A fundus structure three-dimensional reconstruction method and a fundus imaging system

Abstract

The application is suitable for the field of optics, and discloses a fundus structure three-dimensional reconstruction method and a fundus imaging system. The method comprises the following steps: acquiring a plurality of pre-calibrated focal point positions and a micromirror array curvature corresponding to each focal point position, and determining an optimal focusing position for fundus imaging; taking the optimal focusing position as the center, determining a fundus imaging covering focal point interval according to the optimal focusing position, and acquiring a fundus image sequence of the focal point positions covered by the fundus imaging covering focal point interval; for each pixel in the fundus image sequence, calculating the neighborhood features of the pixel corresponding to different focal point positions in the fundus image sequence, obtaining the depth score of the pixel at different focal point positions, and using the focal point position corresponding to the highest depth score as the height information of the pixel; combining the lateral coordinate, longitudinal coordinate and height information of each pixel to form a point cloud set, and splicing the point cloud set to obtain a fundus three-dimensional structure. The fundus three-dimensional structure is accurately reconstructed through micromirror focusing.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optics, in particular to a fundus structure three-dimensional reconstruction method and a fundus imaging system. BACKGROUND

[0002] In the field of clinical ophthalmology, in order to achieve early and accurate judgment of fundus lesions caused by diseases such as diabetes and thrombosis, fundus cameras are often used in clinical practice to detect and image the deep blood vessels and cell tissues of the human retina. However, due to significant individual differences in the refractive states of the eyeballs of different subjects (such as refractive errors such as myopia and hypermetropia), the fundus camera needs to dynamically compensate for the refractive differences through a focusing system to meet the clinical diagnostic needs of eyes with different refractive errors.

[0003] Current clinical application of fundus camera focusing technology mainly falls into two categories: one is manual focusing technology, which is relatively cumbersome and highly dependent on the clinical experience of the operator. Due to the large difference in refractive states of the subjects, the focusing process often needs to be repeated several times, which can easily lead to blurred images of key areas of the fundus (such as the macula and optic disc) due to operational deviations, increasing the risk of missed detection of lesions and potentially affecting the accuracy of diagnosis. The other is a passive focusing scheme that uses an electric motor to drive the focusing lens. This technology collects fundus images at different positions along the motor drive direction, and selects the optimal focusing image through a focusing evaluation operator, which simplifies the operation process to some extent, but introduces mechanical transmission structures, resulting in a relatively large device size and high energy consumption. At the same time, due to the limitations of the quality of the focusing lens and the motor drive performance, the focusing frequency is generally low, which makes it difficult to meet the current clinical needs of miniaturization and high-speed detection of fundus examination devices.

[0004] Therefore, the existing technology needs to be improved and developed. SUMMARY

[0005] The first object of the present application is to provide a fundus structure three-dimensional reconstruction method that accurately reconstructs the three-dimensional structure of the fundus through micro-lens focusing.

[0006] To achieve the above-mentioned object, the present application provides the following scheme:

[0007] The application discloses a fundus structure three-dimensional reconstruction method based on a fundus imaging system, the fundus imaging system comprising a micromirror array, the fundus structure three-dimensional reconstruction method comprising: obtaining a plurality of pre-calibrated focal point positions and a micromirror array curvature corresponding to each focal point position, and determining an optimal focusing position of fundus imaging; taking the optimal focusing position as a center, extending the optimal focusing position to each of a front direction and a rear direction by a preset number of focal point positions to form a fundus imaging covering focal point interval, and obtaining a fundus depth image of the focal point positions covered by the fundus imaging covering focal point interval to form a fundus image sequence; for each pixel in the fundus image sequence, calculating neighborhood features of the pixel corresponding to different focal point positions in the fundus image sequence to obtain a depth score of the pixel at different focal point positions, and using a focal point position corresponding to a highest depth score as height information of the pixel, and traversing all pixels in the fundus image sequence to obtain height information of all pixels; combining a horizontal coordinate, a vertical coordinate and the height information of each pixel to form a point cloud set, and splicing the point cloud set to obtain a fundus three-dimensional structure, wherein each point in the point cloud set represents a three-dimensional space position of a pixel.

[0008] Preferably, the step of obtaining a plurality of pre-calibrated focal point positions and a micromirror array curvature corresponding to each focal point position, and determining an optimal focusing position of fundus imaging comprises: obtaining a plurality of pre-calibrated focal point positions and a micromirror array curvature corresponding to each focal point position; pre-imaging a fundus structure every m focal point positions to obtain a plurality of focusing images; calculating a definition score of the plurality of focusing images, and determining a focal point position corresponding to a focusing image with a highest definition score as the optimal focusing position.

[0009] Preferably, m=10.

[0010] Preferably, the definition score of the plurality of focusing images is calculated by using a Tenengrad operator.

[0011] Preferably, the step of, for each pixel in the fundus image sequence, calculating neighborhood features of the pixel corresponding to different focal point positions in the fundus image sequence to obtain a depth score of the pixel at different focal point positions, and using a focal point position corresponding to a highest depth score as height information of the pixel, and traversing all pixels in the fundus image sequence to obtain height information of all pixels comprises: for each pixel in the fundus image sequence, selecting eight adjacent pixels around the pixel as a neighborhood of the pixel, and calculating neighborhood features of the pixel corresponding to different focal point positions in the fundus image sequence by using a kernel function to obtain a depth score of the pixel at different focal point positions; using a focal point position corresponding to a highest depth score as height information of the pixel, and traversing all pixels in the fundus image sequence to obtain height information of all pixels.

[0012] Preferably, the lateral coordinate, longitudinal coordinate and height information of each pixel are combined to form a point cloud set, and the fundus three-dimensional structure is spliced through the point cloud set, each point in the point cloud set representing a three-dimensional space position of a pixel, comprising: traversing all pixels in the fundus image sequence, sequentially storing the lateral coordinate, longitudinal coordinate and height information of each pixel into a pre-constructed blank dataset to form an initial dataset; preprocessing the initial dataset to obtain a point cloud set, each point in the point cloud set representing a three-dimensional space position of a pixel; based on the point cloud set, the point cloud data of the point cloud set is spliced using a Poisson surface reconstruction algorithm to obtain the fundus three-dimensional structure.

[0013] A second object of the present application is to provide a fundus imaging system for implementing the fundus structure three-dimensional reconstruction method as described above, the fundus imaging system comprising a camera, an imaging objective lens, a first half-mirror, a second half-mirror, an ocular objective lens, a micromirror array and a light source, the light beam emitted by the light source being reflected to the ocular objective lens through the second half-mirror, being focused by the ocular objective lens and then being incident on the eyeball, the light reflected by the eyeball being adjusted in direction by the second half-mirror and the first half-mirror, being incident on the micromirror array, changing the focusing depth by the micromirror array, being emitted to the first half-mirror, being reflected to the imaging objective lens by the first half-mirror, being focused by the imaging objective lens and then being incident on the camera.

[0014] Preferably, the camera is a CMOS camera.

[0015] In this scheme, the pre-calibrated multiple focal point positions and the micromirror array curvature corresponding to each focal point position are obtained, the best focusing position for fundus imaging is determined, and then the best focusing position is taken as the center to expand to form a focal point interval and collect a fundus image sequence, the highest corresponding focal point is selected as the height information by calculating the characteristics of different focal point neighborhoods pixel by pixel, and finally the point cloud is formed by combining the lateral coordinate, longitudinal coordinate and height information, and the three-dimensional structure is spliced, which not only realizes flexible regulation and control of multiple focal points by using the micromirror array to ensure that the imaging covers different depth regions of the fundus, but also realizes accurate restoration of the three-dimensional morphology of the fundus by pixel-level depth analysis and point cloud reconstruction, provides high-quality three-dimensional data support for fine observation and disease diagnosis of the fundus structure, and can shorten the overall time consumption of three-dimensional reconstruction of a single patient. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from the structures shown in the drawings without creative labor.

[0017] Figure 1 is a flow chart of the fundus structure three-dimensional reconstruction method provided by the embodiment of the present application;

[0018] Figure 2 is an optical path diagram of the fundus imaging system provided by the embodiment of the present application.

[0019] Explanation of reference signs:

[0020] 10, camera; 20, imaging objective; 30, first half-mirror; 40, second half-mirror; 50, ocular objective; 60, micromirror array; 70, light source; 80, eyeball. DETAILED DESCRIPTION

[0021] The terms "first", "second", "third", "fourth" and the like (if any) in the description, claims and above drawings of the present application are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variation thereof is intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0022] For the convenience of understanding, the specific flow of the embodiments of the present application will be described below. Please refer to Figure 1 In the embodiments of the present application, a fundus structure three-dimensional reconstruction method is provided, and the fundus structure three-dimensional reconstruction method is realized based on a fundus imaging system. The fundus imaging system includes a micromirror array, and the fundus structure three-dimensional reconstruction method includes:

[0023] S101, obtaining a plurality of focal point positions pre-calibrated and a micromirror array curvature corresponding to each focal point position, and determining a best focusing position of fundus imaging;

[0024] S102. Taking the optimal focus position as the center, extend the optimal focus position forward and backward by a preset number of focus positions to form a fundus imaging coverage focus area, and acquire fundus depth images of the focus positions covered by the fundus imaging coverage focus area to form a fundus image sequence.

[0025] S103. For each pixel in the fundus image sequence, calculate the neighborhood features corresponding to different focal positions of the pixel in the fundus image sequence to obtain the depth score of the pixel at different focal positions, and use the focal position corresponding to the highest depth score as the height information of the pixel. Traverse all pixels in the fundus image sequence to obtain the height information of all pixels.

[0026] S104. Combine the horizontal coordinates, vertical coordinates and height information of each pixel to form a point cloud set, and stitch the point cloud set to obtain the three-dimensional structure of the fundus. Each point in the point cloud set represents the three-dimensional spatial position of a pixel.

[0027] In this embodiment, step S101 involves acquiring multiple pre-calibrated focal positions and the curvature of the micromirror array corresponding to each focal position, and determining the optimal focus position for fundus imaging. This includes: acquiring multiple pre-calibrated focal positions and the curvature of the micromirror array corresponding to each focal position; performing pre-imaging of the fundus structure at intervals of m focal positions to obtain multiple focused images; calculating the sharpness score of the multiple focused images, and determining the focal position corresponding to the focused image with the highest sharpness score as the optimal focus position.

[0028] In this embodiment, using its proprietary calibration technology, the micromirror array system can obtain the mapping relationship between the curvature of the micromirrors and the actual focal position after fixing optical components such as the objective lens. By adjusting the angle and position of each micromirror in the micromirror array, the micromirror array can be equivalent to a reflector with different curvatures. Assuming there are a total of... The curvature of the micromirror array determines the number of fundus imaging systems. The focal position, the set of curvature is denoted as The focus set is denoted as ,in .

[0029] In practical applications, since the fundus imaging system itself has depth of field, it is not necessary to traverse every focal point. Imaging calculations can be performed at intervals of m focal points, where m is usually 10, that is, the fundus structure is pre-imaged at intervals of 10 focal points.

[0030] In this embodiment, pre-imaging the fundus structure at every m focal positions means adjusting the curvature of the micromirror array, focusing on the corresponding focal position, and then triggering the camera of the fundus imaging system to take a picture, thereby obtaining a focused image corresponding to that focal position.

[0031] In the present embodiment, the Tenengrad operator, which is sensitive to edges and robust, is used to calculate the sharpness scores of the plurality of focus images. The greater the sharpness score value, the more clearly the focus image is in focus, and the greater the image gradient value, with rich edge information.

[0032] The Tenengrad operator quantifies the sharpness by calculating the gray level gradient change of the pixels in the image. Assuming that the two-dimensional gray value of the focus image is , wherein respectively represent the horizontal and vertical pixel coordinates (discrete values) of the image, The value of the pixel reflects the light and dark degree of the pixel.

[0033] When using the Tenengrad operator to score the sharpness of the focus image, first calculate the gray level gradient components in two directions, and then calculate the sharpness score of the focus image according to the gray level gradient components in two directions.

[0034] The first gray level gradient component reflects the gray level change of the pixel in the horizontal direction, and is obtained by calculating the gray level difference between the right neighborhood ( row) and the left neighborhood ( row) of the pixel. is expressed as: .

