An ultrasonic navigation system for carotid plaque removal surgery

CN122894049APending Publication Date: 2026-10-13THE SECOND AFFILIATED HOSPITAL OF CHONGQING MEDICAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510399518.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-04-01
Publication Date
2026-10-13

AI Technical Summary

Technical Problem

这不仅限制了超声导航系统在基层医院的推广应用,也使得培养能够熟练运用超声导航进行颈动脉斑块去除手术的医生周期较长、难度较大

Benefits of technology

[0020]本发明一种用于颈动脉斑块去除手术的超声导航系统,具有较高的超声图像解读便利性、精准性和稳定性,能够有效减轻医生操作负担,提高超声导航效果。重点在于:

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122894049A_ABST
    Figure CN122894049A_ABST
Patent Text Reader

Abstract

The application relates to the technical field of medical operation navigation, and discloses an ultrasonic navigation system for carotid plaque removal operation, which comprises an operation subsystem; the operation subsystem comprises a mechanical arm, a medical ultrasonic imaging device, an optical tracking device, a positioning target, an RGBD depth camera, an ultrasonic cutter head and an upper computer; the mechanical arm comprises a No. 1 arm and a No. 2 arm; the No. 1 arm is used for clamping and controlling an ultrasonic probe to realize scanning and imaging; the No. 2 arm is used for clamping and controlling the ultrasonic cutter head to remove plaques; the RGBD depth camera is used for collecting depth images; the medical ultrasonic imaging device is used for scanning and acquiring original images of carotid plaques and the ultrasonic cutter head; the optical tracking device is used for cooperating with the positioning target to acquire spatial position coordinates of the RGBD depth camera and the ultrasonic probe; and the upper computer is used for coordinately controlling the operation subsystem. The application has high ultrasonic image interpretation convenience, precision and stability, can effectively reduce the operation burden of doctors, and improves the ultrasonic navigation effect.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of medical surgical navigation technology, specifically to an ultrasound navigation system for carotid artery plaque removal surgery. Background Technology

[0002] With improved living standards, people's diets have become increasingly high in sugar, salt, and fat, coupled with insufficient daily physical exercise, leading to a year-on-year rise in the incidence of cardiovascular diseases. Thanks to the development and advancement of endovascular instruments and techniques, interventional therapy has become a common treatment for cardiovascular diseases. Interventional therapy is usually performed through small incisions or puncture points in the skin, avoiding large-scale open-chest or open-abdomen surgery, effectively reducing the risk of infection for patients. In addition, interventional therapy has advantages such as rapid postoperative recovery, low risk, and wide applicability. During interventional therapy, doctors typically use real-time medical imaging to observe the location of the lesion and endovascular instruments to ensure the accuracy and safety of the procedure.

[0003] Compared to traditional medical imaging systems such as computed tomography (CT) and digital subtraction angiography (DSA), medical ultrasound imaging systems offer advantages such as no radiation and lower real-time imaging latency, making them more suitable for interventional guidance. Using ultrasound navigation in interventional procedures can avoid the radiation exposure to both doctors and patients. However, despite its many potential advantages, ultrasound navigation faces several challenges in its practical application to carotid artery plaque removal surgery. Firstly, to observe the location of lesions and intracavitary instruments during interventional procedures, doctors need to continuously acquire data by holding the ultrasound probe, which places high demands on their physical strength. Prolonged operation can easily lead to hand fatigue, affecting the stability and accuracy of image acquisition. Secondly, the interpretation of ultrasound images is far more difficult than that of DSA and other imaging modalities. Ultrasound images are affected by various factors, exhibit numerous imaging artifacts, and have complex image characteristics, requiring doctors to possess deep professional knowledge and rich clinical experience for accurate interpretation. This not only limits the widespread application of ultrasound navigation systems in primary hospitals but also makes training doctors proficient in using ultrasound navigation for carotid artery plaque removal surgery a lengthy and challenging process. Therefore, only a few hospitals are able to use ultrasound imaging in interventional procedures; and in the few hospitals that do use ultrasound, ultrasound navigation is mainly in an auxiliary role, and a complete, efficient and widely used surgical navigation system has not yet been formed.

