Method for expanding OCT (optical coherence tomography) scanning range based on assistance of mechanical arm

Abstract

The invention discloses a method for expanding an OCT (optical coherence tomography) scanning range based on assistance of a mechanical arm, and relates to the technical field of biomedical imaging. S200, carrying out OCT scanning and storage; s300, surface detection is carried out; s400, the pose of the mechanical arm is adjusted in a self-adaptive mode; s500, performing global scanning on the scanning area range; and S600, reconstructing a three-dimensional image. According to the invention, the whole process from sample surface detection, attitude adjustment to image reconstruction is realized, the OCT scanning range is expanded, and the application scene of OCT in the medical and industrial fields is expanded.

Description

Technical Field

[0001] The present invention relates to the technical field of biomedical imaging, and in particular to a method for expanding an OCT scanning range based on the assistance of a mechanical arm. Background Art

[0002] Optical coherence tomography (OCT) technology has become an important tool in the field of biomedical imaging since the 1990s. OCT technology uses the principle of low-coherence light interference to provide one-dimensional depth images, two-dimensional cross-sectional images or three-dimensional stereo images with micron-level resolution, and non-invasive, high-resolution imaging of biological tissues. OCT technology plays an important role in the diagnosis and treatment of ophthalmology, cardiovascular disease, respiratory system, digestive system, vaginal disease and other fields. In the industrial field, OCT is also used to check the internal quality of transparent or translucent materials, such as lenses, coatings and films. Although OCT technology has made significant progress in imaging depth and resolution, it still faces challenges in large-scale imaging applications. The imaging range of OCT is limited by the highest frequency that the system can detect, which is usually between a few millimeters and tens of millimeters. Time-domain OCT (TD-OCT) can theoretically achieve large-scale imaging, but it is limited by low sensitivity and acquisition speed. To solve this problem, researchers have proposed a variety of technical solutions, such as cyclic ranging OCT and long-distance OCT, but these methods require expensive new light sources and high-sampling rate digitizers.

[0003] In recent years, with the rapid development of robotics, while its cost has been reduced and its application scope has been expanded, it has also been introduced into OCT to expand the scale of high-resolution tomography. "Robotic-arm-assisted flexible largefield-of-view optical coherence tomography" (Biomed. Opt. Express 12, 4596–4609 (2021), Huang et al.) combines a depth camera for large-scale spatial mapping, an RGB camera for precise target positioning, and a robotic arm with an OCT probe. The system uses real-time 3D OCT imaging to adjust the position of the probe relative to the sample and optimize the imaging posture. Subsequently, "Automated robot-assisted wide-field optical coherencetomography using structured light camera" (Biomed. Opt. Express 14, 4310–4325 (2023), Li et al.) replaced the depth camera with a 3D structured light camera, which is able to generate an accurate 3D point cloud of the sample surface, thereby realizing automatic planning of the robotic arm scanning path. In addition, "Large area robotically assisted optical coherence tomography (LARA-OCT)" (Biomed. Opt. Express 15, 3993–4009 (2024), Lotz et al.) demonstrated a large-scale robotic-assisted OCT imaging using a 7-DOF robotic arm and a 3.3 MHz scanned source OCT. They directly used OCT to track the sample surface and perform image stitching, thereby omitting additional machine vision systems such as cameras, positioning, tracking, or navigation equipment. This method can minimize the hardware cost and complexity of the system by using OCT to track the sample surface, which is conducive to its application and popularization, but it still has limitations when encountering three-dimensional objects with highly irregular surfaces. "Large area kidney imaging for pre-transplant evaluation using real-time robotic optical coherence tomography, Ma et al." (Commun. Eng. 3, 122 (2024)) uses real-time robotic OCT to perform large-area kidney imaging for pre-transplant evaluation and pathological tissue analysis.However, the rule-based feature extraction of the above-mentioned prior art is easily affected when facing imaging artifacts, especially when dealing with objects with complex curvatures. It is necessary to further develop more robust surface detection technology.

[0004] Therefore, technicians in this field are committed to developing a method for expanding the OCT scanning range based on the assistance of a robotic arm. Summary of the invention

[0005] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is how to improve the robustness against OCT imaging artifacts and adaptively adjust the position of the robot arm to optimize imaging.

[0006] By analyzing OCT images through deep learning models, the robustness to OCT imaging artifacts is improved, and the imaging process of objects with irregular upper surfaces is optimized by adaptively adjusting the robotic arm posture. This application reduces the hardware complexity and cost of the robotic arm OCT system by combining deep learning-assisted OCT technology and adaptive adjustment of the robotic arm posture, while improving the robustness and efficiency of imaging large and complex objects, opening up new possibilities for the application of OCT technology in the medical and industrial fields. In this application, the scanning point refers to the point where the robotic arm scans, that is, the position where the robotic arm moves; the center point is the center of the edge of the upper surface of the sample in the OCT image.

[0007] In one embodiment of the present invention, a method for expanding the OCT scanning range based on the assistance of a robotic arm is provided, comprising: S100: Scanning area definition, determining the scanning area range of the upper surface of the sample, initially positioning the robotic arm, and moving the robotic arm to the starting point of the scanning area range as the current scanning point; S200: OCT scanning and storage, using optical coherence tomography (OCT) technology to scan the current scanning point and store the OCT images collected during the scanning process; S300: Surface detection, using deep learning models to analyze OCT images, identify and segment the edges of the sample surface, obtain segmented images, and extract and save key features; S400: adaptively adjust the robot arm posture, calculate the robot arm posture of the next scanning point according to the key features, and then adaptively adjust the robot arm to move to the next scanning point, and use the next scanning point as the current scanning point; S500: Scan the entire scanning area, repeating steps S200 to S500 until all scanning points in the scanning area are scanned; S600: Three-dimensional image reconstruction: splicing the OCT images of all scanning points according to their respective positions and angles to reconstruct a three-dimensional image.

[0008] Optionally, in the method for expanding the OCT scanning range based on the assistance of a robot arm in the above embodiment, step S100 includes: S110, determining the scanning area range; S120, initially positioning the robotic arm, and moving the robotic arm to the starting point of the scanning area; S130, setting the current scanning point, taking the starting point of the scanning area as the current scanning point.

[0009] Optionally, in the method for expanding the OCT scanning range based on robot arm assistance in any of the above embodiments, the scanning area is set as a rectangular area.

[0010] Optionally, in the method for expanding the OCT scanning range based on robot assistance in any of the above embodiments, step S200 includes: S210, obtaining the coordinates of the current scanning point, the coordinates of the current scanning point are obtained as (X0, Y0, Z0); S220, optical coherence tomography, scanning the current scanning point using optical coherence tomography (OCT) technology; S230 , data storage, storing the OCT images collected during the scanning process for processing and analysis.

[0011] Optionally, in the method for expanding the OCT scanning range based on robot arm assistance in any of the above embodiments, the deep learning model in step S300 uses a U-Net model.

[0012] Furthermore, in the method for expanding the OCT scanning range based on robot arm assistance in the above embodiment, the U-Net model is based on a symmetrical encoder-decoder structure, including an encoder and a decoder.

[0013] Furthermore, in the method for expanding the OCT scanning range based on robot-assisted operation in the above embodiment, the encoder includes 8 convolution blocks with the same structure.

[0014] Furthermore, in the method for expanding the OCT scanning range based on robot arm assistance in the above embodiment, the convolution block in the encoder includes two 3x3 convolution layers, followed by a ReLU activation function, and a 2x2 maximum pooling operation for downsampling.

[0015] Furthermore, in the method for expanding the OCT scanning range based on robot-assisted operation in the above embodiment, the number and structure of the convolution blocks in the decoder are the same as those of the convolution blocks in the encoder.

