Optical coherence tomography image speckle denoising method and electronic equipment
By using the unsupervised speckle denoising deep learning network model Double-free Net, speckle noise in optical coherence tomography images is automatically extracted and eliminated, solving the problem of speckle noise affecting diagnostic accuracy and achieving high-quality image acquisition.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHENGDU MUGUANG MEDICAL TECHNOLOGY CO LTD
- Filing Date
- 2023-12-04
- Publication Date
- 2026-04-17
AI Technical Summary
In existing optical coherence tomography techniques, speckle noise leads to a decrease in spatial resolution, obscures the fine structure of biological tissues, and affects diagnostic accuracy. Existing denoising methods are complex or result in images that are too smooth and lose detail.
We employ an unsupervised speckle denoising deep learning network model, Double-free Net, to automatically extract speckle noise features by resampling and synthesizing noisy images, avoiding the need for a baseline image and repeated scanning. We then utilize the RDB-UNet network for noise distribution extraction and correction.
It effectively eliminates speckle noise, obtains high-quality optical coherence tomography images, improves diagnostic accuracy, and avoids the use of reference images and repeated scanning.
Smart Images
Figure CN121883290A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of optical imaging denoising methods, and more specifically, to an optical coherence tomography image speckle denoising method, an electronic device, and a storage medium. Background Technology
[0002] Optical coherence tomography (OCT) enables powerful, non-invasive, high-resolution three-dimensional imaging of biological tissues and materials, and has been widely used in biomedicine and other fields. However, OCT relies on the coherence of backscattered light to distinguish the morphology of biological tissues. This coherence inevitably produces speckle noise, leading to a decrease in spatial resolution. Furthermore, this speckle noise can mask the subtle but crucial microstructures within the tissue, negatively impacting clinical diagnosis and reducing diagnostic accuracy.
[0003] In existing technologies, methods for removing speckle noise are divided into hardware methods and software methods. Hardware methods involve complex hardware systems that require repeated scanning, thus reducing the temporal resolution of the imaging. Software methods often result in overly smooth images that lose important details. In recent years, unsupervised deep learning methods have become increasingly popular for removing speckle noise in OCT, but they still require the use of unpaired noise-reference images or paired noise-noise images. Summary of the Invention
[0004] The purpose of this application is to address the shortcomings of the prior art by providing a novel unsupervised optical coherence tomography (OCT) image speckle denoising method, apparatus, electronic device, and storage medium. This method avoids the use of a reference image and eliminates repeated scanning by resampling and synthesizing the noisy image. Furthermore, it can automatically extract speckle noise features, effectively eliminate speckle noise, and obtain high-quality OCT images.
[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows:
[0006] In a first aspect, embodiments of this application provide a method for speckle denoising in optical coherence tomography images, the method comprising:
[0007] Acquire an initial optical coherence tomography image output by an optical coherence tomography system, wherein speckle noise exists in the initial optical coherence tomography image;
[0008] The initial optical coherence tomography image is input into the pre-trained speckle denoising deep learning network model Double-free Net to perform speckle denoising on the initial optical coherence tomography image, resulting in a target optical coherence tomography image free of speckle noise.
[0009] Optionally, acquiring the initial optical coherence tomography image output by the optical coherence tomography system includes: constructing an OCT speckle image dataset, which includes two publicly available OCT datasets and two self-made OCT datasets. The self-made OCT datasets contain OCT images of various human ex vivo tissues, such as large field-of-view retinal images, styrene microparticle-calibrated images, and ex vivo human stomach, esophagus, and urothelial lining.
[0010] Optionally, the speckle denoising deep learning network model Double-free Net includes three parts: a noise distribution extraction module, a noisy image training module, and an output image correction module;
[0011] The noise distribution extraction module includes: a first RDB-UNet network and two adjacent sub-samplers;
[0012] The noisy image training module includes: a second RDB-UNet network;
[0013] The output and correction module includes: a second RDB-UNet network trained by the noisy image training module and a correction operator;
[0014] Specifically, the training phase of the network is carried out in three steps:
[0015] First, two adjacent sub-samplers are used to extract noisy image pairs from the initial optical coherence tomography image, and the first RDB-UNet network is used to extract the speckle noise distribution. Then, the speckle noise distribution is added to another noisy sub-image as a synthetic noise image. The noise image training module uses the noisy image pairs and the corresponding synthetic noise images to train the second RDB-UNet network. Finally, the initial noisy image is input into the well-trained second RDB-UNet network, and the output is corrected to obtain the denoised image.
