Retinal vessel three-dimensional imaging system and method based on ultra-wide-angle laser fundus camera

Abstract

The invention discloses a retinal vessel three-dimensional imaging system and method based on an ultra-wide-angle laser fundus camera, and relates to the field of medical imaging devices. The problems that layered imaging cannot be achieved through single-wavelength laser in existing fundus imaging, and the scanning angle of an existing dual-wavelength confocal system is limited are solved. The system comprises a multi-wavelength laser generating unit, a semi-transmitting and semi-reflecting element, a galvanometer unit, a chromatic aberration lens, a sample to be detected, a focusing lens and a detector with a pinhole, laser emitted by the multi-wavelength laser generating unit is reflected by the semi-transparent and semi-reflective element, then sequentially passes through the galvanometer unit and the chromatic aberration lens, is emitted to a to-be-detected sample, sequentially passes through the chromatic aberration lens, the galvanometer unit, the semi-transparent and semi-reflective element and the focusing lens after being reflected by the to-be-detected sample, is received by the detector with a pinhole, and reaches the silicon PM; a multi-wavelength retina confocal image is generated from the signals acquired by the silicon PM. The method is also suitable for the field of full-layer three-dimensional imaging from the retina nerve fiber layer to the choroid.

Description

Technical Field

[0001] The present invention relates to the technical field of medical imaging devices, and in particular to a retinal blood vessel three-dimensional imaging system and method based on an ultra-wide-angle laser fundus camera. Background Art

[0002] Existing fundus cameras have a field of view of ≤60°, and a single image captures only about 15% of the retinal area. This means they cannot detect peripheral retinal diseases such as holes and degeneration. If necessary, medications may be used to dilate the pupil to increase the amount of light entering. This can cause blurred vision and photophobia for 4-6 hours after the examination, affecting work and driving.

[0003] The existing technology for fundus imaging has the following bottlenecks: First, a common problem in fundus imaging is uneven light distribution, which results in low image contrast, high noise, and areas with unclear boundaries; the human eye's optical system has wavefront aberrations, which limits the imaging resolution and makes it difficult to achieve cellular-level high resolution.

[0004] Secondly, existing imaging technologies such as 785nm infrared light have problems such as insufficient clarity in displaying choroidal blood vessels with a diameter of less than 100μm.

[0005] Finally, fundus imaging is affected by imaging distortion, mainly including three-dimensional to two-dimensional distortion. When the three-dimensional retinal image is converted to a two-dimensional plane, varying degrees of distortion will occur in the peripheral area. The pupil size also limits the overlap of the illumination and imaging beams, resulting in reflections of the cornea and lens or reduced image contrast. The existing dual-wavelength confocal system has a scanning angle of only 120 degrees, and the existing Optos camera lacks three-dimensional reconstruction capabilities. The OCT technology has a limited penetration depth of ≤2mm, making it difficult to simultaneously acquire choroidal blood vessels. That is, for fundus imaging, a single-wavelength laser cannot achieve layered imaging.

[0006] In summary, in order to obtain more fundus depth and width information and view fundus images more intuitively and clearly, the present invention provides a retinal blood vessel three-dimensional imaging system and method based on an ultra-wide-angle laser fundus camera. Summary of the Invention

[0007] The present invention aims to solve the problems in the prior art of fundus imaging, such as the inability of single-wavelength laser to achieve layered imaging, the limited scanning angle of the existing dual-wavelength confocal system, and the limited penetration depth of the existing OCT technology, which makes it difficult to synchronously acquire choroidal blood vessels.

[0008] To solve the above technical problems, the present invention is achieved through the following technical solutions: Solution 1: The present invention proposes a retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera, the system comprising a multi-wavelength laser generating unit, a semi-transmissive and semi-reflective element, a galvanometer unit, a chromatic aberration lens, a sample to be measured, a focusing lens, and a detector with a pinhole; The laser emitted by the multi-wavelength laser generating unit is reflected by the semi-transparent and semi-reflective element, and then passes through the galvanometer unit and the chromatic aberration lens in sequence to be emitted to the sample to be tested. After being reflected by the sample to be tested, it passes through the chromatic aberration lens, the galvanometer unit, the semi-transparent and semi-reflective element, and the focusing lens in sequence, is received by the detector with a pinhole, and reaches the silicon PM. A multi-wavelength retinal confocal image is generated based on the signal collected by the silicon PM.