[0016] Furthermore, in the method for expanding the OCT scanning range based on robot arm assistance in the above embodiment, the decoder receives the feature map from the encoder, and the feature map is directly connected to the decoder through a jump connection to restore the segmented image.

[0017] Furthermore, in the method for expanding the OCT scanning range based on robot assistance in the above embodiment, the last layer of the decoder is a 1x1 convolution layer.

[0018] Furthermore, in the method for expanding the OCT scanning range based on robot assistance in the above embodiment, the U-Net model is trained using 100 OCT B-scan images with artifacts, optimized using a cross entropy loss function, and using an Adam optimizer.

[0019] Furthermore, in the method for expanding the OCT scanning range based on robot assistance in the above embodiment, the beta parameter of the Adam optimizer is set to (0.9, 0.999).

[0020] Furthermore, in the method for expanding the OCT scanning range based on robot arm assistance in the above embodiment, the initial learning rate of the U-Net model is 0.00001.

[0021] Furthermore, in the method for expanding the OCT scanning range based on robot-assisted operation in the above embodiment, the total loss function of the U-Net model converges after about 30 cycles, and the training is stopped at the 40th cycle.

[0022] Optionally, in the method for expanding the OCT scanning range based on robot assistance in any of the above embodiments, step S300 includes: S310, model input, inputting the OCT image acquired by the robot arm at the current scanning point into the deep learning model; S320, edge segmentation, using a deep learning model to analyze the OCT image input, identify and segment the upper surface edge of the sample, and obtain a segmented image; S330, feature extraction, extracting key features from the segmented image and saving them for adaptively adjusting the robot arm posture.

[0023] Furthermore, in the method for expanding the OCT scanning range based on robot-assisted operation in the above embodiment, the key feature includes upper surface height information.

[0024] Optionally, in the method for expanding the OCT scanning range based on robot assistance in any of the above embodiments, step S400 includes: S410, obtaining the current scanning point information, obtaining the edge features of the upper surface of the sample in the OCT image of the current scanning point and the coordinates of the center point of the current scanning point (m0, n0), wherein the center point of the current scanning point is the center of the edge of the upper surface of the sample in the OCT image of the current scanning point; S420, determine the next scanning point, start searching backward from the center point of the current scanning point, until a scanning point with a height difference equal to the set height threshold Hthreshold is found, and set it as the next scanning point; if a point equal to the set height threshold Hthreshold is not found during the search process, it indicates that the upper surface of the sample is relatively flat, and the current scanning point is horizontally moved by the set step size step_size, and the scanning point at this time is set as the next scanning point; S430, calculating the coordinates of the next scanning point, converting the coordinate difference between the center point coordinates (m0, n0) of the current scanning point and the center point coordinates (m1, n1) of the next scanning point into an actual physical distance, and calculating the coordinates (X1, Y1, Z1) of the next scanning point; S440, calculate the inclination angle α of the robot arm. When there are grooves or protrusions on the upper surface of the sample, calculate the inclination angle α of the upper surface of the sample, and obtain the robot arm posture (X1, Y1, Z1, RX0 + α, RY0, RZ0) of the next scanning point, where RX0, RY0, and RZ0 represent the rotation angles of the current scanning point in the X, Y, and Z axis directions, respectively. When the current scanning point is the starting point, RX0, RY0, and RZ0 are 0°, 0°, and 180°, respectively; S450, dynamically adjust the robot arm posture, move the robot arm to the next scanning point, and use the next scanning point as the current scanning point.

[0025] Furthermore, in the method for expanding the OCT scanning range based on the assistance of a robot arm in the above embodiment, the height threshold Hthreshold is set to be one quarter of the height of the OCT image.

[0026] Furthermore, in the method for expanding the OCT scanning range based on the assistance of a robot arm in the above embodiment, the step size step_size is set to 2 mm.

[0027] Furthermore, in the method for expanding the OCT scanning range based on robot-assisted operation in the above-mentioned embodiment, the robot arm tilt angle α is calculated in step S440, and a straight line is fitted using the leftmost and rightmost points on the upper surface edge of the segmented image. The inclination angle of the straight line relative to the horizontal line in the segmented image is fed back to the robot arm to adjust the posture, that is, the robot arm tilt angle α is adjusted based on the RX0 parameter.

[0028] Furthermore, in the method for expanding the OCT scanning range based on the assistance of a robotic arm in the above embodiment, the dynamic adjustment of the posture of the robotic arm in step S450 includes adjusting the position (X, Y, Z) and rotation angle (RX0+α, RY0, RZ0) of the robotic arm to adapt to the irregularities of the upper surface of the sample.

[0029] This application uses a U-Net model to process OCT images to achieve accurate segmentation of the upper surface of the sample, and can resist artifacts that may occur during OCT imaging, such as side lobes and mirror artifacts. The U-Net model trained by a large number of OCT B-scan images with artifacts can accurately identify the true upper surface of the sample in the presence of artifacts, providing a reliable basis for the precise positioning of the robot arm. This application adaptively adjusts the posture of the robot arm, and can dynamically adjust the posture of the robot arm according to the real-time detection results of the upper surface of the sample, including not only the posture of the robot arm, but also its shooting angle, to ensure that the OCT probe always maintains the best imaging distance with the upper surface of the sample, significantly improving the imaging coverage of complex curved surfaces, and avoiding imaging problems caused by irregularities of the upper surface. This application realizes the entire process from sample surface detection, posture adjustment to image reconstruction, expands the OCT scanning range, and expands the application scenarios of OCT in the medical and industrial fields.

[0030] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a flow chart illustrating a method for expanding an OCT scanning range based on robot-assisted operation according to an exemplary embodiment; Figure 2 is a flowchart illustrating adaptively adjusting the posture of a robot arm in a method for expanding an OCT scanning range based on robot arm assistance according to an exemplary embodiment; Figure 3 is a diagram illustrating a comparison of the effects of surface detection in a method for expanding an OCT scanning range based on robot assistance according to an exemplary embodiment; Figure 4 is a comparison diagram illustrating the effect of adaptively adjusting the position and posture of a robot arm in a method for expanding an OCT scanning range based on robot arm assistance according to an exemplary embodiment; Figure 5 3D reconstruction is a comparative diagram illustrating the effect of three-dimensional reconstruction in a method for expanding an OCT scanning range based on robot assistance according to an exemplary embodiment. DETAILED DESCRIPTION

[0032] The following describes several preferred embodiments of the present invention with reference to the drawings in the specification, so that the technical content is clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned in the text.

[0033] In the drawings, components with the same structure are indicated by the same numerical reference numerals, and components with similar structures or functions are indicated by similar numerical reference numerals. The size and thickness of each component shown in the drawings are arbitrarily shown, and the present invention does not limit the size and thickness of each component. In order to make the illustration clearer, the thickness of the components is schematically exaggerated in some places in the drawings.

[0034] This application analyzes OCT images through a deep learning model to improve the robustness against OCT imaging artifacts, and at the same time optimizes the imaging process of objects with irregular upper surfaces by adaptively adjusting the position of the robotic arm. This application reduces the hardware complexity and cost of the robotic arm OCT system by combining deep learning-assisted OCT technology and adaptive adjustment of the robotic arm position, while improving the robustness and efficiency of imaging large and complex objects, opening up new possibilities for the application of OCT technology in the medical and industrial fields. In this application, the scanning point refers to the point where the robotic arm scans, that is, the position where the robotic arm moves; the center point is the center of the edge of the upper surface of the sample in the OCT image.

