CN112419324A - Medical image data expansion method based on semi-supervised task driving - Google Patents
Medical image data expansion method based on semi-supervised task driving Download PDFInfo
- Publication number
- CN112419324A CN112419324A CN202011329403.2A CN202011329403A CN112419324A CN 112419324 A CN112419324 A CN 112419324A CN 202011329403 A CN202011329403 A CN 202011329403A CN 112419324 A CN112419324 A CN 112419324A
- Authority
- CN
- China
- Prior art keywords
- data set
- data
- medical image
- expansion
- semi
- 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
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
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4007—Interpolation-based scaling, e.g. bilinear interpolation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- 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
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
Abstract
The invention belongs to the field of image data expansion, and particularly relates to a medical image data expansion method based on semi-supervised task driving, which comprises the following steps: data acquisition: acquiring a related medical image data set, and labeling the category of the medical image data set; data expansion: the data expansion of the medical image is realized by using a task-driven data expansion method; the construction method comprises the following steps: constructing a data expansion method for data set expansion of medical images; data set saving: and combining the expanded data set with the original data set, and storing the data set. The invention solves the problem of obtaining robust segmentation in limited training data setting by using a semi-supervised task-driven data expansion method, and enables the generated enhanced image to improve the segmentation of the medical image. The invention is used for the expansion of medical image data.
Description
Technical Field
The invention belongs to the field of image data expansion, and particularly relates to a medical image data expansion method based on semi-supervised task driving.
Background
The data expansion is a simple technology for expanding a training set based on generation of a synthetic image-label pair, the idea is to convert an image in a mode that a label is kept unchanged, or to define conversion for the image and the label, but the segmentation performance of a medical image cannot be obviously improved by expansion methods such as random affine transformation, random elastic transformation, random contrast transformation and the like, limited labeled data can be learned by a new task-driven data expansion method, and the synthetic data generator can improve the optimization performance for segmentation tasks.
Accurate medical image segmentation is crucial for current clinical applications, but it is difficult to acquire a large number of annotation examples for medical images, resulting in poor segmentation accuracy of medical images. At present, data expansion of medical images is carried out by methods such as random affine transformation, random elastic transformation, random contrast transformation and the like, but experimental results show that data generated by the current method is not true in nature and the effect is not ideal.
Disclosure of Invention
Aiming at the technical problem that the data of the current medical image data expansion is unreal, the invention provides a medical image data expansion method based on semi-supervised task driving, which has high efficiency, high segmentation accuracy and low cost.
In order to solve the technical problems, the invention adopts the technical scheme that:
a medical image data expansion method based on semi-supervised task driving comprises the following steps:
s1, data acquisition: acquiring a related medical image data set, and labeling the category of the medical image data set;
s2, data expansion: the data expansion of the medical image is realized by using a task-driven data expansion method;
s3, the construction method comprises the following steps: constructing a data expansion method for data set expansion of medical images;
s4, data set storage: and combining the expanded data set with the original data set, and storing the data set.
In the data acquisition in S1, the data set is screened and divided by acquiring a related medical image common data set, so as to construct a medical image data set.
In S3, the construction method is used to generate an extended data set, (X)G,YG)=G((XL,YL),z;wG) Wherein G (·, ·; w is aG) Transformation method for data expansion, z being the random component of the transformation, wGFor the transformed parameters, wherein a data expansion method G, by defining two condition generators, constructing shape and intensity variations for the columns of deformation field generators and intensity field generators, respectively, by which method an image transformation of the input image is performed to obtain an expanded data set, wherein the deformation field generator G is trainedvTransformation information for outputting deformation fields, GvHas a transformation parameter of wGvBy rendering the data set image XLAnd a randomly decimated z-vector as input for generating a dense pixel-by-pixel deformation fieldAccording to the generated deformation field v, carrying out bilinear interpolation on the input image and the corresponding label to obtain an extended data set XGvAnd corresponding tag set YGvThe expression mode is as follows:training intensity field generator GIFor outputting additive intensity mask transformations, GIHas a transformation parameter of wGIInputting training data set image XLAnd a randomly extracted unit Gaussian distributed z-vector as input, inputting an additive strength maskThen add Δ I to XLIn (1), obtaining an extended image set XGIAnd corresponding tag set YGIThe expression mode is as follows:data expansion by a deformation field generator and intensity field generator method:
the data combination method in the step S4 includes: combining the expanded data set with the original data set, and setting XLFor training the data set, YLFor a corresponding tag dataset, (X)N,YN)=(XL∪XG,YL∪YG) Said X isG、YGRepresenting the augmented data set and the corresponding tag data set.
