WO2020252898A1 - 眼底oct影像增强方法、装置、设备及存储介质 - Google Patents

眼底oct影像增强方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2020252898A1
WO2020252898A1 PCT/CN2019/102532 CN2019102532W WO2020252898A1 WO 2020252898 A1 WO2020252898 A1 WO 2020252898A1 CN 2019102532 W CN2019102532 W CN 2019102532W WO 2020252898 A1 WO2020252898 A1 WO 2020252898A1
Authority
WO
WIPO (PCT)
Prior art keywords
oct image
fundus oct
target
target fundus
feature
Prior art date
Application number
PCT/CN2019/102532
Other languages
English (en)
French (fr)
Inventor
成冠举
高鹏
谢国彤
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2020252898A1 publication Critical patent/WO2020252898A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/90—Dynamic range modification of images or parts thereof
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30041—Eye; Retina; Ophthalmic

Definitions

  • This application relates to the field of image enhancement, in particular to methods, devices, equipment, and storage media for fundus OCT image enhancement.
  • OCT optical coherence tomography
  • This application provides a fundus OCT image enhancement method, device, equipment, and storage medium, which are used to improve the authenticity of the new fundus OCT image generated, avoid excessive differences from the original fundus OCT image, and solve the problem of too little real data and insufficient data Balance problems and improve image processing efficiency.
  • the first aspect of the embodiments of the present application provides a method for enhancing fundus OCT images, including: taking original fundus optical coherence tomography OCT images; constructing a generator and a discriminator through a preset deep learning network model; and using the generator
  • the preset random noise is converted into a target fundus OCT image; the discriminator determines whether the target fundus OCT image is true; if the target fundus OCT image is true, the target fundus OCT image is retained.
  • the second aspect of the embodiments of the present application provides a fundus OCT image intensification device, including: an acquisition unit for acquiring original fundus optical coherence tomography technology OCT images; a construction unit for constructing a preset deep learning network model A generator and a discriminator; a conversion unit for converting preset random noise into a target fundus OCT image by the generator; a judgment unit for judging whether the target fundus OCT image is true or not by the discriminator; reserved The unit is used to retain the target fundus OCT image if the target fundus OCT image is real.
  • the third aspect of the embodiments of the present application provides a fundus OCT image intensification device, including a memory, a processor, and a computer program stored on the memory and running on the processor, and the processor executes the The computer program realizes the above-mentioned fundus OCT image enhancement method.
  • the fourth aspect of the embodiments of the present application provides a computer-readable storage medium having instructions stored in the computer-readable storage medium.
  • the computer executes the above-mentioned fundus OCT image enhancement method. step.
  • the original fundus optical coherence tomography technology OCT image is obtained; the generator and discriminator are constructed through the preset deep learning network model; the preset random noise is converted into the target fundus OCT through the generator Image: Determine whether the target fundus OCT image is real through the discriminator; if the target fundus OCT image is true, then keep the target fundus OCT image.
  • a new fundus OCT image is generated based on the original fundus OCT image, which improves the authenticity of the new fundus OCT image, avoids excessive differences with the original fundus OCT image, and solves the problem of too little real data and data imbalance, and improves Image processing efficiency.
  • FIG. 1 is a schematic diagram of an embodiment of a method for enhancing a fundus OCT image in an embodiment of the application
  • FIG. 2 is a schematic diagram of another embodiment of a method for enhancing fundus OCT images in an embodiment of this application;
  • FIG. 3 is a schematic diagram of an embodiment of a fundus OCT image intensifier in an embodiment of the application
  • FIG. 4 is a schematic diagram of another embodiment of the fundus OCT image intensifying device in the embodiment of the application.
  • FIG. 5 is a schematic diagram of an embodiment of a fundus OCT image intensification device in an embodiment of the application.
  • This application provides a fundus OCT image enhancement method, device, equipment, and storage medium, which are used to improve the authenticity of the new fundus OCT image generated, avoid excessive differences from the original fundus OCT image, and solve the problem of too little real data and insufficient data Balance problems and improve image processing efficiency.
  • the flowchart of the OCT image enhancement method of the fundus includes:
  • the fundus OCT image intensifier device obtains the original fundus optical coherence tomography technology OCT image.
  • the fundus OCT image enhancement device obtains original fundus optical coherence tomography (OCT) images, and the original fundus OCT images are directly obtained by the OCT device without any processing.
  • OCT optical coherence tomography
  • Time-domain OCT is to superimpose and interfere with the optical signal reflected from the tissue at the same time and the optical signal reflected from the reference mirror, and then image.
  • Frequency domain OCT means that the reference mirror of the reference arm is fixed, and the signal interference is realized by changing the frequency of the light wave of the light source.
  • the original fundus OCT image can be obtained in a variety of ways, which can be obtained in a TD-OCT mode, or can also be obtained in an FD-OCT mode. The specific obtaining method is not limited here.
  • the execution subject of this application may be a fundus OCT image intensification device, or a terminal or a server, which is not specifically limited here.
  • the embodiment of the present application takes the fundus OCT image intensifier as the execution subject as an example for description.
  • the fundus OCT image enhancement device constructs a generator and a discriminator through a preset deep learning network model, where the generator is used to generate a target fundus OCT image, and the discriminator is used to discriminate whether the generated target fundus OCT image is authentic.
  • the essence of deep learning is to learn more useful features by constructing machine learning models with many hidden layers and massive training data, thereby ultimately improving the accuracy of classification or prediction.
  • the difference of deep learning is: 1) The depth of the model structure is emphasized, usually there are 5 layers, 6 layers, or even 10 layers of hidden nodes; 2) The importance of feature learning is clearly highlighted In other words, by layer-by-layer feature transformation, the feature representation of the sample in the original space is transformed into a new feature space, thus making classification or prediction easier.
  • using big data to learn features can better describe the rich internal information of the data. How to build a deep learning network model can refer to the existing technology, which will not be repeated here.
  • the fundus OCT image enhancement device converts the preset random noise into the target fundus OCT image through the generator.
  • the generator generates a target fundus OCT image that obeys the original data distribution and characteristics based on the original data distribution and characteristics.
  • the original data distribution and characteristics can be obtained from the original fundus OCT image, which will not be repeated here.
  • both the generator and the discriminator are components of a generative adversarial net (GAN), where the generator takes random noise as input and tries to generate sample data.
  • the discriminator takes real data or generated data as input, and predicts whether the current input is real data or generated data.
  • the generator needs to create as much samples as possible to confuse the discriminator, and the discriminator as much as possible to identify the samples from the generator.
  • the generator and the discriminator can finally reach a balance.
  • the preset random noise may be currently acquired, or may be generated in advance and stored on the fundus OCT image intensifier, and the specifics are not limited here.
  • the fundus OCT image enhancement device judges whether the target fundus OCT image is real through the discriminator.
  • the fundus OCT image intensifier device inputs the acquired original fundus OCT image and the target fundus OCT image generated by the generator to the discriminator, and the discriminator will obtain the distribution characteristics of the original fundus OCT image and the distribution of the target fundus OCT image Features:
  • the discriminator compares the distribution characteristics of the target fundus OCT image with that of the original fundus OCT image; judges whether the target fundus OCT image is true.
  • the fundus OCT image intensifier device can also compare other parameters of the target fundus OCT image and the original fundus OCT image.
  • the discriminator uses the cross-entropy loss function to evaluate the distribution difference between the target fundus OCT image and the original fundus OCT image. The smaller the difference between the two data distributions, the smaller the cross-entropy value, indicating that the generated target fundus OCT image and the original fundus OCT image are more similar. When the cross entropy is less than the preset threshold, it can be determined that the target fundus OCT image is true.
  • Other parameters can also be compared to determine the similarity between the target fundus OCT image and the original fundus OCT image, which is not limited here.
  • the target fundus OCT image is real, then keep the target fundus OCT image.
