Imaging methods, devices, equipment, and media for multi-aperture synthetic optical coherence tomography

Abstract

This application provides an imaging method, apparatus, device, and medium for multi-aperture synthetic optical coherence tomography (OCT). The imaging method includes: scanning and detecting a target object using an aperture synthetic optical coherence tomography module within a speckle-free multi-aperture synthetic optical coherence tomography system to determine an OCT image of the target object; inputting the OCT image into a generator network within the speckle-free multi-aperture synthetic optical coherence tomography system to denoise the OCT image and outputting a noise-removed OCT image; wherein the speckle-free multi-aperture synthetic optical coherence tomography system is obtained by combining the multi-aperture synthetic optical coherence tomography module with a generator network obtained through iterative training of a generative adversarial network. By utilizing the generator network within the speckle-free multi-aperture synthetic optical coherence tomography system to denoise the OCT image, a noise-removed and resolution-enhanced OCT image is obtained, thereby improving the resolution of the OCT image.

Description

Technical Field

[0001] This application relates to the field of OCT technology, and in particular to imaging methods, apparatus, equipment and media for multi-aperture synthetic optical coherence tomography. Background Technology

[0002] Optical coherence tomography (OCT) is a non-invasive sectional three-dimensional imaging technique widely used in ophthalmology, cardiology, and endoscopy. In OCT imaging, axial and lateral resolutions are independent: the former is controlled by the optical bandwidth of the laser source, while the latter is controlled by the numerical aperture (NA) and wavelength of the objective lens. Specifically, lateral resolution is inversely proportional to the objective lens's NA and depth of focus (DOF), which is proportional to the square of the lateral resolution. Laterally, when the objective lens's NA is high, the resolution in the defocused area decreases significantly. Simultaneously, speckle noise reduces OCT imaging resolution and limits potential resolution enhancement techniques. Therefore, generating noise-free, high-resolution OCT images has become a significant technical challenge. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide an imaging method, apparatus, device and medium for multi-aperture synthetic optical coherence tomography, which uses a generator network in a speckle-free multi-aperture synthetic optical coherence tomography system to denoise OCT detection images, thereby obtaining noise-removed and resolution-enhanced OCT detection images to improve the resolution of OCT detection images.

[0004] This application provides an imaging method for multi-aperture synthetic optical coherence tomography, the imaging method comprising:

[0005] The aperture synthesis optical coherence tomography module in the speckle-free multi-aperture synthesis optical coherence tomography system scans and detects the object to be measured, and determines the OCT detection image of the object to be measured.

[0006] The OCT detection image is input into the generator network of the speckle-free multi-aperture synthetic optical coherence tomography system to perform noise reduction processing on the OCT detection image and output a noise-removed OCT detection image.

[0007] The speckle-free multi-aperture synthetic optical coherence tomography system is obtained by combining the multi-aperture synthetic optical coherence tomography module with the generator network obtained by iteratively training a generative adversarial network.

[0008] In one possible implementation, the generator network is determined through the following steps:

[0009] Based on the multi-aperture synthetic optical coherence tomography module, multiple OCT sample images were determined;

[0010] Multiple OCT sample images are synthesized to determine multiple synthesized images;

[0011] The initial deep learning network is iteratively trained based on multiple synthesized images to determine the target deep learning network and the noise-free image corresponding to the synthesized image.

[0012] The generator network is determined by iteratively training the initial generator network and the initial discriminator network in the generative adversarial network based on multiple OCT sample images and the noise-free images.

[0013] In one possible implementation, the initial deep learning network includes a first initial deep learning network and a second initial deep learning network, and the target deep learning network includes a first target deep learning network and a second target deep learning network. The step of iteratively training the initial deep learning network based on multiple synthesized images to determine the target deep learning network and the noise-free image corresponding to the synthesized image includes:

[0014] The synthesized image is sampled from adjacent pixels to determine the first noise image and the second noise image;

[0015] The first noisy image is input into the first initial deep learning network for denoising processing, and the first image is output. The first initial deep learning network is iteratively trained based on the first image and the second noisy image to determine the first target deep learning network.

[0016] The synthesized image is input into the first target deep learning network for denoising processing, and a second image is output; the second image is processed to determine a third noisy image;

[0017] The third noisy image is input into the second initial deep learning network for denoising processing, and the third image is output. The second initial deep learning network is iteratively trained based on the third image and the first noisy image to determine the second target deep learning network.

[0018] The synthesized image is input into the second target deep learning network, and the noise-free image is output.