[0004] Considering the above problems, it is urgent to design an ultrasonic navigation system to solve them. Summary of the Invention

[0005] The present invention aims to provide an ultrasound navigation system for carotid artery plaque removal surgery, which has high convenience, accuracy and stability in ultrasound image interpretation, can effectively reduce the workload of doctors and improve the ultrasound navigation effect.

[0006] The basic solution provided by this invention is: an ultrasound navigation system for carotid artery plaque removal surgery, including an operation subsystem; the operation subsystem includes a robotic arm, a medical ultrasound imaging device, an optical tracking device, a positioning target, an RGBD depth camera, and an ultrasonic scalpel head; it also includes a host computer;

[0007] The robotic arm includes a first arm and a second arm; the first arm is used to grip and control the ultrasound probe for scanning and imaging; the second arm is used to grip and control the ultrasound scalpel head for removing plaque; the RGBD depth camera is used to acquire depth images; the medical ultrasound imaging device is used to scan and acquire the original images of the carotid artery plaque and the ultrasound scalpel head; the optical tracking device is used in conjunction with the positioning target to acquire the spatial coordinates of the RGBD depth camera and the ultrasound probe.

[0008] The host computer is used to coordinate and control the operation subsystem. In the coordinated control, based on the depth image, the scanning area of ​​the ultrasound probe is identified, and the first arm is guided to control the ultrasound probe to contact the skin of the target area and start imaging. Then, the second arm is guided to locate and remove the plaque.

[0009] Furthermore, the host computer is equipped with a region information acquisition module; the region information acquisition module is used to identify target regions based on depth images and through a deep learning network.

[0010] Furthermore, the deep learning network includes a YOLOv5 model, a DeepSORT model, an AlphaPose network, and a ROI localization algorithm. The YOLOv5 model identifies the depth image and combines it with the DeepSORT model to obtain the human body region bounding box. Then, the AlphaPose network extracts the human skeleton based on the human body region bounding box to obtain human body landmarks, so as to roughly locate the scanning area of ​​the ultrasound probe, i.e., the target area. Finally, the center point of the target area is confirmed based on the ROI localization algorithm.

[0011] Furthermore, the host computer is equipped with a patch information acquisition module; the patch information acquisition module is used to identify and segment patches based on imaging information.

[0012] Furthermore, the patch information acquisition module includes an image processing network; the image processing network includes a Unet model and a BiConvLSTM module; and the encoder and decoder of the Unet model are integrated with an ASPP module.

[0013] Furthermore, the imaging information includes ultrasound video corresponding to the target area; the image processing network, when identifying and segmenting patches, includes the following steps:

[0014] Ultrasound video is input as video frames into the encoder of the Unet network. Combined with full-size skip connections, high-level semantic information is extracted, and features useful for segmentation are obtained. This high-level semantic information is then input into the ASPP module to extract multi-scale features. The high-level semantic information containing multi-scale features is then input into the decoder and merged with the corresponding low-level semantic information from the encoder, outputting a feature map of the same size as the original image. This feature map is then input into the BiConvLSTM module, which combines feature information from previous and subsequent frames to optimize the segmentation result of the current frame, outputting the final segmentation result, namely, patchy and non-patchy regions.

[0015] Furthermore, when guiding the second arm to locate the plaque, the spatial coordinates of the ultrasonic probe are first obtained by positioning and tracking the ultrasonic probe. Then, based on the pose matrix of the ultrasonic probe, the ultrasonic scalpel head, and the plaque region, the spatial coordinates of the plaque region are solved to guide the second arm to locate the plaque.