[0035] This application designs a method to expand the OCT scanning range based on the assistance of a robotic arm, such as Figure 1 As shown, the following steps are included: S100: Scanning area definition, determining the scanning area range of the upper surface of the sample, initially positioning the robotic arm, and moving the robotic arm to the starting point of the scanning area range as the current scanning point; specifically including: S110, determining the scanning area range, and setting the scanning area range to a rectangular area; S120, initially positioning the robotic arm, and moving the robotic arm to the starting point of the scanning area; S130, setting the current scanning point, taking the starting point of the scanning area as the current scanning point.

[0036] S200: OCT scanning and storage, using optical coherence tomography (OCT) technology to scan the current scanning point and store the OCT images collected during the scanning process; specifically including: S210, obtaining the coordinates of the current scanning point, the coordinates of the current scanning point are obtained as (X0, Y0, Z0); S220, optical coherence tomography, scanning the current scanning point using optical coherence tomography (OCT) technology; S230 , data storage, storing the OCT images collected during the scanning process for processing and analysis.

[0037] S300: Surface detection, using the U-Net model to analyze OCT images, identify and segment the edges of the sample surface, obtain the segmented image, extract and save key features; the U-Net model is based on a symmetrical encoder-decoder structure, including an encoder and a decoder. The encoder includes 8 convolution blocks of the same structure. The convolution block in the encoder contains two 3x3 convolution layers, followed by a ReLU activation function, and a 2x2 maximum pooling operation for downsampling; the number and structure of the convolution blocks in the decoder are the same as those of the encoder. The decoder ends with a 1x1 convolution layer. The decoder receives the feature map from the encoder, which is directly connected to the decoder through a jump connection to restore the segmented image; the U-Net model is trained using 100 OCT B-scan images with artifacts, optimized using the cross entropy loss function, and using the Adam optimizer. The beta parameter of the Adam optimizer is set to (0.9, 0.999), the initial learning rate of the U-Net model is 0.00001, the total loss function of the U-Net model converges after about 30 cycles, and the training stops at the 40th cycle; specifically: S310, model input, inputting the OCT image acquired by the robot arm at the current scanning point into the deep learning model; S320, edge segmentation, using a deep learning model to analyze the OCT image input, identify and segment the upper surface edge of the sample, and obtain a segmented image; S330, feature extraction, extracting key features from the segmented image and saving them for adaptively adjusting the robot arm posture.

[0038] S400: Adaptively adjust the robot arm posture, calculate the robot arm posture of the next scanning point according to the key features, and then perform adaptive adjustment to move the robot arm to the next scanning point and use the next scanning point as the current scanning point, wherein the key features include the upper surface height information; Figure 4 As shown, specifically including: S410, obtaining the current scanning point information, obtaining the edge features of the upper surface of the sample in the OCT image of the current scanning point and the coordinates of the center point of the current scanning point (m0, n0), wherein the center point of the current scanning point is the center of the edge of the upper surface of the sample in the OCT image of the current scanning point; S420, determine the next scanning point, start searching backward from the center point of the current scanning point, until a scanning point with a height difference equal to the set height threshold Hthreshold is found, and set it as the next scanning point; if a point equal to the set height threshold Hthreshold is not found during the search process, it indicates that the upper surface of the sample is relatively flat, and the current scanning point is horizontally moved by the set step size step_size, and the scanning point at this time is set as the next scanning point; wherein the set height threshold Hthreshold is one quarter of the height of the OCT image, and the set step size step_size is 2mm; S430, calculating the coordinates of the next scanning point, converting the coordinate difference between the center point coordinates (m0, n0) of the current scanning point and the center point coordinates (m1, n1) of the next scanning point into an actual physical distance, and calculating the coordinates (X1, Y1, Z1) of the next scanning point; S440, calculate the inclination angle α of the robot arm. When there are grooves or protrusions on the upper surface of the sample, calculate the inclination angle α of the upper surface of the sample, and obtain the robot arm posture (X1, Y1, Z1, RX0 + α, RY0, RZ0) of the next scanning point, where RX0, RY0, and RZ0 represent the rotation angles of the current scanning point in the X, Y, and Z axis directions, respectively. When the current scanning point is the starting point, RX0, RY0, and RZ0 are 0°, 0°, and 180°, respectively. The leftmost and rightmost points on the edge of the upper surface of the segmented image are used to fit a straight line. The inclination angle of the straight line relative to the horizontal line in the segmented image is fed back to the robot arm to adjust the posture, that is, the robot arm inclination angle α is adjusted based on the RX0 parameter; S450, dynamically adjust the robot arm posture, move the robot arm to the next scanning point, adjust the position (X, Y, Z) and rotation angle (RX0+α, RY0, RZ0) of the robot arm to adapt to the irregularity of the sample surface, and use the next scanning point as the current scanning point.

