CN111539467A - GAN network architecture and method for data augmentation of medical image data set based on generation of countermeasure network - Google Patents
GAN network architecture and method for data augmentation of medical image data set based on generation of countermeasure network Download PDFInfo
- Publication number
- CN111539467A CN111539467A CN202010304146.0A CN202010304146A CN111539467A CN 111539467 A CN111539467 A CN 111539467A CN 202010304146 A CN202010304146 A CN 202010304146A CN 111539467 A CN111539467 A CN 111539467A
- Authority
- CN
- China
- Prior art keywords
- data
- training
- network
- samples
- medical image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a GAN network architecture and a method for data augmentation of medical image data sets based on generation of a confrontation network, which comprises the following steps: acquiring a real data set of the existing medical image; taking out a sample containing a focus and a sample without the focus from the samples, inputting the samples as a group, and operating a cycle generation type confrontation network to obtain an artificial sample similar to real data; adding the artificial sample into the real data set to obtain a mixed data set; the mixed dataset is taken as input and a classifier is used for classification tasks. The invention introduces the constraint condition of the reconstruction consistency loss function to realize the conversion from source distribution to target distribution and then reconstruct the source distribution; finally, a stable normalization layer is added in the discriminator, the distribution characteristics of real data are effectively simulated, an image is generated through a generated countermeasure network for data enhancement, then a large number of medical image samples are simulated, and the influence of insufficient data samples on the medical image data classification task is effectively improved.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a network architecture based on a generative countermeasure network and a convolutional neural network and an image conversion method.
Background
In recent years, with the rapid development of information technology and the continuous improvement of computer application level, deep learning technology and medical imaging technology, such as Computed Tomography (CT), Positron Emission Tomography (PET), Single Photon Emission Computed Tomography (SPECT), Magnetic Resonance Imaging (MRI), ultrasound imaging and other images obtained by medical imaging equipment, are widely used in various medical links such as medical diagnosis, anatomical structure learning, treatment planning, functional imaging data and post-operative monitoring. The rapid development of medical imaging technology has greatly advanced the modern medical diagnosis level and has become one of the main motivations for promoting medical progress.
Compared with common images, the medical images have more textures, higher resolution and stronger correlation, and meanwhile, data are difficult to acquire, and the number of available samples is far smaller than the number of features, so that some machine learning algorithms cannot be used. Therefore, in medical imaging, due to the influence of natural factors and the limitation of data recording conditions, it is generally impractical to obtain large-scale labeled data sets, often with only a small number of labeled data samples. In the conventional data augmentation method, image samples are usually added through operations of rotation, scaling, folding and the like in image research, so that new samples do not appear in the added samples actually, and the problems cannot be solved effectively. However, deep learning requires a large amount of data, which is not readily available, especially for medical image data, which limits the development of deep learning in medical image processing.
To address this problem, in 2014, Goodfellow et al first proposed generating a antagonistic network model (GAN), known by LeCun as the "best exciting point in the field of machine learning over the past decade". The original GAN model is composed of a generator and a discriminator, wherein the generator generates new data samples by capturing potential distribution of real data samples; the discriminator is a two-classifier which discriminates whether the input is real data or a generated sample. Under the current artificial intelligence hot tide, the proposal of the GAN meets the research and application requirements of a plurality of fields, and simultaneously injects new development power for the fields. At present, the most widely used field of GAN is the image and vision field. It has shown unique advantages in image generation, image segmentation, generation of high resolution images from low resolution images, etc. In recent years, with the rapid development of GAN, more and more scholars begin to apply GAN to medical image processing. This brings new methods and ideas to the field.
To date, traditional image transformation methods have been trained under supervised learning, requiring large sets of paired data with specific annotations. However, in practice it is time consuming and difficult to obtain such training samples. Therefore, it is important to design a network framework that is capable of transforming images from a source domain to a target domain using an unpaired, unlabeled image dataset.
However, when the CycleGAN is applied to the image pair data set, the CycleGAN does not fully utilize the monitoring information implied by the image pair, but directly learns the domain information represented between the image sets, so that the CycleGAN is difficult to accurately ensure that the synthesized medical image maintains sufficient image quality. On the other hand, because the CycleGAN contains a self-reconstruction process, neglecting to effectively constrain the reconstruction result leads the CycleGAN to easily lose balance between two generators on a difficult task and generate a result with poor quality. Therefore, how to find a method that has good quality of generated images and original features of the generated images becomes one of the technical problems to be solved in the art.
