CN111145200B - Blood vessel center line tracking method combining convolutional neural network and cyclic neural network - Google Patents
Blood vessel center line tracking method combining convolutional neural network and cyclic neural network Download PDFInfo
- Publication number
- CN111145200B CN111145200B CN201911108172.XA CN201911108172A CN111145200B CN 111145200 B CN111145200 B CN 111145200B CN 201911108172 A CN201911108172 A CN 201911108172A CN 111145200 B CN111145200 B CN 111145200B
- Authority
- CN
- China
- Prior art keywords
- sequence
- neural network
- blood vessel
- tracking
- image block
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/04—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/143—Segmentation; Edge detection involving probabilistic approaches, e.g. Markov random field [MRF] modelling
-
- 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
Abstract
The invention belongs to the technical field of image processing and machine learning, and discloses a blood vessel center line tracking method combining a convolutional neural network and a cyclic neural network, wherein in an original image, a center line is sampled at equal intervals, image blocks are extracted and form a sequence, and the direction category of the sequence is given; selecting non-central line points, extracting image blocks and generating an expansion sequence, and giving direction categories of the expansion sequence; training a convolution-circulation neural network by taking the sequence and the direction category as inputs; segmenting the center line of the blood vessel in the test image and fitting to obtain an initial sequence for tracking; starting from the start sequence, tracking is achieved by predicting the direction through a convolutional-recurrent neural network. According to the method, the starting point and the end point of tracking are automatically determined, so that the problem that manual intervention is needed in the traditional method is solved, and the global and local information of the blood vessel is effectively utilized; the automatic tracking of the central line is realized, and the method has the characteristics of no need of pre-segmenting the blood vessel and high accuracy.
Description
Technical Field
The invention belongs to the technical field of image processing and machine learning, and particularly relates to a blood vessel center line tracking method combining a convolutional neural network and a cyclic neural network.
Background
The extraction of the center line of the blood vessel has important significance for the accurate positioning and quantitative analysis of the blood vessel diseases. With the vessel centerline, the physician can more conveniently diagnose. Clinical surgery will produce a tremendous leap from empirical to digital clinics after accurate visualization of the vessel centerline and proper assessment of the morphology of the vessel. After the central line of the blood vessel is extracted, the connection condition of the branches of the blood vessel and the mutual connection condition, the shape of the blood vessel and the like are clearly displayed in front of a doctor, and the auxiliary means can make up for the shortage of part of the experience of the doctor and assist the doctor to accurately predict and judge the cardiovascular disease. Manual labeling by an experienced physician is the most straightforward way to obtain the vessel centerline. However, with the continuous development of medical imaging, the imaging device can acquire images with different sizes, high dimensionality and high resolution, and the number of medical images also shows geometric progression increase, so that the method of manual labeling is time-consuming and labor-consuming. Meanwhile, because the blood vessel structure is complex, the number of background points in the image is far greater than that of target points, and the method for automatically extracting the blood vessel center line has certain limitations all the time. The center line is extracted by calculating the minimum cost path between the manually defined starting point and the manually defined end point, the method has small calculation amount and high speed, but the method needs manual intervention, and the extracted center line is generally poor in smoothness in a local area; the method is a full-automatic method, but is easily affected by vascular diseases and has low accuracy; in recent years, methods based on machine learning are applied to extraction of blood vessel center lines, but the design complexity and application limitation of feature extraction algorithms and the diversity of combination of feature extraction algorithms and classifiers limit the application of traditional machine learning methods in the field.