[0016] Furthermore, the image processing network is pre-trained using a self-made dataset; the self-made dataset contains pre-processed ultrasound images of the carotid artery and its plaque, and the ultrasound images are labeled; the ultrasound images and label files are saved in JPG and JSON file formats respectively, each JSON file represents a slice, and the carotid artery and its plaque are stored in a masked manner; the preprocessing includes image cropping, median filtering, and grayscale normalization.

[0017] Furthermore, the host computer is equipped with a three-dimensional reconstruction module; the three-dimensional reconstruction module is used to construct a three-dimensional reconstruction model of the carotid artery and its plaque and to visualize it.

[0018] Furthermore, when constructing the 3D reconstruction model, the 3D reconstruction module uses the Marching Cubes algorithm to perform 3D reconstruction based on the segmentation results of the patch information acquisition module.

[0019] The working principle and advantages of this invention are as follows:

[0020] This invention discloses an ultrasound navigation system for carotid artery plaque removal surgery, which offers high convenience, accuracy, and stability in ultrasound image interpretation, effectively reducing the workload of physicians and improving ultrasound navigation results. The key points are:

[0021] First, this solution enables deep collaboration between multimodal perception and intelligent decision-making, achieving high accuracy and stability in ultrasound image interpretation, and constructing a precise and dynamic closed loop for automated surgical operations, effectively reducing the workload of doctors.

[0022] Specifically, this system integrates three major sensing modules: an optical tracking device, an RGBD depth camera, and a medical ultrasound imaging device, forming a comprehensive data acquisition system covering preoperative localization, intraoperative navigation, and postoperative verification. Firstly, the optical tracking device acquires the spatial coordinates of the ultrasound probe and the RGBD camera in real time through a positioning target, ensuring the spatial registration accuracy between the robotic arm's trajectory and the patient's anatomical structure. Secondly, through the collaborative multi-level models of YOLOv5, DeepSORT, and AlphaPose, key points of the human skeleton (such as shoulder and neck landmarks) are extracted from the depth image, and the ultrasound scanning area is dynamically delineated using a ROI localization algorithm, solving the deviation problem of traditional manual localization relying on physician experience. Thirdly, through a composite network architecture of UNet+ASPP+BiConvLSTM, real-time plaque segmentation can be achieved in the ultrasound video stream—the UNet encoder extracts multi-scale features, the ASPP module enhances adaptability to plaque size changes, BiConvLSTM utilizes temporal information to suppress single-frame noise interference, and the final output plaque mask generates a three-dimensional vascular model through the Marching Cubes algorithm, providing physicians with three-dimensional surgical field visualization support.

[0023] Secondly, this system organically integrates dynamic closed-loop control of the robotic arm with three-dimensional visualization navigation, completely changing the passive operation mode of traditional ultrasound-guided surgery. Specifically, the first arm uses deep learning-guided ultrasound probe positioning to ensure stable ultrasound image quality; the second arm, based on plaque segmentation results and pose matrix calculations, can precisely guide the ultrasonic scalpel head to the target area, achieving high operational accuracy. Furthermore, the collaborative three-dimensional reconstruction module uses the Marching Cubes algorithm to transform two-dimensional ultrasound sequences into three-dimensional models, intuitively displaying the spatial distribution of plaques and providing doctors with richer and more intuitive visual information. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the system operation process of an ultrasound navigation system for carotid artery plaque removal surgery according to a first embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram of the operation flow of the region information acquisition module in an embodiment of an ultrasound navigation system for carotid artery plaque removal surgery according to the present invention.

[0026] Figure 3This is a schematic diagram of the camera tracking principle based on an infrared optical tracking system, which is an embodiment of the ultrasound navigation system for carotid artery plaque removal surgery according to the present invention.