Disclosure of Invention
Based on the above, it is necessary to provide a method for identifying and simulating medical image data based on loop generation versus network data enhancement, which addresses the problems of the conventional technology. Specifically, the invention provides an emotion classification method based on a method for increasing data of a generation confrontation network. In the process, CycleGAN is used for data enhancement, one type of data can be converted into another type of data, in the invention, the image without the focus is converted into the image with the focus, so that the problem that the image data with the focus has few samples can be increased, and finally, the CycleGAN is used for pre-training the convolutional neural network classifier, so that the accuracy of the model is improved.
In a first aspect, an embodiment of the present invention provides a medical image lesion modality identification method based on generation of confrontation network data enhancement, where the method includes: acquiring a medical image data set as training data for training to generate a countermeasure network, and preprocessing the training data; constructing a cycleGAN model and a convolutional neural network model according to a training target; combining the loss function of the CycleGAN model, the loss function of the convolutional neural network model and the training target; training the cycleGAN model, and performing data enhancement by using the trained cycleGAN model; training the convolutional neural network model, and verifying the accuracy of the convolutional neural network classifier on a test set by using the trained convolutional neural network classifier; the model is used for data enhancement, and the convolutional neural network model is used for data classification.
In one embodiment, the acquiring the medical image data set as training data for generating the countermeasure network includes: and acquiring pre-marked medical image data.
In one embodiment, the preprocessing the training data comprises: judging whether the sizes of the obtained multiple medical image data samples are consistent; and when the sizes of the focus areas in the sample are different, cutting and correcting the inconsistent medical image data according to a preset size, and adjusting the length-width ratio of the corrected sample data.
In one embodiment, the method further comprises the following steps: and defining a loss function of the CycleGAN model and defining a loss function of the convolutional neural network model.
In one embodiment, training the CycleGAN model comprises: initializing parameters of each layer of network, uninterruptedly inputting training samples within a preset time period, and calculating a loss value of the network according to the loss function; calculating the gradient of the parameters of each layer of network through back propagation, and optimizing the parameters of each layer of network through an Adam optimization algorithm.
In one embodiment, the data enhancement using the trained CycleGAN model includes: and converting the image without the focus information into an image with the focus information of a preset modality by using the cycleGAN model.
In one embodiment, the CycleGAN model consists of two generators and two discriminators.
In a second aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the method for identifying and classifying a lesion modality based on generation of anti-network data enhancement according to the first aspect.
In a third aspect, an embodiment of the present invention provides a computer program product containing instructions, which when run on a computer, causes the computer to perform the method according to the first aspect.
In a fourth aspect, an embodiment of the present invention further provides an apparatus for identifying a lesion modality based on medical imaging with data enhancement generated by an anti-network, where the apparatus includes: the acquisition and preprocessing module is used for acquiring a medical image sample data set as training data for training to generate an confrontation network and preprocessing the training data; the construction module is used for constructing a CycleGAN model and a convolutional neural network model according to a training target; the combination module is used for combining the loss function of the CycleGAN model, the loss function of the convolutional neural network model and a training target; the training and enhancing module is used for training the cycleGAN model and enhancing data by using the trained cycleGAN model; the training and verifying module is used for training the convolutional neural network model and verifying the accuracy of the convolutional neural network classifier on a test set by using the trained convolutional neural network classifier; the training and enhancing module is used for performing data enhancement on the data, and the convolutional neural network model in the training and verifying module is used for performing data classification on the data.
In one embodiment, the acquisition and preprocessing module includes: and the marking unit is used for acquiring a medical image data sample marked in advance.
In one embodiment, the acquiring and preprocessing module further includes: the judging unit is used for judging whether the acquired focus regions and the acquired modes have the same size; and the cutting and adjusting unit is used for cutting the inconsistent medical image data sample according to a preset size and adjusting the length-width ratio of the cut data sample when the size of the focus area is inconsistent.
In one embodiment, the method further comprises the following steps: and the combination module is used for defining the loss function of the CycleGAN model and defining the loss function of the convolutional neural network model.
In one embodiment, the training and enhancement module comprises: the first calculation unit is used for initializing parameters of each layer of network, inputting training samples uninterruptedly within a preset time period, and calculating a loss value of the network according to a loss function; the second calculation unit is used for calculating the gradient of the parameters of each layer of network through back propagation; and the optimization unit is used for optimizing the parameters of each layer of network through an Adam optimization algorithm.