  • the fundus OCT image intensifier device retains the target fundus OCT image, that is, the target fundus OCT image is regarded as a qualified image. Specifically, if the fundus OCT image intensifier determines that the generated target fundus OCT image is real, then the target fundus OCT image and the original fundus OCT image are placed in the same image set to facilitate the next call.
  • a new fundus OCT image is generated based on the original fundus OCT image, which improves the authenticity of the new fundus OCT image, avoids excessive differences with the original fundus OCT image, and solves the problem of too little real data and data imbalance, and improves Image processing efficiency.
  • FIG. 2 another flowchart of the fundus OCT image enhancement method provided by the embodiment of the present application, which specifically includes:
  • the fundus OCT image intensifier device obtains the original fundus optical coherence tomography technology OCT image.
  • the fundus OCT image enhancement device obtains original fundus optical coherence tomography (OCT) images, and the original fundus OCT images are directly obtained by the OCT device without any processing.
  • OCT optical coherence tomography
  • Time-domain OCT is to superimpose and interfere with the optical signal reflected from the tissue at the same time and the optical signal reflected from the reference mirror, and then image.
  • Frequency domain OCT means that the reference mirror of the reference arm is fixed, and the signal interference is realized by changing the frequency of the light wave of the light source.
  • the original fundus OCT image can be obtained in a variety of ways, which can be obtained in a TD-OCT mode, or can also be obtained in an FD-OCT mode. The specific obtaining method is not limited here.
  • the execution subject of this application may be a fundus OCT image intensification device, or a terminal or a server, which is not specifically limited here.
  • the embodiment of the present application takes the fundus OCT image intensifier as the execution subject as an example for description.
  • the fundus OCT image enhancement device constructs a generator and a discriminator through a preset deep learning network model, where the generator is used to generate a target fundus OCT image, and the discriminator is used to discriminate whether the generated target fundus OCT image is authentic.
  • the essence of deep learning is to learn more useful features by constructing machine learning models with many hidden layers and massive training data, thereby ultimately improving the accuracy of classification or prediction.
  • the difference of deep learning is: 1) The depth of the model structure is emphasized, usually there are 5 layers, 6 layers, or even 10 layers of hidden nodes; 2) The importance of feature learning is clearly highlighted In other words, by layer-by-layer feature transformation, the feature representation of the sample in the original space is transformed into a new feature space, thus making classification or prediction easier.
  • using big data to learn features can better describe the rich internal information of the data. How to build a deep learning network model can refer to the existing technology, which will not be repeated here.
  • the fundus OCT image enhancement device converts the preset random noise into the target fundus OCT image through the generator.
  • the generator generates a target fundus OCT image that obeys the original data distribution and characteristics based on the original data distribution and characteristics.
  • the original data distribution and characteristics can be obtained from the original fundus OCT image, which will not be repeated here.
  • both the generator and the discriminator are components of a generative adversarial net (GAN), where the generator takes random noise as input and tries to generate sample data.
  • the discriminator takes real data or generated data as input, and predicts whether the current input is real data or generated data.
  • the generator needs to create as much samples as possible to confuse the discriminator, and the discriminator as much as possible to identify the samples from the generator.
  • the generator and the discriminator can finally reach a balance.
  • the preset random noise may be currently acquired, or may be generated in advance and stored on the fundus OCT image intensifier, and the specifics are not limited here.
  • the fundus OCT image enhancement device analyzes the target fundus OCT image through the discriminator to determine the distribution characteristics of the target fundus OCT image.
  • Distribution characteristics include shape characteristics, horizontal characteristics and difference characteristics.
  • the shape features include skewness and kurtosis. For example, when the skewness is 0, the distribution is symmetric; when the skewness is greater than 0, it means that there is a long tail on the right side of the distribution; when the skewness is less than 0, it means that the distribution is symmetrical. There is a long tail on the left.
  • the level features include descriptive statistics, which include the mean, median, quartile, and mode.
  • the mean includes simple average and weighted average.
  • the median can be sorted and then viewed.
  • the parity is calculated, the quartiles are the values at the quarter and three-quarter positions after the data is sorted, and the mode is the value with the most occurrences.
  • the difference characteristics include range, interquartile, variance, standard deviation, dispersion coefficient, and standard score. For example, the smaller the value of the interquartile, the more concentrated the distribution. Other characteristics of the difference feature can refer to the existing technology. I won't repeat them here.
  • the fundus OCT image intensifier device determines whether the target fundus OCT image is real according to the distribution characteristics of the target fundus OCT image. Specifically, the fundus OCT image intensifier device obtains the distribution characteristics of the original fundus OCT image.
  • the distribution characteristics include shape characteristics, horizontal characteristics, and difference characteristics; the shape, horizontal, and difference characteristics of the original fundus OCT image are compared with those of the target fundus OCT image.
  • the shape feature, the horizontal feature and the difference feature are compared separately to obtain the shape feature similarity value, the horizontal feature similarity value and the difference feature similarity value; the shape feature similarity value, the horizontal feature similarity value and the difference feature similarity value are The respective weights are calculated to obtain the distribution feature similarity; if the distribution feature similarity is greater than the threshold, the discriminator determines that the target fundus OCT image is true.
  • the fundus OCT image intensifier device can also compare other parameters of the target fundus OCT image and the original fundus OCT image.
  • the discriminator uses the cross-entropy loss function to evaluate the distribution difference between the target fundus OCT image and the original fundus OCT image. The smaller the difference between the two data distributions, the smaller the cross-entropy value, indicating that the generated target fundus OCT image and the original fundus OCT image are more similar.
  • the fundus OCT image enhancement device generates the cross entropy loss function of the distribution feature of the target fundus OCT image and the distribution feature of the original fundus OCT image; the discriminator determines whether the value of the cross entropy loss function is less than the threshold; if the cross entropy loss function is If the value is less than the threshold, the discriminator determines that the target fundus OCT image is true.
  • Other parameters can also be compared to determine the similarity between the target fundus OCT image and the original fundus OCT image, which is not limited here.
  • the target fundus OCT image is real, keep the target fundus OCT image.
  • the fundus OCT image intensifier device retains the target fundus OCT image, that is, the target fundus OCT image is regarded as a qualified image. Specifically, if the fundus OCT image intensifier determines that the generated target fundus OCT image is real, then the target fundus OCT image and the original fundus OCT image are placed in the same image set to facilitate the next call.
  • the fundus OCT image intensifier device stores the real target fundus OCT image and the original fundus OCT image in the same fundus OCT image collection, so that the image collection can be called next time.
  • the target fundus OCT image is forged by the generator.
  • the discriminator considers the target fundus OCT image to be real, but because it is the target
  • the fundus OCT image is forged (generated), so it is more or less different from the real original fundus OCT image (original data), which ultimately enhances the data. In other words, it enriches the data diversity of the original image collection.
  • the target fundus OCT image is not real, re-input the target fundus OCT image to the generator for image optimization.
  • the fundus OCT image intensifier device re-inputs the target fundus OCT image to the generator for image optimization, until the target fundus OCT image generated by the generator is determined as a real image by the discriminator.
  • the fundus OCT image intensifier device re-inputs the target fundus OCT image to the generator for further modification to improve the similarity between the target fundus OCT image and the original fundus OCT image.
  • the generator first converts the preset random noise into intermediate data of the same size as the original fundus OCT image. Based on this intermediate data, it generates the target fundus OCT image and inputs it to the discriminator to determine if the discriminator considers the target fundus OCT The image is different from the original fundus OCT image.
  • the generator will continue to generate the target fundus OCT image based on the target fundus OCT image until the discriminator determines that the generated target fundus OCT image is the same as the original fundus OCT image or the similarity meets the requirements.