Compared with the prior art, the invention has the following beneficial effects:
the invention solves the problem of obtaining robust segmentation in limited training data setting by using a semi-supervised task-driven data expansion method, and enables the generated enhanced image to improve the segmentation of the medical image.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
A semi-supervised task driven medical image data expansion method, as shown in fig. 1, comprising the following steps:
step 1, data acquisition: a related medical image dataset is acquired and its categories are labeled.
Step 2, data expansion: data expansion of medical images is achieved using a task-driven data expansion method.
And step 3, a construction method comprises the following steps: a data expansion method is constructed for data set expansion of medical images.
And 4, data set storage: and combining the expanded data set with the original data set, and storing the data set.
Further, in the data acquisition in the step 1, the data set is screened and divided by acquiring a related medical image public data set, so as to construct a medical image data set.
Further, in step 3, a construction method is used to generate the augmented data set, (X)G,YG)=G((XL,YL),z;wG) Wherein G (·,. cndot.). w is aG) Transformation method for data expansion, z being the random component of the transformation, wGFor the transformed parameters, wherein a data expansion method G, by defining two condition generators, constructing shape and intensity variations for the columns of deformation field generators and intensity field generators, respectively, by which method an image transformation of the input image is performed to obtain an expanded data set, wherein the deformation field generator G is trainedvTransformation information for outputting deformation fields, GvHas a transformation parameter of wGvBy rendering the data set image XLAnd a randomly decimated z-vector as input for generating a dense pixel-by-pixel deformation fieldAccording to the generated deformation field v, carrying out bilinear interpolation on the input image and the corresponding label to obtain an extended data set XGvAnd corresponding tag set YGvThe expression mode is as follows:training intensity field generator GIFor output additive strength mask transformation,GIHas a transformation parameter of wGIInputting training data set image XLAnd a randomly extracted unit Gaussian distributed z-vector as input, inputting an additive strength maskThen add Δ I to XLIn (1), obtaining an extended image set XGIAnd corresponding tag set YGIThe expression mode is as follows:data expansion by a deformation field generator and intensity field generator method:
further, the data combination method in step 4 is as follows: combining the expanded data set with the original data set, and setting XLFor training the data set, YLFor a corresponding tag dataset, (X)N,YN)=(XL∪XG,YL∪YG),XG、YGRepresenting the augmented data set and the corresponding tag data set.
Although only the preferred embodiments of the present invention have been described in detail, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art, and all changes are encompassed in the scope of the present invention.
Claims (4)
1. A medical image data expansion method based on semi-supervised task driving is characterized in that: comprises the following steps:
s1, data acquisition: acquiring a related medical image data set, and labeling the category of the medical image data set;
s2, data expansion: the data expansion of the medical image is realized by using a task-driven data expansion method;
s3, the construction method comprises the following steps: constructing a data expansion method for data set expansion of medical images;
s4, data set storage: and combining the expanded data set with the original data set, and storing the data set.
2. The semi-supervised task driven medical image data augmentation method of claim 1, wherein: in the data acquisition in S1, the data set is screened and divided by acquiring a related medical image common data set, so as to construct a medical image data set.
