CN108615251B - Line sample enhancement method and coronary sample enhancement method applying same - Google Patents
Line sample enhancement method and coronary sample enhancement method applying same Download PDFInfo
- Publication number
- CN108615251B CN108615251B CN201810300148.5A CN201810300148A CN108615251B CN 108615251 B CN108615251 B CN 108615251B CN 201810300148 A CN201810300148 A CN 201810300148A CN 108615251 B CN108615251 B CN 108615251B
- Authority
- CN
- China
- Prior art keywords
- point
- new
- line
- tail
- sample
- 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
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/20—Drawing from basic elements, e.g. lines or circles
- G06T11/203—Drawing of straight lines or curves
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2210/00—Indexing scheme for image generation or computer graphics
- G06T2210/41—Medical
Abstract
Hair brushThe invention discloses a line sample enhancement method, which takes one end of a line as a head point and the other end as a tail point, and respectively gives a moving range threshold value to the head point and the tail point so as to randomly generate a new head point and a new tail point; any point p of the original line point except the head point and the tail point is concentratediIncrements are added to generate a new line sample. The invention also provides a coronary sample data enhancement method, which is based on the line sample enhancement method and comprises the following steps: s1, extracting the central line of each branch blood vessel from the volume data of the coronary artery to obtain a central line point set; s2, generating a new centerline point set based on the extracted centerline point set; and S3, performing 3D deformation on the volume data of the coronary artery based on the new central line point set, and generating a new coronary artery sample. The sample data enhancement method is different from the traditional data enhancement method, not only enriches sample data, but also does not deviate from the real sample condition, and can effectively cover the real sample space.
Description
Technical Field
The invention relates to the field of sample data processing of artificial neural networks, in particular to a line sample enhancement method and a coronary sample enhancement method applying the same.
Background
The automatic coronary reconstruction has important clinical value and practical significance for doctors. Different from the traditional coronary artery reconstruction method, people are more inclined to adopt the artificial neural network to complete various kinds of work in the process of coronary artery reconstruction, such as blood vessel segmentation and the like, due to the advantages of the artificial neural network. But an artificial neural network. For example, deep learning neural networks have high requirements on the number of training samples, and for the case of few sample data, it is necessary to provide an effective sample data enhancement method.
Most of the traditional data enhancement methods are translation, rotation, scaling and the like, and cannot effectively cover the real sample space, so that the enhancement effect is very limited, and overfitting is usually caused. For the elastic deformation method, most of the elastic deformation methods are in 2d or 3d data, the deformation center position is given, the defects are that other points far away from the deformation center position have large deformation, and points near the deformation center position have small deformation, so that the method is not a true sample condition.
Disclosure of Invention
The invention aims to provide a line sample enhancement method capable of effectively covering a real sample space and a coronary sample enhancement method applying the line sample enhancement method.
In order to achieve the purpose, the invention adopts the following technical scheme:
a method of line sample enhancement comprising:
s1, generating a head point and a tail point of a new line:
giving a head point moving range threshold value r1 and a tail point moving range threshold value r2 to one end of the line as a head point and the other end as a tail point so as to randomly generate a new head point and a new tail point;
s2, generating other points in the new line:
any point p except head point and tail point is concentrated on the point of the original lineiAdding an increment D (i-1), wherein D is the increment between two adjacent points, D/(N-1), D is the increment between a new tail point and a new head point, N is the number of point set points of the original line, i belongs to [2, N-1 ]];
And S3, generating a new line based on the new line point set.
Further, in step S2, if d is smaller than the threshold a, increments d × i, d × (i-1), or d × (i-2) are randomly added to two adjacent dots.
The invention also provides a coronary sample data enhancement method, which comprises the following steps:
s1, extracting the central line of each branch blood vessel from the volume data of the coronary artery to obtain a central line point set;
s2, generating a new centerline point set based on the extracted centerline point set:
s21, generating a head point and a tail point of a new center line point set:
taking the end point of the central line connected with the main coronary artery as the head point of the central line, taking the end point at the other end of the central line as the tail point, giving a head point moving range threshold value r1, giving a tail point moving range threshold value r2, wherein r1 is less than r2, generating a new head point and a new tail point, and constraining the new tail point by utilizing constraint conditions, wherein the constraint conditions are that the tail point of the central line of each branch blood vessel and a middle region are not intersected;
s22, generating other points in the new center line point set:
any point p except the head point and the tail point is concentrated on the original central line pointiAdding an increment D (i-1), wherein D is the increment between two adjacent points, D is D/(N-1), D is the increment between a new tail point and a head point, N is the number of central line points, i belongs to [2, N-1 ]];
And S3, performing 3D deformation on the volume data of the coronary artery based on the new central line point set, and generating a new coronary artery sample.
Further, in step S2, if d is smaller than the threshold a, increments d × i, d × (i-1), or d × (i-2) are randomly added to two adjacent points.
After adopting the technical scheme, compared with the background technology, the invention has the following advantages:
the invention provides a novel sample data enhancement method, which is different from the traditional data enhancement method, enriches sample data, does not deviate from the real sample condition, can effectively cover the real sample space, and brings considerable enhancement effect.