[0035] The second gray level gradient component reflects the gray level change of the pixel in the vertical direction, and is obtained by calculating the gray level difference between the lower neighborhood ( column) and the upper neighborhood ( column) of the pixel, is expressed as:

[0036] .

[0037] The sharpness score of the focus image is calculated according to the gray level gradient components in two directions is expressed as: .

[0038] In other embodiments, focus evaluation operators such as the Laplace operator and the Brenner operator can also be used to score the sharpness of the plurality of focus images, to obtain a plurality of sharpness scores.

[0039] In this embodiment, in step S102, the best focus position is taken as the center, and the best focus position is extended by k focus positions in front and back directions respectively. The value of k is estimated by the statistical law of the population lens thickness, which uses a large amount of clinical data to analyze the depth interval that needs to be covered by the fundus imaging corresponding to the lens thickness fluctuation, to ensure that the traversal range is complete and covers the depth information of the fundus tissue, and does not increase the number of invalid focus too much. That is, the fundus imaging covers the focus interval [z j-k , z j+k ].

[0040] In this embodiment, based on 1000 cases of lens thickness detection of people in different age groups, the thickness fluctuation range is 2.5-5.5mm, and k=20 focus points.

[0041] In this embodiment, the fundus imaging system controls the micromirror array to switch to all focus positions of the fundus imaging cover focus interval , triggers the camera of the fundus imaging system to collect the fundus depth image corresponding to each focus point, and obtains the fundus image sequence with depth information , each fundus depth image contains the depth information of the fundus tissue structure, such as the texture and blood vessel cross section of a certain layer of retina.

[0042] In this embodiment, in step S103, for each pixel in the fundus image sequence, 8 adjacent pixels around the pixel are selected as the neighborhood of the pixel, and the neighborhood features of the pixel corresponding to different focus positions in the fundus image sequence are calculated by the kernel function to obtain the score of the pixel under different focus points.

[0043] For example, for a pixel in the fundus image sequence, its neighborhood is taken as , which is expressed as:

[0044] .

[0045] In this embodiment, the neighborhood feature score of the pixel under different focus points is obtained by the kernel function . The height information of the pixel is determined as by solving . , and traversing all pixels in the fundus image sequence, the height information of each pixel is obtained, and the mapping of two-dimensional pixels to three-dimensional coordinates in the height dimension is completed.

[0046] In the embodiment, the kernel function can be based on the weighted calculation of the gray difference to quantitatively score the neighborhood features (edge definition, texture consistency), so as to obtain the depth score of the pixel at each focal position. The higher the score is, the clearer the imaging of the pixel at the focal position is, and the closer the pixel is to the real depth.

[0047] In the embodiment, in step S104, in the embodiment, the lateral coordinate, the longitudinal coordinate and the height information of each pixel are combined to form a point cloud set, and the fundus three-dimensional structure is obtained by splicing the point cloud set. Each point in the point cloud set represents a three-dimensional space position of a pixel, including: traversing all pixels in the fundus image sequence, storing the lateral coordinate, the longitudinal coordinate and the height information of each pixel into a pre-constructed blank dataset in turn to form an initial dataset; preprocessing the initial dataset to obtain a point cloud set, each point in the point cloud set representing a three-dimensional space position of a pixel; based on the point cloud set, the point cloud data of the point cloud set is spliced by using a Poisson surface reconstruction algorithm to obtain the fundus three-dimensional structure. The fundus three-dimensional structure can directly display the spatial distribution of the fundus tissue, such as the direction of blood vessels and the depth position of lesions, and provide a quantitative three-dimensional imaging basis for clinical diagnosis.

[0048] In the embodiment, the lateral coordinate , the longitudinal coordinate and the height information of each pixel are combined to form a three-dimensional coordinate point . All three-dimensional coordinate points of the pixels form an initial dataset, and each point in the initial dataset accurately corresponds to the spatial coordinates of a certain position of the fundus, such as the cross-sectional position of a certain blood vessel or the cell distribution of a certain layer of the retina.