In one embodiment, the training and enhancement module further comprises: and the conversion unit is used for converting the image without the focus information into an image of a preset focus modality by using a CycleGAN model.
In one embodiment, the CycleGAN model consists of two generators and two discriminators.
The invention provides a medical image focus modal information identification method based on generation of confrontation network data enhancement and a corresponding depth network architecture, a medical image data set is obtained to serve as training data for training generation of a confrontation network, and the training data is preprocessed; constructing a cycleGAN model and a convolutional neural network model according to a training target; combining a loss function of a CycleGAN model, a loss function of a convolutional neural network model and a training target; training a CycleGAN model, and performing data enhancement by using the trained CycleGAN model; training the convolutional neural network model, and verifying the accuracy of the convolutional neural network classifier on a test set by using the trained convolutional neural network classifier; the CycleGAN model is used for data enhancement, and the convolutional neural network model is used for data classification. According to the method, the problem of unbalanced data categories or the problem of small data sets is solved by generating the confrontation network generated image, data enhancement is carried out, and then the classifier is trained, so that the model accuracy rate is improved. In the process, CycleGAN is used for data enhancement, one type of data can be converted into the other type of data, in the invention, the image without a focus region is converted into the image with the preset focus modal region, so that the problem that the image with focus information has little data volume can be solved, and finally, the images are used for pre-training the convolutional neural network classifier, so that the accuracy of the model is improved.
Compared with the prior art, the invention has the following beneficial effects:
the medical image data augmentation method based on the cycle consistency loss constraint can convert the focus region images of different data sets to obtain medical image images containing focus information in a unified mode; the negative influence on model training caused by the modal difference between different data sets is avoided, and therefore the accuracy of the medical image detection algorithm is improved.
Drawings
In order to make the purpose of the present invention more comprehensible, the present invention will be further described with reference to the accompanying drawings.
FIG. 1 is a flow chart of task execution of the present invention;
fig. 2 is a flowchart of an image transformation method based on generative countermeasure network and cyclic consistency loss according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, 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 some, but not all, embodiments of the present invention.
Examples
The problem of insufficient data sample size is a common problem in the process of acquiring a medical image data set, and is specifically represented as follows: the number of lesion information-containing samples in the data set is far from the number of healthy samples; the number of samples in a certain modality is not equal to the number of samples in the other modalities, and the number of data samples with obvious lesion information is too small. If the data set or the algorithm is not improved correspondingly, the classification training is directly carried out, and as a result, a small number of types of sample data cannot be paid sufficient attention, and in a severe case, the sample data can be even ignored as noise by a classifier, so that the classification result is seriously biased.
Under such a background, how to perform data expansion on data samples with small sample size in a medical image data set and obtain an ideal result in a classifier becomes a problem requiring deep research. The problems that the medical image data volume is unbalanced and the focus sample data volume is small mainly include two major optimization methods: (1) changing the original distribution of the data set, and performing down sampling on most types of samples, or performing over sampling on few types of samples, or combining the two to balance the number of different types of the data set as much as possible; (2) and optimizing the classifier, such as improving the weight of the classifier on a few classes of samples during training, so that the classifier is fully paid attention. Many studies and experiments show that the accuracy of the classification result is improved well by the two methods.
The embodiment starts from the idea of changing the original distribution of the data set, utilizes the connection and difference between the data of different modes in the original data set and the advantages of a generative confrontation network to grasp the same parts between the data of different modes, analyzes the difference parts, manufactures the artificial data expansion data set, and improves the accuracy of the classifier.
As can be seen from fig. 1, the classification method includes:
step 1: acquiring a medical image data set S of an initial version;
in this step, the class Imbalance Ratio (IR) represents the severity of imbalance between the number of samples in different classes, which is defined as the ratio of the number of samples in the majority class to the number of samples in the minority class. For the example of two classification problems, the class imbalance ratio of the training data set S should be in the range of 100: 1-2: 1.