[0027] Figure 4 This is a schematic diagram of the image processing network in an embodiment of the ultrasound navigation system for carotid artery plaque removal surgery of the present invention, when identifying and segmenting plaques;

[0028] Figure 5 This is a schematic diagram of the three-dimensional reconstruction process of an ultrasound navigation system for carotid artery plaque removal surgery according to a first embodiment of the present invention. Detailed Implementation

[0029] The following detailed explanation illustrates the specific implementation methods:

[0030] Example 1

[0031] The basic implementation examples are as follows: Figure 1 As shown: An ultrasound navigation system for carotid artery plaque removal surgery includes an operation subsystem; the operation subsystem includes a robotic arm, a medical ultrasound imaging device, an optical tracking device, a positioning target, an RGBD depth camera, an ultrasonic scalpel head, and a host computer.

[0032] The robotic arm includes a first arm and a second arm; the first arm is used to grasp and control the ultrasound probe for scanning and imaging; the second arm is used to grasp and control the ultrasound scalpel head to remove plaque; the RGBD depth camera is used to acquire depth images; the medical ultrasound imaging device is used to scan and acquire the original images of the carotid artery plaque and the ultrasound scalpel head; the optical tracking device is used in conjunction with the positioning target to acquire the spatial coordinates of the RGBD depth camera and the ultrasound probe. In this embodiment, the optical tracking device adopts an infrared light tracking system. The positioning target is a rigid body positioning target.

[0033] The host computer is used to coordinate and control the operation subsystem. In the coordinated control, based on the depth image, the scanning area of ​​the ultrasound probe is identified, and the first arm is guided to control the ultrasound probe to contact the skin of the target area and start imaging. Then, the second arm is guided to locate and remove the plaque.

[0034] Specifically, the host computer includes a region information acquisition module, a neck information acquisition module, a patch information acquisition module, a blade information acquisition module, and a three-dimensional reconstruction module.

[0035] The region information acquisition module is used to identify target regions based on depth images and through a deep learning network. In this embodiment, the target region is the neck region.

[0036] The deep learning network includes a YOLOv5 model, a DeepSORT model, an AlphaPose network, and a ROI localization algorithm. The YOLOv5 model identifies depth images and, combined with the DeepSORT model, obtains the human body bounding box. Then, the AlphaPose network extracts the human skeleton based on the bounding box, obtaining human landmarks to roughly locate the scanning area of ​​the ultrasound probe, i.e., the target area. Finally, based on the ROI localization algorithm, the center point of the target area is confirmed, such as... Figure 2 As shown.

[0037] Here, this setup allows multiple models in the deep learning network to collaborate. The YOLOv5 model can quickly detect human body regions, balancing speed and accuracy (suitable for real-time processing). The DeepSORT model can perform human body tracking based on the detected bounding boxes, solving problems related to short-term occlusion or motion blur. The AlphaPose network can infer the approximate neck area from key points of the human skeleton (such as the neck and shoulders), enhancing the accuracy of localization. The ROI localization algorithm can further refine the localization of the center point based on key points, improving the accuracy of ultrasound scans. Furthermore, during network operation, the multi-stage processing through multi-model collaboration effectively mitigates the risk of single-model failure (e.g., if YOLOv5 misses a detection, DeepSORT can maintain tracking through motion prediction), exhibiting high reliability.

[0038] The neck information acquisition module is used to obtain the spatial coordinates of the target area. The neck information acquisition module is linked with the optical tracking device, the positioning target, and the RGBD depth camera, specifically including the following steps: using an infrared optical tracking system and a rigid positioning target, the spatial coordinate information of the camera is obtained. Specifically, the infrared optical tracking system identifies the pose matrix of the rigid positioning target relative to a fixed infrared sensor coordinate system through infrared sensors at both ends; then, the rigid positioning target is fixed on the RGBD depth camera, enabling positioning and tracking of the RGBD depth camera. Further, since the coordinate systems of the rigid positioning target and the camera do not coincide, the transformation matrix X of the rigid positioning target relative to the camera coordinate system is calculated using a hand-eye calibration algorithm.