  • a new fundus OCT image is generated based on the original fundus OCT image, which improves the authenticity of the new fundus OCT image, avoids excessive differences with the original fundus OCT image, and solves the problem of too little real data and data imbalance, and improves Image processing efficiency.
  • the performance of the training model and the generalization effect can be improved by adding the enhanced image data generated by the generation confrontation network.
  • an embodiment of the fundus OCT image enhancement device in the embodiment of the application includes:
  • the obtaining unit 301 is used to obtain original OCT images of the fundus optical coherence tomography
  • the construction unit 302 is used to construct a generator and a discriminator through a preset deep learning network model
  • a conversion unit 303 configured to convert preset random noise into a target fundus OCT image through the generator
  • the judging unit 304 is configured to judge whether the target fundus OCT image is real through the discriminator;
  • the retaining unit 305 is configured to retain the target fundus OCT image if the target fundus OCT image is real.
  • a new fundus OCT image is generated based on the original fundus OCT image, which improves the authenticity of the new fundus OCT image, avoids excessive differences with the original fundus OCT image, and solves the problem of too little real data and data imbalance, and improves Image processing efficiency.
  • FIG. 4 another embodiment of the fundus OCT image intensifier in the embodiment of the present application includes:
  • the obtaining unit 301 is used to obtain original OCT images of the fundus optical coherence tomography
  • the construction unit 302 is used to construct a generator and a discriminator through a preset deep learning network model
  • a conversion unit 303 configured to convert preset random noise into a target fundus OCT image through the generator
  • the judging unit 304 is configured to judge whether the target fundus OCT image is real through the discriminator;
  • the retaining unit 305 is configured to retain the target fundus OCT image if the target fundus OCT image is real.
  • the fundus OCT image enhancement device further includes:
  • the optimization unit 306 is configured to re-input the target fundus OCT image to the generator for image optimization if the target fundus OCT image is not real.
  • the judging unit 304 includes:
  • the first determining module 3041 is configured to analyze the target fundus OCT image through the discriminator to determine the distribution feature of the target fundus OCT image;
  • the first determining module 3042 is configured to determine whether the target fundus OCT image is true based on the distribution feature of the target fundus OCT image by the discriminator.
  • the first judgment module 3042 is specifically configured to:
  • the distribution characteristics of the original fundus OCT image including shape feature, horizontal feature, and difference feature; compare the shape feature, horizontal feature, and difference feature of the original fundus OCT image with the shape feature of the target fundus OCT image , Horizontal feature and difference feature are respectively compared to obtain shape feature similarity value, horizontal feature similarity value and difference feature similarity value; compare the shape feature similarity value, the horizontal feature similarity value and the difference feature The similarity value is calculated according to the respective weight to obtain the distribution feature similarity; if the distribution feature similarity is greater than the threshold, it is determined that the target fundus OCT image is true.
  • the judging unit 304 further includes:
  • a generating module 3043 configured to generate a cross-entropy loss function of the distribution feature of the target fundus OCT image and the distribution feature of the original fundus OCT image;
  • the second judgment module 3044 is configured to judge whether the value of the cross-entropy loss function is less than a threshold value through the discriminator;
  • the second determining module 3045 if the value of the cross-entropy loss function is less than the threshold, is used to determine that the target fundus OCT image is true through the discriminator.
  • the generating module 3043 is specifically used for:
  • the fundus OCT image enhancement device further includes:
  • the storage unit 307 is configured to store the real target fundus OCT image and the original fundus OCT image in the same fundus OCT image set.
  • a new fundus OCT image is generated based on the original fundus OCT image, which improves the authenticity of the new fundus OCT image, avoids excessive differences with the original fundus OCT image, and solves the problem of too little real data and data imbalance, and improves Image processing efficiency.
  • the performance of the training model and the generalization effect can be improved by adding the enhanced image data generated by the generation confrontation network.
  • FIG. 5 is a schematic structural diagram of a fundus OCT image intensification device provided by an embodiment of the present application.
  • the fundus OCT image intensification device 500 may have relatively large differences due to different configurations or performance, and may include one or more processors (central Processing units, CPU) 501 (for example, one or more processors) and memory 509, and one or more storage media 508 for storing application programs 507 or data 506 (for example, one or one storage device with a large amount of storage).
  • the memory 509 and the storage medium 508 may be short-term storage or persistent storage.
  • the program stored in the storage medium 508 may include one or more modules (not shown in the figure), and each module may include a series of command operations in the fundus OCT image enhancement device.
  • the processor 501 may be configured to communicate with the storage medium 508, and execute a series of instruction operations in the storage medium 508 on the fundus OCT image intensification device 500.
  • the fundus OCT image enhancement device 500 may also include one or more power supplies 502, one or more wired or wireless network interfaces 503, one or more input and output interfaces 504, and/or one or more operating systems 505, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc.
  • operating systems 505 such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc.
  • the processor 501 can execute the functions of the acquisition unit 301, the construction unit 302, the conversion unit 303, the judgment unit 304, and the optimization unit 306 in the foregoing embodiment.
  • the processor 501 is the control center of the fundus OCT image enhancement device, and can perform processing according to the set fundus OCT image enhancement method.
  • the processor 501 uses various interfaces and lines to connect various parts of the entire fundus OCT image intensification device, and executes fundus OCT by running or executing the software programs and/or modules stored in the memory 509 and calling the data stored in the memory 509.
  • the various functions and processing data of the image enhancement equipment improve the authenticity of the new fundus OCT images generated, avoid too much difference with the original fundus OCT images, solve the problem of too little real data and data imbalance, and improve the efficiency of image processing.
  • the storage medium 508 and the memory 509 are both carriers for storing data. In the embodiment of the present application, the storage medium 508 may refer to an internal memory with a small storage capacity but high speed, and the storage 509 may have a large storage capacity but a slow storage speed. External memory.
  • the memory 509 may be used to store software programs and modules, and the processor 501 executes various functional applications and data processing of the fundus OCT image intensification device 500 by running the software programs and modules stored in the memory 509.
  • the memory 509 may mainly include a program storage area and a data storage area.
  • the storage program area may store an operating system, at least one application program required by a function (for example, a generator converts preset random noise into a target fundus OCT image), etc. ;
  • the storage data area can store data (such as generators and discriminators) created based on the use of fundus OCT image enhancement equipment.
  • the memory 509 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices.
  • a non-volatile memory such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices.
  • the computer-readable storage medium may be a non-volatile computer-readable storage medium.
  • the computer-readable storage medium stores instructions. When the instructions run on a computer At the time, the computer is caused to execute the following steps of the fundus OCT image enhancement method:
  • the target fundus OCT image is real, then the target fundus OCT image is retained.
  • the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.
  • the computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium.
  • the computer instructions may be transmitted from a website, computer, server, or data center. Transmission to another website site, computer, server or data center via wired (such as coaxial cable, optical fiber, twisted pair) or wireless (such as infrared, wireless, microwave, etc.).
  • the computer-readable storage medium may be any available medium that can be stored by a computer or a data storage device such as a server or data center integrated with one or more available media.
  • the usable medium may be a magnetic medium (for example, a floppy disk, a hard disk, and a magnetic tape), an optical medium (for example, an optical disc), or a semiconductor medium (for example, a solid state disk (SSD)).
  • the disclosed system, device, and method may be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division, and there may be other divisions in actual implementation, for example, multiple units or components can be combined or It can be integrated into another system, or some features can be ignored or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Eye Examination Apparatus (AREA)
  • Image Analysis (AREA)

