AU2020102667A4 - Adversarial training for large scale healthcare data using machine learning system - Google Patents
Adversarial training for large scale healthcare data using machine learning system Download PDFInfo
- Publication number
- AU2020102667A4 AU2020102667A4 AU2020102667A AU2020102667A AU2020102667A4 AU 2020102667 A4 AU2020102667 A4 AU 2020102667A4 AU 2020102667 A AU2020102667 A AU 2020102667A AU 2020102667 A AU2020102667 A AU 2020102667A AU 2020102667 A4 AU2020102667 A4 AU 2020102667A4
- Authority
- AU
- Australia
- Prior art keywords
- images
- medical images
- image
- training
- machine learning
- 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.)
- Ceased
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V10/7753—Incorporation of unlabelled data, e.g. multiple instance learning [MIL]
-
- 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
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- 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
- G06N3/094—Adversarial learning
-
- 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
- G06T7/0014—Biomedical image inspection using an image reference approach
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Molecular Biology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Multimedia (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Radiology & Medical Imaging (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
ADVERSARIAL TRAINING FOR LARGE SCALE HEALTHCARE DATA USING
MACHINE LEARNING SYSTEM
ABSTRACT
The mechanisms are presented to execute the machine learning training model. The Image
generator which is implemented in the generative adversarial network is trained by using the
machine learning training model in order to produce the medical images by estimating the realistic
medical images. The trained medical images are expanded by the machine learning training model
in order to incorporate one or more generated medical images which is generated by the image
generator present in the generalized adversarial network (GAN). The real and fake medical images
are classified by applying the trained machine learning model to the new medical image inputs.
11 P a g e
ADVERSARIAL TRAINING FOR LARGE SCALE HEALTHCARE DATA USING
MACHINE LEARNING SYSTEM
Drawings:
Real Images
(Labelled and
Unlabelled) Real
Discriminator -, (normal/abnormal)/
fake label
Fake
Images
Generated
Fake Images
Random Generator
Vector Noise
FineTuning
Training Logic
Figure: Block diagram of Generalized Adversarial Network Architecture
11 P a g e
Description
Drawings:
Real Images (Labelled and Unlabelled) Real Discriminator -, (normal/abnormal)/ fake label
Fake Images Generated Fake Images
Random Generator Vector Noise FineTuning Training Logic
Figure: Block diagram of Generalized Adversarial Network Architecture
11 P a g e
Description
Field of Invention:
This field of invention addresses the adversarial training for the large-scale healthcare data by using the machine learning system. The proposed approach is to classify the real normal or abnormal images and fake images on basis of the health records.
Background of invention:
In recent years, the acquisition of electronic health record (EHR) systems in hospitals has become common. This digital report in the large-scale healthcare data granted itself to the modern machine learning tools which is used to perform different tasks like detection of disease, data augmentation, etc. The convolutional neural networks are successful in detecting the different disease and also includes the classification tasks like lung disease, and colonoscopy frames. This method is efficient in finding the disease and results it as positive or negative. But it also struggles in finding the result as positive or negative for rare disease patients like Ebola, because less patients will be tested as positive. In this case, the generative adversarial networks (GANs) are utilized because they have capability to learn in order to produce fake examples for the inadequate representation of the data and better training of the model. In addition to that, the generative adversarial networks can also be used for de-identifying the data, that will prevent patient personal information from vulnerability. In healthcare field, the de-identification of the data is one of the challenging tasks because the conventional method is not better to withstand re-identification of data. In current research, the GAN models are the promising solution to many of the difficulties facing in the healthcare. The Generative Adversarial Networks are also known as set of deep neural network models which is used to generate the synthetic data. The synthetic data is used to refer the data which is applicable to the particular situation that will not be directly obtained via the real-world measurements. The main objective of the generative adversarial network is to train the discriminator in order to show the difference between real and fake data. The simple convolutional neural network which is named as discriminator has capability to differentiate the original and synthetic images and at the same time, the generator will be trained to generate the synthetic data which will practice the discriminator. The modification of convolutional neural network is named as generator that will be trained to generate the original fake images. The GANs will train the ordinary as well as modified convolutional neural network together in order to enhance the ability of ordinary CNN for the reason of marking the fake images and to enhance the ability of modified CNN to generate the practical or reasonable output. The modified CNN is trained on the random noise pixel data and then produces the fake brain CT images. The Fake images which is made by the modified CNN is later fed to the original CNN next to the real images. The modified CNN will target to generate the practical fake images that can 'fetch the original CNN. As the training
1| P a g e continues, the modified CNN will get better improvement to produce the fake images, and the ordinary CNN will be better at differentiating the real and fake images until the modified CNN generates the images which is close to the original images. After completion of training, the GAN will generate the practical images that can enlarge the existing data or creates the new data. Data augmentation is useful in the situations like rare diseases. Later, creation of synthetic data set is very much useful in protecting the privacy of the patient's information and also medical data can be purchased from the expensive clinically interpreted data.