In summary, the problems of the prior art are as follows:
(1) The existing semi-automatic method for extracting the center line of the blood vessel needs manual intervention;
(2) The traditional machine learning method is limited in application in the field and cannot achieve ideal precision;
the difficulty of solving the technical problems is as follows:
(1) The method for semi-automatically extracting the center line of the blood vessel needs to manually specify the starting point and the end point of the blood vessel, but with the development of medical imaging equipment and the increase of the number of medical images, the method for manually marking wastes time and labor;
(2) The method for extracting the center line of the blood vessel based on machine learning has the problems that due to the complexity of feature selection and the diversity of classifiers, how to select effective features and a proper classifier is a difficult problem;
the significance of solving the technical problems is as follows:
(1) The starting point and the end point of the tracking of the blood vessel center line are automatically determined, so that the problem that manual intervention is needed for center line extraction can be solved, and the speed of processing images is greatly improved;
(2) The uncertainty brought by a feature selection algorithm can be effectively avoided by automatically selecting the useful features, the accuracy of the result is increased, and the complexity of tracking the center line of the blood vessel is reduced.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a blood vessel center line tracking method combining a convolutional neural network and a cyclic neural network.
The invention is realized in such a way that a blood vessel center line tracking method combining a convolutional neural network and a cyclic neural network comprises the following steps: sampling the central line in the original image at equal intervals, extracting image blocks and forming a sequence, and giving the direction category of the sequence; selecting non-center line points, extracting image blocks and generating an expansion sequence, and giving the direction category of the expansion sequence; training a convolution-circulation neural network by taking the sequence and the direction category as input; segmenting the center line of the blood vessel in the test image and fitting to obtain an initial sequence for tracking; starting from the start sequence, tracking is achieved by predicting the direction through a convolutional-recurrent neural network.
Further, the blood vessel centerline tracking method combining the convolutional neural network and the cyclic neural network comprises the following steps:
firstly, extracting a training image block sequence, dividing a blood vessel central line into a plurality of line segments without branches by taking a blood vessel bifurcation point of an original three-dimensional image as a starting point, sampling all the line segments according to a distance R to obtain a discrete central line sampling point, and extracting a three-dimensional image block by taking the sampling point as a center; respectively selecting N adjacent image blocks along two directions of a blood vessel central line from each sampling point, thereby obtaining an image block sequence with the length of N; uniformly dividing the directions of the three-dimensional space into K types according to angles, and giving the direction types according to the angles from each sequence to the adjacent image blocks;
secondly, amplifying the image block sequence samples, randomly selecting T non-central line points at the distance R of each sampling point as expansion points, and extracting a three-dimensional image block by taking the expansion points as centers; for each sequence, replacing the Nth image block of the sequence with an expansion image block and generating an expansion sequence, wherein each sequence corresponds to T expansion sequences, and the direction category of the expansion sequence is given according to the angle from the expansion sequence to an adjacent image block;
step three, training a convolution-circulation neural network, and taking the sequence and the corresponding direction category as input training convolution-circulation neural network; the network is formed by combining a convolutional neural network and a cyclic neural network, the convolutional neural network extracts the characteristics of the image blocks, and the cyclic neural network combines the information of the whole image block sequence; the output of the network is the probability P (D) of the sequence to neighboring point direction category 1 )，P(D 2 )，P(D 3 )...P(D K ) Determining the direction category from the sequence to the adjacent points according to the magnitude of the probability value;
fourthly, determining a tracking starting sequence, and segmenting the center line of the blood vessel in the original test image based on the existing segmentation algorithm; fitting all discrete points with the line segments to obtain L continuous central line segments; sampling each section of central line according to the distance R and selecting N continuous sampling points; extracting image blocks with sampling points as centers to form L image block sequences with the length of N, and recording the image block sequences as S 0 ,S 1 ,S 2 ...S L ；
Fifth step, tracking of vessel centerline, from sequence S 0 Initially, the direction D of the sequence to the neighboring point is predicted by a convolutional-recurrent neural network, and the neighboring point M is determined from the direction D and the distance R 0 The position of (a); at point M 0 Extracting a three-dimensional image block for the center, adding a sequence, deleting the image block at the initial position of the sequence, and keeping the length of the sequence to be N;updating the sequence continuously and determining the sequence neighbors M through the network 1 ,M 2 ,M 3 ...M x When the entropy value H corresponding to the adjacent point is larger than the threshold value U, the tracking is stopped; for sequence S 1 ,S 2 ,S 3 ...S L Performing the same processing, judging whether the obtained blood vessel central line covers the sequence when the sequence is tracked, and if so, not processing the sequence; after all sequence tracking processes, the complete vessel centerline is obtained.