[0009] Furthermore, a preferred embodiment is provided, wherein the multi-wavelength laser generating unit is a continuous wavelength laser generator or a discontinuous laser generator.

[0010] Furthermore, a preferred embodiment is provided, wherein the chromatic aberration of the chromatic aberration lens is axial chromatic aberration.

[0011] Furthermore, a preferred embodiment is provided, wherein the multi-wavelength laser generating unit includes a laser and a laser collimating unit, wherein the laser is used to diverge the laser light source, and the laser collimating unit converts the diverged laser beam into a parallel beam.

[0012] Solution 2: A method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera, the method being implemented based on any system described in Solution 1, the method comprising the following steps: Step 1: registering the generated multi-wavelength retinal confocal images; Step 2: Segment the four-channel image after registration in step 1 to obtain a complete fundus image of the vascular choroid map. Step 3: Generate a three-dimensional vascular model based on the vascular choroid map using the vascular segmentation mask to achieve three-dimensional retinal vascularization. Image visualization.

[0013] Furthermore, a preferred embodiment is provided, in which the method for registering the generated multi-wavelength retinal confocal images in step 1 is: Construct Gaussian pyramid and DoG pyramid, detect DoG extreme points, find positioning key points, assign main directions based on the positioned key points, generate feature descriptors, and complete SIFT feature point extraction.

[0014] Furthermore, a preferred embodiment is provided, in which the method for segmenting the registered four-channel image in step 2 is: using an Attention U-Net model to segment the registered four-channel image.

[0015] Furthermore, a preferred embodiment is provided, wherein step 2 also includes the steps of performing data enhancement and normalization processing on the vascular choroid map.

[0016] Furthermore, a preferred embodiment is provided, in step 3, the method for generating a three-dimensional vascular model using a vascular segmentation mask according to the vascular choroid map to realize visualization of retinal vascular three-dimensional imaging is: The segmented vascular mask is first subjected to skeleton extraction to obtain a set of vascular centerlines, the local principal axis direction is estimated, and the skeleton point cloud is sparsely connected and optimized. Continuity is maintained based on the principal direction, and isolated noise points are removed. Intensity sampling is performed in the normal direction to obtain the radius of each skeleton point. 3D point coordinates are then constructed to achieve visualization of 3D imaging of retinal vessels.

[0017] Furthermore, a preferred embodiment is provided, wherein the method for constructing 3D point coordinates is: Vascular mask and confocal Z-axis slice information, each slice spacing , for each blood vessel pixel (x, y), the relative depth is calculated based on the signal intensity I(x, y):

[0018] Convert 2D pixels to 3D physical coordinates: .

[0019] The present invention is beneficial in that: The present invention provides a retinal blood vessel three-dimensional imaging system and method based on an ultra-wide-angle laser fundus camera, which combines four-channel depth scanning confocal and 3D image reconstruction algorithms to provide more intuitive and three-dimensional image information.

[0020] The retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera described in the present invention adopts a multi-wavelength laser generating unit that can be used to switch lasers of different wavelengths. Combined with a chromatic aberration lens, it can quickly capture images of samples to be tested at different depths, reducing operational difficulty and improving depth image acquisition efficiency. Due to its fast shooting speed, photos of different depths are less likely to be misaligned, especially for objects with low-frequency vibrations. That is, the present invention can achieve better scanning effects than the existing technology.

[0021] The present invention is also applicable to the application field of full-layer three-dimensional imaging from the retinal nerve fiber layer to the choroid. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Schematic diagram of the architecture of a retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera as described in embodiment 1.

[0023] Figure 2Schematic diagram of the optical path of a retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera as described in embodiment 1.

[0024] Figure 3 Schematic diagram of the principle of focal plane translation of a retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera as described in embodiment eleven.

[0025] Figure 4 These are four sets of wavelength penetration characteristic curves in a retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera as described in the eleventh embodiment.

[0026] Figure 5 This is a flow chart of a method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera as described in embodiment five.

[0027] Figure 6 This is a model structure diagram described in the eleventh embodiment.

[0028] Figure 7 This is a 3D schematic diagram of the eleventh embodiment.

[0029] Among them, there are a multi-wavelength laser generating unit 1, a semi-transmissive and semi-reflective element 2, a galvanometer unit 3, a chromatic aberration lens 4, a sample to be measured 5, a focusing lens 6, a detector with a pinhole 7, a pinhole 7-1, a detector 7-2, photos of samples at different depths 8, a first wavelength 10-1, a second wavelength 10-2, an n-th wavelength 10-n, an ideal imaging focal plane 5-1-0 corresponding to the first wavelength 10-1, an ideal imaging focal plane 5-1-1 corresponding to the second wavelength 10-2, an ideal imaging focal plane 5-1-n corresponding to the n-th wavelength 10-n, a first reflecting mirror 11, a second reflecting mirror 12, a scanning mirror group 13, a beam splitter 14, and an eyepiece objective lens 15. DETAILED DESCRIPTION In order to make the purpose, technical solutions and advantages of the implementation methods of this application clearer, the technical solutions in the implementation methods of this application will be clearly and completely described below in combination with the drawings in the implementation methods of this application. Obviously, the described implementation methods are only part of the implementation methods of this application, not all of the implementation methods.

[0030] Embodiment 1: This embodiment proposes a retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera, the system comprising a multi-wavelength laser generating unit 1, a semi-transmissive and semi-reflective element 2, a galvanometer unit 3, a chromatic aberration lens 4, a sample to be measured 5, a focusing lens 6, and a detector with a pinhole 7; The laser emitted by the multi-wavelength laser generating unit 1 is reflected by the semi-transparent and semi-reflective element 2 and then passes through the galvanometer unit 3 and the chromatic aberration lens 4 in sequence to be emitted to the sample to be tested 5. After being reflected by the sample to be tested 5, it passes through the chromatic aberration lens 4, the galvanometer unit 3, the semi-transparent and semi-reflective element 2 and the focusing lens 6 in sequence, is received by the detector with a pinhole 7, and reaches the silicon PM. A multi-wavelength retinal confocal image is generated based on the signal collected by the silicon PM.

[0031] Implementation 2: This implementation is a further limitation of the retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera described in Implementation 1. The multi-wavelength laser generating unit 1 is a continuous wavelength laser generator or a discontinuous laser generator. Embodiment 3: This embodiment further limits the retinal blood vessel three-dimensional imaging system based on the ultra-wide-angle laser fundus camera described in Embodiment 1. The chromatic aberration of the chromatic aberration lens 4 is axial chromatic aberration.

[0032] Implementation method 4. This implementation method further limits the retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera described in implementation method 1. The multi-wavelength laser generating unit 1 includes a laser and a laser collimation unit. The laser is used to diverge the laser light source and convert the diverged laser beam into a parallel beam through the laser collimation unit.

[0033] Implementation 5: This implementation proposes a method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera. The method is implemented based on the system described in any one of Implementation 1, and the method includes the following steps: Step 1: registering the generated multi-wavelength retinal confocal images; Step 2: Segment the four-channel image after registration in step 1 to obtain a complete fundus image of the vascular choroid map. Step 3: Generate a three-dimensional vascular model based on the vascular choroid map using the vascular segmentation mask to achieve three-dimensional retinal vascularization. Image visualization.

[0034] Embodiment 6: This embodiment further limits the method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera described in Embodiment 5. The method for registering the generated multi-wavelength retinal confocal images in step 1 is: Construct Gaussian pyramid and DoG pyramid, detect DoG extreme points, find positioning key points, assign main directions based on the positioned key points, generate feature descriptors, and complete SIFT feature point extraction.

[0035] Implementation method seven. This implementation method further limits the retinal vascular three-dimensional imaging method based on an ultra-wide-angle laser fundus camera described in implementation method five. The method for segmenting the aligned four-channel image in step 2 is: using the Attention U-Net model to segment the aligned four-channel image.

[0036] Implementation 8. This implementation further limits the method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera described in Implementation 5. Step 2 also includes the steps of data enhancement and normalization processing of the blood vessel choroid map.

[0037] Embodiment 9. This embodiment further defines the method for three-dimensional retinal vascular imaging based on an ultra-wide-angle laser fundus camera described in embodiment 5. In step 3, a three-dimensional vascular model is generated based on the vascular choroid map using a vascular segmentation mask to achieve visualization of the three-dimensional retinal vascular imaging. The segmented vascular mask is first subjected to skeleton extraction to obtain a set of vascular centerlines, the local principal axis direction is estimated, and the skeleton point cloud is sparsely connected and optimized. Continuity is maintained based on the principal direction, and isolated noise points are removed. Intensity sampling is performed in the normal direction to obtain the radius of each skeleton point. 3D point coordinates are then constructed to achieve visualization of 3D imaging of retinal vessels.

[0038] Embodiment 10: This embodiment further defines the method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera described in Embodiment 5. The method for constructing 3D point coordinates is as follows: Vascular mask and confocal Z-axis slice information, each slice spacing , for each blood vessel pixel (x, y), the relative depth is calculated based on the signal intensity I(x, y):

[0039] Convert 2D pixels to 3D physical coordinates: .

[0040] Implementation 11: This implementation provides an example, which is used to explain the above implementations 1 to 8. Specifically, the example is as follows: See also Figures 1 to 7The present embodiment describes a workflow of a retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera as described in the present embodiment. The workflow includes the following steps: a 4-wavelength laser is coupled to the output end of an optical fiber, passes through a reflective collimator to reach a first reflector 11, and after reflection, passes through a beam splitter prism to reach a second reflector 12. The second reflector 12 reflects the incident collimated light beam to a two-dimensional scanning module composed of a low-speed galvanometer Y axis and a high-speed galvanometer X axis. The divergent light beam coming out of the scanning module then passes through a scanning mirror group 13, a beam splitter prism, and a second reflector 12. After passing through the optical mirror 14 and the eyepiece objective lens 15, the light is converged and incident on the fundus of the human eye. The retina reflects or scatters the laser light incident on the fundus back to the original light path, and passes through the eyepiece objective lens 15, the beam splitter 14, the scanning mirror group 13, and the reflector in sequence to reach the beam splitter prism. The beam splitter prism reflects the returning signal light beam to the dichroic mirror or the reflector. The dichroic mirror or the reflector reflects the signal light according to the corresponding wavelength to the collecting and focusing lens 6 and the pinhole 7-1, and then reaches the silicon PM. Finally, an image of the retina is generated based on the signal collected by the silicon PM.

[0041] In this embodiment, the laser emission array is composed of four groups of semiconductor lasers (488nm±10nm / 520nm±10nm / 660nm±10nm / 785nm±10nm) The scanning mirror assembly achieves 190° wide-angle scanning in the eye by combining a resonant galvanometer (scanning frequency ≥ 8kHz) and a polygonal mirror.

[0042] The confocal optical path adopts a pinhole array, wherein the diameter of the pinhole array is adjustable in the range of 30-100 μm. Among them, the laser wavelengths in Table 1 are described as follows:

[0043] like Figure 2 As shown in the figure, it is a schematic diagram of the optical path of the deep scanning laser confocal microscope of the present invention. The laser emitted by the multi-wavelength laser generating unit 1 is reflected by the semi-transparent and semi-reflective element 2, and then passes through the galvanometer unit 3 and the chromatic aberration lens 4 in sequence to be emitted to the sample to be tested 5. Then, after being reflected by the sample to be tested 5, it passes through the chromatic aberration lens 4, the galvanometer unit 3, the semi-transparent and semi-reflective element 2, and the focusing lens 6 in sequence, and is received by the detector 7 with a pinhole.

[0044] By switching the laser beams of different wavelengths through the multi-wavelength laser generating unit 1 and cooperating with the chromatic aberration lens 4, multiple images of different depths can be obtained, such as Figure 1 As shown: When the wavelength is 10-1, the chromatic aberration lens makes the imaging focal plane at 5-1-0. At this time, the galvanometer completes a scanning cycle and obtains the image at that depth.

[0045] When the wavelength is switched to 10-2, the chromatic aberration lens makes the imaging focal plane at 5-1-1. At this time, the galvanometer completes a scanning cycle and obtains the image at this depth.

[0046] Similarly, when the wavelength is switched to 10-n, the chromatic aberration lens makes the imaging focal plane at 5-1-n. At this time, the galvanometer completes a scanning cycle and obtains the image at that depth, so that n images at different depths can be obtained, thereby realizing depth scanning.