Abstract

一种眼底OCT影像增强方法、装置、设备及存储介质，用于提高生成的新眼底OCT影像的真实性，避免和原始眼底OCT影像差异过大，解决真实数据太少及数据不均衡问题，提高影像处理效率。该方法包括：获取原始眼底光学相干断层扫描技术OCT影像（101）；通过预置的深度学习网络模型构建生成器和鉴别器（102）；通过生成器将预置的随机噪声转换成目标眼底OCT影像（103）；通过鉴别器判断目标眼底OCT影像是否真实（104）；若目标眼底OCT影像真实，则将目标眼底OCT影像保留（105）。

Description

眼底OCT影像增强方法、装置、设备及存储介质
本申请要求于2019年6月18日提交中国专利局、申请号为201910524224.5、发明名称为“眼底OCT影像增强方法、装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在申请中。
技术领域
本申请涉及图像增强领域，尤其涉及眼底OCT影像增强方法、装置、设备及存储介质。
背景技术
随着人工智能的快速发展，人工智能的应用领域也越来越多，人工智能在医疗领域也得到了广泛应用。人工智能在医疗领域的应用经常面临医疗影像数据少，各类医疗影像数据不均衡的问题。
目前对医疗领域中光学相干断层扫描技术(optical coherence tomography，OCT)影像的数据增强通常使用传统图像处理方法，比如对原始图像进行翻转、平移、扭曲、灰度处理等，并将处理后的图像保存以达到增强数据的目的。
发明人意识到这种方法有两个缺点，一是通过灰度处理、平移等方法增加的数据和原始数据相似度过高，导致模型训练出现过拟合现象，且模型在实际使用中泛化效果不好；二是通过翻转、扭曲等方法增加的数据和原始数据差异过大，不真实，影像处理效率低。
发明内容
本申请提供了一种眼底OCT影像增强方法、装置、设备及存储介质，用于提高生成的新眼底OCT影像的真实性，避免和原始眼底OCT影像差异过大，解决真实数据太少及数据不均衡问题，提高影像处理效率。
本申请实施例的第一方面提供一种眼底OCT影像增强方法，包括：取原始眼底光学相干断层扫描技术OCT影像；通过预置的深度学习网络模型构建生成器和鉴别器；通过所述生成器将预置的随机噪声转换成目标眼底OCT影像；通过所述鉴别器判断所述目标眼底OCT影像是否真实；若所述目标眼底OCT影像真实，则将所述目标眼底OCT影像保留。
本申请实施例的第二方面提供了一种眼底OCT影像增强装置，包括：获取单元，用于获取原始眼底光学相干断层扫描技术OCT影像；构建单元，用于通过预置的深度学习网络模型构建生成器和鉴别器；转换单元，用于通过所述生成器将预置的随机噪声转换成目标眼底OCT影像；判断单元，用于通过所述鉴别器判断所述目标眼底OCT影像是否真实；保留单元，若所述目标眼底OCT影像真实，则用于将所述目标眼底OCT影像保留。
本申请实施例的第三方面提供了一种眼底OCT影像增强设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现上述眼底OCT影像增强方法。
本申请实施例的第四方面提供了一种计算机可读存储介质，所述计算机可读存储介质中存储有指令，当所述指令在计算机上运行时，使得计算机执行上述眼底OCT影像增强方法的步骤。
本申请实施例提供的技术方案中，获取原始眼底光学相干断层扫描技术OCT影像；通过预置的深度学习网络模型构建生成器和鉴别器；通过生成器将预置的随机噪声转换成目标眼底OCT影像；通过鉴别器判断目标眼底OCT影像是否真实；若目标眼底OCT影像真实，则将目标眼底OCT影像保留。本申请实施例，根据原始眼底OCT影像生成新眼底OCT影像，提高了新眼底OCT影像的真实性，避免和原始眼底OCT影像差异过大，解决了真实数据太少及数据不均衡问题，提高了影像处理效率。
附图说明
图1为本申请实施例中眼底OCT影像增强方法的一个实施例示意图；
图2为本申请实施例中眼底OCT影像增强方法的另一个实施例示意图；
图3为本申请实施例中眼底OCT影像增强装置的一个实施例示意图；
图4为本申请实施例中眼底OCT影像增强装置的另一个实施例示意图；
图5为本申请实施例中眼底OCT影像增强设备的一个实施例示意图。
具体实施方式
本申请提供了一种眼底OCT影像增强方法、装置、设备及存储介质，用于提高生成的新眼底OCT影像的真实性，避免和原始眼底OCT影像差异过大，解决真实数据太少及数据不均衡问题，提高影像处理效率。
为了使本技术领域的人员更好地理解本申请方案，下面将结合本申请实施例中的附图，对本申请实施例进行描述。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”、“第三”、“第四”等(如果存在)是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的实施例能够以除了在这里图示或描述的内容以外的顺序实施。此外，术语“包括”或“具有”及其任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不 必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
请参阅图1，本申请实施例提供的眼底OCT影像增强方法的流程图，具体包括：
101、获取原始眼底光学相干断层扫描技术OCT影像。
眼底OCT影像增强装置获取原始眼底光学相干断层扫描技术OCT影像。其中，眼底OCT影像增强装置获取原始眼底光学相干断层扫描技术(optical coherence tomography，OCT)影像，该原始眼底OCT影像由OCT设备直接得到，未经过任何处理。
目前OCT分为两大类：时域OCT(time domain optical coherence tomography，TD-OCT)和频域OCT(frequency domain optical coherence tomography，FD-OCT)。时域OCT是把在同一时间从组织中反射回来的光信号与参照反光镜反射回来的光信号叠加、干涉，然后成像。频域OCT是参考臂的参照反光镜固定不动，通过改变光源光波的频率来实现信号的干涉。本申请实施例可以通过多种方式获取原始眼底OCT影像，可以通过TD-OCT方式获取，还可以通过FD-OCT方式获取，具体采用何种获取方式此处不做限定。
可以理解的是，本申请的执行主体可以为眼底OCT影像增强装置，还可以是终端或者服务器，具体此处不做限定。本申请实施例以眼底OCT影像增强装置为执行主体为例进行说明。
102、通过预置的深度学习网络模型构建生成器和鉴别器。
眼底OCT影像增强装置通过预置的深度学习网络模型构建生成器和鉴别器，其中，该生成器用于生成目标眼底OCT影像，该鉴别器用于鉴别生成的目标眼底OCT影像是否真实。
需要说明的是，深度学习的实质，是通过构建具有很多隐层的机器学习模型和海量的训练数据，来学习更有用的特征，从而最终提升分类或预测的准确性。区别于传统的浅层学习，深度学习的不同在于：1)强调了模型结构的深度，通常有5层、6层，甚至10多层的隐层节点；2)明确突出了特征学习的重要性，也就是说，通过逐层特征变换，将样本在原空间的特征表示变换到一个新特征空间，从而使分类或预测更加容易。与人工规则构造特征的方法相比，利用大数据来学习特征，更能够刻画数据的丰富内在信息。如何构建深度学习网络模型可以参考现有技术，此处不再赘述。
103、通过生成器将预置的随机噪声转换成目标眼底OCT影像。
眼底OCT影像增强装置通过生成器将预置的随机噪声转换成目标眼底OCT影像。例如，生成器基于原始数据分布及特征生成服从原始数据分布及特征的目标眼底OCT影像，原始 数据分布及特征可以从原始眼底OCT影像中获取，此处不再赘述。
需要说明的是，生成器和鉴别器都是生成对抗网络(generative adversarial net，GAN)的组成部分，其中，生成器以随机噪声作为输入并试图生成样本数据。鉴别器以真实数据或者生成数据作为输入，并预测当前输入是真实数据还是生成数据。生成器需要尽可能造出样本迷惑鉴别器，而鉴别器则尽可能识别出来自生成器的样本。可选的，生成器和判别器最终能达到平衡。
可以理解的是，预置的随机噪声可以是当前获取的，也可以是提前生成并存储在眼底OCT影像增强装置上的，具体此处不做限定。
104、通过鉴别器判断目标眼底OCT影像是否真实。
眼底OCT影像增强装置通过鉴别器判断目标眼底OCT影像是否真实。具体的，眼底OCT影像增强装置将获取到的原始眼底OCT影像和生成器生成的目标眼底OCT影像输入到鉴别器，同时鉴别器会获取原始眼底OCT影像的分布特征，和目标眼底OCT影像的分布特征，鉴别器会比较目标眼底OCT影像的分布特征和原始眼底OCT影像的分布特征；判断目标眼底OCT影像是否真实。
需要说明的是，眼底OCT影像增强装置还可以比较目标眼底OCT影像和原始眼底OCT影像的其他参数，例如，鉴别器通过交叉熵损失函数评价目标眼底OCT影像和原始眼底OCT影像的分布差异，当两个数据分布差异越小则交叉熵值越小，说明生成的目标眼底OCT影像和原始眼底OCT影像相似度越高。当交叉熵小于预设的阈值时，则可以确定目标眼底OCT影像真实。还可以比较其他参数，以判断目标眼底OCT影像和原始眼底OCT影像的相似度，具体此处不做限定。
