021-12-07 | 2022-03-11 | 奇安信科技集团股份有限公司 | Optical character recognition method, device, electronic equipment and storage medium |
| WO2023070865A1 (en) * | 2021-10-26 | 2023-05-04 | 深圳先进技术研究院 | Vascular reactivity evaluation method and apparatus, electronic device, and storage medium |
| CN116091515A (en) * | 2023-03-13 | 2023-05-09 | 同心智医科技(北京)有限公司 | Method for establishing cerebrovascular segmentation model, and cerebrovascular segmentation method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222759A (en) * | 2019-06-03 | 2019-09-10 | 中国医科大学附属第一医院 | A kind of Plaque Vulnerability in Coronary Artery automatic recognition system |
| CN110533664A (en) * | 2019-07-26 | 2019-12-03 | 浙江工业大学 | A kind of cranial nerve automatic division method based on big-sample data driving |
| CN110942466A (en) * | 2019-11-22 | 2020-03-31 | 北京灵医灵科技有限公司 | Cerebral artery segmentation method and device based on deep learning technology |
| US20200116808A1 (en) * | 2018-10-15 | 2020-04-16 | Zayed University | Cerebrovascular segmentation from mra images |
-
2020
- 2020-07-09 CN CN202010656252.5A patent/CN111986101B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200116808A1 (en) * | 2018-10-15 | 2020-04-16 | Zayed University | Cerebrovascular segmentation from mra images |
| CN110222759A (en) * | 2019-06-03 | 2019-09-10 | 中国医科大学附属第一医院 | A kind of Plaque Vulnerability in Coronary Artery automatic recognition system |
| CN110533664A (en) * | 2019-07-26 | 2019-12-03 | 浙江工业大学 | A kind of cranial nerve automatic division method based on big-sample data driving |
| CN110942466A (en) * | 2019-11-22 | 2020-03-31 | 北京灵医灵科技有限公司 | Cerebral artery segmentation method and device based on deep learning technology |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112767350A (en) * | 2021-01-19 | 2021-05-07 | 深圳麦科田生物医疗技术股份有限公司 | Method, device, equipment and storage medium for predicting maximum interval of thromboelastogram |
| CN112767350B (en) * | 2021-01-19 | 2024-04-26 | 深圳麦科田生物医疗技术股份有限公司 | Method, device, equipment and storage medium for predicting maximum section of thromboelastography |
| CN112907563A (en) * | 2021-03-17 | 2021-06-04 | 复旦大学附属华山医院 | CTA full-automatic collateral circulation scoring method and system based on deep learning |
| CN112907563B (en) * | 2021-03-17 | 2023-12-01 | 复旦大学附属华山医院 | CTA full-automatic collateral circulation scoring method and system based on deep learning |
| WO2023070865A1 (en) * | 2021-10-26 | 2023-05-04 | 深圳先进技术研究院 | Vascular reactivity evaluation method and apparatus, electronic device, and storage medium |
| CN114170594A (en) * | 2021-12-07 | 2022-03-11 | 奇安信科技集团股份有限公司 | Optical character recognition method, device, electronic equipment and storage medium |
| CN116091515A (en) * | 2023-03-13 | 2023-05-09 | 同心智医科技(北京)有限公司 | Method for establishing cerebrovascular segmentation model, and cerebrovascular segmentation method and device |
| CN116091515B (en) * | 2023-03-13 | 2023-07-04 | 同心智医科技(北京)有限公司 | Method for establishing cerebrovascular segmentation model, and cerebrovascular segmentation method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111986101B (en) | 2024-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111986101B (en) | Cerebrovascular map construction method | |
| JP7143008B2 (en) | Medical image detection method and device based on deep learning, electronic device and computer program | |
| CN109409416B (en) | Feature vector dimension reduction method, medical image identification method, device and storage medium | |
| CN113706487B (en) | Multi-organ segmentation method based on self-supervision characteristic small sample learning | |
| CN111931811B (en) | Calculation method based on super-pixel image similarity | |
| CN111340828A (en) | Brain glioma segmentation based on cascaded convolutional neural networks | |
| CN111932529B (en) | Image classification and segmentation method, device and system | |
| CN107730542B (en) | Cone beam computed tomography image correspondence and registration method | |
| US11972571B2 (en) | Method for image segmentation, method for training image segmentation model | |
| CN112365980A (en) | Brain tumor multi-target point auxiliary diagnosis and prospective treatment evolution visualization method and system | |
| WO2024104035A1 (en) | Long short-term memory self-attention model-based three-dimensional medical image segmentation method and system | |
| CN110097921B (en) | Visualized quantitative method and system for glioma internal gene heterogeneity based on image omics | |
| Wu et al. | A supervoxel classification based method for multi-organ segmentation from abdominal ct images | |
| CN112712532B (en) | Bottleneck structure-based multi-scale DC-CUNets liver tumor segmentation method | |
| CN112070781A (en) | Processing method and device of craniocerebral tomography image, storage medium and electronic equipment | |
| CN111524140A (en) | Medical image semantic segmentation method based on CNN and random forest method | |
| CN117058170A (en) | Carotid plaque segmentation method based on double-branch multi-scale cross fusion network | |
| CN115984555A (en) | Coronary artery stenosis identification method based on depth self-encoder composition | |
| CN115311193A (en) | Abnormal brain image segmentation method and system based on double attention mechanism | |
| CN114972745A (en) | Trigeminal nerve automatic segmentation method based on deep network | |
| CN118212501A (en) | Method, system, equipment and medium for detecting multiple focus of lung cancer | |
| CN112164447B (en) | Image processing method, device, equipment and storage medium | |
| CN111209946B (en) | Three-dimensional image processing method, image processing model training method and medium | |
| CN117038080A (en) | Multi-modal deep learning combined high-grade serous ovarian cancer recurrence prediction method | |
| CN116958705A (en) | Medical image classifying system based on graph neural network |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |