A feature-based automatic registration method for surgical navigation, and a storage medium

Abstract

The application provides a feature-based automatic registration method for surgical navigation and a storage medium, and belongs to the technical field of medical instruments.The technical points are as follows: based on the first space transformation matrix, the preoperative region of interest three-dimensional data and the intraoperative region of interest three-dimensional data are precisely registered based on the intraoperative stable region three-dimensional data weighting to obtain a second space transformation matrix, and the spatial mapping relationship from the intraoperative real space to the preoperative medical image space is obtained based on the first and second space transformation matrices.The feature-based automatic registration method for surgical navigation and the storage medium can further improve the accuracy of spatial registration in surgical navigation, lay a good foundation for realizing accurate surgical navigation, and meet the accuracy requirements of neurosurgical navigation.

Description

Technical Field

[0001] This invention belongs to the technical field of medical devices, specifically relating to a feature-based automatic registration method for surgical navigation and a storage medium. Background Technology

[0002] Currently, surgical navigation systems have been applied in many surgical procedures, especially in neurosurgery, to select the optimal surgical path based on accurate location of the lesion, thereby reducing surgical trauma.

[0003] The basic principle of surgical navigation is to establish a pose transformation mapping relationship between the image space (a virtual anatomical model reconstructed from preoperative medical images) and the patient space (the patient's actual anatomical structure) through spatial registration. Then, the actual pose of surgical instruments is mapped to the image space in real time, enabling instrument operation guided by a visualized image. Therefore, the method of registering the image space with the patient space is both the key and the challenge of surgical navigation systems, directly affecting the accuracy of the entire system.

[0004] Spatial registration in existing surgical navigation systems can be divided into two main types:

[0005] First, there are point-matching methods. Point-matching methods (e.g., CN115363753A, CN115887003A, CN113349930A) include bone anchor points, anatomical anchor points, and anchor points adhered to the skin surface. Spatial registration is achieved by matching anchor points obtained from a 3D model reconstructed from preoperative medical images with intraoperative patient anchor points obtained by capturing probe positions using a positioning and tracking device. Among these, bone anchor points offer the highest accuracy, but require preoperative implantation, introducing additional trauma to the patient. Anatomical and adhered anchor points reduce trauma but are susceptible to surface skin deformation, resulting in lower accuracy.

[0006] Second, surface matching-based methods. Surface matching-based methods (such as CN105658167A) employ the ICP algorithm, which has high requirements for the initial pose between the point clouds to be registered. It requires manual coarse registration based on dissected markers using point matching methods, which makes automatic registration impossible, significantly impacting the registration speed and potentially causing registration failure.

[0007] The navigation accuracy of the methods mentioned above is insufficient and cannot meet the high precision requirements of surgeries such as neurosurgery. Therefore, it is necessary to develop a more accurate automatic registration method for surgical navigation. Summary of the Invention

[0008] The purpose of this invention is to solve the problems existing in the prior art and to provide a feature-based automatic registration method for surgical navigation and a storage medium.

[0009] The technical solution of this application is as follows:

[0010] A feature-based automatic registration method for surgical navigation includes the following steps:

[0011] S100, acquires three-dimensional data of the skin surface before surgery;

[0012] S200, obtains three-dimensional data of the preoperative region of interest;

[0013] S300, acquire 3D data of intraoperative region of interest: process the 3D data acquired by S200 to obtain 3D data of intraoperative region of interest;

[0014] S400, the preoperative three-dimensional data of the region of interest and the intraoperative three-dimensional data of the region of interest are processed to obtain preoperative three-dimensional data of regions with unique single-scale features and continuous multi-scale features, as well as intraoperative three-dimensional data of regions with unique single-scale features and continuous multi-scale features.

[0015] S500, obtain the first spatial transformation matrix TR A The first spatial transformation matrix is ​​obtained by coarsely registering the preoperative three-dimensional data of the region with unique single-scale features and continuous multi-scale features and the intraoperative three-dimensional data of the region with unique single-scale features and continuous multi-scale features using a feature-based automatic registration method.

[0016] S600, obtain the second spatial transformation matrix TR B The first spatial transformation matrix is ​​used to perform fine registration of the preoperative region of interest 3D data and the intraoperative region of interest 3D data based on the weighted intraoperative stable region 3D data to obtain the second spatial transformation matrix.

[0017] S700, obtains the spatial mapping relationship between the intraoperative real space and the preoperative medical imaging space. c ;

[0018] TR c =TR B ·TR A .

[0019] Furthermore, S100 includes the following sub-steps:

[0020] S101, preprocess the acquired preoperative medical images to obtain preprocessed preoperative medical images;

[0021] S102, threshold segmentation is used to segment the preprocessed preoperative medical images to obtain image data of the preoperative skin surface;

[0022] S103, Reconstruct the preoperative skin surface image data to obtain preoperative three-dimensional skin surface data.

[0023] Furthermore, S200 also includes: processing the preoperative three-dimensional data of the skin surface to obtain preoperative three-dimensional data of the region of interest.

[0024] Furthermore, S400 includes the following sub-steps:

[0025] S401, Based on the scale of the three-dimensional data of the preoperative and intraoperative regions of interest, a set of discrete feature descriptor scales within a continuous interval is obtained, and a set of continuous and discrete feature descriptor scales is obtained by using a feature descriptor neighborhood radius calculation method based on global metric.

[0026] S402, based on the radius of the feature descriptor scale obtained in S401, calculate each single-scale feature descriptor of the three-dimensional data of the preoperative and intraoperative regions of interest.

[0027] S403, Calculate the mean μ and standard deviation σ of each single-scale feature descriptor method feature descriptor of the three-dimensional data of the preoperative region of interest based on distance metric;

[0028] S404, based on the mean μ and standard deviation σ of each single-scale feature descriptor, obtain the single-scale feature unique descriptor outside the interval μ±α·σ and the corresponding single-scale feature unique region three-dimensional data.

[0029] S405 performs intersection processing on the feature descriptors that are unique in each single scale to obtain feature descriptors that are unique in single scale and persistent in multi-scale features, as well as the corresponding three-dimensional data of the preoperative and intraoperative regions that are unique in single scale and persistent in multi-scale features.

[0030] Furthermore, S500 includes the following sub-steps:

[0031] S501, perform feature detection on the preoperative single-scale feature unique and multi-scale feature continuous three-dimensional data and the intraoperative single-scale feature unique and multi-scale feature continuous three-dimensional data to obtain preoperative single-scale feature unique and multi-scale feature continuous key three-dimensional data and intraoperative single-scale feature unique and multi-scale feature continuous key three-dimensional point cloud data.

[0032] S502, perform feature description on the key three-dimensional data of the preoperative single-scale feature-unique, multi-scale feature-continuous region and the intraoperative single-scale feature-unique, multi-scale feature-continuous region to obtain feature descriptors of the preoperative single-scale feature-unique, multi-scale feature-continuous region and the intraoperative single-scale feature-unique, multi-scale feature-continuous region.

[0033] S503, feature matching is performed based on the feature descriptors of the key three-dimensional data of the preoperative single-scale feature-unique and multi-scale feature-continuous region and the feature descriptors of the intraoperative single-scale feature-unique and multi-scale feature-continuous region to obtain feature matching relationship.

[0034] S504, perform error matching elimination on the feature matching relationship to obtain an optimized feature matching relationship;

[0035] S505, based on the optimized feature matching relationship, the first spatial transformation matrix is ​​obtained by least squares or singular value decomposition to complete the coarse registration process.

[0036] Furthermore, S600 includes the following sub-steps:

[0037] S601, Process the three-dimensional data of the intraoperative region of interest to obtain three-dimensional data of the intraoperative stable region;

[0038] S602, Based on the first spatial transformation matrix, perform fine registration on the preoperative region of interest 3D data and the intraoperative region of interest 3D data weighted based on the intraoperative stable region 3D data to obtain the second spatial transformation matrix and complete the fine registration process;

[0039] The stable region can be any part or more of the nasal region, brow bone region, and frontal bone region; the stable region is segmented by accumulating differences or a pre-trained stable region detection model to identify the region that does not deform in the three-dimensional data of multiple intraoperative regions of interest.