[0019] In one possible implementation, the step of iteratively training the first initial deep learning network based on the first image and the second noisy image to determine the first target deep learning network includes:

[0020] Determine the first mean error loss value between the first image and the second noisy image;

[0021] The network parameters of the first initial deep learning network are adjusted based on the first mean error loss value until the decrease of the first mean error loss value tends to level off, and then the iterative training is stopped to determine the first target deep learning network.

[0022] In one possible implementation, inputting the synthesized image into the second target deep learning network and outputting the noise-free image includes:

[0023] The synthesized image is input into the second target deep learning network to output a fourth image;

[0024] The noise-free image after denoising is determined based on the difference between twice the pixel value of the fourth image and the pixel value of the synthesized image.

[0025] In one possible implementation, the step of iteratively training the initial generator network and the initial discriminator network in the generative adversarial network based on multiple OCT sample images and the noise-free image to determine the generator network includes:

[0026] The OCT sample image is input into the initial generator network, and the fifth image is output.

[0027] The adversarial loss value and content loss value between the fifth image and the noise-free image were determined;

[0028] The fifth image and the noise-free image are input into the initial discriminator network to determine the discriminator loss value between the fifth image and the noise-free image;

[0029] The network parameters of the initial generator network are adjusted based on the adversarial loss value and the content loss value, and the network parameters of the initial discriminator network are adjusted based on the discriminator loss value. Iterative training is stopped when the adversarial loss value, the content loss value, and the discriminator loss value decrease gradually, and the generator network is determined.

[0030] This application embodiment also provides an imaging device for multi-aperture synthetic optical coherence tomography, the imaging device comprising:

[0031] The first detection module is used to scan and detect the object to be measured based on the aperture synthesis optical coherence tomography module in the speckle-free multi-aperture synthesis optical coherence tomography system, and to determine the OCT detection image of the object to be measured.

[0032] The second detection module is used to input the OCT detection image into the generator network of the speckle-free multi-aperture synthetic optical coherence tomography system, perform noise reduction processing on the OCT detection image, and output a noise-removed OCT detection image; wherein, the speckle-free multi-aperture synthetic optical coherence tomography system is obtained by combining the multi-aperture synthetic optical coherence tomography module and the generator network obtained by iteratively training a generative adversarial network.

[0033] In one possible implementation, the imaging device further includes a training module, which determines the generator network through the following steps:

[0034] Based on the multi-aperture synthetic optical coherence tomography module, multiple OCT sample images were determined;

[0035] Multiple OCT sample images are synthesized to determine multiple synthesized images;

[0036] The initial deep learning network is iteratively trained based on multiple synthesized images to determine the target deep learning network and the noise-free image corresponding to the synthesized image.

[0037] The generator network is determined by iteratively training the initial generator network and the initial discriminator network in the generative adversarial network based on multiple OCT sample images and the noise-free images.

[0038] This application also provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the imaging method of multi-aperture synthetic optical coherence tomography as described above are performed.

[0039] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the imaging method of multi-aperture synthetic optical coherence tomography as described above.

[0040] This application provides an imaging method, apparatus, device, and medium for multi-aperture synthetic optical coherence tomography (OCT). The imaging method includes: scanning and detecting a target object using an aperture synthetic optical coherence tomography module within a speckle-free multi-aperture synthetic optical coherence tomography system to determine an OCT image of the target object; inputting the OCT image into a generator network within the speckle-free multi-aperture synthetic optical coherence tomography system to denoise the OCT image and outputting a noise-removed OCT image; wherein the speckle-free multi-aperture synthetic optical coherence tomography system is obtained by combining the multi-aperture synthetic optical coherence tomography module with a generator network obtained through iterative training of a generative adversarial network. By utilizing the generator network within the speckle-free multi-aperture synthetic optical coherence tomography system to denoise the OCT image, a noise-removed and resolution-enhanced OCT image is obtained, thereby improving the resolution of the OCT image.

[0041] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0042] 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.

[0043] Figure 1 A flowchart of an imaging method for multi-aperture synthetic optical coherence tomography provided in this application embodiment;

[0044] Figure 2 One of the schematic diagrams of the imaging method for multi-aperture synthetic optical coherence tomography provided in the embodiments of this application;

[0045] Figure 3 A second schematic diagram showing the results of an imaging method for multi-aperture synthetic optical coherence tomography provided in an embodiment of this application;

[0046] Figure 4 This is one of the structural schematic diagrams of an imaging device for multi-aperture synthetic optical coherence tomography provided in an embodiment of this application;

[0047] Figure 5 A second schematic diagram of a multi-aperture synthetic optical coherence tomography imaging device provided in an embodiment of this application;

[0048] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0049] 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 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.

[0050] 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.