[0049] In the embodiment, when all valid pixels in the fundus image sequence are traversed, the invalid black border pixels of the image edge can be excluded, and only the valid pixels are retained.

[0050] Exemplarily, a fundus image with a resolution of 1024x1024 can generate about 1 million three-dimensional points, each of which accurately corresponds to the spatial coordinates of a certain microscopic position of the fundus (such as a point of the retina or a point of the blood vessel wall).

[0051] In the embodiment, the initial dataset is preprocessed, including noise removal processing and missing point completion processing.

[0052] In the embodiment, the point cloud data in the initial dataset can contain noise (such as an incorrect depth value caused by imaging blur), redundant points (such as overlapping areas calculated repeatedly) or missing points (such as unimaged pixels in the shadow area of the fundus), and thus the noise removal processing and the missing point completion processing are needed to improve the reliability.

[0053] Specifically, for each three-dimensional coordinate point , calculate the average depth of points in its 5x5 neighborhood ; if the difference between the z of the three-dimensional coordinate point and exceeds a preset threshold, it is determined to be a noise point and is removed. The noise removal process can filter abnormal depth values caused by micromirror array adjustment errors or CMOS sensor noise.

[0054] In this embodiment, the preset threshold is set based on the continuity of the fundus tissue, for example, the preset threshold is ±5 μm.

[0055] In this embodiment, the missing point completion process is specifically: for the missing pixels of the fundus shadow area, the point cloud is completed based on the depth distribution of the surrounding effective points to ensure the continuity of the three-dimensional structure. The fundus shadow area refers to areas such as unimaged areas caused by lens obstruction.

[0056] In this embodiment, for edge points of key structures such as fundus blood vessels and macular areas (whose neighborhood depth changes should be larger), edge detection algorithms are used to mark them to avoid being misjudged as noise.

[0057] In this embodiment, if the point cloud density is too high (such as the depth difference between adjacent pixels being less than 0.1 μm), a voxel downsampling algorithm is used to reduce the data volume while preserving the details of key structures.

[0058] In this embodiment, the point cloud data of the point cloud set is regarded as sampling points in a three-dimensional space, a continuous surface passing through these points is fitted by solving the Poisson equation, and the surface is discretized into triangular meshes through mesh division, thereby generating a visualized fundus three-dimensional structure. In the fundus three-dimensional structure, the spatial positions and shapes of each layer of the retina, blood vessel distribution, and lesions (such as microaneurysms) can be clearly distinguished. The Poisson surface reconstruction algorithm is suitable for the reconstruction of layered structures such as the fundus retina. The Poisson surface reconstruction algorithm can automatically smooth noise while preserving the details of protruding structures such as blood vessels.

[0059] In this embodiment, anatomical priors of the fundus can be combined to impose constraints on the reconstructed surface (such as limiting the thickness of a certain layer of the retina to be between 100-200 μm), to avoid the appearance of biologically unreasonable structure shapes.

[0060] In this embodiment, anatomical priors of the fundus such as the thickness range of each layer of the retina and the trend of blood vessel orientation.

[0061] In this embodiment, constraints are imposed on the reconstructed surface, such as limiting the thickness of a certain layer of the retina to be between 100-200 μm.

[0062] In the embodiment, if the fundus image sequence covers multiple fields of view (such as multi-region imaging realized by micro-mirror array switching), the point clouds of different fields of view need to be aligned through point cloud registration: based on feature points such as blood vessel intersection points and macular fovea, a rotation matrix and a translation vector are calculated to unify the multi-view point clouds to the same three-dimensional coordinate system.

[0063] In the embodiment, by acquiring a plurality of focal point positions pre-calibrated and the micro-mirror array curvature corresponding to each focal point position, the best focusing position for fundus imaging is determined, and then the best focusing position is used as the center to expand to form a focal point interval and collect a fundus image sequence, the highest corresponding focal point is selected as the height information through pixel-by-pixel calculation of different focal point neighborhood features and depth score screening, and finally the point cloud is formed by combining the horizontal and vertical coordinates with the height information, and the three-dimensional structure is spliced, which not only realizes flexible regulation and control of multiple focal points by using the micro-mirror array to ensure that the imaging covers different depth regions of the fundus, but also realizes accurate restoration of the three-dimensional morphology of the fundus by pixel-level depth analysis and point cloud reconstruction, thereby providing high-quality three-dimensional data support for fine observation of the fundus structure and disease diagnosis, and shortening the overall time consumption of three-dimensional reconstruction of a single patient.

