Conjugate mirror image removing method and device based on deep learning

Abstract

The application relates to the field of image processing and discloses an OCT conjugate mirror image removing method and device based on deep learning, which comprises the following steps: constructing a complex conjugate image of spectral data, constructing a true anterior eye segment image of the spectral data, performing image preprocessing on the complex conjugate image and the true anterior eye segment image; analyzing a pseudo-anterior eye segment image corresponding to the preprocessed conjugate image, calculating an absolute value loss between the pseudo-anterior eye segment image and the preprocessed anterior eye segment image; discriminating the falsity between the pseudo-anterior eye segment image and the preprocessed conjugate image, discriminating the authenticity between the preprocessed conjugate image and the preprocessed anterior eye segment image, calculating a generative network loss of a generative network; performing model training on a pre-trained neural network model to obtain a trained neural network model; performing model verification on the trained neural network model to obtain a verified neural network model, calculating a peak signal-to-noise ratio of the verified neural network model, and selecting a final neural network model. The application can reduce the difficulty of removing the OCT conjugate mirror image.

Description

Technical Field

[0001] This invention relates to the field of image processing, and in particular to a method and apparatus for OCT conjugate image removal based on deep learning. Background Technology

[0002] While Fourier domain optical coherence tomography (OCT) offers advantages over temporal domain OCT in terms of higher sensitivity and faster measurement speed, it also has inherent limitations: it is susceptible to the influence of complex conjugate images. Because the system's highest sensitivity is located near zero phase delay, practical processing methods have low utilization of the imaging region with the highest signal-to-noise ratio. Due to conjugate symmetry, the usable imaging range is only about half of the entire imaging range. Eliminating the image effect could significantly expand the detection range. In recent years, researchers both domestically and internationally have proposed a series of methods for removing complex conjugate images. The basic idea behind these methods is to construct complex interference spectra using various techniques. However, regardless of whether it's carrier frequency or phase-shifting methods, only a few can achieve high-speed imaging (>20 kHz). The imaging speed is too slow, making it difficult to perform biopsies on biological samples. Alternatively, it may require expensive or difficult-to-operate devices (such as electro-optic, acousto-optic, or multiple spectrometers), leading to experimental difficulties. For example, Fuzhou University disclosed a fixed-phase-difference dual-channel spectral domain OCT device in its published patent document CN109297597A, which uses two spectrometers to simultaneously acquire interference information from two phases, obtaining two interference signals with a 90° phase difference. Although the system's imaging speed is not reduced, the use of two spectrometers places high demands on equipment calibration and operation. Therefore, OCT conjugate image removal is quite difficult. Summary of the Invention

[0003] To address the aforementioned problems, this invention provides a method and apparatus for OCT conjugate image removal based on deep learning, which can reduce the difficulty of OCT conjugate image removal.

[0004] In a first aspect, the present invention provides a deep learning-based OCT conjugate image removal method, comprising:

[0005] Collect spectral data, construct a complex conjugate image of the spectral data, and construct a true anterior segment image of the spectral data. Perform image preprocessing on the complex conjugate image and the true anterior segment image to obtain a preprocessed conjugate image and a preprocessed anterior segment image.

[0006] The pseudo anterior segment image corresponding to the preprocessed conjugate image is analyzed using the generative network in the pretrained neural network model, and the absolute value loss between the pseudo anterior segment image and the preprocessed anterior segment image is calculated using the absolute value loss function of the generative network.

[0007] The discriminant network in the pre-trained neural network model is used to determine the falsity between the pseudo anterior segment image and the preprocessed conjugate image, and the discriminant network is used to determine the authenticity between the preprocessed conjugate image and the preprocessed anterior segment image. Based on the falsity, the authenticity, and the absolute value loss, the generation network loss of the generation network is calculated.

[0008] Based on the generated network loss, the pre-trained neural network model is trained to obtain a trained neural network model;

[0009] The trained neural network model is validated to obtain a validated neural network model. The peak signal-to-noise ratio (PSNR) between the pseudo anterior segment image and the real anterior segment image corresponding to the validated neural network model is calculated. The final neural network model is selected from the validated neural network models using the PSNR.

[0010] The final neural network model is used to remove the complex conjugation from the current complex conjugated image to obtain an anterior segment image with the complex conjugation removed.

[0011] In one possible implementation of the first aspect, constructing the true anterior segment image of the spectral data includes:

[0012] The spectral data is modulated to obtain modulated data;

[0013] Construct the complex conjugate image corresponding to the modulation data to obtain the modulation complex conjugate image;

[0014] The true anterior segment image corresponding to the modulated complex conjugate image is calculated using the following formula:

[0015]

[0016] Wherein, H(u)(t) represents the true anterior segment image, p*v represents the Cauchy principal value, which is the integral value of the imperfect integral, u represents the modulated complex conjugate image, τ represents the independent variable parameter, and t represents the pixel value in the modulated complex conjugate image u(t).

[0017] In one possible implementation of the first aspect, the image preprocessing of the complex conjugate image and the true anterior segment image to obtain a preprocessed conjugate image and a preprocessed anterior segment image includes:

[0018] The resolution of the complex conjugate image and the anterior segment image of the true eye is scaled to obtain the scaled complex conjugate image and the scaled anterior segment image of the true eye;

[0019] The scaled complex conjugate image and the scaled anterior segment image are normalized to obtain a normalized complex conjugate image and a normalized anterior segment image.

[0020] The normalized complex conjugate image is divided into a training set, a validation set, and a test set corresponding to the normalized complex conjugate image. The normalized complex conjugate image and the corresponding training set, validation set, and test set are used as the preprocessed conjugate image and the preprocessed front section image. The normalized complex conjugate image and the corresponding training set, validation set, and test set are used for subsequent model training, model performance verification, and model testing.

[0021] In one possible implementation of the first aspect, the analysis of the pseudo anterior segment map corresponding to the preprocessed conjugate map using a generative network in a pre-trained neural network model includes:

[0022] The convolutional layer-activation layer-batch normalization layer in the generator network is used to extract features from the training set in the preprocessed conjugate graph to obtain feature vectors;

[0023] The pseudo-front view image corresponding to the feature vector is generated using the deconvolution layer-activation layer-batch normalization layer in the generative network.