[0051] To enable those skilled in the art to use the content of this application and, in conjunction with the specific application scenario of "imaging multi-aperture synthetic optical coherence tomography", the following implementation methods are provided. For those skilled in the art, the general principles defined herein can be applied to other embodiments and application scenarios without departing from the spirit and scope of this application.

[0052] First, the applicable scenarios for this application will be introduced. This application can be applied to the field of OCT technology.

[0053] Research has shown that Optical Coherence Tomography (OCT) is a non-invasive 3D imaging technique widely used in ophthalmology, cardiology, and endoscopy. In OCT imaging, axial and lateral resolutions are independent: the former is controlled by the optical bandwidth of the laser source, while the latter is controlled by the numerical aperture (NA) and wavelength of the objective lens. Specifically, lateral resolution is inversely proportional to the objective lens's NA and depth of focus (DOF), and is proportional to the square of the lateral resolution. Laterally, when the objective lens's NA is high, the resolution in the defocused area decreases significantly. Simultaneously, speckle noise reduces OCT imaging resolution and limits potential resolution enhancement techniques. Therefore, generating high-resolution OCT images has become a significant technical challenge.

[0054] Based on this, this application provides an imaging method for multi-aperture synthetic optical coherence tomography. By utilizing the generator network in the speckle-free multi-aperture synthetic optical coherence tomography system to denoise the OCT detection image, a noise-removed and resolution-enhanced OCT detection image is obtained, thereby improving the resolution of the OCT detection image.

[0055] Please see Figure 1 , Figure 1 This is a flowchart illustrating an imaging method for multi-aperture synthetic optical coherence tomography provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the imaging method includes:

[0056] S101: The aperture synthesis optical coherence tomography module in the speckle-free multi-aperture synthesis optical coherence tomography system scans and detects the object to be measured, and determines the OCT detection image of the object to be measured.

[0057] In this step, the object to be measured is scanned and detected using the aperture synthesis optical coherence tomography module in the speckle-free multi-aperture synthesis optical coherence tomography system, thereby determining the OCT detection image of the object to be measured.

[0058] Here, the speckle-free multi-aperture synthetic optical coherence tomography (OCT) system is pre-designed. This system is obtained by combining the multi-aperture synthetic optical coherence tomography module with a generator network obtained through iterative training of a generative adversarial network. The multi-aperture synthetic optical coherence tomography module can be any existing system capable of performing OCT inspection.

[0059] S102: Input the OCT detection image into the generator network of the speckle-free multi-aperture synthetic optical coherence tomography system, perform noise reduction processing on the OCT detection image, and output the noise-removed OCT detection image.

[0060] In this step, the OCT detection image output by the multipath synthesis optical coherence tomography module is input into the generator network to perform noise reduction processing on the OCT detection image, thereby obtaining a noise-removed OCT detection image.

[0061] Here, the noise-removed OCT detection image is a noise-removed and resolution-enhanced OCT image.

[0062] In one possible implementation, the generator network is determined through the following steps:

[0063] A: Based on the multi-aperture synthetic optical coherence tomography module, multiple OCT sample images are determined.

[0064] Here, multiple OCT sample images are output from the multi-aperture synthetic optical coherence tomography module.

[0065] Among them, the multi-aperture synthetic optical coherence tomography module can output low-resolution OCT sample images of multiple apertures at once, such as five low-resolution OCT sample images of five apertures at the same time.

[0066] B: Perform synthesis processing on multiple OCT sample images to determine multiple synthesized images.

[0067] Here, multiple OCT sample images are synthesized to determine multiple composite images. Specifically, multiple low-resolution OCT sample images with multiple apertures are simultaneously output and synthesized to determine a single composite image. For example, five low-resolution OCT sample images with five apertures are synthesized to obtain a single composite image.

[0068] C: Iteratively train the initial deep learning network based on multiple synthesized images to determine the target deep learning network and the noise-free image corresponding to the synthesized image.

[0069] Here, the initial deep learning network is iteratively trained based on multiple synthesized images to determine the target deep learning network and the noise-free image corresponding to the synthesized image.

[0070] Here, the initial deep learning network includes a first initial deep learning network and a second initial deep learning network, and the target deep learning network includes a first target deep learning network and a second target deep learning network. The first initial deep learning network is constructed as a U-Net structure consisting of an encoder and a decoder. The encoder consists of encoding blocks composed of convolutional layers, densely connected blocks, and max-pooling layers. The decoder consists of deconvolutional layers, densely connected blocks, and convolutional layers. Each densely connected block consists of three 3×3 convolutional layers with activation functions and one 1×1 convolutional layer. The output of each 3×3 convolutional layer is connected to the subsequent convolutional layer via skip connections. The encoder downsamples the image using five encoding blocks to obtain high-level features, while the decoder upsamples the high-level features to recover the image using five decoding blocks. The output of each encoding block is also connected to the corresponding decoding block via skip connections. The network structures of the first and second initial deep learning networks are consistent.