[0039] S500: Scan the entire scanning area, repeating steps S200 to S500 until all scanning points in the scanning area are scanned; S600: Three-dimensional image reconstruction: splicing the OCT images of all scanning points according to their respective positions and angles to reconstruct a three-dimensional image.

[0040] This application conducted experiments to verify the technical effect. The experiment used a spectral domain (SD) OCT interferometer and a flying spot scanning sample arm. The OCT light source was a superluminescent diode (SLD) with a central wavelength of 840nm and a spectral bandwidth of 50nm, providing an axial resolution of approximately 5 microns (µm). The sample arm serves as a global scanner, including a single-axis galvanometer (S8107, SunnyTechnology, China) as a local scanner and a 6-degree-of-freedom (6-DOF) robotic arm (CR3, Dobot Robots, China). The maximum load of the robotic arm is 3 kg, the maximum speed is 2m / s, and the repeatability is ±20µm. First, the Denavit-Hartenberg (DH) parameter method is used to calibrate the coordinate system of the robotic arm. The transformation matrix is ​​used to describe the transformation from one coordinate system to another. For two continuous coordinate systems and , the transformation matrix It can be expressed as: in, represents the rotation matrix, represents the translation matrix, Represents the distance between the Z axes of two consecutive coordinate systems, measured along the X axis, Represents the angle between the X-axes of two consecutive coordinate systems, from the Z-axis to the X-axis, Represents the distance between the X-axes of two consecutive coordinate systems, measured along the Z-axis, It represents the angle between the Z axes of two continuous coordinate systems, from the X axis to the Z axis. Finally, the complete transformation matrix from the base coordinate system to the coordinate system of any link is obtained, which is recursively performed for each link until the end effector of the robot is reached.

[0041] The robot performs serpentine scanning, and the galvanometer performs local scanning to collect OCT B-scan data. A deep learning model (U-Net) is used for surface detection to adjust the posture of the robot. The robot posture is adaptively adjusted to dynamically adjust the scanning path of the robot according to the surface features of the sample. This application uses a variety of samples for experiments, including human brain models, in vitro pig kidneys, and 3D printed models, to demonstrate the effectiveness and applicability of the method.

[0042] Compare surface detection, such as Figure 3As shown, the scale of all images is 1mm. The green lines represent the results obtained by the traditional edge detection method, and the red lines represent the results detected using the U-Net model. It can be seen that the U-Net model shows consistent performance in all test images, maintaining a high level of accuracy even in the presence of severe artifacts. This consistency is attributed to the model's ability to understand the content of the OCT image, thereby distinguishing between true upper surface features and artifacts. In contrast, the edge detection method used in the method of Huber et al. relies on gradient calculation, and its performance degrades when the artifact intensity increases. In addition, the present application utilizes the fast reasoning capabilities of the end-to-end deep learning network to achieve a detection time of approximately 8 milliseconds per OCT scan. This is a significant improvement over the edge detection method, which takes 40 milliseconds to process the same scan.