[0024] In one possible implementation of the first aspect, calculating the absolute loss between the pseudo-anterior segment image and the preprocessed anterior segment image using the absolute value loss function of the generating network includes:

[0025] Image binarization is performed on the training sets corresponding to the pseudo anterior segment image and the preprocessed anterior segment image, respectively, to obtain the binarized pseudo anterior segment image and the binarized preprocessed anterior segment image;

[0026] The absolute value loss between the binarized pseudo-front view image and the binarized preprocessed front view image is calculated using the absolute value loss function, wherein the absolute value loss function includes:

[0027] L L1(G) =0.8*L1 foreground +0.2*L1 background

[0028] Among them, L L1(G) L1 represents the absolute value loss. foreground L1 represents the absolute loss value between the foreground region of the binarized pseudo-foreground image and the foreground region of the binarized preprocessed foreground image. background This represents the absolute value of the loss between the background region of the binarized pseudo-anterior segment image and the background region of the binarized preprocessed anterior segment image; 0.8 represents L1. foregroundThe weight, 0.2 represents L1. background The weight.

[0029] In one possible implementation of the first aspect, calculating the generator network loss based on the falsity, the authenticity, and the absolute value loss includes:

[0030] Based on the falsity, the authenticity, and the absolute value loss, the generation network loss of the generation network is calculated using the following formula:

[0031]

[0032] Among them, G * Let L represent the loss of the generator network, x represent the training set of the preprocessed conjugate graph, y represent the training set of the preprocessed anterior segment graph, z represent the noise added to the generator network when converting the training set of the preprocessed conjugate graph into the pseudo anterior segment graph, E represent the expected value, and λ represent the coefficient used to control L. L1(G) The weights are λ, and the larger λ is, the better, ranging from 300 to 700. D(x,y) represents the output classification probability obtained by simultaneously inputting the training set of the preprocessed conjugate map and the training set of the preprocessed anterior segment map into the discriminant network, i.e., the authenticity. G(x,z) represents the pseudo anterior segment map, and D(x,G(x,z)) represents the output classification probability obtained by simultaneously inputting the training set of the preprocessed conjugate map and the pseudo anterior segment map into the discriminant network, i.e., the falsity.

[0033] In one possible implementation of the first aspect, the step of training the pre-trained neural network model based on the generative network loss to obtain a trained neural network model includes:

[0034] Based on the loss of the generator network, the parameters of the generator network and the discriminator network are iteratively updated by alternating training of the generator network and the discriminator network, resulting in alternatingly trained generator network and alternatingly trained discriminator network;

[0035] When a preset Nash equilibrium is reached between the alternately trained generative network and the alternately trained discriminative network, the trained neural network model is obtained.

[0036] In one possible implementation of the first aspect, the step of performing model validation on the trained neural network model to obtain a validated neural network model includes:

[0037] Obtain the preprocessed conjugate map and the preprocessed anterior segment map, and use the validation set corresponding to the preprocessed conjugate map and the preprocessed anterior segment map to perform model validation on the trained neural network model to obtain the validated neural network model.

[0038] In one possible implementation of the first aspect, calculating the peak signal-to-noise ratio (PSNR) between the pseudo anterior segment image and the real anterior segment image corresponding to the validated neural network model includes:

[0039] Obtain the preprocessed conjugate image and the preprocessed anterior segment image, and use the test set corresponding to the preprocessed conjugate image and the preprocessed anterior segment image to determine the pseudo anterior segment image and the true anterior segment image corresponding to the validated neural network model;

[0040] The peak signal-to-noise ratio (PSNR) between the pseudo anterior segment image and the real anterior segment image corresponding to the validated neural network model is calculated using the following formula:

[0041]

[0042] Wherein, PSNR represents the peak signal-to-noise ratio, MAX represents the largest pixel in the pseudo anterior segment image corresponding to the verified neural network model, with a value of 8 bits, indicating that the highest pixel value is 255, and MSE represents the mean square error between the pseudo anterior segment image and the real anterior segment image corresponding to the verified neural network model.

[0043] Secondly, the present invention provides a deep learning-based OCT conjugate image removal device, the device comprising:

[0044] The image preprocessing module is used to acquire spectral data, construct a complex conjugate image of the spectral data, and construct a true anterior segment image of the spectral data. The complex conjugate image and the true anterior segment image are preprocessed to obtain a preprocessed conjugate image and a preprocessed anterior segment image.

[0045] The absolute value loss calculation module is used to analyze the pseudo anterior segment image corresponding to the preprocessed conjugate image using the generative network in the pretrained neural network model, and to calculate the absolute value loss between the pseudo anterior segment image and the preprocessed anterior segment image using the absolute value loss function of the generative network.

[0046] The network loss calculation module is used to use the discriminant network in the pre-trained neural network model to determine the falsity between the pseudo anterior segment image and the preprocessed conjugate image, and to use the discriminant network to determine the authenticity between the preprocessed conjugate image and the preprocessed anterior segment image. Based on the falsity, the authenticity, and the absolute value loss, the module calculates the generation network loss of the generation network.

[0047] The model training module is used to train the pre-trained neural network model based on the generative network loss to obtain a trained neural network model.

[0048] The model selection module is used to perform model verification on the trained neural network model to obtain a verified neural network model, and to calculate the peak signal-to-noise ratio between the pseudo anterior segment image and the real anterior segment image corresponding to the verified neural network model, and to select the final neural network model from the verified neural network models using the peak signal-to-noise ratio.

[0049] The complex conjugate removal module is used to remove complex conjugates from the current complex conjugate image using the final neural network model, so as to obtain an anterior segment image with complex conjugates removed.