Step 2: screening samples S + of different modes from the S;
in the step, if the S + contains different types, samples of all types are extracted, and the steps 3, 4 and 5 are independently carried out;
and step 3: data samples with lesion information S1 and data samples without lesion information S2 are distinguished in S +. The specific method comprises the following steps: when the data set is preprocessed, marking the data in advance, distinguishing data samples of different modes, selecting the data sample containing focus information as an object for neural network learning S1, and selecting the data sample without the focus information as initial input data S2;
and 4, step 4: training a cycleGAN by taking a data sample S1 containing focus information and a data sample S2 containing no focus information as a pair of inputs to obtain an artificial sample S x;
cyclic GAN (cyclic-coherent adaptive Networks, periodic coherent Generative Networks) is a derivative model based on GAN (Generative adaptive Networks).
GAN is an unsupervised learning method proposed by Ian Goodfellow et al in 2014, which learns by letting two neural networks game each other. The GAN consists of a generation network and a discrimination network. The generation network takes as input a random sampling from the underlying space, and its output needs to mimic as much as possible the real samples in the training set. The input of the discrimination network is the real sample or the output of the generation network, and the purpose is to distinguish the output of the generation network from the real sample as much as possible. The generation network should cheat the discrimination network as much as possible. The two networks resist each other and continuously adjust parameters, and the final purpose is to make the judgment network unable to judge whether the output result of the generated network is real or not.
CycleGAN is a GAN-based network model proposed by Jun-Yan Zhu et al in 2017. The CycleGAN includes two GAN groups, where two generators are mapping from reference domain to target domain and mapping from target domain to reference domain, respectively. The core idea of CycleGAN is that if the result of the mapping of the producer from the reference domain to the target domain is good enough, this result is used for the reverse mapping, i.e. the mapping from the target domain to the reference domain, and the result should be as consistent as possible with the original reference domain.
The CycleGAN performs well on the task of style migration learning, and when the reference domain is a certain type of picture and the target domain is another type of picture, the trained CycleGAN can obtain a first type of picture with a second type of picture style. If the reference field is a landscape shot by a camera and the target field is a Sanskrit painting, the landscape shot with the Sanskrit style can be obtained through the generator finally. The method for classifying the class imbalance problem based on the extended training data set improves the original CycleGAN, and constructs a new network model suitable for the same type of sample data extension.
A diagram of a network model of the improved algorithm based on the cycleGAN is shown in figure 2.
The method for generating the artificial sample by combining the focus information with the cycleGAN specifically comprises the following steps:
in each training round, the same number of samples are extracted from the data sample with focus information S1 and the data sample without focus information S2, respectively, and are used as a reference field and a target field.
The samples in the reference field R are passed through a generator G, resulting in an artificial sample G (R). And inputting the G (R) and the corresponding sample in the target domain T into a discriminator Dt, and calculating to obtain a part of loss of the training of the current round. The calculation formula is as follows:
the samples in the target field T are passed through a generator F, resulting in artificial samples F (T). And F (T) and the corresponding sample in the reference field R are input into a discriminator Dr, and the other part of the loss of the training of the current round is calculated. The calculation formula is as follows:
g (R) passes through a generator F to obtain an artificial sample F (G (R)). Similarly, f (t) goes through generator G, resulting in artificial sample G (f (t)). By calculating the difference between F (G (R)) and the reference domain R, G (F (T)) and the target domain T, a periodic consistency loss can be obtained. The calculation formula is as follows:
the above three loss functions are added to obtain the overall loss function of the training round, wherein the goal of G, F is to minimize the above value, and the goal of Dr, Dt is to maximize the above value. And in the final link of each training, updating the weight of each parameter in the network by using a random gradient descent method.
When the values of L (G, F, Dr, Dt) converge, or the number of training times has reached a pre-designed threshold (1000 iterations), the training of the entire network is stopped, at which point a trained generator G is obtained.
And randomly extracting a pair of samples in the S +, respectively taking the samples as a reference set and a training set, and inputting the samples into a generator G to obtain an artificial sample. This step is repeated several times to obtain a certain number of artificial samples S.
And 5: obtaining a sampling rate (IR-1) according to the unbalanced ratio IR of the real data set, randomly extracting (IR-1) N + samples from the artificial samples and marking the samples as a minority class, wherein N + is the number of real minority class samples, adding the samples into the real data set and disordering the samples to obtain an expanded mixed data set S'; and the mixed data set S' is used for replacing an original real data set to serve as input, a classifier is trained, and the final classification accuracy is obtained.