Drawings
FIG. 1 is a flow chart of a line sample enhancement method according to the present invention;
fig. 2 is a flowchart of a coronary artery sample data enhancement method according to 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 described in further detail below with reference to the accompanying drawings and 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.
Example 1
Referring to fig. 1, the present invention discloses a method for enhancing a line sample, which includes:
s1, generating a head point and a tail point of a new line:
and (3) giving a head point moving range threshold value r1 and a tail point moving range threshold value r2 to one end of the line as a head point and the other end as a tail point so as to randomly generate a new head point and a new tail point.
When the line is a two-dimensional plane line, the thresholds r1 and r2 are circle radii, and when the line is a three-dimensional space line, the thresholds r1 and r2 are sphere radii.
S2, generating other points in the new line:
any point p except head point and tail point is concentrated on the point of the original lineiAdding an increment D (i-1), wherein D is the increment between two adjacent points, D/(N-1), D is the increment between a new tail point and a new head point, N is the number of point set points of the original line, i belongs to [2, N-1 ]]。
Random probability control may also be introduced in step S2 to make the line more variable: if d is less than threshold a, adding increment d i, d (i-1) or d (i-2) to two adjacent points randomly, and adding increment d i, d (i-1) or d (i-2) to two adjacent points randomlyiAnd pi+1The probability of randomness is as follows:
pi,pi+1→pi+d*(i-1),pi+1+d*(i-2);
pi,pi+1→pi+d*(i-1),pi+1+d*(i-2);
pi,pi+1→pi+d*(i-1),pi+1+d*(i);
pi,pi+1→pi+d*(i-2),pi+1+d*(i-1);
pi,pi+1→pi+d*(i-2),pi+1+d*(i);
pi,pi+1→pi+d*(i),pi+1+d*(i-1)。
in this embodiment, the threshold a is 2 pixels, for example: and generating a random number of 0-1, and adjusting the point satisfying that d is smaller than the threshold value a when the number is smaller than 0.5 to ensure that each point is not constant.
And S3, generating a new line based on the new line point set.
Example 2
Referring to fig. 2, a coronary sample data enhancement method includes:
s1, the center line of each branch blood vessel is extracted from the volume data of the coronary artery, and a center line point set is obtained.
In this step, a centerline extraction method based on Lable can be adopted: the method comprises the steps of extracting a center line reference point from volume data by a 3D-thin method, performing spatial sequencing on center line points by using an MST method, and then obtaining a center line and a center line point set by a filtering and smoothing method.
And S2, generating a new centerline point set based on the extracted centerline point set.
S21, generating a head point and a tail point of a new center line point set:
the end point of the central line connected with the main coronary artery is taken as the head point of the central line, the end point of the other end of the central line is taken as the tail point, the head point is connected with the artery, so that a smaller moving range threshold value is given, and the tail point is given with a large moving range threshold value, namely r1 is less than r2, so as to randomly generate the positions of a new head point and a new tail point. In this embodiment, r1 is 3-5 pixels and r2 is 20-30 pixels.
Because a plurality of central lines exist, the tail point and the middle region of each central line do not intersect, a constraint condition is introduced to constrain the tail point, and the constraint condition is that the tail point and the middle region of the central line of each branch blood vessel do not intersect.
The constraint is implemented as follows:
assuming 3 blood vessels A, B and C, blood vessels A, B, and C have original positions;
the random range set of the blood vessel A can reduce the positions of B and C plus Buffer, and the Buffer is a space with the central points of B and C and the sphere expanded by a fixed length;
after the blood vessel A generates a random result, the blood vessel B starts to be random, and the random range set of the blood vessel B can subtract the positions of A, C and buffer. The Buffer is a space with the center points of A and C and the sphere expanded by a fixed length;
after the blood vessel B generates a random result, the blood vessel C starts to be random, and the random range set of the blood vessel C can subtract the positions of A, B + buffer. Buffer is the center point of A and B, and the sphere expands to a space with a fixed length.
S22, generating other points in the new center line point set:
to the original central line point concentrationEach central point p ofiAdding an increment D (i-1) to generate a new central line point set, wherein D is the increment between two adjacent points (including three dimensions of Dx, Dy and Dz), D/(N-1), D is the increment between a new tail point and a new head point (also including three dimensions of Dx, Dy and Dz), N is the number of central line points, i belongs to [2, N-1 ]]。
And S3, performing 3D deformation on the volume data of the coronary artery based on the new central line point set, and generating a new coronary artery sample. This step can be implemented using a nearest neighbor interpolation algorithm.