[0050] Compared with existing technologies, the technical principles and beneficial effects of this solution are as follows:

[0051] This invention first constructs a complex conjugate image of the spectral data to generate training samples for subsequent model training. Further, it constructs a true anterior segment image of the spectral data to generate real data for model prediction, which is then compared with the model prediction results to optimize the predictions. Further, it preprocesses the complex conjugate image and the true anterior segment image to scale the original image resolution to 1024*1536 to fit the network's input size, and then normalizes the pixel values ​​of each image to -1 to 1, making the network more convergent. Finally, it uses a generative network in a pre-trained neural network model to analyze the pseudo anterior segment image corresponding to the preprocessed conjugate image to remove the pseudo anterior segment image. By refining the conjugate image in the conjugate image, a pseudo-anterior segment image with conjugate removed is obtained. Further, this embodiment of the invention calculates the absolute value loss between the pseudo-anterior segment image and the preprocessed anterior segment image using the absolute value loss function of the generator network, to make the loss calculation more accurate. Since the background occupies most of the OCT image and is mostly black, and the focus is mainly on the foreground, which occupies a smaller portion, directly calculating the loss would result in the foreground's small proportion being dominant, leading to inaccurate calculations. Therefore, to enhance the difference between the foreground and background, the absolute value distance is calculated in two parts. Further, this embodiment of the invention calculates the peak signal-to-noise ratio (PSNR) between the pseudo-anterior segment image and the real anterior segment image corresponding to the validated neural network model, to select the optimal model from the tested models based on the PSNR. Therefore, this embodiment of the invention proposes a deep learning-based OCT conjugate image removal method and apparatus, which can reduce the difficulty of OCT conjugate image removal. Attached Figure Description

[0052] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart illustrating a deep learning-based OCT conjugate image removal method according to an embodiment of the present invention.

[0055] Figure 2 As shown in one embodiment of the present invention Figure 1 A schematic diagram illustrating the construction of complex conjugate images using a deep learning-based OCT conjugate image removal method;

[0056] Figure 3a As shown in one embodiment of the present invention Figure 1 A schematic diagram of a complex conjugate image provided by a deep learning-based OCT conjugate image removal method;

[0057] Figure 3b As shown in one embodiment of the present invention Figure 1 A schematic diagram of a true anterior segment image using a deep learning-based OCT conjugate mirror removal method;

[0058] Figure 4 As shown in one embodiment of the present invention Figure 1 A schematic diagram of complex conjugate removal in a deep learning-based OCT conjugate image removal method is provided;

[0059] Figure 5a As shown in one embodiment of the present invention Figure 1 A schematic diagram of the generative network training process for a deep learning-based OCT conjugate image removal method is provided.

[0060] Figure 5b As shown in one embodiment of the present invention Figure 1 A schematic diagram illustrating the training process of the discriminant network for a deep learning-based OCT conjugate image removal method;

[0061] Figure 5c As shown in one embodiment of the present invention Figure 1 A schematic diagram is provided showing how a deep learning-based OCT conjugate mirror removal method achieves a preset Nash equilibrium during training.

[0062] Figure 6 A schematic diagram of a module for a deep learning-based OCT conjugate image removal device according to an embodiment of the present invention;

[0063] Figure 7This is a schematic diagram of the internal structure of an electronic device that implements a deep learning-based OCT conjugate image removal method according to an embodiment of the present invention. Detailed Implementation

[0064] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0065] This invention provides a deep learning-based OCT conjugate image removal method. The execution entity of this deep learning-based OCT conjugate image removal method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this invention: a server, a terminal, etc. In other words, the deep learning-based OCT conjugate image removal method can be executed by software or hardware installed on a terminal device or a server device. The software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0066] See Figure 1 The diagram shown is a flowchart illustrating a deep learning-based OCT conjugate image removal method according to an embodiment of the present invention. Figure 1 The deep learning-based OCT conjugate image removal methods described in the article include:

[0067] S1. Acquire spectral data, construct a complex conjugate image of the spectral data, and construct a true anterior segment image of the spectral data. Perform image preprocessing on the complex conjugate image and the true anterior segment image to obtain a preprocessed conjugate image and a preprocessed anterior segment image.

[0068] In this embodiment of the invention, a complex conjugate image of the spectral data is constructed to generate training samples for subsequent model training.

[0069] Optionally, the process of constructing the complex conjugate image of the spectral data is achieved through a spectral domain optical coherence tomography (SD-OCT) system. Optical coherence tomography is a non-invasive (non-destructive), non-contact imaging technique with micron-level resolution. It can obtain the tomographic structure inside tissues by using optical coherence gates.

[0070] See Figure 2 As shown, this is one embodiment of the present invention. Figure 1 This document provides a schematic diagram illustrating the construction of complex conjugate images using a deep learning-based OCT conjugate mirror removal method. Figure 2 In the spectrum, SLD represents the light source, FC represents the fiber optic coupler, NPBS represents the non-polarizing beam splitter, ND represents the filter, M represents the mirror, GS-xy represents the galvanometer, L represents the lens, S represents the sample to be tested, and SMF represents the single-mode fiber. The spectrometer consists of a DG grating, an LSC linear array, and a CCD camera.

[0071] Furthermore, in this embodiment of the invention, a true anterior segment image of the spectral data is constructed to generate real data for the model prediction results, which is then compared with the model prediction results to optimize the model prediction results.

[0072] In one embodiment of the present invention, constructing the anterior segment image of the spectral data includes: modulating the spectral data to obtain modulated data; constructing a complex conjugate image corresponding to the modulated data to obtain a modulated complex conjugate image; and calculating the anterior segment image of the real eye corresponding to the modulated complex conjugate image using the following formula:

[0073]

[0074] Wherein, H(u)(t) represents the true anterior segment image, p*v represents the Cauchy principal value, which is the integral value of the imperfect integral, u represents the modulated complex conjugate image, τ represents the independent variable parameter, and t represents the pixel value in the modulated complex conjugate image u(t).

[0075] Optionally, the process of constructing the true anterior segment image of the spectral data is as follows: from the modulated spectral data with phase shift introduced, a complex value data with a depth range increased by two times is calculated using an algorithm based on Hilbert transform to obtain the corresponding full-range anterior segment image.

[0076] See Figure 3a As shown, this is one embodiment of the present invention. Figure 1 A schematic diagram of a complex conjugate image is provided, based on a deep learning-based OCT conjugate image removal method.

[0077] See Figure 3b As shown, this is one embodiment of the present invention. Figure 1 A schematic diagram of a true anterior segment image using a deep learning-based OCT conjugate mirror removal method.

[0078] Furthermore, in this embodiment of the invention, image preprocessing is performed on the complex conjugate image and the true anterior segment image to scale the resolution of the original image to 1024*1536 to adapt to the input size of the network. Then, the pixel value range of each image is normalized to -1 to 1, making the network more likely to converge.