The medical image data set with the focus information has the advantages that the problem that the data volume of a focus information sample is seriously insufficient in the medical image data set is effectively solved, and the result deviation brought by the training of a classification task is reduced by using a mixed data set after the data augmentation task is completed. Most of the traditional data set expansion methods are derived based on random repeated sampling, and have the characteristics of simple realization, low time complexity and poor effect. The invention combines the advantages of a generative confrontation network in the field of deep learning, improves the existing model, and enables the method to better fit the distribution characteristics of the original data, so that the accuracy of classification task training by using the expanded data set can be improved. Meanwhile, the invention only expands the data set without modifying the classification algorithm, so that the method can be directly used on different classifiers and has self-adaptability.
Claims (9)
1. A medical image data augmentation method based on generation of antagonistic network data augmentation, the method comprising: acquiring a medical image data set as training data for training to generate an antagonistic network, preprocessing the training data to acquire a real data set, and screening out a sample containing focus information; distinguishing samples containing focus information and samples without focus information; training a generative confrontation network by taking the two samples as input to obtain a series of artificial samples distributed similarly to a real data set; constructing a cycleGAN model and a convolutional neural network model according to a training target; combining the loss function of the CycleGAN model, the loss function of the convolutional neural network model, and the training target; training the cycleGAN model, and performing data enhancement by using the trained cycleGAN model; adding a certain amount of artificially generated samples into the real data set according to the set sampling rate to obtain a mixed data set; training the convolutional neural network model, and verifying the accuracy of the convolutional neural network classifier on a test set by using the trained convolutional neural network classifier; the CycleGAN model is used for data enhancement, and the convolutional neural network model is used for data classification.
2. The method for augmenting medical imaging data based on generation of countermeasure network data according to claim 1, wherein the acquiring of the medical imaging data set as training data for training generation of the countermeasure network comprises: and acquiring a pre-marked medical image data sample.
3. The method for augmenting medical image data based on generation of anti-network data enhancement according to claim 1, wherein preprocessing the training data comprises: judging whether the acquired multiple medical image samples have the same size; and when the sizes of the medical image samples are inconsistent, the inconsistent medical image samples are cut according to the preset size, and the length-width ratio of the cut image is adjusted.
4. The medical image data augmentation method based on generation of countermeasure network data augmentation according to claim 1, further comprising: defining a loss function of the CycleGAN model, and defining a loss function of the convolutional neural network model.
5. The method of claim 1, wherein training the CycleGAN model comprises: initializing parameters of each layer of network, uninterruptedly inputting training samples within a preset time period, and calculating a loss value of the network according to the loss function; calculating the gradient of the parameters of each layer of network through back propagation, and optimizing the parameters of each layer of network through an Adam optimization algorithm.
6. The method for augmenting medical image data based on generation of countermeasure network data enhancement as recited in claim 1, wherein the data enhancement using the trained CycleGAN model comprises: and converting the image sample without the focus into an image sample with a preset focus pattern by using the cycleGAN model.
7. The method for augmenting medical image data based on generation of countermeasure network data enhancement according to any one of claims 4 to 6, wherein the CycleGAN model is composed of two generators and two discriminators.