105、若目标眼底OCT影像真实，则将目标眼底OCT影像保留。
若目标眼底OCT影像真实，则眼底OCT影像增强装置将目标眼底OCT影像保留，即将目标眼底OCT影像作为合格的影像。具体的，若眼底OCT影像增强装置确定生成的目标眼底OCT影像为真实的，那么，就将该目标眼底OCT影像和原始眼底OCT影像放在同一个影像集合中，方便下次调用。
本申请实施例，根据原始眼底OCT影像生成新眼底OCT影像，提高了新眼底OCT影像的真实性，避免和原始眼底OCT影像差异过大，解决了真实数据太少及数据不均衡问题，提高了影像处理效率。
请参阅图2，本申请实施例提供的眼底OCT影像增强方法的另一个流程图，具体包括：
201、获取原始眼底光学相干断层扫描技术OCT影像。
眼底OCT影像增强装置获取原始眼底光学相干断层扫描技术OCT影像。其中，眼底OCT影像增强装置获取原始眼底光学相干断层扫描技术(optical coherence tomography，OCT)影像，该原始眼底OCT影像由OCT设备直接得到，未经过任何处理。
目前OCT分为两大类：时域OCT(time domain optical coherence tomography，TD-OCT)和频域OCT(frequency domain optical coherence tomography，FD-OCT)。时域OCT是把在同一时间从组织中反射回来的光信号与参照反光镜反射回来的光信号叠加、干涉，然后成像。频域OCT是参考臂的参照反光镜固定不动，通过改变光源光波的频率来实现信号的干涉。本申请实施例可以通过多种方式获取原始眼底OCT影像，可以通过TD-OCT方式获取，还可以通过FD-OCT方式获取，具体采用何种获取方式此处不做限定。
可以理解的是，本申请的执行主体可以为眼底OCT影像增强装置，还可以是终端或者服务器，具体此处不做限定。本申请实施例以眼底OCT影像增强装置为执行主体为例进行说明。
202、通过预置的深度学习网络模型构建生成器和鉴别器。
眼底OCT影像增强装置通过预置的深度学习网络模型构建生成器和鉴别器，其中，该生成器用于生成目标眼底OCT影像，该鉴别器用于鉴别生成的目标眼底OCT影像是否真实。
需要说明的是，深度学习的实质，是通过构建具有很多隐层的机器学习模型和海量的训练数据，来学习更有用的特征，从而最终提升分类或预测的准确性。区别于传统的浅层学习，深度学习的不同在于：1)强调了模型结构的深度，通常有5层、6层，甚至10多层的隐层节点；2)明确突出了特征学习的重要性，也就是说，通过逐层特征变换，将样本在原空间的特征表示变换到一个新特征空间，从而使分类或预测更加容易。与人工规则构造特征的方法相比，利用大数据来学习特征，更能够刻画数据的丰富内在信息。如何构建深度学习网络模型可以参考现有技术，此处不再赘述。
203、通过生成器将预置的随机噪声转换成目标眼底OCT影像。
眼底OCT影像增强装置通过生成器将预置的随机噪声转换成目标眼底OCT影像。例如，生成器基于原始数据分布及特征生成服从原始数据分布及特征的目标眼底OCT影像，原始数据分布及特征可以从原始眼底OCT影像中获取，此处不再赘述。
需要说明的是，生成器和鉴别器都是生成对抗网络(generative adversarial net，GAN)的组成部分，其中，生成器以随机噪声作为输入并试图生成样本数据。鉴别器以真实 数据或者生成数据作为输入，并预测当前输入是真实数据还是生成数据。生成器需要尽可能造出样本迷惑鉴别器，而鉴别器则尽可能识别出来自生成器的样本。可选的，生成器和判别器最终能达到平衡。
可以理解的是，预置的随机噪声可以是当前获取的，也可以是提前生成并存储在眼底OCT影像增强装置上的，具体此处不做限定。
204、通过鉴别器对目标眼底OCT影像进行分析，确定目标眼底OCT影像的分布特征。
眼底OCT影像增强装置通过鉴别器对目标眼底OCT影像进行分析，确定目标眼底OCT影像的分布特征。分布特征包括形状特征、水平特征和差异特征。其中，形状特征包括偏度和峰度，例如，当偏度为0时，说明分布对称；当偏度大于0时，说明分布的右侧有长尾；当偏度小于0时，说明分布的左侧有长尾。当峰度为0时，说明为标准正太分布；当峰度大于0时，说明为尖峰分布；当峰度小于0时，说明为扁平分布。其中，水平特征包括描述统计量，该描述统计量包括平均数、中位数、四分位数和众数，例如，平均数包括简单平均数和加权平均数，中位数可以先排序然后看奇偶进行计算得到，四分位数为数据排序后处于四分之一和四分之三位置上的值，众数为出现次数最多的值。其中，差异特征包括极差、四分位差、方差、标准差、离散系数、标准分数，例如，四分位差的值越小分布越集中，其他差异特征的特性可参照现有技术，具体此处不再赘述。
205、通过鉴别器基于目标眼底OCT影像的分布特征判断目标眼底OCT影像是否真实。
眼底OCT影像增强装置根据目标眼底OCT影像的分布特征判断目标眼底OCT影像是否真实。具体的，眼底OCT影像增强装置获取原始眼底OCT影像的分布特征，分布特征包括形状特征、水平特征和差异特征；将原始眼底OCT影像的形状特征、水平特征和差异特征，与目标眼底OCT影像的形状特征、水平特征和差异特征分别进行比较，得到形状特征相似度值、水平特征相似度值和差异特征相似度值；将形状特征相似度值、水平特征相似度值和差异特征相似度值按照各自的权重计算得到分布特征相似度；若分布特征相似度大于阈值，则通过鉴别器确定目标眼底OCT影像真实。
需要说明的是，眼底OCT影像增强装置还可以比较目标眼底OCT影像和原始眼底OCT影像的其他参数，例如，鉴别器通过交叉熵损失函数评价目标眼底OCT影像和原始眼底OCT影像的分布差异，当两个数据分布差异越小则交叉熵值越小，说明生成的目标眼底OCT影像和原始眼底OCT影像相似度越高。具体的，眼底OCT影像增强装置生成目标眼底OCT影像的分布特征和原始眼底OCT影像的分布特征的交叉熵损失函数；通过鉴别器判断交叉熵 损失函数的值是否小于阈值；若交叉熵损失函数的值小于阈值，则通过鉴别器确定目标眼底OCT影像真实。其中，所述生成目标眼底OCT影像的分布特征和原始眼底OCT影像的分布特征的交叉熵损失函数的过程包括：将目标眼底OCT影像的分布特征值作为预测值q(x)；将原始眼底OCT影像的分布特征值作为真实值p(x)；生成预测值q(x)和真实值p(x)的交叉熵函数H(p,q)，其中，x为正整数，具体表达式为：H(p,q)＝-∑p(x)log(q(x))。当交叉熵小于预设的阈值时，则可以确定目标眼底OCT影像真实。还可以比较其他参数，以判断目标眼底OCT影像和原始眼底OCT影像的相似度，具体此处不做限定。
206、若目标眼底OCT影像真实，则将目标眼底OCT影像保留。
若目标眼底OCT影像真实，则眼底OCT影像增强装置将目标眼底OCT影像保留，即将目标眼底OCT影像作为合格的影像。具体的，若眼底OCT影像增强装置确定生成的目标眼底OCT影像为真实的，那么，就将该目标眼底OCT影像和原始眼底OCT影像放在同一个影像集合中，方便下次调用。
207、将真实的目标眼底OCT影像和所述原始眼底OCT影像存储在同一个眼底OCT影像集合中。
眼底OCT影像增强装置将真实的目标眼底OCT影像和所述原始眼底OCT影像存储在同一个眼底OCT影像集合中，方便下一次调用该影像集合。
需要说明的是，目标眼底OCT影像是生成器伪造生成的，只要目标眼底OCT影像和原始眼底OCT影像一样(服从原始数据分布)，那么鉴别器就认为目标眼底OCT影像是真实，但因为是目标眼底OCT影像伪造出来的(生成的)，所以和真实的原始眼底OCT影像(原始数据)或多或少会有差异，最终起到了增强数据的效果。也就是说，丰富了原有影像集合的数据的多样性。
208、若目标眼底OCT影像不真实，则将目标眼底OCT影像重新输入到生成器进行影像优化。
若目标眼底OCT影像不真实，则眼底OCT影像增强装置将目标眼底OCT影像重新输入到生成器进行影像优化，直至生成器生成的改目标眼底OCT影像被鉴别器确定为真实影像。
具体的，当目标眼底OCT影像不真实，则眼底OCT影像增强装置将目标眼底OCT影像重新输入到生成器，进行进一步的修改，提高目标眼底OCT影像与原始眼底OCT影像的相似度。例如，首先生成器会将预置的随机噪声转换成和原始眼底OCT影像同样大小的中间数据，基于这个中间数据生成目标眼底OCT影像并将其输入鉴别器判断，如果鉴别器认为 该目标眼底OCT影像和原始眼底OCT影像不同，生成器会基于该目标眼底OCT影像继续生成，直到鉴别器认为生成的目标眼底OCT影像和原始眼底OCT影像相同或相似度达到要求为止。
本申请实施例，根据原始眼底OCT影像生成新眼底OCT影像，提高了新眼底OCT影像的真实性，避免和原始眼底OCT影像差异过大，解决了真实数据太少及数据不均衡问题，提高了影像处理效率。同时在样本较小且样本不均衡的情况，可以通过加入生成对抗网络生成的增强影像数据，提高了训练模型的性能及泛化效果。