[0079] In one embodiment of the present invention, the image preprocessing of the complex conjugate image and the true anterior segment image to obtain a preprocessed conjugate image and a preprocessed anterior segment image includes: scaling the resolution of the complex conjugate image and the true anterior segment image respectively to obtain a scaled complex conjugate image and a scaled true anterior segment image; normalizing the scaled complex conjugate image and the scaled true anterior segment image to obtain a normalized complex conjugate image and a normalized true anterior segment image; dividing the normalized complex conjugate image and the corresponding training set, validation set, and test set, and using the normalized complex conjugate image and the corresponding training set, validation set, and test set as the preprocessed conjugate image and the preprocessed anterior segment image, wherein the normalized complex conjugate image and the corresponding training set, validation set, and test set are used for subsequent model training, model performance verification, and model testing.

[0080] S2. Analyze the pseudo anterior segment image corresponding to the preprocessed conjugate image using the generative network in the pretrained neural network model, and calculate the absolute value loss between the pseudo anterior segment image and the preprocessed anterior segment image using the absolute value loss function of the generative network.

[0081] This invention utilizes a generative network in a pre-trained neural network model to analyze the pseudo-anterior segment image corresponding to the preprocessed conjugate image, thereby removing complex conjugations from the preprocessed conjugate image and obtaining a pseudo-anterior segment image with complex conjugations removed. The generative network refers to a generator network G, using a U-Net network as the generator. The U-Net framework includes convolutional layers, deconvolutional layers, activation layers, and batch normalization layers.

[0082] In one embodiment of the present invention, the step of analyzing the pseudo anterior segment image corresponding to the preprocessed conjugate image using the generative network in the pretrained neural network model includes: extracting features from the training set in the preprocessed conjugate image using the convolutional layer-activation layer-batch normalization layer in the generative network to obtain a feature vector; and generating the pseudo anterior segment image corresponding to the feature vector using the deconvolutional layer-activation layer-batch normalization layer in the generative network.

[0083] Furthermore, in this embodiment of the invention, the absolute value loss between the pseudo-foreground image and the preprocessed foreground image is calculated using the absolute value loss function of the generator network to make the loss calculation more accurate. Since the background occupies most of the OCT image and is mostly black, and the focus is mainly on the foreground, the foreground occupies a small portion. Directly calculating the loss would result in the foreground accounting for a small proportion, which is equivalent to the background loss dominating, leading to inaccurate calculation. Therefore, in order to enhance the difference between the foreground and the background, the absolute value distance is calculated in two parts.

[0084] In one embodiment of the present invention, calculating the absolute value loss between the pseudo-anterior segment image and the preprocessed anterior segment image using the absolute value loss function of the generator network includes: binarizing the training sets corresponding to the pseudo-anterior segment image and the preprocessed anterior segment image respectively to obtain a binarized pseudo-anterior segment image and a binarized preprocessed anterior segment image; calculating the absolute value loss between the binarized pseudo-anterior segment image and the binarized preprocessed anterior segment image using the absolute value loss function, wherein the absolute value loss function includes:

[0085] L L1(G) =0.8*L1 foreground +0.2*L1 background

[0086] Among them, L L1(G) L1 represents the absolute value loss. foreground L1 represents the absolute loss value between the foreground region of the binarized pseudo-foreground image and the foreground region of the binarized preprocessed foreground image. background This represents the absolute value of the loss between the background region of the binarized pseudo-anterior segment image and the background region of the binarized preprocessed anterior segment image; 0.8 represents L1. foreground The weight, 0.2 represents L1. background The weight.

[0087] See Figure 4 As shown, this is one embodiment of the present invention. Figure 1 This diagram illustrates the complex conjugate removal process of a deep learning-based OCT conjugate image removal method. Figure 4 In this context, 'c' indicates the use of optimized L. L1(G) The model output obtained from the formula, where 'a' and 'b' represent the model output obtained by not calculating the loss values ​​of the background and foreground separately, shows that the model outputs of 'a' and 'b' do not completely remove the complex conjugate in the overlapping area of ​​the image. Optimization L... L1(G) The complex conjugate image of the region within the box after the calculation formula is removed quite well.

[0088] S3. Using the discriminant network in the pre-trained neural network model, determine the falsity between the pseudo anterior segment image and the preprocessed conjugate image, and use the discriminant network to determine the authenticity between the preprocessed conjugate image and the preprocessed anterior segment image. Based on the falsity, the authenticity, and the absolute value loss, calculate the generation network loss of the generation network.

[0089] Optionally, the process of using the discriminant network in the pre-trained neural network model to determine the falsity between the pseudo anterior segment image and the preprocessed conjugate image, and the process of using the discriminant network to determine the authenticity between the preprocessed conjugate image and the preprocessed anterior segment image, both involve first performing feature processing on the training sets of the preprocessed conjugate image and the training sets of the preprocessed anterior segment image, and then using activation functions to calculate their classification probabilities.

[0090] In one embodiment of the present invention, based on the falsity, the authenticity, and the absolute value loss, the generator network loss of the generator network is calculated using the following formula:

[0091]

[0092] Among them, G * Let L represent the loss of the generator network, x represent the training set of the preprocessed conjugate graph, y represent the training set of the preprocessed anterior segment graph, z represent the noise added to the generator network when converting the training set of the preprocessed conjugate graph into the pseudo anterior segment graph, E represent the expected value, and λ represent the coefficient used to control L. L1(G) The weights are λ, and the larger λ is, the better, ranging from 300 to 700. D(x,y) represents the output classification probability obtained by simultaneously inputting the training set of the preprocessed conjugate map and the training set of the preprocessed anterior segment map into the discriminant network, i.e., the authenticity. G(x,z) represents the pseudo anterior segment map, and D(x,G(x,z)) represents the output classification probability obtained by simultaneously inputting the training set of the preprocessed conjugate map and the pseudo anterior segment map into the discriminant network, i.e., the falsity.

[0093] S4. Based on the generated network loss, the pre-trained neural network model is trained to obtain a trained neural network model.