[0071] In one possible implementation, the initial deep learning network includes a first initial deep learning network and a second initial deep learning network, and the target deep learning network includes a first target deep learning network and a second target deep learning network. The step of iteratively training the initial deep learning network based on multiple synthesized images to determine the target deep learning network and the noise-free image corresponding to the synthesized image includes:

[0072] (1): Sampling adjacent pixels of the synthesized image to determine the first noise image and the second noise image.

[0073] Here, adjacent pixels of the synthesized image are sampled to determine the first noise image and the second noise image.

[0074] In this process, adjacent pixels of the synthesized image are sampled to obtain two paired noise images ([g1(y),g2(y)]) with similar content but no noise correlation, where g1(y) and g2(y) are the first noise image and the second noise image, respectively.

[0075] (2): Input the first noisy image into the first initial deep learning network for denoising processing, output the first image, and iteratively train the first initial deep learning network based on the first image and the second noisy image to determine the first target deep learning network.

[0076] Here, the first noisy image is input into the first initial deep learning network for denoising processing, and the first image is output. The first initial deep learning network is then iteratively trained based on the first image and the second noisy image to determine the first target deep learning network.

[0077] In one possible implementation, the step of iteratively training the first initial deep learning network based on the first image and the second noisy image to determine the first target deep learning network includes:

[0078] a: Determine the first mean error loss value between the first image and the second noisy image.

[0079] Here, the first mean error loss value between the first image and the second noisy image is determined according to the following formula:

[0080]

[0081] Among them, f θ (g1(y)) is the first image, g2(y) is the second noisy image, L NBR γ is the first mean error loss value, γ is an adjustable parameter, usually set to 2, θ is the network parameter of the first initial deep learning network, and g(·) is the neighboring pixel sampler.

[0082] b: Adjust the network parameters of the first initial deep learning network based on the first mean error loss value until the decrease of the first mean error loss value tends to level off, and then stop iterative training to determine the first target deep learning network.

[0083] Here, the network parameters of the first initial deep learning network are adjusted according to the first mean error loss value until the decrease of the first mean error loss value tends to level off, and then the iterative training is stopped to determine the first target deep learning network.

[0084] (3): Input the synthesized image into the first target deep learning network for denoising processing and output the second image; perform image processing on the second image to determine the third noisy image.

[0085] Here, the synthesized image is input into the first target deep learning network for denoising processing, and the second image is output; the second image is then processed to determine the third noisy image.

[0086] The second image (f) is obtained by inputting the synthesized image into the first target deep learning network. θ (y)), and obtain g1(f) by sampling adjacent pixels of the second image. θ (y)) and g2(f θ (y)), then through g1(y)+[g2(y)-g2(f) θ [(y))] Construct a third noisy image with content similar to g1(y) but with more noise.

[0087] (4): Input the third noisy image into the second initial deep learning network for denoising processing, output the third image, and iteratively train the second initial deep learning network based on the third image and the first noisy image to determine the second target deep learning network.

[0088] Here, the third noisy image is input into the second initial deep learning network for denoising processing, and the third image is output. The second initial deep learning network is then iteratively trained based on the third image and the first noisy image to determine the second target deep learning network.

[0089] Wherein, the third noisy image Z = g1(y) + [g2(y) - g2(f)] θ (y))] is used as the input to the second initial deep learning network, and the mean square error loss L between the output third image and the first noisy image of the second initial deep learning network is calculated. Noisier2Noise Update the parameters of the second initial deep learning network.

[0090] L is calculated using the following formula. Noisier2Noise Calculation:

[0091]

[0092] Among them, This represents the output of the second initial deep learning network. Let g1(y) represent the parameters of the second initial deep learning network, and g1(y) be the first noisy image.

[0093] (5): Input the synthesized image into the second target deep learning network and output the noise-free image.

[0094] Here, the synthesized image is input into the second target deep learning network, which outputs a noise-free image.

[0095] In one possible implementation, inputting the synthesized image into the second target deep learning network and outputting the noise-free image includes:

[0096] The synthesized image is input into the second target deep learning network to output a fourth image; based on the difference between twice the pixel value of the fourth image and the pixel value of the synthesized image, the noise-free image after denoising is determined.

