Three-dimensional fundus image generation method and device, electronic equipment and storage medium

Abstract

The invention discloses a three-dimensional fundus image generation method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring a first two-dimensional fundus scanning image and a second two-dimensional fundus scanning image of a target object, wherein the scanning directions corresponding to the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image are mutually orthogonal; and based on a preset image generation model, reconstructing the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image to obtain a three-dimensional fundus image of the target object. According to the method, the three-dimensional fundus image can be generated through reconstruction of the orthogonal two-dimensional fundus scanning image, the three-dimensional fundus image does not need to be obtained through scanning of an optical coherence tomography scanner, and therefore the quality of the three-dimensional fundus image and the efficiency of obtaining the three-dimensional fundus image are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method, device, electronic device and storage medium for generating three-dimensional fundus images. Background Art

[0002] Optical Coherence Tomography (OCT) is a common imaging device for ophthalmic examinations. It has been continuously developed and widely used in the field of medical imaging. OCT has the characteristics of structural imaging, high-resolution imaging, radiation-free and non-invasive. However, based on the physical acquisition operation mode of point-by-point scanning of OCT, multiple acquisitions at multiple positions are generally required clinically to effectively obtain the actual three-dimensional fundus image.

[0003] However, in the actual acquisition process, the following two types of problems affect the effect of three-dimensional fundus images:

[0004] First, due to the time-consuming acquisition, OCT basically scans point by point on the fundus retina according to A-line scanning. This itself requires scanning a certain amount of point data in three-dimensional space to form a three-dimensional image, which consumes time. For general frequency-domain OCT, the time is relatively long.

[0005] Second, due to the eye movement of the acquisition object during the acquisition process, eye movement will occur during point-by-point scanning. Then the acquired image is contextually discontinuous, and basically an effective three-dimensional fundus image cannot be obtained. Especially because of the motion artifacts caused by eye movement, which interfere with the results of the fundus image, and the quality of the acquired three-dimensional fundus image is poor. Summary of the Invention

[0006] The present invention provides a method, device, electronic device and storage medium for generating three-dimensional fundus images to solve the above technical problems.

[0007] According to one aspect of the present invention, a method for generating a three-dimensional fundus image is provided, including:

[0008] Obtaining a first two-dimensional fundus scan image and a second two-dimensional fundus scan image of a target object, wherein the scanning directions corresponding to the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are orthogonal to each other;

[0009] Reconstructing the first two-dimensional fundus scan image and the second two-dimensional fundus scan image based on a pre-set image generation model to obtain a three-dimensional fundus image of the target object;

[0010] Wherein, the image generation model includes: a splicing processing module, an information encryption module, a dimension space conversion module and an information decryption module;

[0011] Input the first two-dimensional fundus scan image and the second two-dimensional fundus scan image into the splicing processing module for extended dimension splicing to obtain splicing features; input the splicing features into the information encryption module for feature extraction to obtain visual features at different levels; input the visual features at different levels into the dimension space conversion module for dimension conversion to obtain three-dimensional comprehensive features; input the three-dimensional comprehensive features into the information decryption module to output the three-dimensional fundus image of the target object.

[0012] According to another aspect of the present invention, there is provided a three-dimensional fundus image generation device, including:

[0013] A fundus scan image acquisition module for acquiring a first two-dimensional fundus scan image and a second two-dimensional fundus scan image of a target object, and the scanning directions corresponding to the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are orthogonal to each other;

[0014] A three-dimensional fundus image reconstruction module for reconstructing the first two-dimensional fundus scan image and the second two-dimensional fundus scan image based on a pre-set image generation model to obtain the three-dimensional fundus image of the target object;

[0015] Wherein, the image generation model includes: a splicing processing module, an information encryption module, a dimension space conversion module and an information decryption module;

[0016] Input the first two-dimensional fundus scan image and the second two-dimensional fundus scan image into the splicing processing module for extended dimension splicing to obtain splicing features; input the splicing features into the information encryption module for feature extraction to obtain visual features at different levels; input the visual features at different levels into the dimension space conversion module for dimension conversion to obtain three-dimensional comprehensive features; input the three-dimensional comprehensive features into the information decryption module to output the three-dimensional fundus image of the target object.

[0017] According to another aspect of the present invention, there is provided an electronic device, the electronic device includes:

[0018] At least one processor; and

[0019] A memory communicatively connected to the at least one processor; wherein,

[0020] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the three-dimensional fundus image generation method according to any embodiment of the present invention.

[0021] According to another aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions for causing a processor to implement the three-dimensional fundus image generation method according to any embodiment of the present invention when executed.

[0022] In the technical solution of the embodiment of the present invention, by obtaining a first two-dimensional fundus scan image and a second two-dimensional fundus scan image of a target object, the scanning directions corresponding to the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are orthogonal to each other; based on a pre-set image generation model, the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are reconstructed to obtain a three-dimensional fundus image of the target object; wherein, the image generation model includes: a splicing processing module, an information encryption module, a dimension space conversion module, and an information decryption module; the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are input into the splicing processing module for extended dimension splicing to obtain a spliced feature; the spliced feature is input into the information encryption module for feature extraction to obtain visual features at different levels; the visual features at different levels are input into the dimension space conversion module for dimension conversion to obtain a three-dimensional comprehensive feature; the three-dimensional comprehensive feature is input into the information decryption module to output a three-dimensional fundus image of the target object. A three-dimensional fundus image can be reconstructed and generated from orthogonal two-dimensional fundus scan images without using an optical coherence tomography scanner to scan and obtain a three-dimensional fundus image, thereby improving the quality of the three-dimensional fundus image and the efficiency of obtaining the three-dimensional fundus image.

[0023] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0025] Figure 1 is a flowchart of a three-dimensional fundus image generation method provided in Embodiment 1 of the present invention;

[0026] Figure 2 is a flowchart of a three-dimensional fundus image generation method provided in Embodiment 2 of the present invention;

[0027] Figure 3 is a flowchart of an image generation model training method provided in Embodiment 3 of the present invention;

[0028] Figure 4 It is a schematic structural diagram of a three-dimensional fundus image generation device provided in the fourth embodiment of the present invention;

[0029] Figure 5 It is a schematic structural diagram of an electronic device provided in the fifth embodiment of the present invention. Detailed implementation manners

[0030] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0031] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0032] Embodiment 1

[0033] Figure 1 It is a flowchart of a three-dimensional fundus image generation method provided in the first embodiment of the present invention. This embodiment is applicable to various situations. This method can be executed by a three-dimensional fundus image generation device, which can be implemented in the form of hardware and / or software, and the three-dimensional fundus image generation device can be configured in electronic devices such as computers and servers. As Figure 1 shown, the method includes:

[0034] S110. Obtain a first two-dimensional fundus scan image and a second two-dimensional fundus scan image of a target object, and the scanning directions corresponding to the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are orthogonal to each other.

[0035] Among them, the two-dimensional fundus scan image is a two-dimensional fundus image obtained by scanning the rigor of the target object through Optical Coherence Tomography. The scanning directions of the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are orthogonal to each other. For example, assuming that the scanning direction of the first two-dimensional fundus scan image is the direction where the XZ plane is located, then the scanning direction of the second two-dimensional fundus scan image is the direction where the YZ plane is located, and the XZ plane and the YZ plane are orthogonal to each other.

[0036] S120. Based on a pre-set image generation model, reconstruct the first two-dimensional fundus scan image and the second two-dimensional fundus scan image to obtain the three-dimensional fundus image of the target object.

[0037] Among them, the image generation model includes: a splicing processing module, an information encryption module, a dimension space conversion module, and an information decryption module; input the first two-dimensional fundus scan image and the second two-dimensional fundus scan image into the splicing processing module for extended dimension splicing to obtain splicing features; input the splicing features into the information encryption module for feature extraction to obtain visual features at different levels; input the visual features at different levels into the dimension space conversion module for dimension conversion to obtain three-dimensional comprehensive features; input the three-dimensional comprehensive features into the information decryption module to output the three-dimensional fundus image of the target object.

[0038] In the embodiment of the present invention, input the first two-dimensional fundus scan image and the second two-dimensional fundus scan image into the splicing processing module for extended dimension splicing to obtain splicing features; among them, the splicing features are three-dimensional feature data. Specifically, assume that the first two-dimensional fundus scan image is a two-dimensional image with dimensions (Sz, Sx) and the second two-dimensional fundus scan image is a two-dimensional image with dimensions (Sz, Sy). After splicing, the dimensions of the obtained splicing features are (2Sz, 1 + Sy, Sx + 1). Further, the information encryption module includes multiple network layers at different levels. It can be understood that the lower the feature extraction level, the closer it is to the input layer of the information encryption module; input the splicing features into the information encryption module and perform feature extraction through network layers at different levels to obtain visual features at different levels; among them, the visual features at different levels can be divided into shallow visual features and deep visual features. The shallow visual features include but are not limited to features such as pixel values and color distributions, and the deep visual features include but are not limited to abstract information features such as image textures, shapes, and edges. Further, input the shallow visual features and the deep visual features into the dimension space conversion module for dimension migration and feature transformation to obtain three-dimensional comprehensive features; the three-dimensional comprehensive features include the three-dimensional structure and morphology of the fundus. Further, input the three-dimensional comprehensive features into the information decryption module for mining and summarization, and output the three-dimensional fundus image of the target object.

[0039] Based on the above embodiments, optionally, the information encryption module includes a two-dimensional convolutional block, a two-dimensional scaling sampling block, and a first activation function connected in sequence; the dimension space conversion module includes a convolutional block, a fully connected block, a three-dimensional reorganization processing block, and a second activation function connected in sequence; the information decryption module includes a three-dimensional convolutional block, a three-dimensional expansion sampling block, a three-dimensional transposed convolutional block, and a third activation function connected in sequence; an N - time loop connection is provided between the output end and the input end of the information encryption module; and / or, an M - time loop connection is provided between the output end and the input end of the information decryption module.

[0040] In the embodiment of the present invention, the information encryption module includes a two-dimensional convolutional block, a two-dimensional scaling sampling block, and a first activation function connected in sequence, and an N - time loop connection is provided between the output end and the input end of the information encryption module, that is, N - time feature extraction is performed on the concatenated features input to the information encryption module; among them, the N - time loop connection is set by those skilled in the art, generally set to a power of 2, such as 8.

[0041] Exemplarily, the scaling ratio of the two-dimensional scaling sampling block can be 1 / 2, the first activation function can use the LeakyReLU function, and the parameter of the LeakyReLU function is set to 0.01.

[0042] The dimension space conversion module includes a convolutional block, a fully connected block, a three-dimensional reorganization processing block, and a second activation function connected in sequence. Among them, the convolutional block includes, but is not limited to, using two-dimensional or three-dimensional convolutional functions. All variables output by the convolutional block are converted into a unified one-dimensional variable, and the fully connected block is used to expand the input one-dimensional variable into a one-dimensional variable with n elements; m and n can be set by those skilled in the art, and it should be noted that m is not equal to n. Exemplarily, the expansion multiple after full connection can be a power of 2 greater than 2 and not greater than 64. The three-dimensional reorganization processing block is a tensor dimension expansion operation, used to deform the one-dimensional variable expanded by the above fully connected block into a two-dimensional tensor. The second activation function can be the ReLU function.

[0043] The information decryption module includes a three-dimensional convolutional block, a three-dimensional expansion sampling block, a three-dimensional transposed convolutional block, and a third activation function connected in sequence, and an M - time loop connection is provided between the output end and the input end of the information decryption module, that is, the information decryption module performs M - time mining and summarization on the conversion features of the set dimension output by the dimension space conversion module. Among them, the M - time loop connection is set by those skilled in the art. Exemplarily, the number of loop connections M of the information decryption module can be the same as the number of loop connections N of the information encryption module.

[0044] Specifically, the upsampling ratio of the three-dimensional extended sampling block can be the reciprocal of the scaling ratio in the information encryption block. The third activation function can use the Softmax function.

[0045] In the technical solution of this embodiment, by obtaining a first two-dimensional fundus scan image and a second two-dimensional fundus scan image of a target object, the scanning directions corresponding to the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are orthogonal to each other; based on a pre-set image generation model, the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are reconstructed to obtain a three-dimensional fundus image of the target object; wherein, the image generation model includes: a splicing processing module, an information encryption module, a dimension space conversion module, and an information decryption module; the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are input into the splicing processing module for extended dimension splicing to obtain a spliced feature; the spliced feature is input into the information encryption module for feature extraction to obtain visual features at different levels; the visual features at different levels are input into the dimension space conversion module for dimension conversion to obtain a three-dimensional comprehensive feature; the three-dimensional comprehensive feature is input into the information decryption module to output the three-dimensional fundus image of the target object. The three-dimensional fundus image can be reconstructed and generated from orthogonal two-dimensional fundus scan images without using an optical coherence tomography scanner to scan and obtain the three-dimensional fundus image, thereby improving the quality of the three-dimensional fundus image and the efficiency of obtaining the three-dimensional fundus image.

[0046] Embodiment 2

[0047] Figure 2 FIG. is a flowchart of a method for generating a three-dimensional fundus image provided in Embodiment 2 of the present invention. In this embodiment, after obtaining the first two-dimensional fundus scan image and the second two-dimensional fundus scan image of the target object, "respectively obtaining the minimum length unit between pixels in each dimension direction of the first two-dimensional fundus scan image and the second two-dimensional fundus scan image, and the preset length of the three-dimensional fundus image in the three dimension directions; determining the dimension size of the three-dimensional fundus image based on the preset length in the three dimension directions and the minimum length unit between pixels in each dimension direction, where the dimension size includes the number of pixels in the three dimension directions; limiting the ranges of the first two-dimensional fundus scan image and the second two-dimensional fundus scan image based on the dimension size to obtain the first two-dimensional fundus scan image and the second two-dimensional fundus scan image of the target range; performing gray level update processing on the first two-dimensional fundus scan image and the second two-dimensional fundus scan image of the target range, where the gray level update processing includes one or more of the following: gray level value truncation processing and gray level value normalization processing." are added. Among them, the explanations of the same or corresponding terms as those in the above embodiments will not be repeated here.

[0048] As Figure 2 shown, the method includes:

[0049] S210. Obtain a first two-dimensional fundus scanning image and a second two-dimensional fundus scanning image of a target object, and the scanning directions corresponding to the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image are orthogonal to each other.

[0050] S220. Respectively obtain the minimum length unit between pixels in each dimension direction of the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image, and the preset lengths in three dimension directions of the three-dimensional fundus image; determine the dimension size of the three-dimensional fundus image based on the preset lengths in the three dimension directions and the minimum length unit between pixels in each dimension direction, where the dimension size includes the number of pixels in the three dimension directions; limit the ranges of the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image based on the dimension size to obtain the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image of the target range; perform gray-scale update processing on the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image of the target range, where the gray-scale update processing includes one or more of the following: gray-scale value truncation processing and gray-scale value normalization processing.

[0051] Specifically, assume that the scanning direction of the first two-dimensional fundus scanning image is the direction where the XZ plane is located, then the scanning direction of the second two-dimensional fundus scanning image is the direction where the YZ plane is located. The minimum length unit between pixels in the X direction and the Z direction can be obtained based on the first two-dimensional fundus scanning image, and the minimum length unit between pixels in the Y direction can be obtained based on the second two-dimensional fundus scanning image. At the same time, obtain the preset lengths in the three dimension directions of the three-dimensional fundus image. It should be noted that the preset lengths are set by those skilled in the art according to the empirical range of collecting three-dimensional fundus images, and are not limited here.

[0052] In an embodiment of the present invention, determine the dimension size of the three-dimensional fundus image based on the preset lengths in the three dimension directions and the minimum length unit between pixels in each dimension direction; where the dimension size includes the number of pixels in the three dimension directions. Exemplarily, the calculation formula for the dimension size of the three-dimensional fundus image is as follows:

[0053] Mx = len_x / space_x

[0054] My = len_y / space_y;

[0055] Mz = len_z / space_z

[0056] Wherein, (Mx, My, Mz) represents the number of pixels in the three dimensions x, y, and z of the three-dimensional fundus image, (len_x, len_y, len_z) represents the preset lengths of the three-dimensional fundus image in the x, y, and z dimensions, and (space_x, space_y, space_z) represents the minimum length units of the first two-dimensional fundus scan image and the second two-dimensional fundus scan image in the x, y, and z dimensions.

[0057] In the embodiment of the present invention, based on the dimensional sizes, the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are respectively limited in range, and the first two-dimensional fundus scan image and the second two-dimensional fundus scan image within the target range are cropped; wherein, the target range refers to the range of the regional images in the first two-dimensional fundus scan image and the second two-dimensional fundus scan image corresponding to the dimensional sizes of the three-dimensional fundus image. Further, the first two-dimensional fundus scan image and the second two-dimensional fundus scan image within the target range are subjected to gray-scale update processing, wherein the gray-scale update processing includes one or more of the following: gray-scale value truncation processing and gray-scale value normalization processing.

[0058] Based on the above embodiments, optionally, the gray-scale value truncation processing includes: for each two-dimensional fundus scan image within the target range of the first two-dimensional fundus scan image and the second two-dimensional fundus scan image, determining the gray-scale value distribution data of each two-dimensional fundus scan image within the target range, and respectively matching the preset upper limit distribution data and the preset lower limit distribution data with the gray-scale value distribution data of each two-dimensional fundus scan image within the target range to determine the target lower gray-scale value and the target upper gray-scale value corresponding to each two-dimensional fundus scan image within the target range; updating the image gray-scale values of each two-dimensional fundus scan image within the target range with the target lower gray-scale value and the target upper gray-scale value corresponding to each two-dimensional fundus scan image within the target range; the gray-scale value normalization processing includes: based on the target lower gray-scale value and the target upper gray-scale value of each two-dimensional fundus scan image within the target range, determining the normalized gray-scale value corresponding to the gray-scale value in each two-dimensional fundus scan image within the target range, and updating the gray-scale value in each two-dimensional fundus scan image within the target range to the corresponding normalized gray-scale value.

[0059] In an embodiment of the present invention, for each two-dimensional fundus scan image of a target range in the first two-dimensional fundus scan image and the second two-dimensional fundus scan image of the target range, the gray value distribution data of each two-dimensional fundus scan image of the target range is determined; wherein, the gray value distribution data is the gray value of the pixel points in the two-dimensional fundus scan image of the target range. Further, the preset upper limit distribution data and the preset lower limit distribution data are respectively matched with the gray value distribution data of each two-dimensional fundus scan image of the target range, and the gray value in the gray value distribution data that is closest to the preset upper limit distribution data is used as the target upper limit gray value, and the gray value in the gray value distribution data that is closest to the preset lower limit distribution data is used as the target lower limit gray value. Further, for each two-dimensional fundus scan image of the target range, the gray values in the gray value distribution data of the two-dimensional fundus scan image of the target range that are less than the target lower limit gray value are updated to the target lower limit gray value, and the gray values in the gray value distribution data of the two-dimensional fundus scan image of the target range that are greater than the target upper limit gray value are updated to the target upper limit gray value, thereby realizing gray truncation. Exemplarily, the calculation formulas for the target lower limit gray value and the target upper limit gray value are as follows:

[0060]

[0061] Wherein, H represents the gray cumulative statistical histogram performed on the two-dimensional fundus gray image I of the input target range, which is used to determine the gray value distribution data of the two-dimensional fundus gray image of the target range, and α low represents the preset lower limit distribution data, and α high represents the preset upper limit distribution data, g low represents the target lower limit gray value, and g high represents the target upper limit gray value.

[0062] In an embodiment of the present invention, based on the target lower limit gray value and the target upper limit gray value of each two-dimensional fundus scan image of the target range, a normalized linear transformation is performed on the gray values in each two-dimensional fundus scan image of the target range to obtain a normalized gray value, and the gray values in each two-dimensional fundus scan image of the target range are updated to the corresponding normalized gray values.

[0063] Exemplarily, the gray normalization formula is as follows:

[0064] f = (g - g_min) / (g_max - g_min);

[0065] Wherein, f represents the normalized gray value, g is the gray value of the two-dimensional fundus scan image of the input target range, g_min is the target lower limit gray value, and g_max is the target upper limit gray value.

[0066] S230. Based on a pre-set image generation model, reconstruct the first two-dimensional fundus scan image and the second two-dimensional fundus scan image after gray-scale update processing to obtain a three-dimensional fundus image of the target object.

[0067] The technical solution of this embodiment can further improve the quality of the reconstructed three-dimensional fundus image by performing range limitation, gray-scale truncation processing, and gray-scale normalization on orthogonal two-dimensional fundus scan images.

[0068] Embodiment III

[0069] Figure 3 FIG. is a flowchart of a method for training an image generation model provided in Embodiment III of the present invention. This embodiment elaborates in detail on the method for training an image generation model on the basis of the above embodiments. Among them, the explanations of the same or corresponding terms in the above embodiments will not be repeated here.

[0070] As Figure 3 shown, the method includes:

[0071] S310. Obtain two two-dimensional fundus scan images corresponding to a sample object in orthogonal directions and three-dimensional fundus label data, where the three-dimensional fundus label data includes a three-dimensional fundus scan image of the sample object.

[0072] In the embodiment of the present invention, two two-dimensional fundus scan images corresponding to a sample object in orthogonal directions and three-dimensional fundus label data are obtained to form a training data set. Among them, the three-dimensional fundus label data is obtained by scanning with OCT technology, and the three-dimensional fundus label data provides three-dimensional structural information of the fundus.

[0073] In some embodiments, optionally, preprocess the two two-dimensional fundus scan images and the three-dimensional fundus label data respectively; where the preprocessing includes one or more of the following: perform interpolation processing on the two two-dimensional fundus scan images and the three-dimensional fundus label data respectively to obtain two-dimensional fundus scan images of a preset size and the three-dimensional fundus label data of the preset size; perform gray-scale value truncation processing on the two two-dimensional fundus scan images and the three-dimensional fundus label data respectively; perform gray-scale value normalization processing on the two two-dimensional fundus scan images and the three-dimensional fundus label data respectively.

[0074] In the embodiments of the present invention, preprocessing is respectively performed on two two-dimensional fundus scan images and three-dimensional fundus label data to resample the two two-dimensional fundus scan images and the three-dimensional fundus label data. Specifically, since the sizes of the two two-dimensional fundus scan images and the corresponding three-dimensional fundus scan image corresponding to the obtained sample object in the orthogonal directions are not the same, in order for the model to be normally trained and converge, interpolation processing is respectively performed on the two two-dimensional fundus scan images and the three-dimensional fundus label data to obtain two-dimensional fundus scan images of a preset size and three-dimensional fundus label data of a preset size. In addition, since the pixel values of the three-dimensional fundus scan image in the two two-dimensional fundus scan images and the three-dimensional fundus label data obtained by OCT scanning have a wider distribution range than the pixel values in natural images, in order for the model to find the optimal solution faster, gray value truncation processing is respectively performed on the three-dimensional fundus scan images in the two two-dimensional fundus scan images and the three-dimensional fundus label data, and gray value normalization processing is respectively performed on the three-dimensional fundus scan images in the two two-dimensional fundus scan images and the three-dimensional fundus label data. It should be noted that the gray value truncation processing method and the gray value normalization method are the same as those in the above embodiments and will not be elaborated here.

[0075] S320. Obtain an image generation model to be trained, and reconstruct the two two-dimensional fundus scan images through the image generation model to be trained to obtain a three-dimensional fundus prediction image.

[0076] Among them, the image generation model to be trained can be a generative adversarial network model, a deep learning model, etc. In the embodiments of the present invention, the two two-dimensional fundus scan images corresponding in the orthogonal direction are input into the image generation model to be trained for reconstruction to obtain a three-dimensional fundus prediction image.

[0077] S330. Obtain a discriminator, and respectively discriminate the three-dimensional fundus prediction image and the three-dimensional fundus label data through the discriminator to obtain a first discrimination result corresponding to the three-dimensional fundus prediction image and a second discrimination result corresponding to the three-dimensional fundus label data.

[0078] Among them, the discriminator is used to judge the authenticity of the three-dimensional fundus prediction image and the three-dimensional fundus label data. Specifically, the discriminator includes a three-dimensional convolutional block, a three-dimensional downsampling block, a one-dimensional flattening block, a fully connected block, and a fourth activation function connected in sequence. Exemplarily, the image downsampling scaling ratio can be 1 / 2; the fourth activation function can be a sigmoid function. In the embodiments of the present invention, the three-dimensional fundus prediction image and the three-dimensional fundus label data are respectively input into the discriminator for discrimination to obtain a first discrimination result corresponding to the three-dimensional fundus prediction image and a second discrimination result corresponding to the three-dimensional fundus label data.

[0079] S340. Generate a target loss function based on the three-dimensional fundus prediction image, the three-dimensional fundus label data, the first discrimination result, and the second discrimination result.

[0080] Based on the above embodiments, optionally, generating a target loss function based on the three-dimensional fundus prediction image, the three-dimensional fundus label data, the first discrimination result, and the second discrimination result includes: at least one first loss function generated based on the three-dimensional fundus label data and the three-dimensional fundus prediction image; generating a second loss function based on a preset type of loss function, the first discrimination result, and the second discrimination result, where the second loss function includes one or more of the following: a first loss term generated by the preset type of loss function and the first discrimination result, a second loss term generated by the preset type of loss function and the second discrimination result, and a third loss term generated by the preset type of loss function and the gradient term of the first discrimination result; generating the target loss function based on the at least one first loss function and the second loss function.

[0081] Specifically, the target loss function includes at least one first loss function and a second loss function. The first loss function is used to quantitatively calculate the loss between the three-dimensional fundus prediction image and the three-dimensional fundus label data. The type of the first loss function can be an L1 norm loss function, an L2 Euclidean loss function, etc. At least one first loss function generated based on the three-dimensional fundus label data and the three-dimensional fundus prediction image; further, generating a first loss term based on a preset type of loss function and the first discrimination result, generating a second loss term based on a preset type of loss function and the second discrimination result, and generating a third loss term based on a preset type of loss function and the gradient term of the first discrimination result, generating a second loss function based on the first loss term, the second loss term, and the third loss term; generating the target loss function based on the at least one first loss function and the second loss function. Exemplarily, the target loss function is as follows:

[0082]

[0083] where L represents the target loss function, L a represents the first loss function of the L1 type, L b represents the first loss function of the L2 type, L c represents the second loss function, θ and are weight coefficients, θ and are values between 0 and 1, related to the duration and effect of network training. Exemplarily, the value ranges of θ and can be 0.001 - 0.1.

[0084] The first loss function of the L1 type is:

[0085]

[0086] The first loss function of type L2 is as follows:

[0087]

[0088] Wherein, I represents the three-dimensional fundus prediction image, G represents the three-dimensional fundus label data, and Z represents the number of images. It can be understood that the three-dimensional fundus prediction image and the three-dimensional fundus label data are in one-to-one correspondence, so the numbers of both are the same.

[0089] The second loss function is as follows:

[0090]

[0091] Wherein, L c is used to quantify and calculate the difference between the first discrimination result and the second discrimination result obtained by the discriminator, C represents a loss function of a preset type, for example: L1 norm loss function; represents the first discrimination result, is the first loss term; Dis(x) represents the second discrimination result, is the second loss term, δ is the weight coefficient, is the gradient operator, which can be interpreted as the change amount of the weight trained in each training round of. is the gradient term of the first discrimination result, is the third loss term.

[0092] S350. Taking reducing the target loss function as the goal, training the image generation model to be trained until a trained image generation model is obtained.