[0016] Optionally, the image segmentation deep learning network model RDB-UNet includes: an encoder and a decoder connected in sequence;
[0017] The initial optical coherence tomography image is input into the encoder of the RDB-UNet, which performs downsampling processing on the initial optical coherence tomography image and sends the downsampling result to the decoder of the RDB-UNet. The decoder performs upsampling processing on the downsampling result to obtain the processed optical coherence tomography image.
[0018] In the encoder and decoder, traditional layers are replaced with Residual Dense Blocks (RDBs), and RDBs are used at jumpers between the encoder and decoder to enhance the network. Each RDB consists of three regular convolutional layers, followed by leaky ReLU, and finally a regular convolutional layer as the output layer. The proposed RDB-UNet can also be implemented using other existing network architectures.
[0019] Optionally, the initial optical coherence tomography image is input into a pre-trained speckle denoising deep learning network model, Double-free Net, to perform speckle denoising on the initial optical coherence tomography image, thereby obtaining a target optical coherence tomography image free of speckle noise, including:
[0020] A. Noise Distribution Extraction: Adjacent sub-samplers sample two similar but distinct images G1(y) and G2(y) from the initial optical coherence tomography image y. Their noise N1 and N2 are independent and identically distributed. G1(y) is then input into the first RDB-UNet neural network for processing, attempting to minimize... Where x is the noiseless image corresponding to the noisy image (G1(y), G2(y)), and F θ It is a neural network parameterized by θ, where E is the expectation operator and F is the parameterized parameter. θ (G1(y)) is the prediction of the noisy image G1(y), and the extracted speckle noise is G1(y)-F. θ (G1(y));
[0021] Simultaneously, the initial optical coherence tomography image y is input into the first RDB-UNet network to obtain the predicted F of the noisy image y. θ (y), and input it into the same adjacent subsampler to obtain a denoised image pair {G1(F) with some residual noise. θ (y)),G2(F θ (y))}, to simplify the calculation of loss information;
[0022] B. Noisy Image Training: The speckle noise distribution extracted in step A is added to another noisy sub-image to construct a noisier image Z. This is achieved by minimizing... To train the second RDB-UNet network, where F ω It is a neural network parameterized by ω;
[0023] C. Output and Correction: Input the initial optical coherence tomography image y into the second RDB-UNet network that has been well trained in step B, and perform correction operations: double the output of B and subtract its input to obtain the denoised image;
[0024] Optionally, the loss information of the noise distribution extraction module include:
[0025] Where x is the noise-free image corresponding to the pair of noisy images (G1(y), G2(y)), and σ 2 Let δ represent the variance of G2(y), and let δ be the gap between the clean underlying images of a pair of noisy images (G1(y), G2(y)). 12 By designing a neighborhood subsampler to ensure sufficiently small gaps, a network trained with (G1(y), G2(y)) can be viewed as an approximation of Noise2Noise. When δ 12 When σ = 0, 2 It is a constant value.
[0026] From the above equation, it can be seen that the noise-free image x in the loss information can be eliminated through design, thereby achieving unsupervised training without a reference image and without repeated scanning. Finally, the loss function of the noise distribution extraction module is:
[0027]
[0028] The second term is a regularization term that optimizes the denoising effect, where γ is the weight that controls the regularization term.
[0029] Optionally, the loss information of the noisy image training module include:
[0030] The noisy image training module trains a Noiser2Noise model, which uses a noisy image G2(y) as a label and a noisier image Z as input, where Z = G1(y) - G1(F). θ (y))+G2(y).
[0031] The loss function for the noisy image training module is:
[0032] Secondly, embodiments of this application also provide an electronic device, including: a processor, a storage medium, and a bus. The storage medium stores program instructions executable by the processor. When the application runs, the processor communicates with the storage medium via the bus, and the processor executes the program instructions to perform the steps of the optical coherence tomography image speckle denoising method described in the first aspect.
[0033] Thirdly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which is read and executes the steps of the optical coherence tomography image speckle denoising method described in the first aspect.