[0097] Here, the synthesized image is input into the second target deep learning network, and the fourth image is output. Based on the difference between twice the pixel value of the fourth image and the pixel value of the synthesized image, the noise-free image after denoising is determined.

[0098] The image to be denoised (synthesized image y) is input into the second target deep learning network, and the output of the second target deep learning network is the fourth image. Multiply by 2 and then subtract the image to be denoised to obtain a clean image after denoising, i.e., a clean image.

[0099] D: Based on multiple OCT sample images and the noise-free images, the initial generator network and the initial discriminator network in the generative adversarial network are iteratively trained to determine the generator network.

[0100] Here, the initial generator network and the initial discriminator network in the generative adversarial network are iteratively trained based on multiple OCT sample images and noise-free images to determine the generator network.

[0101] The Generative Adversarial Network (GAN) consists of a generator and a discriminator. The initial generator network structure is consistent with the first objective deep learning network structure. The discriminator uses a multi-scale discriminator network, employing three discriminators of different scales to enhance the GAN's performance. The first discriminator has the same input size as the original, while the second and third discriminators use a max-pooling layer to reduce the input to half and a quarter of its original size, respectively. Each discriminator consists of eight convolutional layers followed by activation functions and two dense layers.

[0102] In one possible implementation, the step of iteratively training the initial generator network and the initial discriminator network in the generative adversarial network based on multiple OCT sample images and the noise-free image to determine the generator network includes:

[0103] i: Input the OCT sample image into the initial generator network and output the fifth image.

[0104] Here, the OCT sample image is input into the initial generator network, and the fifth image is output.

[0105] ii: Determine the adversarial loss value and content loss value between the fifth image and the noise-free image.

[0106] Here, the adversarial loss value between the fifth image and the noise-free image is determined using the following formula:

[0107]

[0108] Among them, L adv To counteract the loss value, where d i (·) represents the i-th discriminator, h and w represent the height and width of the image, respectively. a (I)) is the fifth image.

[0109] Here, the content loss value between the fifth image and the noise-free image is determined using the following formula:

[0110]

[0111] Among them, L C I is the content loss value. r The image is noise-free, I is the fifth image, VGG 19 The image shows a trained 19-layer VGG network model, where β represents the network parameters.

[0112] iii: Input the fifth image and the noise-free image into the initial discriminator network to determine the discriminator loss value between the fifth image and the noise-free image.

[0113] Here, the discriminator loss is calculated using the following formula:

[0114]

[0115] Among them, L D I is the discriminator loss value. r I is the noise-free image, and D is the fifth image. i (·) represents the i-th discriminator, h and w represent the height and width of the image, respectively. a (I)) is the fifth image.

[0116] iv: Adjust the network parameters of the initial generator network based on the adversarial loss value and the content loss value, and adjust the network parameters of the initial discriminator network based on the discriminator loss value, until the adversarial loss value, the content loss value, and the discriminator loss value decrease gradually, and then stop iterative training to determine the generator network.

[0117] Here, the network parameters of the initial generator network are adjusted based on the adversarial loss value and the content loss value, and the network parameters of the initial discriminator network are adjusted based on the discriminator loss value. Iterative training stops when the adversarial loss value, the content loss value, and the discriminator loss value decrease gradually, thus determining the generator network.

[0118] Before training MAS Net (Speckle-Free Multi-Aperture Synthetic Optical Coherence Tomography), image denoising is necessary; otherwise, the distribution of complex scattering patterns will reduce the controllability of training. Here, three methods—BM3D, Neighbor2Neighbor, and the method adopted in this proposal—are compared to demonstrate the effectiveness of the generator network in this proposal. BM3D is a traditional denoising method that removes noise well but severely, destroying details in OCT images and making particles in the image unclear. Neighbor2Neighbor preserves details but does not effectively remove speckle noise. The denoising method proposed in this proposal not only completely eliminates speckle noise but also preserves detailed structure, making particles and interparticle gaps clearly visible, providing high-quality data for subsequent training of MAS Net. The quantitative results for each method are shown in the lower left corner of the image.