Further, the first step is to sample the central line in the original image at equal intervals, extract image blocks and form a sequence, and the direction category of the sequence is given, and the method comprises the following steps:
(1) Dividing a blood vessel central line into a plurality of line segments without branches by taking a blood vessel bifurcation point of an original three-dimensional image as a starting point;
(2) Sampling all line segments according to the distance R to obtain discrete central line sampling points, and extracting a three-dimensional image block by taking the sampling points as the center;
(3) Respectively selecting N adjacent image blocks along two directions of a blood vessel central line from each sampling point, thereby obtaining an image block sequence with the length of N;
(4) The directions of the three-dimensional space are evenly divided into K types according to angles, and the direction types are given according to the angles from each sequence to the adjacent image blocks.
Further, the second step selects non-center line points, extracts image blocks and generates an extended sequence, gives the direction category of the extended sequence, and is performed according to the following steps:
(1) Randomly selecting T non-center line points at the distance R of each sampling point as expansion points;
(2) Extracting a three-dimensional image block by taking an expansion point as a center;
(3) And for each sequence, replacing the Nth image block of the sequence with the expanded image block and generating expanded sequences, wherein each sequence corresponds to T expanded sequences, and the direction category of each expanded sequence is given according to the angle from the expanded sequence to the adjacent image block.
Further, the third step trains the convolutional-cyclic neural network by taking the sequence and the direction category as input, and comprises the following steps:
(1) Training a convolution-circulation neural network by taking the sequence and the corresponding direction category as input;
(2) Output as probability P (D) of sequence to neighbor direction class 1 )，P(D 2 )，P(D 3 )...P(D K )。
Further, the fourth step is to segment the central line in the test image and fit the central line to obtain an initial sequence for tracking, and the fourth step is performed according to the following steps:
(1) Based on the existing segmentation algorithm, segmenting the center line of the blood vessel in the original test image;
(2) Fitting all the discrete central line points and the line segments to obtain L continuous central line segments;
(3) Sampling each section of central line according to the distance R and selecting N continuous sampling points;
(4) Extracting three-dimensional image blocks by taking sampling points as centers to form L image block sequences with the length of N, and recording the L image block sequences as S 0 ,S 1 ,S 2 ...S L 。
Further, the fifth step starts from the start sequence, and the direction is predicted to realize tracking through a convolution-circulation neural network, and the tracking is carried out according to the following steps:
(1) From the sequence S 0 Initially, the direction D of the sequence to the adjacent point is predicted by a convolution-cyclic neural network, and the adjacent point M is determined according to the direction D and the distance R 0 The position of (a);
(2) At point M 0 Extracting a three-dimensional image block for the center, adding a sequence, deleting the image block at the initial position of the sequence, and keeping the length of the sequence to be N;
(3) Updating the sequence continuously and determining the sequence neighbors M through the network 1 ,M 2 ,M 3 ...M X When the entropy H corresponding to the neighboring point is greater than the threshold U, the tracking is stopped, and the formula for calculating the entropy is:
wherein D is a direction category, K is the number of direction categories, P is the probability of belonging to a certain direction category, and H is an entropy value;
(4) For the sequence S 1 ,S 2 ,S 3 ...S L Performing the same processing, judging whether the obtained blood vessel central line covers the sequence when the sequence is tracked, and if so, not processing the sequence;
(5) After all sequence tracking processes, the complete vessel centerline is obtained.
Another object of the present invention is to provide an image segmentation system using the blood vessel centerline tracking method combining the convolutional neural network and the cyclic neural network.