[0034] The beneficial effects of this application are:
[0035] This application provides a novel unsupervised method, apparatus, electronic device, and storage medium for speckle denoising of optical coherence tomography (OCT) images. The method acquires an initial OCT image output from an OCT system, inputs this initial OCT image into a pre-trained speckle denoising deep learning network model (Double-free Net), and performs speckle denoising on the initial OCT image to obtain a target OCT image free of speckle noise. This method avoids the use of a reference image and eliminates re-scanning by resampling and synthesizing the noisy image. Furthermore, it can automatically extract speckle noise features, effectively eliminating speckle noise and obtaining high-quality OCT images. Attached Figure Description
[0036] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 This application provides a schematic diagram of the application scenario and structure of an optical coherence tomography image speckle denoising method.
[0038] Figure 2 A schematic flowchart of an optical coherence tomography image speckle denoising method provided in this application embodiment;
[0039] Figure 3 A schematic flowchart of another optical coherence tomography image speckle denoising method provided in this application embodiment;
[0040] Figure 4 This is a schematic diagram of the structure of the initial model provided in the embodiments of this application;
[0041] Figure 5 This is one of the schematic diagrams showing the results of an optical coherence tomography image speckle denoising method provided in an embodiment of this application;
[0042] Figure 6 A second schematic diagram illustrating the result of an optical coherence tomography image speckle denoising method provided in this application embodiment;
[0043] Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation
[0044] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.
[0045] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
[0046] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.
[0047] Figure 1 This application provides a schematic diagram of the application scenario and structure of an optical coherence tomography image speckle denoising method, as illustrated in the embodiments of this application. Figure 1 As shown, this application applies to scenarios where an OCT image of any sample is acquired. The sample may be, for example, a biological sample. This scenario involves an Optical Coherence Tomography (OCT) system and electronic equipment. First, scanning any sample using the OCT system outputs an OCT image containing speckle noise. Then, this speckle-noise OCT image is input into the aforementioned electronic equipment, which uses a trained speckle denoising deep learning network model to denoise the OCT image, ultimately obtaining an OCT image free of speckle noise.
[0048] Figure 2 This application provides a schematic flowchart of an optical coherence tomography image speckle denoising method, as illustrated in the embodiments below. Figure 2As shown, this method is applied to the aforementioned electronic device, which may be, for example, a desktop computer, a laptop computer, or other terminal device with computing power and display function, or it may be a server. The method includes:
[0049] S101. Obtain the initial optical coherence tomography image output by the optical coherence tomography system.
[0050] Optionally, the OCT system may include a light source, a coupler, a reference mirror, a sample arm, and a spectrometer. The specific principle is as follows: the system can use a broadband light source as the light source, for example, a broadband light source with a center wavelength of 850nm and a full width at half maximum (FWHM) of 165nm. The light emitted by the broadband light source passes through the coupler and then through the sample arm and reference arm to illuminate the sample and reference mirror respectively. The reflected light in the two optical paths merges in the coupler and generates an interference signal under certain conditions. The interference signal is output to the spectrometer, which transmits the detected interference signal to the computer. The computer performs grayscale imaging and outputs an initial OCT image, which contains speckle noise.
[0051] S102. Input the initial optical coherence tomography image into the pre-trained speckle denoising deep learning network model to perform speckle denoising on the initial optical coherence tomography image, and obtain the target optical coherence tomography image without speckle noise.
[0052] This embodiment acquires an initial optical coherence tomography (OCT) image output by an optical coherence tomography system. This initial OCT image is then input into a pre-trained speckle denoising deep learning network model to perform speckle denoising, resulting in a target OCT image free of speckle noise. The pre-trained speckle denoising deep learning network model can automatically extract speckle noise features, effectively eliminating speckle noise and obtaining a high-quality OCT image.
[0053] Optionally, the above speckle denoising deep learning network model may include: a noise distribution extraction module (a first RDB-UNet network and two adjacent sub-samplers), a noisy image training module (a second RDB-UNet network), and an output image correction module (a second RDB-UNet network and a correction operator);
[0054] The initial OCT image is extracted using two adjacent sub-samplers, and the speckle noise distribution is extracted using the first RDB-UNet network. Then, the speckle noise distribution is added to another noisy sub-image as a synthetic noise image. The noise image training module uses the noise image pairs and the corresponding synthetic noise images to train the second RDB-UNet network. Finally, the initial noise image is input into the well-trained second RDB-UNet network, and the output is corrected to obtain the denoised image.