[0119] Please see Figure 2 , Figure 3 , Figure 2 This is one of the schematic diagrams showing the results of an imaging method for multi-aperture synthetic optical coherence tomography provided in an embodiment of this application. Figure 3 This is a second schematic diagram illustrating the results of an imaging method for multi-aperture synthetic optical coherence tomography provided in an embodiment of this application. Figure 2 As shown, the lateral resolution improvement performance of MAS NetOCT was verified by calibrating sample images using polystyrene microparticles. The samples were prepared by mixing agarose solution and polystyrene microparticles with a diameter of 6 μm. Figure 2 (a) is one of the five wells captured in the B scan. Figure 2 (b) is the B-scan after using the MAS algorithm. Figure 2 (c) is the corresponding denoised B-scan. Figure 2 (d) shows the results from MAS Net. The magnified boxes and horizontal outlines shown are illustrated by dashed lines in the right-hand portion of each image. By observation, the magnified... Figure 2 The two particles in (a) are stuck together and cannot be distinguished due to the very low lateral resolution, but they are in... Figure 2 (b) shows a clear separation. Because the high-resolution image is denoised to the ground truth used to train the MAS Net, the lateral resolution of the predicted result ( Figure 2 (d) is improved, and speckle noise is also removed, resulting in a clean, horizontally resolution-enhanced image. Horizontal line contours show a consistent intensity distribution. The horizontal resolution enhancement result ( Figure 2 (d) has almost the same lateral resolution as ground truth. Figure 2 (c)), as shown in the dashed box and its corresponding curve. Figure 3 This is the result of MAS Net's generalization ability on fresh grape samples. Figure 3 (a) is a B scan obtained from a conventional OCT system. Figure 3 (b) is the predicted image from MAS Net.

[0120] This paper proposes a speckle-free MAS-Net OCT to improve lateral resolution and extend the DOF of OCT systems. A novel self-supervised denoising method is presented to remove speckle noise from high-lateral-resolution OCT images from MAS. Pairs of low-resolution and denoised high-resolution images are input into the proposed MAS Net, a GAN constructed from U-Net with RDB blocks and a multi-scale discriminator. Experimental results on homemade microparticle samples and fresh lemon samples demonstrate the excellent denoising performance and feasibility of this method, as well as the application of MAS Net in enhancing lateral resolution.

[0121] This application provides an imaging method for multi-aperture synthetic optical coherence tomography (OCT). The method includes: scanning and detecting a target object using an aperture synthetic optical coherence tomography module within a speckle-free multi-aperture synthetic optical coherence tomography system to determine an OCT image of the target object; inputting the OCT image into a generator network within the speckle-free multi-aperture synthetic optical coherence tomography system to denoise the OCT image and outputting a noise-removed OCT image; wherein the speckle-free multi-aperture synthetic optical coherence tomography system is obtained by combining the multi-aperture synthetic optical coherence tomography module with a generator network obtained through iterative training of a generative adversarial network. By utilizing the generator network within the speckle-free multi-aperture synthetic optical coherence tomography system to denoise the OCT image, a noise-removed and resolution-enhanced OCT image is obtained, thereby improving the resolution of the OCT image.

[0122] Please see Figure 4 , Figure 5 , Figure 4 This is one of the structural schematic diagrams of an imaging device for multi-aperture synthetic optical coherence tomography provided in an embodiment of this application; Figure 4 This is a second schematic diagram of the structure of an imaging device for multi-aperture synthetic optical coherence tomography provided in an embodiment of this application. Figure 4 As shown, the imaging device 400 for multi-aperture synthetic optical coherence tomography includes:

[0123] The first detection module 410 is used to scan and detect the object to be measured based on the aperture synthesis optical coherence tomography module in the speckle-free multi-aperture synthesis optical coherence tomography system, and to determine the OCT detection image of the object to be measured.

[0124] The second detection module 420 is used to input the OCT detection image into the generator network of the speckle-free multi-aperture synthetic optical coherence tomography system, perform noise reduction processing on the OCT detection image, and output a noise-removed OCT detection image; wherein, the speckle-free multi-aperture synthetic optical coherence tomography system is obtained by combining the multi-aperture synthetic optical coherence tomography module and the generator network obtained by iteratively training the generative adversarial network.

[0125] Furthermore, such as Figure 5 As shown, the imaging device further includes a training module 430, which determines the generator network through the following steps:

[0126] Based on the multi-aperture synthetic optical coherence tomography module, multiple OCT sample images were determined;

[0127] Multiple OCT sample images are synthesized to determine multiple synthesized images;

[0128] The initial deep learning network is iteratively trained based on multiple synthesized images to determine the target deep learning network and the noise-free image corresponding to the synthesized image.

[0129] The generator network is determined by iteratively training the initial generator network and the initial discriminator network in the generative adversarial network based on multiple OCT sample images and the noise-free images.

[0130] Furthermore, when the training module 430 is used for iterative training of the initial deep learning network (including a first initial deep learning network and a second initial deep learning network) and the target deep learning network (including a first target deep learning network and a second target deep learning network), to determine the target deep learning network and the noise-free image corresponding to the synthesized image, the training module 430 is specifically used for:

[0131] The synthesized image is sampled from adjacent pixels to determine the first noise image and the second noise image;

[0132] The first noisy image is input into the first initial deep learning network for denoising processing, and the first image is output. The first initial deep learning network is iteratively trained based on the first image and the second noisy image to determine the first target deep learning network.