[0039] Specifically, the pose matrix H1 of the rigid target on the camera in the coordinate system of the fixed sensor can be calculated from the position vector and quaternion vector output by the infrared optical tracking system. The pose matrix H2 of the checkerboard pattern in the camera coordinate system can be obtained by calibrating the RGBD depth camera, as shown below. Figure 3 As shown.

[0040] While keeping the relative pose of the checkerboard and the fixed sensor unchanged, the RGBD depth camera along with the positioning target is moved to another position. At this point, the pose matrix of the positioning target in the fixed sensor coordinate system is H′1, and the pose matrix of the checkerboard in the camera coordinate system is H′2. Since the pose of the checkerboard relative to the fixed sensor and the pose of the camera relative to the fixed target also remain unchanged, we have:

[0041] H1XH2=H′1XH′2 (1-1)

[0042] Multiplying both sides by the inverse matrix yields:

[0043]

[0044] Let A = (H′1) -1 H1, let The original expression is:

[0045] AX = XB (1-3)

[0046] Then, matrix X can be calculated to obtain the spatial position of the binocular camera. By combining the pose matrix between the target area and the camera, the spatial position information of the target area can be obtained.

[0047] The patch information acquisition module is used to identify and segment patches based on imaging information.

[0048] The patch information acquisition module includes an image processing network; the image processing network includes a Unet model and a BiConvLSTM module; the encoder and decoder of the Unet model are integrated with an ASPP module.

[0049] The image processing network is pre-trained using a self-made dataset. This dataset contains pre-processed ultrasound images of the carotid artery and its plaques, with labels affixed to each image. In this embodiment, 200 ultrasound images of the carotid artery and its plaques are included. Each image contains the original ultrasound image of the carotid artery, and the plaques are manually segmented by experts to obtain their ultrasound images. In the labeling process, expert physicians can read and observe the medical images (original ultrasound images) using interactive software (LabelMe), annotate the original images using existing drawing tools, and the software automatically saves the label files.

[0050] Ultrasound images and label files are saved in JPG and JSON file formats, respectively. Each JSON file represents a slice, and the carotid artery and its plaque are stored in a masked manner.

[0051] The preprocessing includes image cropping, median filtering, and grayscale normalization.

[0052] Specifically, when determining the initial acquisition point, the carotid artery is located in the central region of the image, and in subsequent consecutive two-dimensional image sequences, the carotid artery also remains largely within the central region. Therefore, this approach first trims the two-dimensional ultrasound image to remove unnecessary surrounding areas, ultimately obtaining a 512×512 two-dimensional image that can focus on the carotid artery structure while reducing computational complexity.

[0053] Secondly, due to the presence of scatterers within the human body's internal tissues, ultrasound images can generate salt-and-pepper noise (scattered black and white noise) due to tissue scattering, affecting segmentation accuracy (such as blurred blood vessel boundaries and falsely detected holes). This solution uses a median filtering algorithm to address this issue. Specifically, it replaces the center pixel with the median gray value of the pixels within a sliding window (such as 3×3 or 5×5) to suppress impulse noise (salt-and-pepper noise manifests as extreme gray values).

[0054] Furthermore, to eliminate brightness differences between different ultrasound images caused by variations in imaging conditions, grayscale normalization was performed on the carotid artery ultrasound images. Grayscale normalization restricts the image's grayscale values ​​to a smaller range, which is beneficial for subsequent data processing. Simultaneously, narrowing the data range accelerates the convergence speed of the loss function, thereby effectively reducing the time required for model training and prediction.

[0055] Optionally, in practical applications, median filtering can be performed first, followed by image cropping and grayscale normalization, in order to avoid noise interference with target localization and to prevent noise statistics from affecting the normalization parameters.