[0040] Furthermore, the fine registration method based on the weighted three-dimensional data of the intraoperative stable region adopts the weighted ICP algorithm, the three-dimensional data of the preoperative region of interest is three-dimensional mesh data, and the three-dimensional data of the intraoperative region of interest is three-dimensional point cloud data;

[0041] Specifically, it includes:

[0042] (1) For each point in the intraoperative region of interest 3D point cloud data, the nearest corresponding grid in the preoperative region of interest 3D grid data is found to form the nearest correspondence;

[0043] (2) The points in the nearest correspondence are projected onto the three-dimensional mesh data of the preoperative region of interest along the mesh normal direction;

[0044] (3) The point projected onto the preoperative three-dimensional mesh data is located within the ω region of the weighted area based on the intraoperative stable region three-dimensional data. i >1, the point projected onto the preoperative 3D mesh data is located outside the weighted region based on the intraoperative stable region 3D data at ω. i -1;

[0045] (4) After each iteration completes the search for the nearest correspondence, calculate x(R,T) corresponding to the current nearest correspondence according to the function 2, and then use R and T to calculate the error after this iteration using the function 1. Stop iterating when the expected error threshold is met, and obtain the second spatial transformation matrix TR. B ;

[0046] Wherein, function 1 represents the error optimization function of the weighted ICP algorithm as follows:

[0047] Here, function 2 represents: x(R,T) is obtained by calculating all the nearest correspondences using the linear least squares method, that is:

[0048]

[0049] A storage medium storing a program for running the aforementioned surgical navigation automatic registration method.

[0050] The advantages of the technical solution of this invention are mainly reflected in:

[0051] (1) The registration method provided by the present invention can realize fully automated spatial registration without the need for marker point identification, which can improve the efficiency of surgical navigation spatial registration.

[0052] (2) Using preoperative three-dimensional data of regions with unique single-scale features and continuous multi-scale features, and intraoperative three-dimensional data of regions with unique single-scale features and continuous multi-scale features, in feature-based automatic registration methods can improve the robustness of registration and better cope with abnormal situations in the surgical environment.

[0053] (3) Based on the first spatial transformation matrix, the three-dimensional data of the preoperative region of interest and the three-dimensional data of the intraoperative region of interest are precisely registered by weighting the three-dimensional data of the stable intraoperative region to obtain the second spatial transformation matrix. This can further improve the accuracy of spatial registration in surgical navigation, lay a good foundation for achieving accurate surgical navigation, and meet the accuracy requirements of neurosurgical navigation. Attached Figure Description

[0054] The present invention will be further described in detail below with reference to the embodiments shown in the accompanying drawings, but this does not constitute any limitation on the present invention.

[0055] Figure 1 This is a schematic flowchart of a surgical navigation spatial registration method provided in an embodiment of the present invention;

[0056] Figure 2 This is a schematic flowchart of a feature-based automatic registration method provided in an embodiment of the present invention.

[0057] Figure 3 A schematic diagram of the intraoperative stable region is provided for one embodiment of the present invention;

[0058] Figure 4 This invention provides a schematic diagram of three-dimensional point cloud data of regions with unique single-scale features and continuous multi-scale features before and during surgery, as well as an embodiment of the present invention.

[0059] The annotations in the attached figures are explained as follows:

[0060] 1. Patient's head; 2. Stable area during surgery; 3. Preoperative 3D point cloud data of a region with unique single-scale features and continuous multi-scale features; 4. Intraoperative 3D point cloud data of a region with unique single-scale features and continuous multi-scale features. Detailed Implementation

[0061] The surgical navigation spatial registration method proposed in this invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0062] The embodiments of the present invention are described as a processing flow. Although the various operation steps of the flow may be given sequential step numbers, the operation steps may be implemented in parallel, concurrently, or simultaneously.

[0063] Unless otherwise specified, the terms used herein generally have their ordinary meaning in the context of this art, the specific embodiments, and the particular context. Certain terms used to describe these embodiments will be discussed below or elsewhere in this specification to provide additional guidance to those skilled in the art in relation to the description of this case.

[0064] The terms “contains,” “includes,” and “has” used in this article are all open-ended, meaning they include but are not limited to.