[0093] In the embodiment of the present invention, taking reducing the target loss function as the goal, adjusting the parameters of the image generation model by using a preset optimization algorithm; during the training process, continuously and iteratively generating three-dimensional fundus prediction images, evaluating them by using a discriminator, and updating the model parameters according to the evaluation results until the preset training stop condition is met, and a trained image generation model is obtained.

[0094] Embodiment 4

[0095] Figure 4 is a schematic structural diagram of a three-dimensional fundus image generation device provided in Embodiment 4 of the present invention. As Figure 4 shown, the device includes:

[0096] The fundus scan image acquisition module 410 is configured to acquire a first two-dimensional fundus scan image and a second two-dimensional fundus scan image of a target object, and the scanning directions corresponding to the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are orthogonal to each other;

[0097] The three-dimensional fundus image reconstruction module 420 is configured to reconstruct the first two-dimensional fundus scan image and the second two-dimensional fundus scan image based on a pre-set image generation model to obtain a three-dimensional fundus image of the target object;

[0098] Wherein, the image generation model includes: a splicing processing module, an information encryption module, a dimension space conversion module, and an information decryption module; inputting the first two-dimensional fundus scan image and the second two-dimensional fundus scan image into the splicing processing module for extended dimension splicing to obtain a splicing feature; inputting the splicing feature into the information encryption module for feature extraction to obtain visual features at different levels; inputting the visual features at different levels into the dimension space conversion module for dimension conversion to obtain a three-dimensional comprehensive feature; inputting the three-dimensional comprehensive feature into the information decryption module to output the three-dimensional fundus image of the target object.

[0099] The technical solution of this embodiment is to acquire a first two-dimensional fundus scan image and a second two-dimensional fundus scan image of a target object, and the scanning directions corresponding to the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are orthogonal to each other; reconstruct the first two-dimensional fundus scan image and the second two-dimensional fundus scan image based on a pre-set image generation model to obtain a three-dimensional fundus image of the target object; wherein, the image generation model includes: a splicing processing module, an information encryption module, a dimension space conversion module, and an information decryption module; inputting the first two-dimensional fundus scan image and the second two-dimensional fundus scan image into the splicing processing module for extended dimension splicing to obtain a splicing feature; inputting the splicing feature into the information encryption module for feature extraction to obtain visual features at different levels; inputting the visual features at different levels into the dimension space conversion module for dimension conversion to obtain a three-dimensional comprehensive feature; inputting the three-dimensional comprehensive feature into the information decryption module to output the three-dimensional fundus image of the target object. A three-dimensional fundus image can be reconstructed and generated from orthogonal two-dimensional fundus scan images without using an optical coherence tomography scanner to scan and acquire a three-dimensional fundus image, thereby improving the quality of the three-dimensional fundus image and the efficiency of acquiring the three-dimensional fundus image.

[0100] Based on the above embodiments, optionally, the information encryption module includes a two-dimensional convolutional block, a two-dimensional scaling sampling block, and a first activation function connected in sequence; the dimension space conversion module includes a convolutional block, a fully connected block, a three-dimensional restructuring processing block, and a second activation function connected in sequence; the information decryption module includes a three-dimensional convolutional block, a three-dimensional expansion sampling block, a three-dimensional transposed convolutional block, and a third activation function connected in sequence; an N - time loop connection is provided between the output end and the input end of the information encryption module; and / or, an M - time loop connection is provided between the output end and the input end of the information decryption module.

[0101] Based on the above embodiments, optionally, the device further includes a pre - processing module, configured to obtain the minimum length unit between pixels in each dimension direction of the first two - dimensional fundus scan image and the second two - dimensional fundus scan image, and the preset lengths in three dimension directions of the three - dimensional fundus image; determine the dimension size of the three - dimensional fundus image based on the preset lengths in the three dimension directions and the minimum length unit between pixels in each dimension direction, where the dimension size includes the number of pixels in the three dimension directions; limit the ranges of the first two - dimensional fundus scan image and the second two - dimensional fundus scan image based on the dimension size to obtain the first two - dimensional fundus scan image and the second two - dimensional fundus scan image within the target range; perform gray - level update processing on the first two - dimensional fundus scan image and the second two - dimensional fundus scan image within the target range, where the gray - level update processing includes one or more of the following: gray - level value truncation processing and gray - level value normalization processing.

[0102] Based on the above embodiments, optionally, the pre - processing module includes a gray - level value truncation processing unit, configured to: the gray - level value truncation processing includes: for each two - dimensional fundus scan image within the target range of the first two - dimensional fundus scan image and the second two - dimensional fundus scan image within the target range, determine the gray - level value distribution data of each two - dimensional fundus scan image within the target range, match the preset upper - limit distribution data and the preset lower - limit distribution data with the gray - level value distribution data of each two - dimensional fundus scan image within the target range respectively, and determine the target lower - limit gray - level value and the target upper - limit gray - level value corresponding to each two - dimensional fundus scan image within the target range; update the image gray - level value of each two - dimensional fundus scan image within the target range with the target lower - limit gray - level value and the target upper - limit gray - level value corresponding to each two - dimensional fundus scan image within the target range;

[0103] The grayscale update module includes a grayscale value normalization processing unit for: The grayscale value normalization processing includes: based on the target lower grayscale value and the target upper grayscale value of the two-dimensional fundus scan image of each of the target ranges, determining the normalized grayscale value corresponding to the grayscale value in the two-dimensional fundus scan image of each of the target ranges, and updating the grayscale value in the two-dimensional fundus scan image of each of the target ranges to the corresponding normalized grayscale value.