As described in example 1, if d is smaller than the threshold a in step S2, increments d × i, d × (i-1), or d × (i-2) may be added to two adjacent points randomly to increase the variation of the center line, so as to obtain more coronary sample data.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (2)
1. A coronary sample data enhancement method, comprising:
s1, extracting the central line of each branch blood vessel from the volume data of the coronary artery to obtain a central line point set;
s2, generating a new centerline point set based on the extracted centerline point set;
s3, carrying out 3D deformation on the volume data of the coronary artery based on the new central line point set to generate a new coronary artery sample;
step S2 includes the following substeps:
s21, generating a head point and a tail point of a new center line point set:
taking the end point of the central line connected with the main coronary artery as the head point of the central line, taking the end point at the other end of the central line as the tail point, giving a head point moving range threshold value r1, giving a tail point moving range threshold value r2, wherein r1 is less than r2, generating a new head point and a new tail point, and constraining the new tail point by utilizing constraint conditions, wherein the constraint conditions are that the tail point of the central line of each branch blood vessel and a middle region are not intersected;
s22, generating other points in the new center line point set:
any point P except head and tail points is concentrated on the original central line pointiAdding an increment D (i-1), wherein D is the increment between two adjacent points, D comprises three dimensions of Dx, Dy and Dz, D = D/(N-1), D is the increment between a new tail point and a head point, D comprises three dimensions of Dx, Dy and Dz, N is the number of center line points, i belongs to [2, N-1 ]]。
2. The method of claim 1, wherein the coronary sample data enhancement method comprises: in step S2, if d is smaller than the threshold a, increments d × i, d × (i-1), or d × (i-2) are added to two adjacent points randomly.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810300148.5A CN108615251B (en) | 2018-04-04 | 2018-04-04 | Line sample enhancement method and coronary sample enhancement method applying same |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810300148.5A CN108615251B (en) | 2018-04-04 | 2018-04-04 | Line sample enhancement method and coronary sample enhancement method applying same |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108615251A CN108615251A (en) | 2018-10-02 |
CN108615251B true CN108615251B (en) | 2022-03-29 |
Family
ID=63659576
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810300148.5A Active CN108615251B (en) | 2018-04-04 | 2018-04-04 | Line sample enhancement method and coronary sample enhancement method applying same |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108615251B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109872314B (en) * | 2019-02-20 | 2021-04-16 | 数坤(北京)网络科技有限公司 | Centerline-based optimal segmentation method and device |
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 |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101315665A (en) * | 2008-06-27 | 2008-12-03 | 浙江大学 | Identity recognition method based on three-dimensional nose shape |
CN101526996A (en) * | 2009-02-23 | 2009-09-09 | 华旭 | Method of mouse spontaneous behavior motion monitoring and posture image recognition |
WO2015166503A1 (en) * | 2014-05-01 | 2015-11-05 | Yeda Research And Development Co. Ltd. | Multimodal transcranial brain optical imaging |
-
2018
- 2018-04-04 CN CN201810300148.5A patent/CN108615251B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101315665A (en) * | 2008-06-27 | 2008-12-03 | 浙江大学 | Identity recognition method based on three-dimensional nose shape |
CN101526996A (en) * | 2009-02-23 | 2009-09-09 | 华旭 | Method of mouse spontaneous behavior motion monitoring and posture image recognition |
WO2015166503A1 (en) * | 2014-05-01 | 2015-11-05 | Yeda Research And Development Co. Ltd. | Multimodal transcranial brain optical imaging |
Also Published As
Publication number | Publication date |
---|---|
CN108615251A (en) | 2018-10-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11908244B2 (en) | Human posture detection utilizing posture reference maps | |
US11170558B2 (en) | Automatic rigging of three dimensional characters for animation | |
Zhang et al. | DCSR: Dilated convolutions for single image super-resolution | |
CN116051668B (en) | Training method of diffusion model of draft map and image generation method based on text | |
CN109543548A (en) | A kind of face identification method, device and storage medium | |
CN110084238B (en) | Finger vein image segmentation method and device based on LadderNet network and storage medium | |
CN113140018B (en) | Method for training countermeasure network model, method for establishing word stock, device and equipment | |
CN102663399B (en) | Image local feature extracting method on basis of Hilbert curve and LBP (length between perpendiculars) | |
CN109359527B (en) | Hair region extraction method and system based on neural network | |
CN107590775B (en) | Image super-resolution amplification method using regression tree field | |
CN105335967A (en) | Back-of-hand vein line extraction method based on local maximum between-class variance and mathematical morphology | |
CN108615251B (en) | Line sample enhancement method and coronary sample enhancement method applying same | |
CN110674824A (en) | Finger vein segmentation method and device based on R2U-Net and storage medium | |
CN112686898A (en) | Automatic radiotherapy target area segmentation method based on self-supervision learning | |
CN109711411B (en) | Image segmentation and identification method based on capsule neurons | |
CN110458046A (en) | A kind of human body motion track analysis method extracted based on artis | |
CN112528811A (en) | Behavior recognition method and device | |
CN111985372A (en) | Remote sensing image water body extraction system for deep learning | |
Salem et al. | Semantic image inpainting using self-learning encoder-decoder and adversarial loss | |
CN116682120A (en) | Multilingual mosaic image text recognition method based on deep learning | |
CN114581918A (en) | Text recognition model training method and device | |
Liu et al. | Facial image inpainting using multi-level generative network | |
CN105069767A (en) | Image super-resolution reconstruction method based on representational learning and neighbor constraint embedding | |
CN110992320B (en) | Medical image segmentation network based on double interleaving | |
CN112464908B (en) | Face feature extraction method for attracting local second-order gradient contour mode |
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 |