[0047] Figure 2 The two optical paths are actually realized by the same set of devices. Figure 2 The separation is just to compare the imaging effects of different wavelengths.

[0048] The multi-wavelength laser generating unit 1 of the present invention can be a continuous wavelength laser generator or a discontinuous wavelength laser generator.

[0049] If a continuous wavelength laser generator is used, the emitted wavelength is continuous. To ensure that images of a single depth can be stably acquired, the change in wavelength δλ1 within one scanning cycle Fs of the galvanometer will result in a change in depth δd1. δd1 must be less than the depth resolution D1 of the laser confocal microscope.

[0050] If a discontinuous laser generator is used, the emitted wavelength is discontinuous. To ensure that images of a single depth can be stably acquired, the wavelength 10-1 will not change within one scanning cycle Fs of the galvanometer. The only difference between a discontinuous laser and a continuous laser is the control. Discontinuous lasers are based on PWM control to quickly turn on and off the light source, such as Figure 3 As shown, the continuous laser is based on the continuous high and low levels to continuously control the laser to be always on or always off. Among them, the data described in this embodiment are:

[0051] The image processing flow chart of the method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera described in this embodiment is as follows: Figure 5 shown.

[0052] In this study, confocal images of the retina at four wavelengths (488nm, 520nm, 660nm, and 785nm) were acquired as a two-dimensional matrix of 1024 × 1024 pixels. The four wavelengths correspond to the structure of different retinal layers, providing complete image detail. The four wavelength images were registered using the 785nm image as a reference.

[0053] The present invention uses Attention U-Net to segment the registered four-channel image to obtain a complete fundus image of the vascular venous network, and generates a three-dimensional vascular structure based on the vascular segmentation mask to achieve three-dimensional visualization of the blood vessels.

[0054] The specific implementation process is as follows: 1 Image Registration SIFT is used for image registration. SIFT Scale-Invariant Feature Transform is a classic algorithm for image feature extraction. It has scale invariance and rotation invariance, and is also robust to changes in lighting and perspective.

[0055] The input image is constructed, and Gaussian and DoG pyramids are constructed. DoG extreme points are detected, key points are accurately located, fitted, and removed, and a main direction gradient histogram is assigned. A 128-dimensional feature descriptor is generated, and SIFT feature point extraction is completed. After SIFT feature matching is performed on the 488nm, 520nm, and 660nm wavelength images to the 785nm wavelength, the corresponding affine transformation is performed and fused to form a four-channel image.

[0056] 2. Use the Attention U-Net model to segment blood vessels Attention U-Net is a network that introduces the attention gate (AG) based on U-Net to emphasize target structures such as small blood vessels and suppress background interference.

[0057] The model consists of an encoder, a decoder, and an Attention Gate AG. The encoder includes four downsampling blocks, each containing two 3×3 Conv + ReLU + BatchNorm + MaxPool (2x2). The decoder includes three upsampling blocks, each including Up-conv + Attention Gate + Skip + Conv. The Attention Gate AG input is the skip connection feature and the current decoder feature. The final output uses 1×1 convolution to compress the number of channels to 1, and then uses Sigmoid activation.

[0058] The loss function is:

[0059] TverskyLossα=0.3,β=0.7.

[0060] The training set images underwent data augmentation, including random rotation, random flipping, brightness and contrast adjustment, and finally normalized and input into the model.

[0061] The evaluation indicator is Dice Coefficient:

[0062] in: TP: true positive, true positive, the prediction is positive and the actual result is also positive.

[0063] TN: true negative, the prediction is negative and the actual result is negative.

[0064] FP: false positive, the predicted result is positive but the actual result is negative.

[0065] FN: false negative, the prediction is negative, but the actual result is positive.

[0066] 3D vascular structure reconstruction The segmented vessel mask is first subjected to skeleton extraction to obtain a set of vessel centerlines. PCA is then used to estimate the local principal axis orientation. Sparse connectivity optimization is then performed on the skeleton point cloud, maintaining continuity based on the principal orientation and removing isolated noise points. Intensity sampling is performed in the normal direction to obtain the radius of each skeleton point, and then 3D point coordinates are constructed.