[0094] In one embodiment of the present invention, the step of training the pre-trained neural network model based on the generator network loss to obtain a trained neural network model includes: based on the generator network loss, iteratively updating the parameters of the generator network and the discriminator network by alternately training the generator network and the discriminator network to obtain alternately trained generator network and alternately trained discriminator network; and obtaining the trained neural network model when a preset Nash equilibrium is reached between the alternately trained generator network and the alternately trained discriminator network.

[0095] Optionally, the process of obtaining the trained neural network model when a preset Nash equilibrium is reached between the alternately trained generator network and the alternately trained discriminator network is as follows: the preprocessed sample image is fed into the generator network and the discriminator network, and the two are trained alternately, and the parameters are updated iteratively until the network reaches Nash equilibrium, that is, when the discriminator network D cannot distinguish between the authenticity of the real full-range image and the generated full-range image, the equilibrium is reached.

[0096] See Figure 5a As shown, this is one embodiment of the present invention. Figure 1 This diagram illustrates the training process of a generative network for a deep learning-based OCT conjugate image removal method. Figure 5a In the process, the complex conjugate image x is input into the generator network G to output a full-range anterior segment image G(x) through the generator network G. The complex conjugate image x and the full-range anterior segment image G(x) are simultaneously input into the discriminator network D to identify the authenticity between the generated full-range image and the complex conjugate image x.

[0097] See Figure 5b As shown, this is one embodiment of the present invention. Figure 1 This diagram illustrates the training process of a discriminant network for a deep learning-based OCT conjugate mirror removal method. Figure 5b In this process, the complex conjugate image x and the real full-range anterior segment image y are simultaneously input into the discriminator network D, so as to use the discriminator network D to identify the authenticity between the real full-range image y and the complex conjugate image x.

[0098] See Figure 5c As shown, this is one embodiment of the present invention. Figure 1 This diagram illustrates the training of a deep learning-based OCT conjugate mirror image removal method to achieve a preset Nash equilibrium. Figure 5cIn this process, the complex conjugate image x is input into the generator network G to output a full-range anterior segment image G(x). The complex conjugate image x, the generated full-range anterior segment image G(x), and the real full-range anterior segment image y are simultaneously input into the discriminator network D to identify the authenticity of the real full-range image and the generated full-range anterior segment image G(x).

[0099] S5. Perform model verification on the trained neural network model to obtain a verified neural network model, and calculate the peak signal-to-noise ratio between the pseudo anterior segment image and the real anterior segment image corresponding to the verified neural network model. Use the peak signal-to-noise ratio to select the final neural network model from the verified neural network models.

[0100] In one embodiment of the present invention, the step of performing model verification on the trained neural network model to obtain a verified neural network model includes: obtaining the preprocessed conjugate map and the preprocessed anterior segment map, and using the verification set corresponding to the preprocessed conjugate map and the preprocessed anterior segment map to perform model verification on the trained neural network model to obtain the verified neural network model.

[0101] Furthermore, in this embodiment of the invention, the peak signal-to-noise ratio (PSNR) between the pseudo anterior segment image and the real anterior segment image corresponding to the verified neural network model is calculated, so as to select the optimal model from the tested models based on the PSNR.

[0102] In one embodiment of the present invention, calculating the peak signal-to-noise ratio (PSNR) between the pseudo-anterior segment image and the real anterior segment image corresponding to the validated neural network model includes: obtaining the preprocessed conjugate image and the preprocessed anterior segment image; determining the pseudo-anterior segment image and the real anterior segment image corresponding to the validated neural network model using the test set corresponding to the preprocessed conjugate image and the preprocessed anterior segment image; and calculating the PSNR between the pseudo-anterior segment image and the real anterior segment image corresponding to the validated neural network model using the following formula:

[0103]

[0104] Wherein, PSNR represents the peak signal-to-noise ratio, MAX represents the largest pixel in the pseudo anterior segment image corresponding to the verified neural network model, with a value of 8 bits, indicating that the highest pixel value is 255, and MSE represents the mean square error between the pseudo anterior segment image and the real anterior segment image corresponding to the verified neural network model.

[0105] Optionally, the process of selecting the final neural network model from the verified neural network models using the peak signal-to-noise ratio is as follows: the model that performs best in terms of the peak signal-to-noise ratio is selected as the final neural network model.

[0106] S6. Use the final neural network model to remove the complex conjugate from the current complex conjugate image to obtain the anterior segment image with the complex conjugate removed.

[0107] Optionally, the process of removing the complex conjugate from the current complex conjugate image using the final neural network model is a process of generating a complex conjugate-removed image corresponding to the current complex conjugate image using the generator network in the final neural network model.

[0108] As can be seen, this embodiment of the invention first constructs a complex conjugate image of the spectral data to generate training samples for subsequent model training. Further, this embodiment constructs a true anterior segment image of the spectral data to construct real data for the model prediction results, which is then compared with the model prediction results to optimize the model prediction results. Further, this embodiment preprocesses the complex conjugate image and the true anterior segment image to scale the resolution of the original image to 1024*1536 to adapt to the network's input size. Then, it normalizes the pixel value range of each image to -1 to 1, making the network more likely to converge. Further, this embodiment uses the generative network in the pre-trained neural network model to analyze the pseudo anterior segment image corresponding to the preprocessed conjugate image to remove... The preprocessed conjugate image is used to obtain a pseudo-anterior segment image with conjugation removed. Further, this embodiment calculates the absolute loss between the pseudo-anterior segment image and the preprocessed anterior segment image using the absolute loss function of the generator network to improve the accuracy of the loss calculation. Since the background occupies most of the OCT image and is mostly black, and the focus is primarily on the foreground, which occupies a smaller portion, directly calculating the loss would result in a small foreground proportion, meaning the background loss dominates and the calculation is inaccurate. Therefore, to enhance the difference between the foreground and background, the absolute distance is calculated in two parts. Further, this embodiment calculates the peak signal-to-noise ratio (PSNR) between the pseudo-anterior segment image and the real anterior segment image corresponding to the validated neural network model to select the optimal model from the tested models based on the PSNR. Therefore, the deep learning-based OCT conjugate image removal method proposed in this embodiment can reduce the difficulty of OCT conjugate image removal.