[0064] Referring to Figure 2 The embodiment of the present application also provides a fundus imaging system for realizing the fundus structure three-dimensional reconstruction method, and the fundus imaging system comprises a camera 10, an imaging objective 20, a first semi-transparent half mirror 30, a second semi-transparent half mirror 40, an ocular objective 50, a micro-mirror array 60 and a light source 70. The light beam emitted by the light source 70 is reflected by the second semi-transparent half mirror 40 to the ocular objective 50, is focused by the ocular objective 50 and then is incident on the eyeball 80, the light reflected by the eyeball 80 is adjusted in direction by the second semi-transparent half mirror 40 and the first semi-transparent half mirror 30 and then is incident on the micro-mirror array 60, is changed in focusing depth by the micro-mirror array 60, is emitted to the first semi-transparent half mirror 30, is reflected by the first semi-transparent half mirror 30 to the imaging objective 20, is focused by the imaging objective 20 and then is incident on the camera 10.

[0065] It should be noted that the micromirror array 60 is composed of thousands of micro mirrors with a size of 100 μm x 100 μm in a circular array with a diameter of 10 mm. The rotation and translation of the micromirrors on the same circumference can be synchronously controlled by electrical signals, and the array can accurately realize curvature fitting by controlling the rotation angle of the micromirrors on different circumferences. Compared with the Digital Micromirror Devices (DMD) of Texas Instruments (TI), the micromirror array 60 has obvious advantages. The maximum state switching number of the DMD is usually 3, which is commonly used in optical switches and other scenes with simple and discrete state switching requirements. However, the maximum state switching number of the micromirror array 60 can reach 960, and each state corresponds to a focal point position. Therefore, the micromirror array 60 has obvious advantages in the accuracy and flexibility of optical control, and can be directly used for optical focusing. Due to the small size, the rated power of this component is low, and the integration on the fundus camera 10 can realize the miniaturization of the system.

[0066] In this embodiment, the camera 10 is a CMOS camera 10 (Complementary Metal Oxide Semiconductor Camera), and the core component is a complementary metal oxide semiconductor (CMOS) image sensor. The sensor is used to convert optical signals into electrical signals, thereby realizing image shooting and acquisition.

[0067] In this embodiment, the core components such as the double half-transmission half-reflection mirror and the micromirror array 60 are scientifically arranged to construct an efficient light path. The light beam of the light source 70 is accurately incident on the eyeball 80 after being reflected by the second half-transmission half-reflection mirror 40 and focused by the objective lens 50. The reflected light of the eyeball 80 is sequentially turned to the micromirror array 60 by the double half-transmission half-reflection mirror, and the focusing depth is flexibly adjusted by the micromirror array 60. The light is collected by the camera 10 after being reflected by the first half-transmission half-reflection mirror 30 and focused by the imaging objective lens 20. This not only realizes efficient separation and cooperation of the illumination light path and the imaging light path, but also meets the demand for multi-focal point fundus image sequence for three-dimensional reconstruction by dynamically adjusting the focusing depth with the help of the micromirror array 60. In addition, the optical characteristics of the eyeball 80 are adapted by the objective lens 50, and the light path is stably turned by the double half-transmission half-reflection mirror, so as to ensure low light transmission loss and clear imaging. The overall structure is compact and the components cooperate accurately, thereby providing stable, efficient and high-quality hardware support for three-dimensional reconstruction of the fundus structure, and having excellent optical performance, strong adaptability and convenient operation.

[0068] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structural transformation made according to the content of the present application specification and drawings, or direct / indirect application in other related technical fields within the concept of the present application is included in the patent protection scope of the present application.