[0065] In embodiments of the present invention, the data “and / or” may be used, and “and / or” includes any and all combinations of one or more of the listed associated features.

[0066] It should be noted that the accompanying drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clearly illustrate the purpose of the embodiments of the present invention.

[0067] To make the advantages of the technical solutions in the embodiments of the present invention clearer, the exemplary embodiments of the present invention will be described in further detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not an exhaustive list of all embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0068] <Example 1: A feature-based automatic registration method for surgical navigation>

[0069] As Figure 1 shown, a feature-based automatic registration method for surgical navigation, the implementation thereof comprises the following steps:

[0070] <S100, acquiring preoperative skin surface three-dimensional data>

[0071] Segmenting the acquired preoperative medical image (e.g., craniocerebral medical image) to obtain image data of a preoperative skin surface, and reconstructing the preoperative skin surface image data to obtain preoperative skin surface three-dimensional data.

[0072] Specifically, S100 comprises the following sub-steps:

[0073] S101, preprocessing the acquired preoperative medical image to obtain a preprocessed preoperative medical image (preprocessing the preoperative medical image, and then segmenting the processed preoperative medical image by using a pre-trained segmentation model, which can improve segmentation accuracy; results of the preoperative medical image segmentation, in addition to a preoperative skin region, further include a preoperative lesion region, a preoperative adjacent key blood vessel region, other preoperative related tissue regions, etc.);

[0074] S102, segmenting the preprocessed preoperative medical image by using threshold segmentation to obtain image data of a preoperative skin surface;

[0075] S103, reconstructing according to the image data of the preoperative skin surface to obtain preoperative skin surface three-dimensional data.

[0076] <S200, obtaining three-dimensional data of a preoperative region of interest>

[0077] Processing the preoperative skin surface three-dimensional data to obtain three-dimensional data of a preoperative region of interest.

[0078] Specifically, the preoperative skin surface three-dimensional data is processed by using a pre-trained first interest detection model to obtain three-dimensional data of a preoperative region of interest.

[0079] <S300, acquiring three-dimensional data of an intraoperative region of interest>

[0080] Processing the three-dimensional data acquired in S200 to obtain three-dimensional data of an intraoperative region of interest. Specifically, the three-dimensional data acquired in S200 is processed by using a pre-trained second interest detection model to obtain three-dimensional data of an intraoperative region of interest. Optionally, the first interest detection model and the second interest detection model may be deep learning neural network models in the prior art, and no limitation is imposed on specific network structures.

[0081] <S400, processing the pre-operative three-dimensional data of the region of interest and the intra-operative three-dimensional data of the region of interest to obtain pre-operative three-dimensional data of regions with unique single-scale features and persistent multi-scale features, and intra-operative three-dimensional data of regions with unique single-scale features and persistent multi-scale features>

[0082] S400 comprises the following sub-steps:

[0083] Specifically, the pre-operative three-dimensional data of regions with unique single-scale features and persistent multi-scale features refers to a region composed of three-dimensional data that has unique single-scale features and persistent features at multiple scales;

[0084] S401, obtaining discrete feature descriptor scales within a set of continuous intervals based on the scales of the pre-operative and intra-operative three-dimensional data of the region of interest, and obtaining a set of continuous and discrete feature descriptor scales by using a feature descriptor neighborhood radius calculation method based on global metric;

[0085] Wherein, the calculation method of neighborhood radius ρ of feature descriptors based on global metric is adopted, and its formula is:

[0086]

[0087] Wherein, AM is the total surface area of three-dimensional data M, and α is a parameter controlling the size of the neighborhood radius.

[0088] S402, respectively calculating each single-scale feature descriptor of the pre-operative and intra-operative three-dimensional data of the region of interest based on the radius of the feature descriptor scales obtained in S401;

[0089] A fast point feature histograms (FPFH) feature description algorithm is used to calculate each single-scale FPFH feature descriptor of the pre-operative three-dimensional data of the region of interest and each single-scale FPFH feature descriptor of the intra-operative three-dimensional data of the region of interest (the FPFH feature descriptor is a high-dimensional vector).