[0043] Compare the adaptive adjustment of the robot arm posture, such as Figure 4 As shown, the scale of the three-dimensional object is 5mm, the scale of the OCT image is 1mm, and (a) is the result obtained using the method described in Huber. This method sets the height of the rightmost edge point as the reference for the next pre-scan. The process involves linear fitting to calculate the axial displacement and tilt angle of the robot arm, and then feedback to adjust its posture and height. However, its application is limited to flat upper surfaces, because uneven terrain may cause the sample to fall out of the OCT imaging range during pre-scanning, and may also cause controller parameter deviation and linear fitting errors. As shown in Figure (a), the solid line in the trajectory represents the pre-scan, while the dotted line represents the adjustment after linear fitting. The image connected by the three purple arrows in (a) shows that the pre-scan beyond the imaging range appears as a reflection, and the linear fitting fails to solve this problem, resulting in distortion of subsequent images. The image connected by the three blue arrows in (a) shows that rapid slope changes can cause gaps in the OCT image and inaccurate subsequent scans. These problems were not detected, resulting in continuous distortion of the image. In contrast, Figure (b) shows the result of using the present application to adaptively adjust the robot arm posture, selecting subsequent scanning points that all meet the height threshold to ensure normal OCT imaging. The present application further uses deep learning for edge detection, calculates the axial displacement of the robot arm by the vertical coordinate difference between the center point and the next scanning point, and estimates the tilt angle based on the edge slope rather than linear fitting. The present application is capable of orthogonal posture adjustment to achieve precise robot arm positioning and ensure consistency between the probe and the sample center point.

[0044] Imaging experiments using human brain models and isolated porcine kidneys, such as Figure 5 As shown, the imaging results of the human brain model and the isolated pig kidney show the structural and tomographic information, where the image scales of the human brain model and the isolated pig kidney are 50 mm and 20 mm, respectively. Figure 5(a)-(c) are the scan results of the human brain model. The scale of (a) and (b) is 20 mm, and the scale of (c) is 5 mm. As shown in (a), the volume of the model is 102×91×26 mm 3 (c) is the OCT image at the dotted line position in (b). The human brain has a large curvature and a complex groove structure. Therefore, in the upper surface structure with large curvature changes and crisscross grooves, the present application can ensure that the sample is always within the OCT imaging range, significantly improving the imaging coverage of complex curved surfaces and avoiding imaging problems caused by upper surface irregularities. Figure 5 (d) and (e) are the scan results of the in vitro pig kidney. The scale bar of (d) is 10 mm, the scale bar of (e) is 5 mm, and the scale bar of the enlarged OCT image in (e) is 1 mm. As shown in (d), its volume is 80×40×10 mm 3 . (e) is a cross-sectional OCT image at the dotted line position in (d). It can be seen that the present application is able to capture the structural and tomographic information of such biological tissue under the influence of OCT artifacts caused by interface reflection. In summary, the present application can accurately identify the true upper surface of the sample in the presence of artifacts, providing a reliable basis for the precise positioning of the robotic arm.

[0045] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for expanding the OCT scanning range based on the assistance of a robotic arm, characterized in that: include: S100: Scanning area definition, determining the scanning area range of the upper surface of the sample, initially positioning the robotic arm, and moving the robotic arm to the starting point of the scanning area range as the current scanning point; S200: OCT scanning and storage, scanning the current scanning point using optical coherence tomography (OCT) technology, and storing OCT images collected during the scanning process; S300: Surface detection, using a deep learning model to analyze the OCT image, identify and segment the edge of the upper surface of the sample, obtain a segmented image, and extract and save key features; S400: adaptively adjusting the robot arm posture, calculating the robot arm posture of the next scanning point according to the key feature, and then adaptively adjusting the robot arm to move the robot arm to the next scanning point, and taking the next scanning point as the current scanning point; S500: Scanning the entire scanning area, repeating steps S200 to S500 until all scanning points in the scanning area are scanned; S600: Three-dimensional image reconstruction, splicing the OCT images of all the scanning points according to their respective positions and angles to reconstruct a three-dimensional image.