Objects of the invention:
• To train the image generator present in the generative adversarial network (GAN) in order to create the medical images by estimating the original medical images. • To enlarge the set of trained medical images incorporates the at least one or more created images produced by the image generator of the generalized adversarial network. • To train the machine learning model on the basis of the expanding the set of trained medical images in order to find out the deviations in the medical images. • Applying the trained machine learning model to the new input images in order to distinguish the medical images to make sure the presence of deviations in the images. • The Input is given from the noise generator to the image generator.
Summary of the invention:
The method present in the data processing system consists of the processor and the memory. The memory consists of the instructions that will be executed with the help of processor in order to arrange the processor for implementing the models of machine learning. The methods incorporate the training with the use of machine learning training model. An Image generator in the GAN will create the medical images to evaluate the actual medical images. The method also involves the data enlargement by using the training model of machine learning in order to include the one or more created medical images produced by the image generator of the generative adversarial network. The method also insists the training with the help of machine learning training model, on the basis of the expansion of the trained medical images to identify the deviation in medical images. Addition to that, the method also consists of trained machine learning model to the newly generated medical image input in order to distinguish the medical images related to deviations present in the image. The method initiates the training for generative adversarial network on the basis of the labeled image data, unlabeled image data and the generated image which is created by the image generator of the GAN. The loss function is present in the GAN that consists of error components to each of the labeled and unlabeled, generated image data that will be used to train the generative adversarial network. Further, the method insists to find out the new data source to adapt the trained GAN. The method differentiates the image present in the new data source by using the adapted GAN to the data present in the new data source. The trained generative adversarial network which is adapted obtains the set of labeled images and uses the set of images to perform the adaption of trained GAN. The method contains the configuring the difference of
21Page generative adversarial network to distinguish the input of medical images that will represent the normal or abnormal, generated medical condition. The method will further consist of the one or more created images and given as input to the differentiator of the generative adversarial network. The trained medical image consists of the labeled medical image as the first subset, unlabeled medical image as the second subset, and the generated medical image as the third subset. The method consists of the discriminator for classifying the trained medical images in the set of the specified classes. Deep learning algorithms will require the more amount of annotated data to train the models like image classification. In the field of medical imaging, the data is not enough as other fields because of the privacy, standard laws, integration of information in medical system is not enough. These issues are overcome by using deep learning algorithms as they require volume of data for every task like image classification or segmentation. In most of the cases, the availability of medical image data is unstructured and lacks the labeling. To address the image, the labeling is important. Labelling the image is expensive and time-consuming process because of the manual process. If the labeled image set is accessed for training the classifier, it suffers to maintain the performance level. Neural networks help for segmentation of the image into the different segment parts which is easy for computation and to analyze the data. The features of the segmentation technique are used as the image to train the model of classifier. This is the method of learning the data distribution in one class and takes the advantage to learn from the other class. To enhance the abnormalities, the classification is also used to generative machine learning models. The mechanism is provided for the GAN on the basis of framework to create the medical image data and to train the medical image as classifier based on augmentation of dataset of the medical images. The mechanisms are also provided for the selection of training techniques by using the generative adversarial network framework which is used for creation of new sources of the medical image data such as service initiated for the new client of the medical image classifier to be used by the framework of GAN. The training techniques is selected to agree the first method and the classifier will be trained on the basis of the labeled image source and in second method, the classifier trained on the basis of known labeled image source data as well as new image source data, which acts as unlabeled medical image data.