上面对本申请实施例中眼底OCT影像增强方法进行了描述，下面对本申请实施例中眼底OCT影像增强装置进行描述，请参阅图3，本申请实施例中眼底OCT影像增强装置的一个实施例包括：
获取单元301，用于获取原始眼底光学相干断层扫描技术OCT影像；
构建单元302，用于通过预置的深度学习网络模型构建生成器和鉴别器；
转换单元303，用于通过所述生成器将预置的随机噪声转换成目标眼底OCT影像；
判断单元304，用于通过所述鉴别器判断所述目标眼底OCT影像是否真实；
保留单元305，若所述目标眼底OCT影像真实，则用于将所述目标眼底OCT影像保留。
本申请实施例，根据原始眼底OCT影像生成新眼底OCT影像，提高了新眼底OCT影像的真实性，避免和原始眼底OCT影像差异过大，解决了真实数据太少及数据不均衡问题，提高了影像处理效率。
请参阅图4，本申请实施例中眼底OCT影像增强装置的另一个实施例包括：
获取单元301，用于获取原始眼底光学相干断层扫描技术OCT影像；
构建单元302，用于通过预置的深度学习网络模型构建生成器和鉴别器；
转换单元303，用于通过所述生成器将预置的随机噪声转换成目标眼底OCT影像；
判断单元304，用于通过所述鉴别器判断所述目标眼底OCT影像是否真实；
保留单元305，若所述目标眼底OCT影像真实，则用于将所述目标眼底OCT影像保留。
可选的，眼底OCT影像增强装置还包括：
优化单元306，若所述目标眼底OCT影像不真实，则用于将所述目标眼底OCT影像重新输入到所述生成器进行影像优化。
可选的，判断单元304包括：
第一确定模块3041，用于通过所述鉴别器对所述目标眼底OCT影像进行分析，确定目 标眼底OCT影像的分布特征；
第一判断模块3042，用于通过所述鉴别器基于根据所述目标眼底OCT影像的分布特征判断目标眼底OCT影像是否真实。
可选的，第一判断模块3042具体用于：
获取原始眼底OCT影像的分布特征，所述分布特征包括形状特征、水平特征和差异特征；将所述原始眼底OCT影像的形状特征、水平特征和差异特征，与所述目标眼底OCT影像的形状特征、水平特征和差异特征分别进行比较，得到形状特征相似度值、水平特征相似度值和差异特征相似度值；将所述形状特征相似度值、所述水平特征相似度值和所述差异特征相似度值按照各自的权重计算得到分布特征相似度；若所述分布特征相似度大于阈值，则确定所述目标眼底OCT影像真实。
可选的，判断单元304还包括：
生成模块3043，用于生成所述目标眼底OCT影像的分布特征和所述原始眼底OCT影像的分布特征的交叉熵损失函数；
第二判断模块3044，用于通过所述鉴别器判断所述交叉熵损失函数的值是否小于阈值；
第二确定模块3045，若所述交叉熵损失函数的值小于阈值，则用于通过所述鉴别器确定所述目标眼底OCT影像真实。
可选的，生成模块3043具体用于：
将所述目标眼底OCT影像的分布特征值作为预测值q(x)；将所述原始眼底OCT影像的分布特征值作为真实值p(x)；生成所述预测值q(x)和所述真实值p(x)的交叉熵函数H(p,q)，其中，x为正整数，具体表达式为：H(p,q)＝-∑p(x)log(q(x))。
可选的，眼底OCT影像增强装置还包括：
存储单元307，用于将真实的目标眼底OCT影像和所述原始眼底OCT影像存储在同一个眼底OCT影像集合中。
本申请实施例，根据原始眼底OCT影像生成新眼底OCT影像，提高了新眼底OCT影像的真实性，避免和原始眼底OCT影像差异过大，解决了真实数据太少及数据不均衡问题，提高了影像处理效率。同时在样本较小且样本不均衡的情况，可以通过加入生成对抗网络生成的增强影像数据，提高了训练模型的性能及泛化效果。
上面图3至图4从模块化功能实体的角度对本申请实施例中的眼底OCT影像增强装置进行详细描述，下面从硬件处理的角度对本申请实施例中眼底OCT影像增强设备进行详细 描述。
图5是本申请实施例提供的一种眼底OCT影像增强设备的结构示意图，该眼底OCT影像增强设备500可因配置或性能不同而产生比较大的差异，可以包括一个或一个以上处理器(central processing units，CPU)501(例如，一个或一个以上处理器)和存储器509，一个或一个以上存储应用程序507或数据506的存储介质508(例如一个或一个以上海量存储设备)。其中，存储器509和存储介质508可以是短暂存储或持久存储。存储在存储介质508的程序可以包括一个或一个以上模块(图示没标出)，每个模块可以包括对眼底OCT影像增强设备中的一系列指令操作。更进一步地，处理器501可以设置为与存储介质508通信，在眼底OCT影像增强设备500上执行存储介质508中的一系列指令操作。
眼底OCT影像增强设备500还可以包括一个或一个以上电源502，一个或一个以上有线或无线网络接口503，一个或一个以上输入输出接口504，和/或，一个或一个以上操作系统505，例如Windows Serve，Mac OS X，Unix，Linux，FreeBSD等等。本领域技术人员可以理解，图5中示出的眼底OCT影像增强设备结构并不构成对眼底OCT影像增强设备的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。处理器501可以执行上述实施例中获取单元301、构建单元302、转换单元303、判断单元304和优化单元306的功能。
下面结合图5对眼底OCT影像增强设备的各个构成部件进行具体的介绍：
处理器501是眼底OCT影像增强设备的控制中心，可以按照设置的眼底OCT影像增强方法进行处理。处理器501利用各种接口和线路连接整个眼底OCT影像增强设备的各个部分，通过运行或执行存储在存储器509内的软件程序和/或模块，以及调用存储在存储器509内的数据，执行眼底OCT影像增强设备的各种功能和处理数据，提高生成的新眼底OCT影像的真实性，避免和原始眼底OCT影像差异过大，解决真实数据太少及数据不均衡问题，提高影像处理效率。存储介质508和存储器509都是存储数据的载体，本申请实施例中，存储介质508可以是指储存容量较小，但速度快的内存储器，而存储器509可以是储存容量大，但储存速度慢的外存储器。
存储器509可用于存储软件程序以及模块，处理器501通过运行存储在存储器509的软件程序以及模块，从而执行眼底OCT影像增强设备500的各种功能应用以及数据处理。存储器509可主要包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需的应用程序(比如通过生成器将预置的随机噪声转换成目标眼底OCT影像) 等；存储数据区可存储根据眼底OCT影像增强设备的使用所创建的数据(比如生成器和鉴别器)等。此外，存储器509可以包括高速随机存取存储器，还可以包括非易失性存储器，例如至少一个磁盘存储器件、闪存器件、或其他非易失性固态存储器件。在本申请实施例中提供的眼底OCT影像增强方法程序和接收到的数据流存储在存储器中，当需要使用时，处理器501从存储器509中调用。
本申请还提供一种计算机可读存储介质，该计算机可读存储介质可以为非易失性计算机可读存储介质，所述计算机可读存储介质中存储有指令，当所述指令在计算机上运行时，使得计算机执行如下眼底OCT影像增强方法的步骤：
获取原始眼底光学相干断层扫描技术OCT影像；
通过预置的深度学习网络模型构建生成器和鉴别器；
通过所述生成器将预置的随机噪声转换成目标眼底OCT影像；
通过所述鉴别器判断所述目标眼底OCT影像是否真实；
若所述目标眼底OCT影像真实，则将所述目标眼底OCT影像保留。
在计算机上加载和执行所述计算机程序指令时，全部或部分地产生按照本申请实施例所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络、或者其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一计算机可读存储介质传输，例如，所述计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如同轴电缆、光纤、双绞线)或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存储的任何可用介质或者是包含一个或多个可用介质集成的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质，(例如，软盘、硬盘、磁带)、光介质(例如，光盘)、或者半导体介质(例如固态硬盘(solid state disk，SSD))等。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统，装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统，装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显 示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。