[0055] Optionally, RDB-UNet includes: an encoder and a decoder connected in sequence;
[0056] The initial optical coherence tomography (OCT) image is input into the encoder of RDB-UNet, which downsamples the image and sends the downsampled result to the decoder. The decoder then upsamples the downsampled result to obtain the processed OCT image. Residual dense blocks (RDBs) replace traditional layers in both the encoder and decoder, and RDBs are used at jumpers between the encoder and decoder to enhance the network. Each RDB consists of three regular convolutional layers, followed by leaky ReLU, and finally a regular convolutional layer as the output layer. The proposed RDB-UNet can also be implemented using other existing network architectures.
[0057] The following examples illustrate the training process of the above-mentioned speckle denoising deep learning network model.
[0058] Figure 3 A schematic flowchart of another optical coherence tomography image speckle denoising method provided in this application embodiment is shown below. Figure 3 As shown, step S102 above inputs the initial optical coherence tomography image into the pre-trained speckle denoising deep learning network model. The process of training the speckle denoising deep learning network model before performing speckle denoising on the initial optical coherence tomography image may include:
[0059] S201. Construct the dataset.
[0060] Optionally, the dataset includes two publicly available OCT datasets and two self-made OCT datasets. The self-made OCT datasets contain OCT images of various human ex vivo tissues, such as large field-of-view retinal images, styrene microparticle-calibrated images, and ex vivo human stomach, esophagus, and urothelium.
[0061] S202, Construct the initial model.
[0062] Optionally, the initial model may include: a noise distribution extraction module (a first RDB-UNet network and two adjacent subsamplers), a noisy image training module (a second RDB-UNet network), and an output image correction module (a second RDB-UNet network and a correction operator);
[0063] The initial OCT image is extracted using two adjacent sub-samplers, and the speckle noise distribution is extracted using the first RDB-UNet network. Then, the speckle noise distribution is added to another noisy sub-image as a synthetic noise image. The noise image training module uses the noise image pairs and the corresponding synthetic noise images to train the second RDB-UNet network. Finally, the initial noise image is input into the well-trained second RDB-UNet network, and the output is corrected to obtain the denoised image.
[0064] S203. Train the initial model based on the dataset to obtain the speckle denoising deep learning network.
[0065] A. Noise Distribution Extraction: Adjacent sub-samplers sample two similar but distinct images G1(y) and G2(y) from the initial optical coherence tomography image y. Their noise N1 and N2 are independent and identically distributed. G1(y) is then input into the first RDB-UNet neural network for processing, attempting to minimize... Where x is the noiseless image corresponding to the noisy image (G1(y), G2(y)), and F θ It is a neural network parameterized by θ, where E is the expectation operator and F is the parameterized parameter. θ (G1(y)) is the prediction of the noisy image G1(y), and the extracted speckle noise is G1(y)-F. θ (G1(y));
[0066] Simultaneously, the initial optical coherence tomography image y is input into the first RDB-UNet network to obtain the predicted F of the noisy image y. θ (y), and input it into the same adjacent subsampler to obtain a denoised image pair {G1(F) with some residual noise. θ (y)),G2(F θ (y))}, to simplify the calculation of loss information;
[0067] B. Noisy Image Training: The speckle noise distribution extracted in step A is added to another noisy sub-image to construct a noisier image Z. This is achieved by minimizing... To train the second RDB-UNet network, where F ω It is a neural network parameterized by ω;
[0068] C. Output and Correction: Input the initial optical coherence tomography image y into the second RDB-UNet network that has been well trained in step B, and perform correction operations: double the output of B and subtract its input to obtain the denoised image;
[0069] Optionally, the loss information of the noise distribution extraction module include:
[0070]
[0071] Where x is the noise-free image corresponding to the pair of noisy images (G1(y), G2(y)), and σ 2 Let δ represent the variance of G2(y), and let δ be the gap between the clean underlying images of a pair of noisy images (G1(y), G2(y)). 12 By designing neighborhood subsamplers to ensure sufficiently small gaps, a network trained with (G1(y), G2(y)) can be viewed as an approximation of Noiser2Noise. When δ 12 When σ = 0, 2 It is a constant value.