Another object of the present invention is to provide an image detection system using the blood vessel centerline tracking method combining the convolutional neural network and the cyclic neural network.
Another object of the present invention is to provide an information data processing terminal using the blood vessel centerline tracking method combining the convolutional neural network and the cyclic neural network.
In summary, the advantages and positive effects of the invention are: the invention solves the problems of manual intervention, low speed and low accuracy rate in the conventional blood vessel central line extraction. The starting point and the terminal point of the blood vessel central line tracking are automatically determined, so that the problem that manual intervention is needed in the traditional method is solved; the structure of the convolution-circulation neural network is used, so that the global information and the local information of the blood vessel are effectively utilized; the robustness and generalization capability of the model are improved by the expansion of the data set; the automatic tracking of the central line is realized, and the method has the characteristics of no need of pre-segmenting the blood vessel and high accuracy.
Drawings
Fig. 1 is a flowchart of a vessel centerline tracking method combining a convolutional neural network and a recurrent neural network according to an embodiment of the present invention.
Fig. 2 is a schematic structural diagram of a convolutional-cyclic neural network provided in an embodiment of the present invention.
Fig. 3 is a flowchart for predicting and tracking the direction of the centerline of a blood vessel through a convolution-loop network according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
In view of the problems in the prior art, the present invention provides a method for tracking a centerline of a blood vessel by combining a convolutional neural network and a cyclic neural network, which is described in detail below with reference to the accompanying drawings.
As shown in fig. 1, a blood vessel centerline tracking method combining a convolutional neural network and a cyclic neural network according to an embodiment of the present invention includes the following steps:
s101: sampling the central line in the original image at equal intervals, extracting image blocks and forming a sequence, and giving the direction category of the sequence;
s102: selecting non-central line points, extracting image blocks and generating an expansion sequence, and giving direction categories of the expansion sequence;
s103: training a convolution-circulation neural network by taking the sequence and the direction category as input;
s104: segmenting the center line of the blood vessel in the test image and fitting to obtain an initial sequence for tracking;
s105: starting from the start sequence, tracking is achieved by predicting the direction through a convolutional-recurrent neural network.
The technical solution of the present invention is further described below with reference to the accompanying drawings.
The blood vessel center line tracking method combining the convolutional neural network and the cyclic neural network provided by the embodiment of the invention specifically comprises the following steps:
(1) In an original image, sampling a central line at equal intervals, extracting image blocks and forming a sequence, giving direction categories of the sequence, and comprising the following specific processes:
(1a) The data used in this example was a cardiac CTA image, with the coronary centerline tracked using a convolution-recurrent neural network;
(1b) 48 cardiac CTA images were taken, each having a size of 512 × 512 × C and a number of centerline points of about 800. And resampling all the images to ensure that the image resolution is consistent. 42 images are used for making a training set, and 6 images are used for testing;
(1c) Dividing the coronary artery central line into 17 segments according to medical standards;
(1b) Sampling all line segments according to the distance of 1.5mm to obtain discrete central line sampling points, and taking the sampling points as the center to extract a three-dimensional image block with the size of 15 multiplied by 15;
(1c) Respectively selecting 5 adjacent image blocks along two directions of a blood vessel central line from each sampling point to obtain an image block sequence with the length of 5;
(1d) Uniformly dividing the directions of the three-dimensional space into 20 types according to angles, and giving the direction types according to the angles from each sequence to the adjacent image blocks;
(2) Selecting non-central line points, extracting image blocks and generating an expansion sequence, and giving direction categories of the expansion sequence, wherein the specific process comprises the following steps:
(2a) Randomly selecting 3 non-center line points at the position of 1.5mm away from each sampling point as expansion points;
(2b) Extracting a three-dimensional image block with the size of 15 multiplied by 15 by taking an expansion point as a center;