2. The method for expanding the OCT scanning range based on the assistance of a robotic arm as claimed in claim 1, characterized in that: The step S100 includes: S110, determining the scanning area range; S120, initially positioning the robotic arm, and moving the robotic arm to the starting point of the scanning area; S130: Set the current scanning point, and take the starting point of the scanning area as the current scanning point.

3. The method for expanding the OCT scanning range based on the assistance of a robot arm as claimed in claim 1, characterized in that: The step S200 includes: S210, obtaining the coordinates of the current scanning point, wherein the coordinates of the current scanning point are obtained as (X0, Y0, Z0); S220, optical coherence tomography, scanning the current scanning point using optical coherence tomography (OCT) technology; S230 , data storage: storing the OCT images collected during the scanning process for processing and analysis.

4. The method for expanding the OCT scanning range based on the assistance of a robotic arm as claimed in claim 1, characterized in that: The deep learning model in step S300 uses a U-Net model.

5. The method for expanding the OCT scanning range based on the assistance of a robot arm as claimed in claim 4, characterized in that: The U-Net model is based on a symmetrical encoder-decoder structure, including an encoder and a decoder, wherein the encoder includes 8 convolution blocks with the same structure.

6. The method for expanding the OCT scanning range based on the assistance of a robot arm as claimed in claim 5, characterized in that: The convolutional block in the encoder contains two 3x3 convolutional layers followed by a ReLU activation function and a 2x2 max pooling operation for downsampling.

7. The method for expanding the OCT scanning range based on the assistance of a robot arm as claimed in claim 6, characterized in that: The number and structure of the convolution blocks in the decoder are the same as those of the encoder.

8. The method for expanding the OCT scanning range based on the assistance of a robot arm as claimed in claim 7, characterized in that: The step S300 includes: S310, model input, inputting the OCT image acquired by the robot arm at the current scanning point into the deep learning model; S320, edge segmentation, using the deep learning model to analyze the OCT image input, identify and segment the upper surface edge of the sample, and obtain a segmented image; S330, feature extraction, extracting key features from the segmented image and saving them for adaptively adjusting the robot arm posture.

9. The method for expanding the OCT scanning range based on the assistance of a robot arm as claimed in claim 8, characterized in that: The step S400 includes: S410, obtaining information of the current scanning point, obtaining edge features of the upper surface of the sample in the OCT image of the current scanning point and the coordinates (m0, n0) of the center point of the current scanning point; S420, determining the next scanning point, searching backward from the center point of the current scanning point until a scanning point with a height difference equal to the set height threshold Hthreshold is found, and setting it as the next scanning point; if a point equal to the set height threshold Hthreshold is not found during the search process, it indicates that the upper surface of the sample is relatively flat, and the current scanning point is horizontally moved by the set step size step_size, and the scanning point at this time is set as the next scanning point; S430, calculating the coordinates of the next scanning point, converting the coordinate difference between the center point coordinates (m0, n0) of the current scanning point and the center point coordinates (m1, n1) of the next scanning point into an actual physical distance, and calculating the coordinates (X1, Y1, Z1) of the next scanning point; S440, calculating the inclination angle α of the robot arm. When there are grooves or protrusions on the upper surface of the sample, the inclination angle α of the upper surface of the sample is calculated to obtain the robot arm posture (X1, Y1, Z1, RX0+α, RY0, RZ0) of the next scanning point. When the current scanning point is the starting point, RX0, RY0, and RZ0 are 0°, 0°, and 180° respectively; S450, dynamically adjust the posture of the robotic arm, move the robotic arm to the next scanning point, and use the next scanning point as the current scanning point.

10. The method for expanding the OCT scanning range based on the assistance of a robot arm as claimed in claim 8, characterized in that: The set height threshold Hthreshold is one quarter of the height of the OCT image, and the set step size step_size is 2 mm.

Images