[0072] From the above equation, it can be seen that the noise-free image x in the loss information can be eliminated through design, thereby achieving unsupervised training without a reference image and without repeated scanning. Finally, the loss function of the noise distribution extraction module is:
[0073]
[0074] The second term is a regularization term that optimizes the denoising effect, where γ is the weight that controls the regularization term.
[0075] Optionally, the loss information of the noisy image training module include:
[0076] The noisy image training module trains a Noiser2Noise model, which uses a noisy image G2(y) as a label and a noisier image Z as input, where Z = G1(y) - G1(F). θ (y))+G2(y).
[0077] The loss function for the noisy image training module is:
[0078] Figure 4 This is a schematic diagram of the structure of the initial model provided in the embodiments of this application.
[0079] Optionally, multiple sample objects can be scanned using an OCT system to obtain multiple initial images corresponding to each sample. The sample objects may include vegetables, fruits, meat, etc.
[0080] Figure 5 One of the schematic diagrams showing the results of an optical coherence tomography image speckle denoising method provided in the embodiments of this application. Figure 6 This is a second schematic diagram illustrating the result of an optical coherence tomography image speckle denoising method provided in this embodiment of the application. The original noisy image and the corresponding denoised image are shown below. Figure 5 As shown. Figure 5 (a) is a noisy OCT image, classified as AMD with edema. After denoising processing, Figure 5 In (d), small, highly reflective foci near the edema can be clearly observed. This is a biomarker that helps in the early decision-making process for retinal diseases, but it is easily masked by speckle noise and difficult to distinguish. Figure 5 As shown in (e) and 5(f), the optical coherence tomography image speckle denoising method provided in this application removes a large amount of speckle noise, making edema areas easier to observe. In particular, the boundaries of the edema become clearer after denoising, which helps to quickly determine the quantity and shape of the edema, and has practical value for clinical diagnosis. Although Figure 5 The data in (a), (b), and (c) were acquired by different OCT devices and have different noise levels. However, the optical coherence tomography image speckle denoising method provided in this application can effectively remove speckle noise and display the edges and shape of cystic lesions. Figure 6 As shown in the embodiments of this application, the optical coherence tomography image speckle denoising method has a very good denoising effect on esophageal mucosa sample images, and can simultaneously retain the microstructure information at the location indicated by the arrow.
[0081] Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of this application. Figure 7 As shown, the electronic device may include: a processor 301 and a memory 302.
[0082] Optionally, a bus 303 may also be included, wherein the memory 302 is used to store machine-readable instructions that can be executed by the processor 301. When the electronic device 300 is running, the processor 301 and the memory 302 communicate with each other through the bus 303. When the machine-readable instructions are executed by the processor 301, the method steps in the above method embodiments are performed.
[0083] This application also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the method steps described in the above-described optical coherence tomography image speckle denoising method embodiments.
[0084] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division; in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.
[0085] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.
[0086] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A method for speckle denoising in optical coherence tomography images, characterized in that, The method includes: Acquire an initial optical coherence tomography image output by an optical coherence tomography system, wherein speckle noise exists in the initial optical coherence tomography image; The initial optical coherence tomography image is input into the pre-trained speckle denoising deep learning network model Double-free Net to perform speckle denoising on the initial optical coherence tomography image, resulting in a target optical coherence tomography image free of speckle noise.
2. The optical coherence tomography image speckle denoising method according to claim 1, characterized in that, Before performing speckle denoising on the initial optical coherence tomography image using the speckle denoising deep learning network model, the method further includes: constructing an optical coherence tomography speckle image dataset; The dataset includes two publicly available optical coherence tomography datasets and two self-made optical coherence tomography datasets. The self-made optical coherence tomography datasets contain optical coherence tomography images of large field-of-view retina, styrene microparticle-calibrated images, and optical coherence tomography images of various human ex vivo tissues such as the stomach, esophagus, and urothelium.