Detailed Description of the Invention:
Figure 1 about the block diagram of GAN architecture. The Generative adversarial network is the method of neural networks which is used in the unsupervised learning. It is made of the two models. They have capability to generate the practical images from the images of input with its similarities. These are good for manipulation and generating the image. These are used as tasks in understanding the risks and retaining of the healthcare. The generative adversarial network will achieve by combining the two neural networks such as discriminator network that will help to differentiate the original images from the generated artificial output image. In case, if the discriminator finds out the image as fake in correct way, it will reward in the form of positive feedback or else it receives the negative feedback. The generator network that learns to create the practical images form the random noise input images. The fake images output which is generated
31Page is given to the discriminator in order to classify it as either the real or fake. If the discriminator is not classified properly for fake images as the real image, the generator will be tuned as fine way with the positive feedback or else the techniques of back propagation will be applied with the negative feedback as the method present in the discriminator network. The models of generator and discriminator of the neural networks will compete to each other and along with the multiple steps, both the models will be trained accurately in the training phase. The working of GAN network will distribute the data image and trains in the way so that will maximize the probability to make the mistake and the same discriminator will approximate the probability of the distributed data but not from the artificial output. From the number of steps in the training phase, the discriminator will try to maximize the loss until it gains the more confidence and it will misclassify the produced image to reach its loss until it gains the confidence. In the generative adversarial network, the realistic data will be trained by the discriminator network whereas the generator network will remain in idle. The generator will begin the training, whereas the discriminator network will remain in idle and the discriminator will train up to the generated fake data and will use the prediction result in the feedback for the purpose of tuning the generator to reach the better performance in order to fool the discriminator. The GAN models will generate the new samples of dataset. The two main category of deep generative models are distribution of data or the learning of function in order to transform the sample of existing data distribution. The training of GAN to generate and discriminate the medical images and to use the generator G for creation of practical medical image for data expansion technique to train the abnormal detector that will operate on the medical images. Data expansion is widely used for deep learning to improve the data in the limited scenarios. The modifications lead to produce the new medical images to train and does not improve the training. The discriminator trains on the basis of labeled and unlabeled real images and also generates the fake images to discriminate the type of images. The generator is trained in the GAN to produce fake images that will estimate the real images and the discriminator is fooled by the produced medical images and allows the generator as additional source of the unlabeled image and expands the trained medical data image. The discriminator is trained to perform the feature of images as real normal, abnormal, fake normal and abnormal medical images for high accuracy in the classifying input images. Figure 2 illustrates the GAN operates on the labeled and unlabeled medical images input by using the classifications and it is further configured by loss function to insert the error components for each of the labeled, unlabeled and fake images on the basis of the noise input. The generator of GAN will produce the fake image with the help of input noise. The real normal and abnormal images is given to the discriminator to insert the components of all type's images. The GAN will generate the fake image on the basis of input noise vector.
41Page
Claims (6)
1. The machine learning training model is implemented with the help of data processing system. This method compromises: - By using machine learning training model, Image generator present in the generalized adversarial network will helps to generate the medical images by estimating the original medical images. - By using machine learning training model, the set of trained medical images incorporates the one or more created medical images produced by the image generator of the GAN in order to identify deviations in medical images.
2. The method of claim, the generator in the GAN is trained to produce the medical images estimated by the original medical images. - One of the medical images shows a deviation which is generated by image generator. - A noise input is given to the image generator by a noise generator. - One of the generated images on the basis of the combination of at least one medical image and the noise input. - Analyzing by the discriminator, at least one medical image and at least generated image labels the at least one of the medical images and generated images. - Modification of operational parameter of the generator on the basis of the results by analyzing at least one of the medical images and the generated images by the discriminator.
3. The method of claim, the GAN is trained in order to classify the medical images regarding the abnormalities present in the medical images.
4. The method of claim, the discriminator is trained by using the techniques of adversarial training on the basis of expanding the set of training medical images, to distinguish between the medical images related to deviations present in the image and the normal medical conditions of the images.
5. The method of claim, training the image generator on the basis of the both labelled and non-labelled medical images and also training based on the feedback obtains from the output of the discriminator.
6. The method of claim, the image generator generates the medical image data of the GAN on the basis of the input noise, and training by the machine learning model, the image generator generates the medical image by estimating the original medical images compromising: - The discriminator of the GAN trains to separately classifying the generated medical images from the original medical images and separately classifies the original medical images that shows the normal conditions in image from the original image showing the abnormal conditions.