[0109] like Figure 6 The diagram shown is a functional block diagram of the OCT conjugate image removal device based on deep learning according to the present invention.

[0110] The deep learning-based OCT conjugate image removal device 600 described in this invention can be installed in an electronic device. Depending on the functions implemented, the deep learning-based OCT conjugate image removal device may include an image preprocessing module 601, an absolute value loss calculation module 602, a network loss calculation module 603, a model training module 604, a model selection module 605, and a complex conjugate removal module 606. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0111] In this embodiment of the invention, the functions of each module / unit are as follows:

[0112] The image preprocessing module 601 is used to acquire spectral data, construct a complex conjugate image of the spectral data, and construct a true anterior segment image of the spectral data. The complex conjugate image and the true anterior segment image are preprocessed to obtain a preprocessed conjugate image and a preprocessed anterior segment image.

[0113] The absolute value loss calculation module 602 is used to analyze the pseudo anterior segment image corresponding to the preprocessed conjugate image using the generative network in the pretrained neural network model, and to calculate the absolute value loss between the pseudo anterior segment image and the preprocessed anterior segment image using the absolute value loss function of the generative network.

[0114] The network loss calculation module 603 is used to use the discriminant network in the pre-trained neural network model to determine the falsity between the pseudo anterior segment image and the preprocessed conjugate image, and to use the discriminant network to determine the authenticity between the preprocessed conjugate image and the preprocessed anterior segment image. Based on the falsity, the authenticity, and the absolute value loss, the generator network loss of the generator network is calculated.

[0115] The model training module 604 is used to train the pre-trained neural network model based on the generator network loss to obtain a trained neural network model.

[0116] The model selection module 605 is used to perform model verification on the trained neural network model to obtain a verified neural network model, and to calculate the peak signal-to-noise ratio between the pseudo anterior segment image and the real anterior segment image corresponding to the verified neural network model, and to select the final neural network model from the verified neural network models using the peak signal-to-noise ratio.

[0117] The complex conjugate removal module 606 is used to remove complex conjugates from the current complex conjugate image using the final neural network model, to obtain an anterior segment image with complex conjugates removed.

[0118] In detail, the modules in the deep learning-based OCT conjugate image removal device 400 described in this embodiment of the invention employ the same methods as described above. Figure 1 The technique used is the same as the deep learning-based OCT conjugate image removal method described above, and it can produce the same technical effect, so it will not be repeated here.

[0119] like Figure 7 The diagram shown is a schematic representation of the structure of an electronic device that implements the deep learning-based OCT conjugate image removal method of the present invention.

[0120] The electronic device may include a processor 70, a memory 71, a communication bus 72, and a communication interface 73. It may also include a computer program stored in the memory 71 and capable of running on the processor 70, such as a deep learning-based OCT conjugate image removal program.

[0121] In some embodiments, the processor 70 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 70 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 71 (e.g., executing a deep learning-based OCT conjugate image removal program) and calls data stored in the memory 71 to perform various functions of the electronic device and process data.

[0122] The memory 71 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 71 can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory 71 can be an external storage device of the electronic device, such as a plug-in portable hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. Furthermore, the memory 71 can include both internal and external storage units of the electronic device. The memory 71 can be used not only to store application software and various types of data installed on the electronic device, such as the code of a database configuration connection program, but also to temporarily store data that has been output or will be output.

[0123] The communication bus 72 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable communication between the memory 71 and at least one processor 70, etc.

[0124] The communication interface 73 is used for communication between the electronic device 7 and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, Bluetooth interface, etc.), typically used to establish communication connections between the electronic device and other electronic devices. The user interface may be a display, an input unit (such as a keyboard), or, optionally, a standard wired or wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device and to display a visual user interface.

[0125] Figure 7 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 7The structure shown does not constitute a limitation on the electronic device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0126] For example, although not shown, the electronic device may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 70 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0127] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in terms of the scope of the patent invention.

[0128] The database configuration connection program stored in the memory 71 of the electronic device is a combination of multiple computer programs, which, when run in the processor 70, can achieve the following:

[0129] Collect spectral data, construct a complex conjugate image of the spectral data, and construct a true anterior segment image of the spectral data. Perform image preprocessing on the complex conjugate image and the true anterior segment image to obtain a preprocessed conjugate image and a preprocessed anterior segment image.

[0130] The pseudo anterior segment image corresponding to the preprocessed conjugate image is analyzed using the generative network in the pretrained neural network model, and the absolute value loss between the pseudo anterior segment image and the preprocessed anterior segment image is calculated using the absolute value loss function of the generative network.

[0131] The discriminant network in the pre-trained neural network model is used to determine the falsity between the pseudo anterior segment image and the preprocessed conjugate image, and the discriminant network is used to determine the authenticity between the preprocessed conjugate image and the preprocessed anterior segment image. Based on the falsity, the authenticity, and the absolute value loss, the generation network loss of the generation network is calculated.

[0132] Based on the generated network loss, the pre-trained neural network model is trained to obtain a trained neural network model;

[0133] The trained neural network model is validated to obtain a validated neural network model. The peak signal-to-noise ratio (PSNR) between the pseudo anterior segment image and the real anterior segment image corresponding to the validated neural network model is calculated. The final neural network model is selected from the validated neural network models using the PSNR.

[0134] The final neural network model is used to remove the complex conjugation from the current complex conjugated image to obtain an anterior segment image with the complex conjugation removed.

[0135] Specifically, the processor 70's implementation method of the above-mentioned computer program can be found in [reference needed]. Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0136] Furthermore, if the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium. The storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0137] The present invention also provides a storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following:

[0138] Collect spectral data, construct a complex conjugate image of the spectral data, and construct a true anterior segment image of the spectral data. Perform image preprocessing on the complex conjugate image and the true anterior segment image to obtain a preprocessed conjugate image and a preprocessed anterior segment image.

[0139] The pseudo anterior segment image corresponding to the preprocessed conjugate image is analyzed using the generative network in the pretrained neural network model, and the absolute value loss between the pseudo anterior segment image and the preprocessed anterior segment image is calculated using the absolute value loss function of the generative network.