3. The optical coherence tomography image speckle denoising method according to claim 1, characterized in that, The speckle denoising deep learning network model Double-free Net consists of three parts: a noise distribution extraction module, a noisy image training module, and an output image correction module. The noise distribution extraction module includes: a first RDB-UNet network and two adjacent sub-samplers; The noisy image training module includes: a second RDB-UNet network; The output and correction module includes: a second RDB-UNet network trained by the noisy image training module and a correction operator; Specifically, the training phase of the network is carried out in three steps: First, two adjacent sub-samplers are used to extract noisy image pairs from the initial optical coherence tomography image, and the first RDB-UNet network is used to extract the speckle noise distribution. Then, the speckle noise distribution is added to another noisy sub-image as a synthetic noise image. The noise image training module uses the noisy image pairs and the corresponding synthetic noise images to train the second RDB-UNet network. Finally, the initial noisy image is input into the well-trained second RDB-UNet network, and the output is corrected to obtain the denoised image.
4. The optical coherence tomography image speckle denoising method according to claim 3, characterized in that, The step of inputting the initial optical coherence tomography image into a pre-trained speckle denoising deep learning network model, Double-free Net, to perform speckle denoising on the initial optical coherence tomography image to obtain a target optical coherence tomography image free of speckle noise includes: A. Noise Distribution Extraction: Adjacent sub-samplers sample two similar but distinct images G1(y) and G2(y) from the initial optical coherence tomography image y. Their noise N1 and N2 are independent and identically distributed. G1(y) is then input into the first RDB-UNet neural network for processing to minimize... Where x is the noiseless image corresponding to the noisy image (G1(y), G2(y)), and F θ It is a neural network parameterized by θ, where E is the expectation operator and F is the parameterized parameter. θ (G1(y)) is the prediction of the noisy image G1(y), and the extracted speckle noise is G1(y)-F. θ (G1(y)); Simultaneously, the initial optical coherence tomography image y is input into the first RDB-UNet network to obtain the predicted F of the noisy image y. θ (y), and input it into the same adjacent subsampler to obtain a denoised image pair {G1(F)} with residual noise. θ (y)),G2(F θ (y))}, to simplify the calculation of loss information; B. Noisy Image Training: The speckle noise distribution extracted in step A is added to another noisy sub-image to construct a noisier image Z. This is achieved by minimizing... To train the second RDB-UNet network, where F ω It is a neural network parameterized by ω; C. Output and Correction: Input the initial optical coherence tomography image y into the second RDB-UNet network that has been well trained in step B, and perform correction operations: double the output of B and subtract its input to obtain the denoised image.
5. The optical coherence tomography image speckle denoising method according to claim 3, characterized in that, The RDB-UNet includes: an encoder and a decoder connected in sequence; The initial optical coherence tomography image is input into the encoder of the RDB-UNet. The encoder performs downsampling processing on the initial optical coherence tomography image and sends the downsampling result to the decoder of the RDB-UNet. The decoder performs upsampling processing on the downsampling result to obtain the processed optical coherence tomography image. In the encoder and decoder, dense residual blocks are used to replace traditional layers, and dense residual blocks are used at the jump connections between the encoder and decoder to enhance the network; each dense residual block consists of three regular convolutional layers, followed by a LeakyReLU activation function, and finally a regular convolutional layer as the output layer.
6. The optical coherence tomography image speckle denoising method according to claim 3, characterized in that, Loss information of the noise distribution extraction module include: Where x is the noise-free image corresponding to the pair of noisy images (G1(y), G2(y)), and σ 2 Let δ represent the variance of G2(y), and let δ be the gap between the clean underlying images of a pair of noisy images (G1(y), G2(y)). 12 By designing neighborhood subsamplers to ensure sufficiently small gaps, a network trained with (G1(y), G2(y)) can be viewed as an approximation of Noise2Noise. When δ 12 When σ = 0, 2 It is a constant value. The loss function of the noise distribution extraction module is: The second term is a regularization term that optimizes the denoising effect, where γ is the weight that controls the regularization term.
7. The optical coherence tomography image speckle denoising method according to claim 3, characterized in that, Loss information of the noisy image training module include: The noisy image training module trains a Noiser2Noise model, which uses a noisy image G2(y) as a label and a noisier image Z as input, where Z = G1(y) - G1(F). θ (y))+G2(y); The loss function for the noisy image training module is:
8. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program executable by the processor, and the processor executes the computer program to implement the steps of the optical coherence tomography image speckle denoising method according to any one of claims 1-7.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the optical coherence tomography image speckle denoising method as described in any one of claims 1-7.