(2c) For each sequence, replacing the 5 th image block in the sequence with an expansion image block and generating expansion sequences, wherein each sequence corresponds to 3 expansion sequences, and the direction category of each expansion sequence is given according to the angle from the expansion sequence to the adjacent image block;
(3) Training the convolutional-cyclic neural network by taking the sequence and the direction category as input, as shown in fig. 2, the specific process is as follows:
(3a) Training a convolution-circulation neural network by taking the sequence and the corresponding direction category as input;
(3b) Output as probability P (D) of sequence to neighbor direction class 1 )，P(D 2 )，P(D 3 )...P(D 20 )；
(4) Segmenting the center line of the blood vessel in the test image and fitting to obtain a starting sequence for tracking, wherein the specific process is as follows:
(4a) The embodiment uses a segmentation network U-Net to segment the center line of the blood vessel in an original test image;
(4b) Fitting all the discrete central line points and the line segments to obtain 60 continuous central line segments;
(4c) Sampling each section of central line according to the distance of 1.5mm and selecting 5 continuous sampling points;
(4d) Extracting three-dimensional image blocks of 15 × 15 × 15 size with sampling points as centers to form 60 image block sequences with length of 5, and recording the sequences as S 0 ,S 1 ,S 2 ...S 59 ；
(5) Starting from a starting sequence, predicting the direction through a convolution-circulation neural network to realize tracking, and the specific process is as follows:
(5a) From the sequence S 0 Initially, the direction D of the sequence to neighboring points is predicted by a convolutional-recurrent neural network, according to which the sequence S is interpolated 0 Determining adjacent points M at a distance of 1.5mm 0 The position of (a);
(5b) At point M 0 Extracting a three-dimensional image block for the center, adding a sequence, deleting the image block at the initial position of the sequence, and keeping the length of the sequence to be 5;
(5c) Updating the sequence continuously and determining the sequence neighbors M through the network 1 ,M 2 ,M 3 ...M X When the entropy H corresponding to the neighboring point is greater than the threshold 3.0, the tracking is stopped, and the formula for calculating the entropy is:
wherein D is a direction category, K is the number of direction categories, P is the probability of belonging to a certain direction category, and H is an entropy value;
(5d) For the sequence S 1 ,S 2 ,S 3 ...S 59 Performing the same processing, judging whether the obtained blood vessel central line covers the sequence when the sequence is tracked, and if so, not processing the sequence;
(5e) After all sequence tracking processes, the complete vessel centerline is obtained.
The application effect of the present invention will be described in detail with reference to specific application examples.
Evaluation examples evaluation criteria OV (average total overlap), OF (overlap incomplete first error), OT (clinical dependent overlap), AI (inner accuracy) OF the method proposed in the present invention were respectively defined as follows:
OV: tracking the overlapping degree of the obtained coronary artery central line and the reference central line;
OF: the overlapping degree between the obtained central line and the reference central line when a first error is generated in the tracking process;
OT: the overlapping degree of the central line part which has important diagnostic value clinically and the corresponding reference central line in the tracked coronary artery central lines;
AI: tracking the average distance between the obtained coronary artery central line and the reference central line;
the OV, OF, OT indices are all between [0,1], and closer to 1 indicates better centerline tracking results. The AI index is in millimeters, with smaller values indicating greater accuracy of tracking. In 6 test images OV were all between [0.87,0.94], OF were all between [0.65,0.75], OT were all between [0.91,0.97], and AI were all between [0.2,0.4 ].
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (10)
1. A blood vessel centerline tracking method combining a convolutional neural network and a cyclic neural network, the blood vessel centerline tracking method combining the convolutional neural network and the cyclic neural network comprising: sampling the central line in the original image at equal intervals, extracting image blocks and forming a sequence, and giving the direction category of the sequence; selecting non-central line points, extracting image blocks and generating an expansion sequence, and giving direction categories of the expansion sequence; training a convolution-circulation neural network by taking the sequence and the direction category as input; segmenting the center line of the blood vessel in the test image and fitting to obtain an initial sequence for tracking; starting from the start sequence, tracking is achieved by predicting the direction through a convolutional-recurrent neural network.