3. The semi-supervised task driven medical image data augmentation method of claim 1, wherein: in S3, the construction method is used to generate an extended data set, (X)G,YG)=G((XL,YL),z;wG) Wherein G (·, ·; w is aG) Transformation method for data expansion, z being the random component of the transformation, wGFor the transformed parameters, wherein a data expansion method G, by defining two condition generators, constructing shape and intensity variations for the columns of deformation field generators and intensity field generators, respectively, by which method an image transformation of the input image is performed to obtain an expanded data set, wherein the deformation field generator G is trainedvTransformation information for outputting deformation fields, GvHas a transformation parameter of wGvBy rendering the data set image XLAnd a randomly decimated z-vector as input for generating a dense pixel-by-pixel deformation fieldAccording to the generated deformation field v, carrying out bilinear interpolation on the input image and the corresponding label to obtain an extended data set XGvAnd corresponding tag set YGvThe expression mode is as follows:training intensity field generator GIFor outputting additive intensity mask transformations, GIHas a transformation parameter of wGITo transportImage X of training data setLAnd a randomly extracted unit Gaussian distributed z-vector as input, inputting an additive strength maskThen add Δ I to XLIn (1), obtaining an extended image set XGIAnd corresponding tag set YGIThe expression mode is as follows:data expansion by a deformation field generator and intensity field generator method:
4. the semi-supervised task driven medical image data augmentation method of claim 1, wherein: the data combination method in the step S4 includes: combining the expanded data set with the original data set, and setting XLFor training the data set, YLFor a corresponding tag dataset, (X)N,YN)=(XL∪XG,YL∪YG) Said X isG、YGRepresenting the augmented data set and the corresponding tag data set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011329403.2A CN112419324B (en) | 2020-11-24 | 2020-11-24 | Medical image data expansion method based on semi-supervised task driving |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011329403.2A CN112419324B (en) | 2020-11-24 | 2020-11-24 | Medical image data expansion method based on semi-supervised task driving |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112419324A true CN112419324A (en) | 2021-02-26 |
CN112419324B CN112419324B (en) | 2022-04-19 |
Family
ID=74778709
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011329403.2A Active CN112419324B (en) | 2020-11-24 | 2020-11-24 | Medical image data expansion method based on semi-supervised task driving |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112419324B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107403446A (en) * | 2016-05-18 | 2017-11-28 | 西门子保健有限责任公司 | Method and system for the image registration using intelligent human agents |
CN107885854A (en) * | 2017-11-14 | 2018-04-06 | 山东师范大学 | A kind of semi-supervised cross-media retrieval method of feature based selection and virtual data generation |
CN109522973A (en) * | 2019-01-17 | 2019-03-26 | 云南大学 | Medical big data classification method and system based on production confrontation network and semi-supervised learning |
CN109690554A (en) * | 2016-07-21 | 2019-04-26 | 西门子保健有限责任公司 | Method and system for the medical image segmentation based on artificial intelligence |
CN109886388A (en) * | 2019-01-09 | 2019-06-14 | 平安科技(深圳)有限公司 | A kind of training sample data extending method and device based on variation self-encoding encoder |
WO2020014477A1 (en) * | 2018-07-11 | 2020-01-16 | The University Of North Carolina At Chapel Hill | Methods, systems, and computer readable media for image analysis with deep learning to predict breast cancer classes |
CN110852227A (en) * | 2019-11-04 | 2020-02-28 | 中国科学院遥感与数字地球研究所 | Hyperspectral image deep learning classification method, device, equipment and storage medium |
CN111797885A (en) * | 2019-04-05 | 2020-10-20 | 三星显示有限公司 | System and method for classification |
-
2020
- 2020-11-24 CN CN202011329403.2A patent/CN112419324B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107403446A (en) * | 2016-05-18 | 2017-11-28 | 西门子保健有限责任公司 | Method and system for the image registration using intelligent human agents |