8. The method as claimed in claims 4-6, wherein the generator and the arbiter are trained alternately in equal steps, i.e. in one training, the generator is updated first, the arbiter is updated again, and the updating is stopped until the overall loss function converges or a specified number of training times is reached.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the method of any of claims 1-8 are implemented when the program is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010304146.0A CN111539467A (en) | 2020-04-17 | 2020-04-17 | GAN network architecture and method for data augmentation of medical image data set based on generation of countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010304146.0A CN111539467A (en) | 2020-04-17 | 2020-04-17 | GAN network architecture and method for data augmentation of medical image data set based on generation of countermeasure network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111539467A true CN111539467A (en) | 2020-08-14 |
Family
ID=71952283
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010304146.0A Pending CN111539467A (en) | 2020-04-17 | 2020-04-17 | GAN network architecture and method for data augmentation of medical image data set based on generation of countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111539467A (en) |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112102276A (en) * | 2020-09-10 | 2020-12-18 | 西安电子科技大学 | Low-field-intensity MR stomach segmentation method based on transfer learning image enhancement |
| CN112132211A (en) * | 2020-09-20 | 2020-12-25 | 郑州精铖能源技术有限公司 | Efficient cascade energy management method and system |
| CN112215202A (en) * | 2020-10-30 | 2021-01-12 | 上海禾儿盟智能科技有限公司 | Near-infrared image data offline augmentation device, system and method |
| CN112365464A (en) * | 2020-11-09 | 2021-02-12 | 成都信息工程大学 | GAN-based medical image lesion area weak supervision positioning method |
| CN112396110A (en) * | 2020-11-20 | 2021-02-23 | 南京大学 | Method for generating anti-cascade network augmented image |
| CN112462001A (en) * | 2020-11-17 | 2021-03-09 | 吉林大学 | Gas sensor array model calibration method for data amplification based on condition generation countermeasure network |
| CN112488937A (en) * | 2020-11-27 | 2021-03-12 | 河北工业大学 | Medical image feature enhancement method for segmentation task |
| CN112699726A (en) * | 2020-11-11 | 2021-04-23 | 中国科学院计算技术研究所数字经济产业研究院 | Image enhancement method, genuine-fake commodity identification method and equipment |
| CN112731327A (en) * | 2020-12-25 | 2021-04-30 | 南昌航空大学 | HRRP radar target identification method based on CN-LSGAN, STFT and CNN |
| CN112837317A (en) * | 2020-12-31 | 2021-05-25 | 无锡祥生医疗科技股份有限公司 | Focus classification method and device based on breast ultrasound image enhancement and storage medium |
| CN113012821A (en) * | 2021-03-18 | 2021-06-22 | 日照职业技术学院 | Implementation method of multi-modal rehabilitation diagnosis and treatment cloud platform based on machine learning |
| CN113052171A (en) * | 2021-03-24 | 2021-06-29 | 浙江工业大学 | Medical image augmentation method based on progressive generation network |
| CN113222114A (en) * | 2021-04-22 | 2021-08-06 | 北京科技大学 | Image data augmentation method and device |
| CN113344044A (en) * | 2021-05-21 | 2021-09-03 | 北京工业大学 | Cross-species medical image classification method based on domain self-adaptation |
| CN113378991A (en) * | 2021-07-07 | 2021-09-10 | 上海联影医疗科技股份有限公司 | Medical data generation method and device, electronic equipment and storage medium |
| CN113421212A (en) * | 2021-06-23 | 2021-09-21 | 华侨大学 | Medical image enhancement method, device, equipment and medium |
| CN113658036A (en) * | 2021-08-23 | 2021-11-16 | 平安科技（深圳）有限公司 | Data augmentation method, device, computer and medium based on countermeasure generation network |
| CN114022724A (en) * | 2021-10-08 | 2022-02-08 | 郑州大学 | Pipeline disease image data enhancement method for generating countermeasure network based on style migration |
| CN114176549A (en) * | 2021-12-23 | 2022-03-15 | 杭州电子科技大学 | Fetal heart rate signal data enhancement method and device based on generative countermeasure network |
| CN114782443A (en) * | 2022-06-22 | 2022-07-22 | 深圳科亚医疗科技有限公司 | Device and storage medium for data-based enhanced aneurysm risk assessment |
| CN114974567A (en) * | 2022-05-25 | 2022-08-30 | 浙江大学 | Liver tumor interventional response post-evaluation method and system based on antagonistic generation network |
| CN115481694A (en) * | 2022-09-26 | 2022-12-16 | 南京星环智能科技有限公司 | Data enhancement method, device, equipment and storage medium for training sample set |
| CN115934484A (en) * | 2022-11-29 | 2023-04-07 | 广东技术师范大学 | Diffusion model data enhancement-based anomaly detection method, storage medium and equipment |