[0133] The synthesized image is input into the first target deep learning network for denoising processing, and a second image is output; the second image is processed to determine a third noisy image;

[0134] The third noisy image is input into the second initial deep learning network for denoising processing, and the third image is output. The second initial deep learning network is iteratively trained based on the third image and the first noisy image to determine the second target deep learning network.

[0135] The synthesized image is input into the second target deep learning network, and the noise-free image is output.

[0136] Furthermore, when the training module 430 iteratively trains the first initial deep learning network based on the first image and the second noisy image to determine the first target deep learning network, the training module 430 is specifically used for:

[0137] Determine the first mean error loss value between the first image and the second noisy image;

[0138] The network parameters of the first initial deep learning network are adjusted based on the first mean error loss value until the decrease of the first mean error loss value tends to level off, and then the iterative training is stopped to determine the first target deep learning network.

[0139] Furthermore, when the training module 430 is used to input the synthesized image into the second target deep learning network and output the noise-free image, the training module 430 is specifically used for:

[0140] The synthesized image is input into the second target deep learning network to output a fourth image;

[0141] The noise-free image after denoising is determined based on the difference between twice the pixel value of the fourth image and the pixel value of the synthesized image.

[0142] Furthermore, when the training module 430 iteratively trains the initial generator network and the initial discriminator network in the generative adversarial network based on multiple OCT sample images and the noise-free image to determine the generator network, the training module 430 is specifically used for:

[0143] The OCT sample image is input into the initial generator network, and the fifth image is output.

[0144] The adversarial loss value and content loss value between the fifth image and the noise-free image were determined;

[0145] The fifth image and the noise-free image are input into the initial discriminator network to determine the discriminator loss value between the fifth image and the noise-free image;

[0146] The network parameters of the initial generator network are adjusted based on the adversarial loss value and the content loss value, and the network parameters of the initial discriminator network are adjusted based on the discriminator loss value. Iterative training is stopped when the adversarial loss value, the content loss value, and the discriminator loss value decrease gradually, and the generator network is determined.

[0147] This application provides an imaging device for multi-aperture synthetic optical coherence tomography (OCT). The imaging device includes: a first detection module, used to scan and detect a measurement object based on an aperture synthetic optical coherence tomography module in a speckle-free multi-aperture synthetic optical coherence tomography system, and determine an OCT detection image of the measurement object; and a second detection module, used to input the OCT detection image into a generator network in the speckle-free multi-aperture synthetic optical coherence tomography system, perform denoising processing on the OCT detection image, and output a noise-removed OCT detection image. The speckle-free multi-aperture synthetic optical coherence tomography system is obtained by combining the multi-aperture synthetic optical coherence tomography module with a generator network obtained through iterative training of a generative adversarial network. By using the generator network in the speckle-free multi-aperture synthetic optical coherence tomography system to denoise the OCT detection image, a noise-removed and resolution-enhanced OCT detection image is obtained, thereby improving the resolution of the OCT detection image.

[0148] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device 600 includes a processor 610, a memory 620, and a bus 630.

[0149] The memory 620 stores machine-readable instructions executable by the processor 610. When the electronic device 600 is running, the processor 610 and the memory 620 communicate via the bus 630. When the machine-readable instructions are executed by the processor 610, they can perform the operations described above. Figure 1 The steps of the imaging method for multi-aperture synthetic optical coherence tomography in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0150] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The steps of the imaging method for multi-aperture synthetic optical coherence tomography in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0151] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0152] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0153] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0154] In addition, 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.

[0155] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, 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 a portion 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 described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0156] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method of imaging for multi-aperture synthetic optical coherence tomography, characterized by, The imaging method comprises: scanning and detecting a detection measurement object based on a multi-aperture synthetic optical coherence tomography module in the speckle-free multi-aperture synthetic optical coherence tomography system, to determine an OCT detection image of the detection measurement object; inputting the OCT detection image into a generator network in the speckle-free multi-aperture synthetic optical coherence tomography system, to perform denoising processing on the OCT detection image, and outputting a noise-removed OCT detection image; wherein the speckle-free multi-aperture synthetic optical coherence tomography system is obtained by combining the multi-aperture synthetic optical coherence tomography module and the generator network obtained by iteratively training a generative adversarial network; the generator network is determined by: determining a plurality of OCT sample images based on the multi-aperture synthetic optical coherence tomography module; performing synthesis processing on the plurality of OCT sample images to determine a plurality of synthetic images; iteratively training an initial deep learning network based on the plurality of synthetic images to determine a target deep learning network and a noise-free image corresponding to the synthetic image; iteratively training an initial generator network and an initial discriminator network in the generative adversarial network based on the plurality of OCT sample images and the noise-free image to determine the generator network.