Claims (20)

  1. 一种眼底OCT影像增强方法，包括：
    获取原始眼底光学相干断层扫描技术OCT影像；
    通过预置的深度学习网络模型构建生成器和鉴别器；
    通过所述生成器将预置的随机噪声转换成目标眼底OCT影像；
    通过所述鉴别器判断所述目标眼底OCT影像是否真实；
    若所述目标眼底OCT影像真实，则将所述目标眼底OCT影像保留。
  2. 根据权利要求1所述的眼底OCT影像增强方法，在所述通过所述鉴别器判断所述目标眼底OCT影像是否真实之后，所述方法还包括：
    若所述目标眼底OCT影像不真实，则将所述目标眼底OCT影像重新输入到所述生成器进行影像优化。
  3. 根据权利要求1所述的眼底OCT影像增强方法，所述通过所述鉴别器判断所述目标眼底OCT影像是否真实包括：
    通过所述鉴别器对所述目标眼底OCT影像进行分析，确定目标眼底OCT影像的分布特征；
    通过所述鉴别器基于所述目标眼底OCT影像的分布特征判断目标眼底OCT影像是否真实。
  4. 根据权利要求3所述的眼底OCT影像增强方法，所述通过所述鉴别器基于所述目标眼底OCT影像的分布特征判断目标眼底OCT影像是否真实包括：
    获取原始眼底OCT影像的分布特征，所述分布特征包括形状特征、水平特征和差异特征；
    将所述原始眼底OCT影像的形状特征、水平特征和差异特征，与所述目标眼底OCT影像的形状特征、水平特征和差异特征分别进行比较，得到形状特征相似度值、水平特征相似度值和差异特征相似度值；
    将所述形状特征相似度值、所述水平特征相似度值和所述差异特征相似度值按照各自的权重计算得到分布特征相似度；
    若所述分布特征相似度大于阈值，则通过所述鉴别器确定所述目标眼底OCT影像真实。
  5. 根据权利要求3所述的眼底OCT影像增强方法，所述通过所述鉴别器基于所述目标眼底OCT影像的分布特征判断目标眼底OCT影像是否真实包括：
    生成所述目标眼底OCT影像的分布特征和所述原始眼底OCT影像的分布特征的交叉熵损失函数；
    通过所述鉴别器判断所述交叉熵损失函数的值是否小于阈值；
    若所述交叉熵损失函数的值小于阈值，则通过所述鉴别器确定所述目标眼底OCT影像真实。
  6. 根据权利要求5所述的眼底OCT影像增强方法，所述生成所述目标眼底OCT影像的分布特征和所述原始眼底OCT影像的分布特征的交叉熵损失函数包括：
    将所述目标眼底OCT影像的分布特征值作为预测值q(x)；
    将所述原始眼底OCT影像的分布特征值作为真实值p(x)；
    生成所述预测值q(x)和所述真实值p(x)的交叉熵函数H(p,q)，其中，x为正整数，具体表达式为：H(p,q)＝-∑p(x)log(q(x))。
  7. 根据权利要求1-6中任一所述的眼底OCT影像增强方法，所述方法还包括：
    将真实的目标眼底OCT影像和所述原始眼底OCT影像存储在同一个眼底OCT影像集合中。
  8. 一种眼底OCT影像增强装置，包括：
    获取单元，用于获取原始眼底光学相干断层扫描技术OCT影像；
    构建单元，用于通过预置的深度学习网络模型构建生成器和鉴别器；
    转换单元，用于通过所述生成器将预置的随机噪声转换成目标眼底OCT影像；
    判断单元，用于通过所述鉴别器判断所述目标眼底OCT影像是否真实；
    保留单元，若所述目标眼底OCT影像真实，则用于将所述目标眼底OCT影像保留。
  9. 根据权利要求8所述的眼底OCT影像增强装置，眼底OCT影像增强装置还包括
    优化单元，若所述目标眼底OCT影像不真实，则用于将所述目标眼底OCT影像重新输入到所述生成器进行影像优化。
  10. 根据权利要求8所述的眼底OCT影像增强装置，判断单元包括：
    第一确定模块，用于通过所述鉴别器对所述目标眼底OCT影像进行分析，确定目标眼底OCT影像的分布特征；
    第一判断模块，用于通过所述鉴别器基于所述目标眼底OCT影像的分布特征判断目标眼底OCT影像是否真实。
  11. 根据权利要求10所述的眼底OCT影像增强装置，第一判断模块具体用于：
    获取原始眼底OCT影像的分布特征，所述分布特征包括形状特征、水平特征和差异特征；
    将所述原始眼底OCT影像的形状特征、水平特征和差异特征，与所述目标眼底OCT影像的形状特征、水平特征和差异特征分别进行比较，得到形状特征相似度值、水平特征相似度值和差异特征相似度值；
    将所述形状特征相似度值、所述水平特征相似度值和所述差异特征相似度值按照各自的权重计算得到分布特征相似度；
    若所述分布特征相似度大于阈值，则确定所述目标眼底OCT影像真实。
  12. 根据权利要求10所述的眼底OCT影像增强装置，判断单元还包括：
    生成模块，用于生成所述目标眼底OCT影像的分布特征和所述原始眼底OCT影像的分布特征的交叉熵损失函数；
    第二判断模块，用于通过所述鉴别器判断所述交叉熵损失函数的值是否小于阈值；
    第二确定模块，若所述交叉熵损失函数的值小于阈值，则用于通过所述鉴别器确定所述目标眼底OCT影像真实。
  13. 根据权利要求12所述的眼底OCT影像增强装置，生成模块具体用于：
    将所述目标眼底OCT影像的分布特征值作为预测值q(x)；
    将所述原始眼底OCT影像的分布特征值作为真实值p(x)；
    生成所述预测值q(x)和所述真实值p(x)的交叉熵函数H(p,q)，其中，x为正整数，具体表达式为：H(p,q)＝-∑p(x)log(q(x))。
  14. 根据权利要求8-13中任一所述的眼底OCT影像增强装置，眼底OCT影像增强装置还包括：
    存储单元，用于将真实的目标眼底OCT影像和所述原始眼底OCT影像存储在同一个眼底OCT影像集合中。
  15. 一种眼底OCT影像增强设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现如下步骤：
    获取原始眼底光学相干断层扫描技术OCT影像；
    通过预置的深度学习网络模型构建生成器和鉴别器；
    通过所述生成器将预置的随机噪声转换成目标眼底OCT影像；
    通过所述鉴别器判断所述目标眼底OCT影像是否真实；
    若所述目标眼底OCT影像真实，则将所述目标眼底OCT影像保留。
  16. 根据权利要求15所述的眼底OCT影像增强设备，所述处理器执行所述计算机程序实现所述通过所述鉴别器判断所述目标眼底OCT影像是否真实之后，还包括以下步骤：
    若所述目标眼底OCT影像不真实，则将所述目标眼底OCT影像重新输入到所述生成器进行影像优化。
  17. 根据权利要求15所述的眼底OCT影像增强设备，所述处理器执行所述计算机程序实现所述通过所述鉴别器判断所述目标眼底OCT影像是否真实时，包括以下步骤：
    通过所述鉴别器对所述目标眼底OCT影像进行分析，确定目标眼底OCT影像的分布特征；
    通过所述鉴别器基于所述目标眼底OCT影像的分布特征判断目标眼底OCT影像是否真实。
  18. 根据权利要求17所述的眼底OCT影像增强装置，所述处理器执行所述计算机程序实现所述通过所述鉴别器基于所述目标眼底OCT影像的分布特征判断目标眼底OCT影像是否真实时，包括以下步骤：
    获取原始眼底OCT影像的分布特征，所述分布特征包括形状特征、水平特征和差异特征；
    将所述原始眼底OCT影像的形状特征、水平特征和差异特征，与所述目标眼底OCT影像的形状特征、水平特征和差异特征分别进行比较，得到形状特征相似度值、水平特征相似度值和差异特征相似度值；
    将所述形状特征相似度值、所述水平特征相似度值和所述差异特征相似度值按照各自的权重计算得到分布特征相似度；
    若所述分布特征相似度大于阈值，则通过所述鉴别器确定所述目标眼底OCT影像真实。
  19. 根据权利要求15所述的眼底OCT影像增强装置，所述处理器执行所述计算机程序实现所述通过所述鉴别器基于所述目标眼底OCT影像的分布特征判断目标眼底OCT影像是否真实时，包括以下步骤：
    生成所述目标眼底OCT影像的分布特征和所述原始眼底OCT影像的分布特征的交叉熵损失函数；
    通过所述鉴别器判断所述交叉熵损失函数的值是否小于阈值；
    若所述交叉熵损失函数的值小于阈值，则通过所述鉴别器确定所述目标眼底OCT影像真实。
  20. 一种计算机可读存储介质，所述计算机可读存储介质中存储有指令，当所述指令在计算机上运行时，使得计算机执行如下步骤：
    获取原始眼底光学相干断层扫描技术OCT影像；
    通过预置的深度学习网络模型构建生成器和鉴别器；
    通过所述生成器将预置的随机噪声转换成目标眼底OCT影像；
    通过所述鉴别器判断所述目标眼底OCT影像是否真实；
    若所述目标眼底OCT影像真实，则将所述目标眼底OCT影像保留。
PCT/CN2019/102532 2019-06-18 2019-08-26 眼底oct影像增强方法、装置、设备及存储介质 WO2020252898A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910524224.5 2019-06-18
CN201910524224.5A CN110415182B (zh) 2019-06-18 2019-06-18 眼底oct影像增强方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2020252898A1 true WO2020252898A1 (zh) 2020-12-24