Claims

1. A method of three-dimensional reconstruction of an ocular fundus structure, characterized by, The fundus structure three-dimensional reconstruction method is realized based on a fundus imaging system, and the fundus imaging system comprises a microlens array. A plurality of focal point positions and a microlens array curvature corresponding to each focal point position are acquired, and an optimal focusing position for fundus imaging is determined. The optimal focusing position is taken as a center, and the optimal focusing position is respectively extended to a preset number of focal point positions in front and back directions, so that a fundus imaging covering focal point interval is formed, and a fundus depth image of the focal point interval covered by the fundus imaging is acquired, and a fundus image sequence is formed. For each pixel in the fundus image sequence, neighborhood features corresponding to different focal point positions of the pixel in the fundus image sequence are calculated, a depth score of the pixel at different focal point positions is obtained, and a focal point position corresponding to a highest depth score is used as height information of the pixel, and all pixels in the fundus image sequence are traversed to obtain height information of all pixels. The fundus three-dimensional structure is obtained by splicing a point cloud set formed by combining lateral coordinates, longitudinal coordinates and height information of each pixel.

2. The method of three-dimensional reconstruction of the fundus structure according to claim 1, characterized in that, The plurality of focal point positions and the microlens array curvature corresponding to each focal point position are acquired, and the optimal focusing position for fundus imaging is determined. The plurality of focal point positions and the microlens array curvature corresponding to each focal point position are acquired. The fundus structure is pre-imaged every m focal point positions to obtain a plurality of focusing images. The clarity scores of the plurality of focusing images are calculated, and a focal point position corresponding to a focusing image with a highest clarity score is determined as the optimal focusing position.

3. The method of three-dimensional reconstruction of the fundus structure according to claim 2, characterized in that, The m = 10.

4. The method of three-dimensional reconstruction of the fundus structure according to claim 2, wherein The clarity scores of the plurality of focusing images are calculated using a Tenengrad operator.

5. The method of three-dimensional reconstruction of the fundus structure according to claim 1, wherein For each pixel in the fundus image sequence, neighborhood features corresponding to different focal point positions of the pixel in the fundus image sequence are calculated, a depth score of the pixel at different focal point positions is obtained, and a focal point position corresponding to a highest depth score is used as height information of the pixel, and all pixels in the fundus image sequence are traversed to obtain height information of all pixels. For each pixel in the fundus image sequence, eight adjacent pixels around the pixel are selected as a neighborhood of the pixel, and neighborhood features corresponding to different focal point positions of the pixel in the fundus image sequence are calculated through a kernel function to obtain a depth score of the pixel at different focal point positions. A focal point position corresponding to a highest depth score is used as height information of the pixel, and all pixels in the fundus image sequence are traversed to obtain height information of all pixels.

6. The method of three-dimensional reconstruction of the fundus structure according to claim 1, wherein The fundus three-dimensional structure is obtained by splicing a point cloud set formed by combining lateral coordinates, longitudinal coordinates and height information of each pixel. All pixels in the fundus image sequence are traversed, and lateral coordinates, longitudinal coordinates and height information of each pixel are sequentially stored in a pre-constructed blank data set to form an initial data set. Preprocessing the initial data set to obtain a point cloud set, each point in the point cloud set representing a three-dimensional spatial position of a pixel; Based on the point cloud set, the point cloud data of the point cloud set is spliced by using a Poisson surface reconstruction algorithm to obtain an eye fundus three-dimensional structure.

7. A fundus imaging system characterized by, The eye fundus imaging system is used to implement the eye fundus structure three-dimensional reconstruction method according to any one of claims 1-6, and the eye fundus imaging system comprises a camera, an imaging objective lens, a first half-transmission half-reflection mirror, a second half-transmission half-reflection mirror, an ocular objective lens, a micromirror array and a light source, the light beam emitted by the light source is reflected to the ocular objective lens through the second half-transmission half-reflection mirror, is focused by the ocular objective lens and is incident to an eyeball, the light reflected by the eyeball is adjusted in direction by the second half-transmission half-reflection mirror and the first half-transmission half-reflection mirror, is incident to the micromirror array, changes the focusing depth by the micromirror array, is emitted to the first half-transmission half-reflection mirror, is reflected to the imaging objective lens by the first half-transmission half-reflection mirror, is focused by the imaging objective lens and is incident to the camera.

8. The fundus imaging system of claim 7, wherein, The camera is a CMOS camera.

Images