[0104] Based on the above embodiments, optionally, the device further includes an image generation model training module for obtaining two two-dimensional fundus scan images corresponding to a sample object in orthogonal directions and three-dimensional fundus label data, where the three-dimensional fundus label data includes a three-dimensional fundus scan image of the sample object; obtaining an image generation model to be trained, reconstructing the two two-dimensional fundus scan images through the image generation model to be trained to obtain a three-dimensional fundus prediction image; obtaining a discriminator, and respectively discriminating the three-dimensional fundus prediction image and the three-dimensional fundus label data through the discriminator to obtain a first discrimination result corresponding to the three-dimensional fundus prediction image and a second discrimination result corresponding to the three-dimensional fundus label data; generating a target loss function based on the three-dimensional fundus prediction image, the three-dimensional fundus label data, the first discrimination result, and the second discrimination result; and training the image generation model to be trained with the aim of reducing the target loss function until a trained image generation model is obtained.

[0105] Based on the above embodiments, optionally, the image generation model training module includes a loss function determination unit for generating at least one first loss function based on the three-dimensional fundus label data and the three-dimensional fundus prediction image; generating a second loss function based on a preset type of loss function, the first discrimination result, and the second discrimination result, where the second loss function includes one or more of the following: a first loss term generated by the preset type of loss function and the first discrimination result, a second loss term generated by the preset type of loss function and the second discrimination result, and a third loss term generated by the preset type of loss function and the gradient term of the first discrimination result; and generating the target loss function based on the at least one first loss function and the second loss function.

[0106] Based on the above embodiments, optionally, the image generation model training module further includes a preprocessing unit for respectively preprocessing the two two-dimensional fundus scan images and the three-dimensional fundus label data;

[0107] where the preprocessing includes one or more of the following:

[0108] Interpolate the two two-dimensional fundus scanning images and the three-dimensional fundus label data respectively to obtain two-dimensional fundus scanning images of a preset size and the three-dimensional fundus label data of the preset size;

[0109] Perform gray value truncation processing on the two two-dimensional fundus scanning images and the three-dimensional fundus label data respectively;

[0110] Perform gray value normalization processing on the two two-dimensional fundus scanning images and the three-dimensional fundus label data respectively.

[0111] The three-dimensional fundus image generation device provided by the embodiments of the present invention can execute the three-dimensional fundus image generation method provided by any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method.

[0112] Embodiment Five

[0113] Figure 5 FIG. 16 is a schematic structural diagram of an electronic device provided by Embodiment Five of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0114] As Figure 5 shown, the electronic device 10 includes at least one processor 11, and a memory communicatively connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. The memory stores a computer program executable by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0115] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0116] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the three-dimensional fundus image generation method.

[0117] In some embodiments, the three-dimensional fundus image generation method can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the three-dimensional fundus image generation method described above can be executed. Alternatively, in other embodiments, the processor 11 can be configured to execute the three-dimensional fundus image generation method by any other suitable means (e.g., by means of firmware).

[0118] The various embodiments of the systems and technologies described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs, which can be executed and / or interpreted on a programmable system including at least one programmable processor, the programmable processor can be a dedicated or general-purpose programmable processor, and can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0119] The computer program for implementing the three-dimensional fundus image generation method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to the processors of general-purpose computers, special-purpose computers, or other programmable data processing devices, so that when the computer programs are executed by the processors, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs can be executed entirely on the machine, partially on the machine, partially on the machine as an independent software package and partially on a remote machine, or entirely on a remote machine or server.

[0120] Embodiment Six

[0121] Embodiment Six of the present invention further provides a computer-readable storage medium. The computer-readable storage medium stores computer instructions for causing a processor to execute a three-dimensional fundus image generation method, and the method includes:

[0122] Obtain a first two-dimensional fundus scan image and a second two-dimensional fundus scan image of a target object, and the scanning directions corresponding to the first two-dimensional fundus scan image and the second two-dimensional fundus scan image are orthogonal to each other;

[0123] Based on a pre-set image generation model, reconstruct the first two-dimensional fundus scan image and the second two-dimensional fundus scan image to obtain a three-dimensional fundus image of the target object;

[0124] Among them, the image generation model includes: a splicing processing module, an information encryption module, a dimension space conversion module, and an information decryption module;

[0125] Input the first two-dimensional fundus scan image and the second two-dimensional fundus scan image into the splicing processing module for extended dimension splicing to obtain splicing features; input the splicing features into the information encryption module for feature extraction to obtain visual features at different levels; input the visual features at different levels into the dimension space conversion module for dimension conversion to obtain three-dimensional comprehensive features; input the three-dimensional comprehensive features into the information decryption module to output the three-dimensional fundus image of the target object.

[0126] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0127] To provide for interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can also be used to provide for interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, speech input, or tactile input).

[0128] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0129] A computing system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The client-server relationship is created by computer programs running on respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, and solves the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0130] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is imposed herein.

[0131] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for generating a three-dimensional fundus image, characterized in that: include: Acquire a first two-dimensional fundus scanning image and a second two-dimensional fundus scanning image of the target object, wherein scanning directions corresponding to the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image are orthogonal to each other; Based on a preset image generation model, reconstructing the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image to obtain a three-dimensional fundus image of the target object; The image generation model includes: a splicing processing module, an information encryption module, a dimensional space conversion module and an information decryption module; The first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image are input into the stitching processing module for extended dimensional stitching to obtain stitching features; the stitching features are input into the information encryption module for feature extraction to obtain visual features at different levels; the visual features at different levels are input into the dimensional space conversion module for dimensional conversion to obtain three-dimensional comprehensive features; the three-dimensional comprehensive features are input into the information decryption module to output the three-dimensional fundus image of the target object.