[0067] Vascular mask and confocal Z-axis slice information each layer spacing , for each blood vessel pixel (x, y), the relative depth is calculated based on the signal intensity I(x, y):

[0068] Convert 2D pixels to 3D physical coordinates:

[0069] Finally, the 3D vascular structure is rendered using VTK Visualization Toolkit. Figure 7 shown.

[0070] Those skilled in the art will understand that the above description is only a preferred embodiment of the present invention, and the features described in the various embodiments and / or claims of the present disclosure may be combined or coupled in various ways, even if such a combination or coupling is not explicitly described in the present disclosure. It is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art may still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

[0071] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, the present invention is intended to include such changes and modifications as fall within the scope of the claims and their equivalents.

Claims

1. A retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera, characterized in that: The system comprises a multi-wavelength laser generating unit (1), a semi-transmissive and semi-reflective element (2), a galvanometer unit (3), a chromatic aberration lens (4), a sample to be measured (5), a focusing lens (6), and a detector with a pinhole (7); The laser light emitted by the multi-wavelength laser generating unit (1) is reflected by the semi-transparent and semi-reflective element (2) and then sequentially passes through the galvanometer unit (3) and the chromatic aberration lens (4) to be emitted to the sample to be tested (5). The laser light is reflected by the sample to be tested (5) and then sequentially passes through the chromatic aberration lens (4), the galvanometer unit (3), the semi-transparent and semi-reflective element (2) and the focusing lens (6). The laser light is received by the detector with a pinhole (7) and reaches the silicon PM. A multi-wavelength retinal confocal image is generated based on the signal collected by the silicon PM.

2. The retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera according to claim 1, characterized in that: The multi-wavelength laser generating unit (1) is a continuous wavelength laser generator or a discontinuous laser generator.

3. The retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera according to claim 1, characterized in that: The chromatic aberration form of the chromatic aberration lens (4) is axial chromatic aberration.

4. The retinal vascular three-dimensional imaging system based on an ultra-wide-angle laser fundus camera according to claim 1, characterized in that: The multi-wavelength laser generating unit (1) comprises a laser and a laser collimating unit. The laser is used to diverge the laser light source, and the laser collimating unit converts the diverged laser beam into a parallel beam.

5. A method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera, characterized in that: The method is implemented based on the system according to any one of claims 1 to 4, and the method comprises the following steps: Step 1: registering the generated multi-wavelength retinal confocal images; Step 2: Segment the four-channel image after registration in step 1 to obtain a complete fundus image of the vascular choroid map. Step 3: Generate a three-dimensional vascular model based on the vascular choroid map using the vascular segmentation mask to achieve three-dimensional retinal vascularization. Image visualization.

6. The method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera according to claim 5, characterized in that: The method for registering the multi-wavelength retinal confocal images generated in step 1 is: Construct Gaussian pyramid and DoG pyramid, detect DoG extreme points, find positioning key points, assign main directions based on the positioned key points, generate feature descriptors, and complete SIFT feature point extraction.

7. The method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera according to claim 5, characterized in that: The method for segmenting the registered four-channel image in step 2 is: using the Attention U-Net model to segment the registered four-channel image.

8. The method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera according to claim 5, characterized in that: Step 2 also includes the steps of data enhancement and normalization processing on the vascular choroid map.

9. The method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera according to claim 5, characterized in that: In step 3, a 3D vascular model is generated based on the vascular choroid map using a vascular segmentation mask to achieve visualization of retinal vascular 3D imaging as follows: The segmented vascular mask is first subjected to skeleton extraction to obtain a set of vascular centerlines, the local principal axis direction is estimated, and the skeleton point cloud is sparsely connected and optimized. Continuity is maintained based on the principal direction, and isolated noise points are removed. Intensity sampling is performed in the normal direction to obtain the radius of each skeleton point. 3D point coordinates are then constructed to achieve visualization of 3D imaging of retinal vessels.

10. The method for three-dimensional imaging of retinal blood vessels based on an ultra-wide-angle laser fundus camera according to claim 8, characterized in that: The method to construct 3D point coordinates is: Vascular mask and confocal Z-axis slice information, each slice spacing , for each blood vessel pixel (x, y), the relative depth is calculated based on the signal intensity I(x, y): Convert 2D pixels to 3D physical coordinates: 。

Images