[0056] The imaging information includes ultrasound video corresponding to the target area; the image processing network, when identifying and segmenting patches, includes the following steps:

[0057] Ultrasound video is input as video frames into the encoder of the Unet network. Using full-size skip connections, high-level semantic information is extracted, and useful features for segmentation are obtained. This high-level semantic information is then input into the ASPP module to extract multi-scale features. The high-level semantic information containing multi-scale features is then input into the decoder and merged with the corresponding low-level semantic information from the encoder, outputting a feature map of the same size as the original image. This feature map is then input into the BiConvLSTM module, combining feature information from previous and subsequent frames to optimize the segmentation result of the current frame, and finally outputting the final segmentation result, i.e., patchy regions and non-patchy regions, such as... Figure 4 As shown.

[0058] When guiding the second arm to locate the plaque, the spatial coordinates of the ultrasonic probe are first obtained by positioning and tracking the ultrasonic probe. Then, based on the pose matrix of the ultrasonic probe, ultrasonic scalpel head and plaque area, the spatial coordinates of the plaque area are solved to guide the second arm to locate the plaque.

[0059] Based on the above steps, this scheme achieves accurate patch segmentation and robotic arm guidance through a multi-scale temporal segmentation network and spatial coordinate mapping. The UNet encoder with skip connections extracts high-level semantic information, while the skip connections preserve low-level details (such as patch edges), preventing information loss. Combined with the ASPP module, dilated convolutions are used to extract multi-scale features (enhancing adaptability to patches of different sizes), effectively improving robustness to changes in patch size. The BiConvLSTM module further optimizes the current frame's segmentation result by incorporating features from multiple frames, suppressing single-frame noise interference (such as ultrasound artifacts) and enhancing segmentation consistency (patch motion tracking). Secondly, when guiding the second arm to locate the patch, precise mapping from the patch region to the robotic arm coordinate system is achieved through ultrasound probe positioning and pose matrix calculation, reducing robotic arm operation errors and achieving high spatial registration accuracy. Furthermore, analysis and processing based on the ultrasound video stream can meet the real-time intraoperative positioning requirements (such as the dynamic guidance requirements of patch resection surgery).

[0060] The ultrasonic scalpel head information acquisition module is used to obtain the spatial position of the ultrasonic scalpel head. Specifically, this module uses a YOLOv5 target recognition network to perform image segmentation and localization of the interventional ultrasonic scalpel head. The YOLOv5 target recognition network is pre-trained using a corresponding dataset and annotations. This dataset contains multiple ultrasonic scalpel head images annotated with the ultrasonic scalpel head segmentation and localization results.

[0061] The 3D reconstruction module is used to construct and visualize a 3D reconstruction model of the carotid artery and its plaques.

[0062] The 3D reconstruction module, when constructing the 3D reconstruction model, uses the Marching Cubes algorithm for 3D reconstruction based on the segmentation results of the patch information acquisition module. For example, Figure 5 As shown, the 3D reconstruction system first reads the carotid artery ultrasound video and preprocesses it by splitting the ultrasound video into ultrasound images and removing black borders from the images, thus obtaining a preprocessed ultrasound image sequence of the carotid artery and its plaque. Then, the preprocessed ultrasound image sequence of the carotid artery and its plaque is input into the image processing network of the plaque information acquisition module to obtain the segmentation results of the carotid artery and its plaque for 3D reconstruction.

[0063] In practical applications, the overall operation flow of this system is as follows:

[0064] Preoperatively, the neck information acquisition module guides the first arm of the ultrasound probe to the neck region, where it autonomously completes the scan and acquires the original ultrasound image of the carotid artery. The original ultrasound image (i.e., imaging information) is then input into the plaque information acquisition module to obtain plaque segmentation and location information. The segmentation sequence image is then input into the 3D reconstruction module to obtain a 3D model of the carotid artery plaque.

[0065] During the procedure, the first arm controls the ultrasound probe to scan and acquire the original image of the ultrasonic scalpel head during the intervention, and places it into the scalpel head information acquisition module to obtain the spatial position of the ultrasonic scalpel head. The position information of the ultrasonic scalpel head is transmitted to the second arm that grasps the ultrasonic scalpel head, and combined with the plaque position information acquired before the procedure, the second arm guides the ultrasonic scalpel head to reach the plaque area.