2. The method for vessel centerline tracking in combination with a convolutional neural network and a cyclic neural network as claimed in claim 1, wherein the method for vessel centerline tracking in combination with a convolutional neural network and a cyclic neural network comprises the steps of:
firstly, training image block sequence extraction, dividing a blood vessel central line into a plurality of line segments without branches by taking a blood vessel bifurcation point of an original three-dimensional image as a starting point, sampling all the line segments according to a distance R to obtain a discrete central line sampling point, and extracting a three-dimensional image block by taking the sampling point as a center; respectively selecting N adjacent image blocks along two directions of a blood vessel central line from each sampling point, thereby obtaining an image block sequence with the length of N; uniformly dividing the directions of the three-dimensional space into K types according to angles, and giving the direction types according to the angles from each sequence to the adjacent image blocks;
secondly, amplifying the image block sequence samples, randomly selecting T non-central line points at the distance R of each sampling point as expansion points, and extracting a three-dimensional image block by taking the expansion points as centers; for each sequence, replacing the Nth image block of the sequence with an expansion image block to generate expansion sequences, wherein each sequence corresponds to T expansion sequences, and the direction category of the expansion sequences is given according to the angle from the expansion sequence to the adjacent image block;
step three, training a convolution-circulation neural network, and taking the sequence and the corresponding direction category as input training convolution-circulation neural network; the network is formed by combining a convolutional neural network and a cyclic neural network, the convolutional neural network extracts the characteristics of the image blocks, and the cyclic neural network combines the information of the whole image block sequence; the output of the network is the probability P (D) of the sequence to neighboring point direction category 1 )，P(D 2 )，P(D 3 )...P(D K ) Determining the direction category from the sequence to the adjacent points according to the magnitude of the probability value;
the fourth step, tracing the start sequenceDetermining columns, namely segmenting the center line of a blood vessel in an original test image based on the existing segmentation algorithm; fitting all discrete points with the line segments to obtain L continuous central line segments; sampling each section of central line according to the distance R and selecting N continuous sampling points; extracting image blocks by taking sampling points as centers to form L image block sequences with the length of N, and recording the L image block sequences as S 0 ,S 1 ,S 2 …S L ；
Fifth step, tracking of vessel centerline, from sequence S 0 Initially, the direction D of the sequence to the adjacent point is predicted by a convolution-cyclic neural network, and the adjacent point M is determined according to the direction D and the distance R 0 The position of (a); at point M 0 Extracting a three-dimensional image block for the center, adding a sequence, deleting the image block at the initial position of the sequence, and keeping the length of the sequence to be N; updating the sequence continuously and determining the sequence neighbors M through the network 1 ,M 2 ,M 3 …M x When the entropy value H corresponding to the neighboring point is greater than the threshold U, the tracking is stopped; for the sequence S 1 ,S 2, S 3 …S L Performing the same processing, judging whether the obtained blood vessel central line covers the sequence when the sequence is tracked, and if so, not processing the sequence; after all sequence tracking processes, the complete vessel centerline is obtained.
3. The method for tracking the center line of the blood vessel by combining the convolutional neural network and the cyclic neural network as claimed in claim 2, wherein the first step is to sample the center line equidistantly in the original image in the training image, extract the image blocks and form a sequence, and the direction classes of the given sequence are performed according to the following steps:
(1) Dividing a blood vessel central line into a plurality of line segments without branches by taking a blood vessel bifurcation point of an original three-dimensional image as a starting point;
(2) Sampling all line segments according to the distance R to obtain discrete central line sampling points, and extracting a three-dimensional image block by taking the sampling points as centers;
(3) Respectively selecting N adjacent image blocks along two directions of a blood vessel central line from each sampling point, thereby obtaining an image block sequence with the length of N;
(4) The directions of the three-dimensional space are evenly divided into K types according to angles, and the direction types are given according to the angles from each sequence to the adjacent image blocks.