2. The method according to claim 1, characterized in that The information encryption module includes a two-dimensional convolution block, a two-dimensional scaling sampling block and a first activation function connected in sequence; The dimensional space conversion module includes a convolution block, a fully connected block, a three-dimensional reorganization processing block and a second activation function connected in sequence; The information decryption module includes a three-dimensional convolution block, a three-dimensional extended sampling block, a three-dimensional deconvolution block and a third activation function connected in sequence; An N-fold loop connection is provided between the output end and the input end of the information encryption module; and / or an M-fold loop connection is provided between the output end and the input end of the information decryption module.

3. The method according to claim 1, characterized in that After acquiring the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image of the target object, the method further includes: Respectively obtain the minimum length unit between pixels in each dimensional direction of the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image, and the preset length of the three-dimensional fundus image in three dimensional directions; Determining the dimensional size of the three-dimensional fundus image based on the preset lengths in the three-dimensional directions and the minimum length unit between pixels in each dimensional direction, the dimensional size including the number of pixels in the three-dimensional directions; Based on the dimension, the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image are limited in range to obtain the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image in the target range; Grayscale updating processing is performed on the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image in the target range, wherein the grayscale updating processing includes one or more of the following: grayscale value truncation processing and grayscale value normalization processing.

4. The method according to claim 3, characterized in that The gray value truncation process includes: For each two-dimensional fundus scanning image of the target range in the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image of the target range, grayscale value distribution data of each two-dimensional fundus scanning image of the target range is determined, and based on the preset upper limit distribution data and the preset lower limit distribution data, the grayscale value distribution data of each two-dimensional fundus scanning image of the target range are matched respectively to determine the target lower limit grayscale value and the target upper limit grayscale value corresponding to each two-dimensional fundus scanning image of the target range; Update the image grayscale value of each two-dimensional fundus scan image in the target range by using the target lower limit grayscale value and the target upper limit grayscale value corresponding to each two-dimensional fundus scan image in the target range; The gray value normalization process includes: Based on the target lower limit grayscale value and the target upper limit grayscale value of the two-dimensional fundus scanning image of each target range, determine the normalized grayscale value corresponding to the grayscale value in the two-dimensional fundus scanning image of each target range, and update the grayscale value in the two-dimensional fundus scanning image of each target range to the corresponding normalized grayscale value.

5. The method according to claim 1, characterized in that The method further comprises: Acquire two two-dimensional fundus scan images and three-dimensional fundus label data corresponding to the sample object in orthogonal directions, wherein the three-dimensional fundus label data includes the three-dimensional fundus scan image of the sample object; Acquire an image generation model to be trained, and reconstruct the two two-dimensional fundus scan images using the image generation model to be trained to obtain a three-dimensional fundus prediction image; Acquire a discriminator, and discriminate the three-dimensional fundus prediction image and the three-dimensional fundus label data respectively by the discriminator to obtain a first discrimination result corresponding to the three-dimensional fundus prediction image and a second discrimination result corresponding to the three-dimensional fundus label data; generating a target loss function based on the three-dimensional fundus prediction image, the three-dimensional fundus label data, the first discrimination result, and the second discrimination result; With the goal of reducing the target loss function, the image generation model to be trained is trained until a trained image generation model is obtained.

6. The method according to claim 5, characterized in that Generating a target loss function based on the three-dimensional fundus prediction image, the three-dimensional fundus label data, the first discrimination result, and the second discrimination result includes: at least one first loss function generated based on the three-dimensional fundus label data and the three-dimensional fundus prediction image; Generate a second loss function based on a preset type of loss function, the first discrimination result, and the second discrimination result, wherein the second loss function includes one or more of the following: a first loss term generated by a preset type of loss function and the first discrimination result, a second loss term generated by a preset type of loss function and the second discrimination result, and a third loss term generated by a preset type of loss function and a gradient term of the first discrimination result; The target loss function is generated based on the at least one first loss function and the second loss function.

7. The method according to claim 5, characterized in that The method further comprises: Preprocessing the two two-dimensional fundus scan images and the three-dimensional fundus label data respectively; Wherein, the preprocessing includes one or more of the following: Interpolating the two two-dimensional fundus scanning images and the three-dimensional fundus label data respectively to obtain a two-dimensional fundus scanning image of a preset size and the three-dimensional fundus label data of a preset size; Performing grayscale value truncation processing on the two two-dimensional fundus scan images and the three-dimensional fundus label data respectively; Grayscale value normalization processing is performed on the two two-dimensional fundus scanning images and the three-dimensional fundus label data respectively.

8. A three-dimensional fundus image generating device, characterized in that: include: A fundus scanning image acquisition module, used to acquire a first two-dimensional fundus scanning image and a second two-dimensional fundus scanning image of a target object, wherein the scanning directions corresponding to the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image are orthogonal to each other; A three-dimensional fundus image reconstruction module, used to reconstruct the first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image based on a preset image generation model to obtain a three-dimensional fundus image of the target object; The image generation model includes: a splicing processing module, an information encryption module, a dimensional space conversion module and an information decryption module; The first two-dimensional fundus scanning image and the second two-dimensional fundus scanning image are input into the stitching processing module for extended dimensional stitching to obtain stitching features; the stitching features are input into the information encryption module for feature extraction to obtain visual features at different levels; the visual features at different levels are input into the dimensional space conversion module for dimensional conversion to obtain three-dimensional comprehensive features; the three-dimensional comprehensive features are input into the information decryption module to output the three-dimensional fundus image of the target object.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the three-dimensional fundus image generating method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the three-dimensional fundus image generation method according to any one of claims 1 to 7 when executed.

Images