| CN116402812A (en) * | 2023-06-07 | 2023-07-07 | 江西业力医疗器械有限公司 | Medical image data processing method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108470187A (en) * | 2018-02-26 | 2018-08-31 | 华南理工大学 | A kind of class imbalance question classification method based on expansion training dataset |
| CN109493308A (en) * | 2018-11-14 | 2019-03-19 | 吉林大学 | The medical image synthesis and classification method for generating confrontation network are differentiated based on condition more |
| CN110009028A (en) * | 2019-03-28 | 2019-07-12 | 北京科技大学 | A kind of micro-image data enhancement methods and device |
-
2020
- 2020-04-17 CN CN202010304146.0A patent/CN111539467A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108470187A (en) * | 2018-02-26 | 2018-08-31 | 华南理工大学 | A kind of class imbalance question classification method based on expansion training dataset |
| CN109493308A (en) * | 2018-11-14 | 2019-03-19 | 吉林大学 | The medical image synthesis and classification method for generating confrontation network are differentiated based on condition more |
| CN110009028A (en) * | 2019-03-28 | 2019-07-12 | 北京科技大学 | A kind of micro-image data enhancement methods and device |
Cited By (37)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112102276A (en) * | 2020-09-10 | 2020-12-18 | 西安电子科技大学 | Low-field-intensity MR stomach segmentation method based on transfer learning image enhancement |
| CN112102276B (en) * | 2020-09-10 | 2023-03-28 | 西安电子科技大学 | Low-field-intensity MR stomach segmentation method based on transfer learning image enhancement |
| CN112132211A (en) * | 2020-09-20 | 2020-12-25 | 郑州精铖能源技术有限公司 | Efficient cascade energy management method and system |
| CN112215202A (en) * | 2020-10-30 | 2021-01-12 | 上海禾儿盟智能科技有限公司 | Near-infrared image data offline augmentation device, system and method |
| CN112215202B (en) * | 2020-10-30 | 2023-06-30 | 上海禾儿盟智能科技有限公司 | Near infrared image data offline augmentation device, system and method |
| CN112365464A (en) * | 2020-11-09 | 2021-02-12 | 成都信息工程大学 | GAN-based medical image lesion area weak supervision positioning method |
| CN112699726A (en) * | 2020-11-11 | 2021-04-23 | 中国科学院计算技术研究所数字经济产业研究院 | Image enhancement method, genuine-fake commodity identification method and equipment |
| CN112462001A (en) * | 2020-11-17 | 2021-03-09 | 吉林大学 | Gas sensor array model calibration method for data amplification based on condition generation countermeasure network |
| CN112462001B (en) * | 2020-11-17 | 2021-07-23 | 吉林大学 | Gas sensor array model calibration method for data amplification based on condition generation countermeasure network |
| CN112396110B (en) * | 2020-11-20 | 2024-02-02 | 南京大学 | Method for generating augmented image of countermeasure cascade network |
| CN112396110A (en) * | 2020-11-20 | 2021-02-23 | 南京大学 | Method for generating anti-cascade network augmented image |
| CN112488937A (en) * | 2020-11-27 | 2021-03-12 | 河北工业大学 | Medical image feature enhancement method for segmentation task |
| CN112731327A (en) * | 2020-12-25 | 2021-04-30 | 南昌航空大学 | HRRP radar target identification method based on CN-LSGAN, STFT and CNN |
| CN112731327B (en) * | 2020-12-25 | 2023-05-23 | 南昌航空大学 | HRRP radar target identification method based on CN-LSGAN, STFT and CNN |
| CN112837317A (en) * | 2020-12-31 | 2021-05-25 | 无锡祥生医疗科技股份有限公司 | Focus classification method and device based on breast ultrasound image enhancement and storage medium |
| CN113012821A (en) * | 2021-03-18 | 2021-06-22 | 日照职业技术学院 | Implementation method of multi-modal rehabilitation diagnosis and treatment cloud platform based on machine learning |
| CN113052171A (en) * | 2021-03-24 | 2021-06-29 | 浙江工业大学 | Medical image augmentation method based on progressive generation network |
| CN113222114B (en) * | 2021-04-22 | 2023-08-15 | 北京科技大学 | Image data augmentation method and device |
| CN113222114A (en) * | 2021-04-22 | 2021-08-06 | 北京科技大学 | Image data augmentation method and device |
| CN113344044A (en) * | 2021-05-21 | 2021-09-03 | 北京工业大学 | Cross-species medical image classification method based on domain self-adaptation |
| CN113344044B (en) * | 2021-05-21 | 2024-05-28 | 北京工业大学 | Cross-species medical image classification method based on field self-adaption |
| CN113421212B (en) * | 2021-06-23 | 2023-06-02 | 华侨大学 | Medical image enhancement method, device, equipment and medium |
| CN113421212A (en) * | 2021-06-23 | 2021-09-21 | 华侨大学 | Medical image enhancement method, device, equipment and medium |
| CN113378991A (en) * | 2021-07-07 | 2021-09-10 | 上海联影医疗科技股份有限公司 | Medical data generation method and device, electronic equipment and storage medium |
| CN113658036A (en) * | 2021-08-23 | 2021-11-16 | 平安科技（深圳）有限公司 | Data augmentation method, device, computer and medium based on countermeasure generation network |