4. The method for tracking the centerline of a blood vessel in combination with a convolutional neural network and a cyclic neural network as claimed in claim 2, wherein said second step selects non-centerline points, extracts image blocks and generates extended sequences, and gives the direction category of the extended sequences, the following steps are performed:
(1) Randomly selecting T non-center line points at the distance R of each sampling point as expansion points;
(2) Extracting a three-dimensional image block by taking the expansion point as a center;
(3) For each sequence, the nth image block of the sequence is replaced by the extended image block to generate extended sequences, each sequence corresponds to T extended sequences, and the direction category of the extended sequences is given according to the angle from the extended sequences to the adjacent image block.
5. The method for tracking the centerline of a blood vessel by combining a convolutional neural network and a recurrent neural network as claimed in claim 2, wherein said third step trains the convolutional-recurrent neural network with the sequence and direction classes as input, and is performed as follows:
(1) Training a convolution-circulation neural network by taking the sequence and the corresponding direction category as input;
(2) Output as probability P (D) of sequence to neighbor direction class 1 )，P(D 2 )，P(D 3 )…P(D K )。
6. The method for tracking the centerline of a blood vessel by combining a convolutional neural network and a cyclic neural network as claimed in claim 2, wherein said fourth step is to segment the centerline in the test image and fit it to obtain the starting sequence for tracking, and the following steps are performed:
(1) Based on the existing segmentation algorithm, segmenting the center line of the blood vessel in the original test image;
(2) Fitting all the discrete central line points and the line segments to obtain L continuous central line segments;
(3) Sampling each section of central line according to the distance R and selecting N continuous sampling points;
(4) Extracting three-dimensional image blocks by taking sampling points as centers to form L image block sequences with the length of N, and recording the sequences as S 0 ,S 1 ,S 2 …S L 。
7. The method for tracking the centerline of a blood vessel by combining a convolutional neural network and a recurrent neural network as claimed in claim 2, wherein the fifth step starts from the start sequence and realizes tracking by predicting the direction through the convolutional-recurrent neural network, and comprises the following steps:
(1) From the sequence S 0 Initially, the direction D of the sequence to the neighboring point is predicted by a convolutional-recurrent neural network, and the neighboring point M is determined from the direction D and the distance R 0 The position of (a);
(2) At point M 0 Extracting a three-dimensional image block for the center, adding a sequence, deleting the image block at the initial position of the sequence, and keeping the length of the sequence to be N;
(3) Updating the sequence continuously and determining the sequence neighbors M through the network 1 ,M 2 ,M 3 …M X When the entropy H corresponding to the neighboring point is greater than the threshold U, the tracking is stopped, and the formula for calculating the entropy is:
wherein D is a direction category, K is the number of direction categories, P is the probability of belonging to a certain direction category, and H is an entropy value;
(4) For the sequence S 1 ,S 2, S 3 …S L Performing the same processing, judging whether the obtained blood vessel central line covers the sequence when the sequence is tracked, and if so, not processing the sequence;
(5) After all sequence tracking processes, the complete vessel centerline is obtained.
8. An image segmentation system using the method of tracking the centerline of a blood vessel according to any one of claims 1 to 7 in combination with a convolutional neural network and a recurrent neural network.
9. An image detection system using the blood vessel centerline tracking method combining the convolutional neural network and the cyclic neural network as set forth in any one of claims 1 to 7.