2. The imaging method of claim 1, wherein, The initial deep learning network comprises a first initial deep learning network and a second initial deep learning network, and the target deep learning network comprises a first target deep learning network and a second target deep learning network. The iteratively training an initial deep learning network based on the plurality of synthetic images to determine a target deep learning network and a noise-free image corresponding to the synthetic image comprises: sampling adjacent pixels of the synthetic image to determine a first noise image and a second noise image; wherein the first noise image and the second noise image are content-similar but noise-uncorrelated paired noise images; inputting the first noise image into the first initial deep learning network for denoising processing to output a first image, and iteratively training the first initial deep learning network based on the first image and the second noise image to determine the first target deep learning network; inputting the synthetic image into the first target deep learning network for denoising processing to output a second image; performing image processing on the second image to determine a third noise image; inputting the third noise image into the second initial deep learning network for denoising processing to output a third image, and iteratively training the second initial deep learning network based on the third image and the first noise image to determine the second target deep learning network; inputting the synthetic image into the second target deep learning network to output the noise-free image.

3. The imaging method of claim 2, wherein, The iteratively training the first initial deep learning network based on the first image and the second noise image to determine the first target deep learning network comprises: determining a first mean error loss value between the first image and the second noise image; Adjust network parameters of the first initial deep learning network based on the first mean error loss value until the first mean error loss value stops decreasing and tends to be flat, and then stop iterative training, to determine the first target deep learning network.

4. The imaging method of claim 2, wherein, The inputting the synthetic image into the second target deep learning network and outputting the noise-free image comprises: inputting the synthetic image into the second target deep learning network and outputting a fourth image; Based on the difference between the two times pixel value of the fourth image and the pixel value of the synthetic image, the noise-free image after removing noise is determined.

5. The imaging method of claim 1, wherein, The iterative training of the initial generator network and the initial discriminator network in the generative adversarial network based on multiple OCT sample images and the noise-free image comprises: inputting the OCT sample image into the initial generator network to output a fifth image; determining the adversarial loss value and the content loss value between the fifth image and the noise-free image; inputting the fifth image and the noise-free image into the initial discriminator network to determine the discriminator loss value between the fifth image and the noise-free image; Adjust the network parameters of the initial generator network based on the adversarial loss value and the content loss value, and adjust the network parameters of the initial discriminator network based on the discriminator loss value, until the adversarial loss value, the content loss value and the discriminator loss value stop decreasing and tend to be flat, and then stop iterative training, to determine the generator network.

6. An imaging apparatus of a multi-aperture synthetic optical coherence tomography, characterized by, The imaging device comprises: The first detection module is used for scanning and detecting the detection measurement object based on the aperture synthesis optical coherence tomography module in the speckle-free multi-aperture synthesis optical coherence tomography system, to determine the OCT detection image of the detection measurement object. The second detection module is used for inputting the OCT detection image into the generator network in the speckle-free multi-aperture synthesis optical coherence tomography system to perform noise removal processing on the OCT detection image and output the noise-removed OCT detection image; wherein the speckle-free multi-aperture synthesis optical coherence tomography system is obtained by combining the multi-aperture synthesis optical coherence tomography module and the generator network obtained by iterative training of the generative adversarial network. The imaging device further comprises a training module which determines the generator network by the following steps: Based on the multi-aperture synthesis optical coherence tomography module, multiple OCT sample images are determined. Synthetic processing is performed on multiple OCT sample images to determine multiple synthetic images. Iterative training is performed on an initial deep learning network based on multiple synthetic images to determine a target deep learning network and a noise-free image corresponding to the synthetic image. Iterative training is performed on an initial generator network and an initial discriminator network in the generative adversarial network based on multiple OCT sample images and the noise-free image to determine the generator network.

7. An electronic device, comprising: It comprises: A processor, a memory, and a bus, the memory storing machine readable instructions executable by the processor, when the electronic device is running, the processor and the memory communicating through the bus, the machine readable instructions being executed by the processor to perform the steps of the method for imaging of multi-aperture synthetic optical coherence tomography according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, A computer readable storage medium storing a computer program, the computer program being executed by a processor to perform the steps of the method for imaging of multi-aperture synthetic optical coherence tomography according to any one of claims 1 to 5.

Images