1 Pag e
ADVERSARIAL TRAINING FOR LARGE SCALE HEALTHCARE DATA USING 11 Oct 2020
MACHINE LEARNING SYSTEM
Drawings: 2020102667
Figure1: Block diagram of Generalized Adversarial Network Architecture
1|Page
Figure 2: Flowchart for training the GAN
2|Page
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
AU2020102667A AU2020102667A4 (en) | 2020-10-11 | 2020-10-11 | Adversarial training for large scale healthcare data using machine learning system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
AU2020102667A AU2020102667A4 (en) | 2020-10-11 | 2020-10-11 | Adversarial training for large scale healthcare data using machine learning system |
Publications (1)
Publication Number | Publication Date |
---|---|
AU2020102667A4 true AU2020102667A4 (en) | 2021-01-14 |
Family
ID=74103573
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
AU2020102667A Ceased AU2020102667A4 (en) | 2020-10-11 | 2020-10-11 | Adversarial training for large scale healthcare data using machine learning system |
Country Status (1)
Country | Link |
---|---|
AU (1) | AU2020102667A4 (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112807000A (en) * | 2021-02-04 | 2021-05-18 | 首都师范大学 | Robust electroencephalogram signal generation method and device |
CN113627576A (en) * | 2021-10-08 | 2021-11-09 | 平安科技(深圳)有限公司 | Code scanning information detection method, device, equipment and storage medium |
CN113706583A (en) * | 2021-09-01 | 2021-11-26 | 上海联影医疗科技股份有限公司 | Image processing method, image processing device, computer equipment and storage medium |
DE202023101305U1 (en) | 2023-03-16 | 2023-05-23 | Lulwah Mohammed Alkwai | An intelligent health and fitness data management system using artificial intelligence with IoT devices |
WO2024008043A1 (en) * | 2022-07-05 | 2024-01-11 | 浙江大学 | Automated clinical data generation method and system based on causal relationship mining |
-
2020
- 2020-10-11 AU AU2020102667A patent/AU2020102667A4/en not_active Ceased
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112807000A (en) * | 2021-02-04 | 2021-05-18 | 首都师范大学 | Robust electroencephalogram signal generation method and device |
CN112807000B (en) * | 2021-02-04 | 2023-02-28 | 首都师范大学 | Method and device for generating robust electroencephalogram signals |
CN113706583A (en) * | 2021-09-01 | 2021-11-26 | 上海联影医疗科技股份有限公司 | Image processing method, image processing device, computer equipment and storage medium |
CN113706583B (en) * | 2021-09-01 | 2024-03-22 | 上海联影医疗科技股份有限公司 | Image processing method, device, computer equipment and storage medium |
CN113627576A (en) * | 2021-10-08 | 2021-11-09 | 平安科技(深圳)有限公司 | Code scanning information detection method, device, equipment and storage medium |
CN113627576B (en) * | 2021-10-08 | 2022-01-18 | 平安科技(深圳)有限公司 | Code scanning information detection method, device, equipment and storage medium |
WO2024008043A1 (en) * | 2022-07-05 | 2024-01-11 | 浙江大学 | Automated clinical data generation method and system based on causal relationship mining |
DE202023101305U1 (en) | 2023-03-16 | 2023-05-23 | Lulwah Mohammed Alkwai | An intelligent health and fitness data management system using artificial intelligence with IoT devices |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
AU2020102667A4 (en) | Adversarial training for large scale healthcare data using machine learning system | |
US11645833B2 (en) | Generative adversarial network medical image generation for training of a classifier | |
Kukleva et al. | Unsupervised learning of action classes with continuous temporal embedding | |
US10540578B2 (en) | Adapting a generative adversarial network to new data sources for image classification | |
US10937540B2 (en) | Medical image classification based on a generative adversarial network trained discriminator | |
US11210781B2 (en) | Methods and devices for reducing dimension of eigenvectors and diagnosing medical images | |
CN109741332B (en) | Man-machine cooperative image segmentation and annotation method | |
US10929708B2 (en) | Deep learning network for salient region identification in images | |
Yan et al. | Modeling annotator expertise: Learning when everybody knows a bit of something | |
EP3486838A1 (en) | System and method for semi-supervised conditional generative modeling using adversarial networks | |
CN109102490B (en) | Automatic image registration quality assessment | |
Guan et al. | Discriminative feature learning for thorax disease classification in chest X-ray images | |
KR20180025093A (en) | A method and apparatus for machine learning based on weakly supervised learning | |
JP7282212B2 (en) | Method for learning deep learning network by AI and learning device using the same | |
Viji et al. | RETRACTED ARTICLE: An improved approach for automatic spine canal segmentation using probabilistic boosting tree (PBT) with fuzzy support vector machine | |
Puch et al. | Few-shot learning with deep triplet networks for brain imaging modality recognition | |
CN110543912A (en) | Method for automatically acquiring cardiac cycle video in fetal key section ultrasonic video | |
Zamani et al. | Towards Applicability of Information Communication Technologies in Automated Disease Detection. | |
Zhang et al. | Multiple sclerosis lesion segmentation-a survey of supervised CNN-based methods | |
Ji et al. | PRSNet: part relation and selection network for bone age assessment | |
Gu et al. | Efficient echocardiogram view classification with sampling-free uncertainty estimation | |
Wollek et al. | A knee cannot have lung disease: out-of-distribution detection with in-distribution voting using the medical example of chest X-ray classification | |
CN113160135A (en) | Intelligent colon lesion identification method, system and medium based on unsupervised migration image classification | |
Deshpande et al. | A method for evaluating deep generative models of images via assessing the reproduction of high-order spatial context | |
Huberman-Spiegelglas et al. | Single image object counting and localizing using active-learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
FGI | Letters patent sealed or granted (innovation patent) | ||
MK22 | Patent ceased section 143a(d), or expired - non payment of renewal fee or expiry |