[0066] This embodiment provides an ultrasound navigation system for carotid artery plaque removal surgery, which has high convenience, accuracy and stability in ultrasound image interpretation, can effectively reduce the workload of doctors and improve the ultrasound navigation effect.

[0067] Example 2

[0068] An ultrasound navigation system for carotid artery plaque removal surgery has been modified based on Example 1 as follows.

[0069] The region information acquisition module includes a YOLOv8n model, a DeepSORT model, a Lite-HRNet network, and a ROI localization algorithm within its deep learning network. The YOLOv8n model identifies the depth image and, combined with the DeepSORT model, obtains the human body region bounding box. Then, the Lite-HRNet network extracts the human skeleton based on the bounding box, obtaining human landmarks to roughly locate the scanning area of ​​the ultrasound probe, i.e., the target region. Finally, the ROI localization algorithm confirms the center point of the target region. This configuration in this embodiment, by locally using lightweight models, reduces computational redundancy, achieves higher real-time performance, and provides superior navigation results.

[0070] Furthermore, in the training process of deep learning networks, adopting an end-to-end multi-task joint training approach helps to improve model training efficiency.

[0071] Furthermore, the ROI positioning algorithm also presets the size range of the ROI based on human anatomical proportions (such as a width of 0.6 times the distance between the two clavicles and a height of 1.2 times the vertical distance from the cervical spine to the clavicle). In practical applications, doctors can modify it according to actual needs.

[0072] Furthermore, after confirming the center point of the target area based on the ROI positioning algorithm, feedback verification is performed. This feedback verification includes: converting the depth map within the ROI into a 3D point cloud, with each pixel corresponding to 3D coordinates; calculating the surface normals of the point cloud and analyzing the consistency of the normal directions (e.g., the normals in the neck region should be approximately perpendicular to the coronal plane of the human body), and verifying whether the normal directions are consistent. Then, principal component analysis (PCA) is used to extract the principal axis direction of the point cloud, verifying whether the neck is approximately cylindrical (i.e., the cylindricity error must be less than the corresponding threshold). Next, a surface fitting method is used to calculate the average curvature of the point cloud, verifying whether the curvature of the neck region is lower than that of surrounding tissues (such as the shoulder or chest). Finally, the point cloud is mirrored symmetrically along the central axis, and the root mean square error (RMSE) of the distance between symmetrical points is calculated, verifying whether the RMSE exceeds the threshold. If any of the first three criteria are incorrect, or if the RMSE exceeds the threshold, it indicates an error in the area positioning, and the point location is recalculated based on the depth map.

[0073] In the patch information acquisition module, the image processing network includes a Mobile-UNet model and a Transformer-based temporal module. The encoder and decoder of the Mobile-UNet model integrate an ASPP module. This embodiment, by using a more lightweight UNet variant model and a more efficient temporal module, further optimizes the operating efficiency of the image processing network and enables it to capture long-term temporal correlations, resulting in more accurate patch identification.

[0074] This embodiment provides an ultrasound navigation system for carotid artery plaque removal surgery. Based on Embodiment 1, the network structures of the deep learning network and image processing network have been optimized, resulting in higher analysis efficiency and accuracy.

[0075] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.

Claims

1. An ultrasound navigation system for carotid artery plaque removal surgery, comprising an operating subsystem; said operating subsystem comprising a robotic arm, a medical ultrasound imaging device, an optical tracking device, a positioning target, an RGBD depth camera, and an ultrasonic scalpel head; characterized in that, It also includes the host computer; The robotic arm includes a first arm and a second arm; the first arm is used to grip and control the ultrasound probe for scanning and imaging; the second arm is used to grip and control the ultrasound scalpel head for removing plaque; the RGBD depth camera is used to acquire depth images; the medical ultrasound imaging device is used to scan and acquire the original images of the carotid artery plaque and the ultrasound scalpel head; the optical tracking device is used in conjunction with the positioning target to acquire the spatial coordinates of the RGBD depth camera and the ultrasound probe. The host computer is used to coordinate and control the operation subsystem. In the coordinated control, based on the depth image, the scanning area of ​​the ultrasound probe is identified, and the first arm is guided to control the ultrasound probe to contact the skin of the target area and start imaging. Then, the second arm is guided to locate and remove the plaque.