[0140] The discriminant network in the pre-trained neural network model is used to determine the falsity between the pseudo anterior segment image and the preprocessed conjugate image, and the discriminant network is used to determine the authenticity between the preprocessed conjugate image and the preprocessed anterior segment image. Based on the falsity, the authenticity, and the absolute value loss, the generation network loss of the generation network is calculated.

[0141] Based on the generated network loss, the pre-trained neural network model is trained to obtain a trained neural network model;

[0142] The trained neural network model is validated to obtain a validated neural network model. The peak signal-to-noise ratio (PSNR) between the pseudo anterior segment image and the real anterior segment image corresponding to the validated neural network model is calculated. The final neural network model is selected from the validated neural network models using the PSNR.

[0143] The final neural network model is used to remove the complex conjugation from the current complex conjugated image to obtain an anterior segment image with the complex conjugation removed.

[0144] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0145] The modules described as separate components may or may not be physically separate. The components shown as modules 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 modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0146] Furthermore, the functional modules in the various embodiments of the present invention 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. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0147] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0148] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0149] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0150] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A deep learning-based OCT conjugate image removal method, characterized in that, The method includes: Collect spectral data, construct a complex conjugate image of the spectral data, and construct a true anterior segment image of the spectral data. Perform image preprocessing on the complex conjugate image and the true anterior segment image to obtain a preprocessed conjugate image and a preprocessed anterior segment image. The step of preprocessing the complex conjugate image and the real anterior segment image to obtain a preprocessed conjugate image and a preprocessed anterior segment image includes: scaling the resolution of the complex conjugate image and the real anterior segment image to 1024*1536 pixels respectively to obtain scaled complex conjugate images and scaled real anterior segment images; normalizing the scaled complex conjugate images and scaled real anterior segment images to normalize the pixel value range to -1~1 to obtain normalized complex conjugate images and normalized real anterior segment images; dividing the normalized complex conjugate images and normalized real anterior segment images into training sets, validation sets, and test sets, and using the training sets, validation sets, and test sets corresponding to the normalized complex conjugate images and normalized real anterior segment images as the preprocessed conjugate images and preprocessed anterior segment images. The construction of the true anterior segment image of the spectral data includes: The spectral data is subjected to data modulation that introduces a phase shift to obtain modulated data; Construct the complex conjugate image corresponding to the modulation data to obtain the modulation complex conjugate image; The true anterior segment image corresponding to the modulated complex conjugate image is calculated using the following formula: in, This represents the anterior segment image of the true eye. This represents the Cauchy principal value, which is the integral value of the improper integral. This represents the modulated complex conjugate image. Indicates the independent variable parameter. Represents modulated complex conjugate image The values ​​of the pixels in the image; the true anterior segment image is a complex full-range anterior segment image with twice the depth range; The pseudo anterior segment image corresponding to the preprocessed conjugate image is analyzed using the generative network in the pretrained neural network model, and the absolute value loss between the pseudo anterior segment image and the preprocessed anterior segment image is calculated using the absolute value loss function of the generative network. The step of calculating the absolute loss between the pseudo-anterior segment image and the preprocessed anterior segment image using the absolute loss function of the generator network includes: Image binarization is performed on the training sets corresponding to the pseudo anterior segment image and the preprocessed anterior segment image, respectively, to obtain the binarized pseudo anterior segment image and the binarized preprocessed anterior segment image; The absolute value loss between the binarized pseudo-front view image and the binarized preprocessed front view image is calculated using the absolute value loss function, wherein the absolute value loss function includes: in, This represents the absolute value loss. This represents the absolute loss value between the foreground region of the binarized pseudo-foreground image and the foreground region of the binarized preprocessed foreground image. This represents the absolute loss value between the background region of the binarized pseudo-foreground image and the background region of the binarized preprocessed foreground image. express The weight, express The weights; the foreground region is the anterior segment tissue region, and the background region is the black area of ​​the image; The discriminant network in the pre-trained neural network model is used to determine the falsity between the pseudo anterior segment image and the preprocessed conjugate image, and the discriminant network is used to determine the authenticity between the preprocessed conjugate image and the preprocessed anterior segment image. Based on the falsity, the authenticity, and the absolute value loss, the generation network loss of the generation network is calculated. The calculation of the generator network loss based on the falsity, the authenticity, and the absolute value loss includes: Based on the falsity, the authenticity, and the absolute value loss, the generation network loss of the generation network is calculated using the following formula: in, This represents the loss of the generating network. This represents the training set of the preprocessed conjugate graph. This represents the training set of the preprocessed anterior segment image. This represents the noise added to the generator network when converting the training set of the preprocessed conjugate map into the pseudo anterior segment map. Represents the mathematical expectation. Represents coefficients used for control. The weight, and The larger the better, within the range of 300-700. The output classification probability, i.e., the realism, is obtained by simultaneously inputting the training set of the preprocessed conjugate map and the training set of the preprocessed anterior segment map into the discriminant network. This represents the pseudo-anterior segment diagram. The output classification probability obtained by simultaneously inputting the training set of the preprocessed conjugate graph and the pseudo anterior segment graph into the discriminant network is the falsehood. Based on the generated network loss, the pre-trained neural network model is trained to obtain a trained neural network model; The trained neural network model is validated to obtain a validated neural network model. The peak signal-to-noise ratio (PSNR) between the pseudo anterior segment image and the real anterior segment image corresponding to the validated neural network model is calculated. The final neural network model is selected from the validated neural network models using the PSNR. The final neural network model is used to remove the complex conjugation from the current complex conjugated image to obtain an anterior segment image with the complex conjugation removed.

2. The method according to claim 1, characterized in that, The step of analyzing the pseudo anterior segment image corresponding to the preprocessed conjugate image using a generative network in a pre-trained neural network model includes: The convolutional layer-activation layer-batch normalization layer in the generator network is used to extract features from the training set in the preprocessed conjugate graph to obtain feature vectors; The pseudo-front view image corresponding to the feature vector is generated using the deconvolution layer-activation layer-batch normalization layer in the generative network.