CN109690554A (en) * | 2016-07-21 | 2019-04-26 | 西门子保健有限责任公司 | Method and system for the medical image segmentation based on artificial intelligence |
CN107885854A (en) * | 2017-11-14 | 2018-04-06 | 山东师范大学 | A kind of semi-supervised cross-media retrieval method of feature based selection and virtual data generation |
WO2020014477A1 (en) * | 2018-07-11 | 2020-01-16 | The University Of North Carolina At Chapel Hill | Methods, systems, and computer readable media for image analysis with deep learning to predict breast cancer classes |
CN109886388A (en) * | 2019-01-09 | 2019-06-14 | 平安科技(深圳)有限公司 | A kind of training sample data extending method and device based on variation self-encoding encoder |
CN109522973A (en) * | 2019-01-17 | 2019-03-26 | 云南大学 | Medical big data classification method and system based on production confrontation network and semi-supervised learning |
CN111797885A (en) * | 2019-04-05 | 2020-10-20 | 三星显示有限公司 | System and method for classification |
CN110852227A (en) * | 2019-11-04 | 2020-02-28 | 中国科学院遥感与数字地球研究所 | Hyperspectral image deep learning classification method, device, equipment and storage medium |
Non-Patent Citations (4)
Title |
---|
AMERICO OLIVEIRA 等: "Augmenting data when training a CNN for retinal vessel segmentation: How to warp?", 《2017 IEEE 5TH PORTUGUESE MEETING ON BIOENGINEERING (ENBENG)》 * |
AMY ZHAO 等: "Data Augmentation Using Learned Transformations for One-Shot Medical Image Segmentation", 《PROCEEDINGS OF THE IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
付晓 等: "基于半监督编码生成对抗网络的图像分类模型", 《自动化学报》 * |
张稀珏: "阿尔茨海默病与癫痫MRI自动识别研究", 《中国优秀硕士学位论文全文数据库 医疗卫生科技辑》 * |
Also Published As
Publication number | Publication date |
---|---|
CN112419324B (en) | 2022-04-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Ge et al. | Preserve your own correlation: A noise prior for video diffusion models | |
Vahdat et al. | Lion: Latent point diffusion models for 3d shape generation | |
CN106127684A (en) | Image super-resolution Enhancement Method based on forward-backward recutrnce convolutional neural networks | |
JP2017500651A (en) | A method for processing an input low resolution (LR) image into an output high resolution (HR) image | |
CN111062329B (en) | Unsupervised pedestrian re-identification method based on augmented network | |
Shen et al. | Adaptive dynamic filtering network for image denoising | |
Ma et al. | Symmetric transformer-based network for unsupervised image registration | |
CN111353938A (en) | Image super-resolution learning method based on network feedback | |
US20230162409A1 (en) | System and method for generating images of the same style based on layout | |
Li et al. | Hierarchical opacity propagation for image matting | |
CN116309913A (en) | Method for generating image based on ASG-GAN text description of generation countermeasure network | |
CN112419324B (en) | Medical image data expansion method based on semi-supervised task driving | |
Singh | A Survey of AI Text-to-Image and AI Text-to-Video Generators | |
Wu et al. | Bridging component learning with degradation modelling for blind image super-resolution | |
CN103390266A (en) | Image super-resolution method and device | |
JP2008152619A (en) | Data processor and data processing program | |
CN115100044B (en) | Endoscope super-resolution method and system based on three-generator generation countermeasure network | |
CN111931566B (en) | Human face cartoon image design method based on image processing | |
JP2011070283A (en) | Face image resolution enhancement device and program | |
Wu et al. | Multi-scale non-local attention network for image super-resolution | |
Zong et al. | Single image super-resolution based on self-attention | |
Shen et al. | Itsrn++: Stronger and better implicit transformer network for continuous screen content image super-resolution | |
Yue et al. | An external learning assisted self-examples learning for image super-resolution | |
Chen et al. | A study of style migration generation of traditional Chinese portraits based on DualStyleGAN | |
Chen et al. | DensER: Density-imbalance-Eased Representation for LiDAR-based Whole Scene Upsampling |
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 |