2. The ultrasound navigation system for carotid artery plaque removal surgery according to claim 1, characterized in that, The host computer is equipped with a region information acquisition module; the region information acquisition module is used to identify target regions based on depth images and through a deep learning network.

3. The ultrasound navigation system for carotid artery plaque removal surgery according to claim 2, characterized in that, The deep learning network includes a YOLOv5 model, a DeepSORT model, an AlphaPose network, and a ROI localization algorithm. The YOLOv5 model identifies the depth image and combines it with the DeepSORT model to obtain the human body region bounding box. Then, the AlphaPose network extracts the human skeleton based on the human body region bounding box to obtain human body landmarks, so as to roughly locate the scanning area of ​​the ultrasound probe, i.e., the target area. Finally, the center point of the target area is confirmed based on the ROI localization algorithm.

4. The ultrasound navigation system for carotid artery plaque removal surgery according to claim 1, characterized in that, The host computer is equipped with a patch information acquisition module; the patch information acquisition module is used to identify and segment patches based on imaging information.

5. The ultrasound navigation system for carotid artery plaque removal surgery according to claim 4, characterized in that, The patch information acquisition module includes an image processing network; the image processing network includes a Unet model and a BiConvLSTM module; the encoder and decoder of the Unet model are integrated with an ASPP module.

6. The ultrasound navigation system for carotid artery plaque removal surgery according to claim 5, characterized in that, The imaging information includes ultrasound video corresponding to the target area; the image processing network, when identifying and segmenting patches, includes the following steps: Ultrasound video is input as video frames into the encoder of the Unet network. Combined with full-size skip connections, high-level semantic information is extracted, and features useful for segmentation are obtained. This high-level semantic information is then input into the ASPP module to extract multi-scale features. The high-level semantic information containing multi-scale features is then input into the decoder and merged with the corresponding low-level semantic information from the encoder, outputting a feature map of the same size as the original image. This feature map is then input into the BiConvLSTM module, which combines feature information from previous and subsequent frames to optimize the segmentation result of the current frame, outputting the final segmentation result, namely, patchy and non-patchy regions.

7. The ultrasound navigation system for carotid artery plaque removal surgery according to claim 6, characterized in that, When guiding the second arm to locate the plaque, the spatial coordinates of the ultrasonic probe are first obtained by positioning and tracking the ultrasonic probe. Then, based on the pose matrix of the ultrasonic probe, ultrasonic scalpel head and plaque area, the spatial coordinates of the plaque area are solved to guide the second arm to locate the plaque.

8. The ultrasound navigation system for carotid artery plaque removal surgery according to claim 5, characterized in that, The image processing network is pre-trained using a self-made dataset; the self-made dataset contains pre-processed ultrasound images of the carotid artery and its plaque, and the ultrasound images are labeled; the ultrasound images and label files are saved in JPG and JSON file formats respectively, each JSON file represents a slice, and the carotid artery and its plaque are stored in a masked manner; the preprocessing includes image cropping, median filtering, and grayscale normalization.

9. The ultrasound navigation system for carotid artery plaque removal surgery according to claim 4, characterized in that, The host computer is equipped with a three-dimensional reconstruction module; the three-dimensional reconstruction module is used to construct a three-dimensional reconstruction model of the carotid artery and its plaque and to visualize it.

10. An ultrasound navigation system for carotid artery plaque removal surgery according to claim 8, characterized in that, When constructing a 3D reconstruction model, the 3D reconstruction module uses the Marching Cubes algorithm to perform 3D reconstruction based on the segmentation results of the patch information acquisition module.