| CN113658036B (en) * | 2021-08-23 | 2024-05-31 | 平安科技（深圳）有限公司 | Data augmentation method, device, computer and medium based on countermeasure generation network |
| CN114022724A (en) * | 2021-10-08 | 2022-02-08 | 郑州大学 | Pipeline disease image data enhancement method for generating countermeasure network based on style migration |
| CN114176549B (en) * | 2021-12-23 | 2024-04-16 | 杭州电子科技大学 | Fetal heart rate signal data enhancement method and device based on generation type countermeasure network |
| CN114176549A (en) * | 2021-12-23 | 2022-03-15 | 杭州电子科技大学 | Fetal heart rate signal data enhancement method and device based on generative countermeasure network |
| CN114974567A (en) * | 2022-05-25 | 2022-08-30 | 浙江大学 | Liver tumor interventional response post-evaluation method and system based on antagonistic generation network |
| CN114782443A (en) * | 2022-06-22 | 2022-07-22 | 深圳科亚医疗科技有限公司 | Device and storage medium for data-based enhanced aneurysm risk assessment |
| CN115481694B (en) * | 2022-09-26 | 2023-09-05 | 南京星环智能科技有限公司 | Data enhancement method, device and equipment for training sample set and storage medium |
| CN115481694A (en) * | 2022-09-26 | 2022-12-16 | 南京星环智能科技有限公司 | Data enhancement method, device, equipment and storage medium for training sample set |
| CN115934484B (en) * | 2022-11-29 | 2024-02-09 | 广东技术师范大学 | Diffusion model data enhancement-based anomaly detection method, storage medium and apparatus |
| CN115934484A (en) * | 2022-11-29 | 2023-04-07 | 广东技术师范大学 | Diffusion model data enhancement-based anomaly detection method, storage medium and equipment |
| CN116402812B (en) * | 2023-06-07 | 2023-09-19 | 江西业力医疗器械有限公司 | Medical image data processing method and system |
| CN116402812A (en) * | 2023-06-07 | 2023-07-07 | 江西业力医疗器械有限公司 | Medical image data processing method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111539467A (en) | GAN network architecture and method for data augmentation of medical image data set based on generation of countermeasure network | |
| CN109753992B (en) | Unsupervised domain adaptive image classification method based on condition generation countermeasure network | |
| Ji et al. | Learning calibrated medical image segmentation via multi-rater agreement modeling | |
| CN108491874B (en) | Image list classification method based on generation type countermeasure network | |
| CN113724880A (en) | Abnormal brain connection prediction system, method and device and readable storage medium | |
| WO2019015466A1 (en) | Method and apparatus for verifying person and certificate | |
| CN110728729A (en) | Unsupervised CT projection domain data recovery method based on attention mechanism | |
| CN110910377B (en) | Cerebral infarction MRI image identification method based on neural network | |
| CN111861924B (en) | Cardiac magnetic resonance image data enhancement method based on evolutionary GAN | |
| CN109376787A (en) | Manifold learning network and computer visual image collection classification method based on it | |
| Ma et al. | Face image illumination processing based on generative adversarial nets | |
| Zhou et al. | Pro-pulse: Learning progressive encoders of latent semantics in gans for photo upsampling | |
| CN117253277A (en) | Method for detecting key points of face in complex environment by combining real and synthetic data | |
| Zi et al. | Steganography with convincing normal image from a joint generative adversarial framework | |
| Xing et al. | The Beauty or the Beast: Which Aspect of Synthetic Medical Images Deserves Our Focus? | |
| CN116243680A (en) | Black box domain adaptive industrial equipment diagnosis method, system and storage medium | |
| Liu et al. | Generate realistic traffic sign image using deep convolutional generative adversarial networks | |
| Sünkel et al. | Hybrid quantum machine learning assisted classification of COVID-19 from computed tomography scans | |
| Melo et al. | OCT Image Synthesis through Deep Generative Models | |
| Abdulwahab et al. | Promising depth map prediction method from a single image based on conditional generative adversarial network | |
| CN112992304A (en) | High-resolution pinkeye case data generation method, equipment and storage medium | |
| Jia et al. | Everyone could be artist: sketch to style-specified artwork | |
| CN111860618B (en) | Bidirectional GAN model for pathological data conversion and construction and application methods thereof | |
| Dornadula et al. | AI driven shadow model detection in agropv farms | |
| Zhuang et al. | Synthetic power analyses: empirical evaluation and application to cognitive neuroimaging |
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 |