10. An information data processing terminal using the blood vessel center line tracking method combining the convolutional neural network and the cyclic neural network according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911108172.XA CN111145200B (en) | 2019-11-13 | 2019-11-13 | Blood vessel center line tracking method combining convolutional neural network and cyclic neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911108172.XA CN111145200B (en) | 2019-11-13 | 2019-11-13 | Blood vessel center line tracking method combining convolutional neural network and cyclic neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111145200A CN111145200A (en) | 2020-05-12 |
| CN111145200B true CN111145200B (en) | 2023-04-18 |
Family
ID=70517203
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911108172.XA Active CN111145200B (en) | 2019-11-13 | 2019-11-13 | Blood vessel center line tracking method combining convolutional neural network and cyclic neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111145200B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111640124B (en) * | 2020-05-25 | 2023-06-02 | 浙江同花顺智能科技有限公司 | Blood vessel extraction method, device, equipment and storage medium |
| CN111768411B (en) * | 2020-05-29 | 2023-12-08 | 深圳睿心智能医疗科技有限公司 | Coronary centerline extraction method, device, computer equipment and storage medium |
| CN111724413B (en) * | 2020-06-19 | 2021-09-28 | 上海杏脉信息科技有限公司 | Blood vessel tracking method, coronary image segmentation method, medium and electronic equipment |
| CN113222964B (en) * | 2021-05-27 | 2021-11-12 | 推想医疗科技股份有限公司 | Method and device for generating coronary artery central line extraction model |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10762637B2 (en) * | 2017-10-27 | 2020-09-01 | Siemens Healthcare Gmbh | Vascular segmentation using fully convolutional and recurrent neural networks |
| CN108776779B (en) * | 2018-05-25 | 2022-09-23 | 西安电子科技大学 | Convolutional-circulation-network-based SAR sequence image target identification method |
| CN108986124A (en) * | 2018-06-20 | 2018-12-11 | 天津大学 | In conjunction with Analysis On Multi-scale Features convolutional neural networks retinal vascular images dividing method |
-
2019
- 2019-11-13 CN CN201911108172.XA patent/CN111145200B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111145200A (en) | 2020-05-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111145200B (en) | Blood vessel center line tracking method combining convolutional neural network and cyclic neural network | |
| CN107016665B (en) | CT pulmonary nodule detection method based on deep convolutional neural network | |
| CN108464840B (en) | Automatic detection method and system for breast lumps | |
| CN111985536B (en) | Based on weak supervised learning gastroscopic pathology image Classification method | |
| CN111553892B (en) | Lung nodule segmentation calculation method, device and system based on deep learning | |
| CN107122735B (en) | Multi-target tracking method based on deep learning and conditional random field | |
| CN111598875A (en) | Method, system and device for building thyroid nodule automatic detection model | |
| JPH0638274B2 (en) | Image recognition apparatus and image recognition method | |
| CN110246579B (en) | Pathological diagnosis method and device | |
| CN114782307A (en) | Enhanced CT image colorectal cancer staging auxiliary diagnosis system based on deep learning | |
| CN114581375A (en) | Method, device and storage medium for automatically detecting focus of wireless capsule endoscope | |
| CN114037663A (en) | Blood vessel segmentation method, device and computer readable medium | |
| CN112102332A (en) | Cancer WSI segmentation method based on local classification neural network | |
| CN110738702B (en) | Three-dimensional ultrasonic image processing method, device, equipment and storage medium | |
| CN114155193B (en) | Blood vessel segmentation method and device based on feature enhancement | |
| CN116958147B (en) | Target area determining method, device and equipment based on depth image characteristics | |
| CN111401102A (en) | Deep learning model training method and device, electronic equipment and storage medium | |
| CN112837325A (en) | Medical image processing method, device, electronic equipment and medium | |
| CN113222051A (en) | Image labeling method based on small intestine focus characteristics | |
| CN115880266B (en) | Intestinal polyp detection system and method based on deep learning | |
| CN112017161A (en) | Pulmonary nodule detection method and device based on central point regression | |
| CN112801940A (en) | Model evaluation method, device, equipment and medium | |
| CN115661152B (en) | Target development condition analysis method based on model prediction | |
| Durán López et al. | Polyp detection in gastrointestinal images using faster regional convolutional neural network | |
| CN114010227B (en) | Right ventricle characteristic information identification method and device |
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 |