Family

ID=68359213

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/102532 WO2020252898A1 (zh) 2019-06-18 2019-08-26 眼底oct影像增强方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN110415182B (zh)
WO (1) WO2020252898A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113449774A (zh) * 2021-06-02 2021-09-28 北京鹰瞳科技发展股份有限公司 眼底图像质量控制方法、装置、电子设备及存储介质
CN116912203B (zh) * 2023-07-13 2024-04-05 桂林电子科技大学 基于多智能模型结合的异常眼底图像低耗检测方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107392312A (zh) * 2017-06-01 2017-11-24 华南理工大学 一种基于dcgan性能的动态调整算法
CN108932534A (zh) * 2018-07-15 2018-12-04 瞿文政 一种基于深度卷积生成对抗网络的图片生成方法
CN108961272A (zh) * 2018-07-02 2018-12-07 浙江工业大学 一种基于深度卷积对抗生成网络的皮肤病图像的生成方法
CN109658466A (zh) * 2018-12-10 2019-04-19 苏州大学 一种疾病视网膜光学相干断层影像仿真生成方法
US10275473B2 (en) * 2017-04-27 2019-04-30 Sk Telecom Co., Ltd. Method for learning cross-domain relations based on generative adversarial networks

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10474929B2 (en) * 2017-04-25 2019-11-12 Nec Corporation Cyclic generative adversarial network for unsupervised cross-domain image generation

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10275473B2 (en) * 2017-04-27 2019-04-30 Sk Telecom Co., Ltd. Method for learning cross-domain relations based on generative adversarial networks
CN107392312A (zh) * 2017-06-01 2017-11-24 华南理工大学 一种基于dcgan性能的动态调整算法
CN108961272A (zh) * 2018-07-02 2018-12-07 浙江工业大学 一种基于深度卷积对抗生成网络的皮肤病图像的生成方法
CN108932534A (zh) * 2018-07-15 2018-12-04 瞿文政 一种基于深度卷积生成对抗网络的图片生成方法
CN109658466A (zh) * 2018-12-10 2019-04-19 苏州大学 一种疾病视网膜光学相干断层影像仿真生成方法

Also Published As

Publication number Publication date
CN110415182A (zh) 2019-11-05
CN110415182B (zh) 2024-03-05

Similar Documents

Publication Publication Date Title
Razzano et al. Image-based deep learning for classification of noise transients in gravitational wave detectors
Lin et al. Transforming retinal photographs to entropy images in deep learning to improve automated detection for diabetic retinopathy
JP2020508522A (ja) 監視されていないクロスドメイン画像生成のための周期的敵対的生成ネットワーク
CN109902665A (zh) 相似人脸检索方法、装置及存储介质
US20220222925A1 (en) Artificial intelligence-based image processing method and apparatus, device, and storage medium
TW202030640A (zh) 一種跨模態訊息檢索方法、裝置和儲存介質
Abd Elaziz et al. Medical Image Classification Utilizing Ensemble Learning and Levy Flight‐Based Honey Badger Algorithm on 6G‐Enabled Internet of Things
WO2021051987A1 (zh) 神经网络模型训练的方法和装置
WO2019200702A1 (zh) 去网纹系统训练方法、去网纹方法、装置、设备及介质
WO2020252898A1 (zh) 眼底oct影像增强方法、装置、设备及存储介质
JP2023531350A (ja) サンプル画像を増分する方法、画像検出モデルの訓練方法及び画像検出方法
CN113505883A (zh) 一种神经网络训练方法以及装置
WO2023050868A1 (zh) 融合模型的训练方法、图像融合方法、装置、设备及介质
US20230252070A1 (en) Method and apparatus for training retrieval model, retrieval method and apparatus, device and medium
CN109620293A (zh) 一种图像识别方法、装置以及存储介质
US11929871B2 (en) Method for generating backbone network, apparatus for generating backbone network, device, and storage medium
CN114298997B (zh) 一种伪造图片检测方法、装置及存储介质
CN116664930A (zh) 基于自监督对比学习的个性化联邦学习图像分类方法及系统
Jhansi et al. Improved Accuracy in Speech Recognition System for Detection of COVID-19 using Support Vector Machine and Comparing with Convolution Neural Network Algorithm
CN116628507B (zh) 数据处理方法、装置、设备及可读存储介质
CN112101438A (zh) 一种左右眼分类方法、装置、服务器和存储介质
US11734389B2 (en) Method for generating human-computer interactive abstract image
CN116958020A (zh) 异常图像检测方法、模型训练方法、装置、设备和介质
CN112529772B (zh) 一种零样本设置下的无监督图像转换方法
Liu et al. A complex chained P system based on evolutionary mechanism for image segmentation

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19933804

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19933804

Country of ref document: EP

Kind code of ref document: A1