3. The method according to claim 1, characterized in that, The step of training the pre-trained neural network model based on the generated network loss to obtain a trained neural network model includes: Based on the loss of the generator network, the parameters of the generator network and the discriminator network are iteratively updated by alternating training of the generator network and the discriminator network, resulting in alternatingly trained generator network and alternatingly trained discriminator network; When a preset Nash equilibrium is reached between the alternately trained generative network and the alternately trained discriminative network, the trained neural network model is obtained; the Nash equilibrium is when the discriminator network cannot distinguish between the authenticity of the real full-range anterior segment image and the generated full-range anterior segment image.

4. The method according to claim 1, characterized in that, The step of validating the trained neural network model to obtain a validated neural network model includes: Obtain the preprocessed conjugate map and the preprocessed anterior segment map, and use the validation set corresponding to the preprocessed conjugate map and the preprocessed anterior segment map to perform model validation on the trained neural network model to obtain the validated neural network model.

5. The method according to claim 1, characterized in that, The calculation of the peak signal-to-noise ratio (PSNR) between the pseudo anterior segment image and the real anterior segment image corresponding to the validated neural network model includes: Obtain the preprocessed conjugate image and the preprocessed anterior segment image, and use the test set corresponding to the preprocessed conjugate image and the preprocessed anterior segment image to determine the pseudo anterior segment image and the true anterior segment image corresponding to the validated neural network model; The peak signal-to-noise ratio (PSNR) between the pseudo anterior segment image and the real anterior segment image corresponding to the validated neural network model is calculated using the following formula: in, The peak signal-to-noise ratio is represented by MAX, which represents the largest pixel in the pseudo anterior segment image corresponding to the verified neural network model. MAX is 8 bits, representing the highest pixel value of 255. MSE represents the mean square error between the pseudo anterior segment image and the real anterior segment image corresponding to the verified neural network model.

6. A deep learning-based OCT conjugate image removal device, characterized in that, The device includes: The image preprocessing module is used to acquire spectral data, construct a complex conjugate image of the spectral data, and construct a true anterior segment image of the spectral data. The complex conjugate image and the true anterior segment image are preprocessed to obtain a preprocessed conjugate image and a preprocessed anterior segment image. The step of preprocessing the complex conjugate image and the real anterior segment image to obtain a preprocessed conjugate image and a preprocessed anterior segment image includes: scaling the resolution of the complex conjugate image and the real anterior segment image to 1024*1536 pixels respectively to obtain scaled complex conjugate images and scaled real anterior segment images; normalizing the scaled complex conjugate images and scaled real anterior segment images to normalize the pixel value range to -1~1 to obtain normalized complex conjugate images and normalized real anterior segment images; dividing the normalized complex conjugate images and normalized real anterior segment images into training sets, validation sets, and test sets, and using the training sets, validation sets, and test sets corresponding to the normalized complex conjugate images and normalized real anterior segment images as the preprocessed conjugate images and preprocessed anterior segment images. The construction of the true anterior segment image of the spectral data includes: The spectral data is subjected to data modulation that introduces a phase shift to obtain modulated data; Construct the complex conjugate image corresponding to the modulation data to obtain the modulation complex conjugate image; The true anterior segment image corresponding to the modulated complex conjugate image is calculated using the following formula: in, This represents the anterior segment image of the true eye. This represents the Cauchy principal value, which is the integral value of the improper integral. This represents the modulated complex conjugate image. Indicates the independent variable parameter. Represents modulated complex conjugate image The values ​​of the pixels in the image; the true anterior segment image is a complex full-range anterior segment image with twice the depth range; The absolute value loss calculation module is used to analyze the pseudo anterior segment image corresponding to the preprocessed conjugate image using the generative network in the pretrained neural network model, and to calculate the absolute value loss between the pseudo anterior segment image and the preprocessed anterior segment image using the absolute value loss function of the generative network. The step of calculating the absolute loss between the pseudo-anterior segment image and the preprocessed anterior segment image using the absolute loss function of the generator network includes: Image binarization is performed on the training sets corresponding to the pseudo anterior segment image and the preprocessed anterior segment image, respectively, to obtain the binarized pseudo anterior segment image and the binarized preprocessed anterior segment image; The absolute value loss between the binarized pseudo-front view image and the binarized preprocessed front view image is calculated using the absolute value loss function, wherein the absolute value loss function includes: in, This represents the absolute value loss. This represents the absolute loss value between the foreground region of the binarized pseudo-foreground image and the foreground region of the binarized preprocessed foreground image. This represents the absolute loss value between the background region of the binarized pseudo-foreground image and the background region of the binarized preprocessed foreground image. express The weight, express The weights; the foreground region is the anterior segment tissue region, and the background region is the black area of ​​the image; The network loss calculation module is used to use the discriminant network in the pre-trained neural network model to determine the falsity between the pseudo anterior segment image and the preprocessed conjugate image, and to use the discriminant network to determine the authenticity between the preprocessed conjugate image and the preprocessed anterior segment image. Based on the falsity, the authenticity, and the absolute value loss, the module calculates the generation network loss of the generation network. Calculating the generator network loss of the generator network includes: Based on the falsity, the authenticity, and the absolute value loss, the generation network loss of the generation network is calculated using the following formula: in, This represents the loss of the generating network. This represents the training set of the preprocessed conjugate graph. This represents the training set of the preprocessed anterior segment image. This represents the noise added to the generator network when converting the training set of the preprocessed conjugate map into the pseudo anterior segment map. Represents the mathematical expectation. Represents coefficients used for control. The weight, And the larger the better, within the range of 300-700. The output classification probability, i.e., the realism, is obtained by simultaneously inputting the training set of the preprocessed conjugate map and the training set of the preprocessed anterior segment map into the discriminant network. This represents the pseudo-anterior segment diagram. The output classification probability obtained by simultaneously inputting the training set of the preprocessed conjugate graph and the pseudo anterior segment graph into the discriminant network is the falsehood. The model training module is used to train the pre-trained neural network model based on the generative network loss to obtain a trained neural network model. The model selection module is used to perform model verification on the trained neural network model to obtain a verified neural network model, and to calculate the peak signal-to-noise ratio between the pseudo anterior segment image and the real anterior segment image corresponding to the verified neural network model, and to select the final neural network model from the verified neural network models using the peak signal-to-noise ratio. The complex conjugate removal module is used to remove complex conjugates from the current complex conjugate image using the final neural network model, so as to obtain an anterior segment image with complex conjugates removed.

Images