[0090] S403, calculating the mean value μ and standard deviation σ of each single-scale feature descriptor of the pre-operative three-dimensional data of the region of interest based on distance metric;

[0091] S404, obtaining feature descriptors with unique single-scale features outside the interval μ±α·σ and the corresponding three-dimensional data of unique single-scale feature regions according to the mean value μ and standard deviation σ of each single-scale feature descriptor;

[0092] S405, performing intersection processing on the feature descriptors unique to each single-scale feature to obtain feature descriptors that are unique in single-scale features and persistent in multi-scale features, as well as the corresponding three-dimensional data of the pre-operative and intra-operative regions with unique single-scale features and persistent multi-scale features (such as Figure 4 shown).

[0093] <Obtain the first spatial transformation matrix in S500>

[0094] Performing coarse registration on the three-dimensional data of the pre-operative region with unique single-scale features and persistent multi-scale features and the three-dimensional data of the intra-operative region with unique single-scale features and persistent multi-scale features by a feature-based automatic registration method, so as to obtain a first spatial transformation matrix.

[0095] Specifically, the first spatial transformation matrix is a rough pose transformation mapping relationship between the image space and the patient space established based on the three-dimensional data of the pre-operative region with unique single-scale features and persistent multi-scale features and the three-dimensional data of the intra-operative region with unique single-scale features and persistent multi-scale features, denoted as TR A ;

[0096] Specifically, is a homogeneous transformation matrix, wherein, is a rotation transformation matrix, T A =[t 11 t 21 t 31 T is a translation vector.

[0097] S500 includes the following sub-steps:

[0098] S501, performing feature detection on the three-dimensional data of the pre-operative region with unique single-scale features and persistent multi-scale features and the three-dimensional data of the intra-operative region with unique single-scale features and persistent multi-scale features, so as to obtain key three-dimensional data of the pre-operative region with unique single-scale features and persistent multi-scale features and key three-dimensional point cloud data of the intra-operative region with unique single-scale features and persistent multi-scale features;

[0099] S502, performing feature description on the key three-dimensional data of the pre-operative region with unique single-scale features and persistent multi-scale features and the key three-dimensional data of the intra-operative region with unique single-scale features and persistent multi-scale features, so as to obtain the feature descriptor of the key three-dimensional data of the pre-operative region with unique single-scale features and persistent multi-scale features and the feature descriptor of the key three-dimensional data of the intra-operative region with unique single-scale features and persistent multi-scale features;

[0100] ​S503, performing feature matching based on the feature descriptors of the key three-dimensional data of the persistent region of unique single-scale features and multi-scale features in the preoperative image and the feature descriptors of the key three-dimensional data of the persistent region of unique single-scale features and multi-scale features in the intraoperative image, so as to obtain a feature matching relationship;

[0101] S504, eliminating mismatches from the feature matching relationship to obtain an optimized feature matching relationship;

[0102] S505, processing by least square method or singular value decomposition based on the optimized feature matching relationship to obtain a first spatial transformation matrix, and completing a coarse registration process.

[0103] <S600, obtaining a second spatial transformation matrix>

[0104] Performing fine registration weighted based on the intraoperative stable region three-dimensional data on the preoperative region of interest three-dimensional data and the intraoperative region of interest three-dimensional data based on the first spatial transformation matrix to obtain a second spatial transformation matrix.

[0105] S600, which comprises the following sub-steps:

[0106] S601, processing the intraoperative region of interest three-dimensional data to obtain intraoperative stable region three-dimensional data;

[0107] S602, performing fine registration weighted based on the intraoperative stable region three-dimensional data on the preoperative region of interest three-dimensional data and the intraoperative region of interest three-dimensional data based on the first spatial transformation matrix to obtain a second spatial transformation matrix, and completing a fine registration process.

[0108] Optionally, the fine registration method weighted based on the intraoperative stable region three-dimensional data adopts a weighted ICP algorithm (the ICP algorithm belongs to the prior art), and specifically comprises:

[0109] The preoperative region of interest three-dimensional data is used as target data, denoted as Q = {q i | i = 1, 2, ..., M};

[0110] The intraoperative region of interest three-dimensional data is used as source data, denoted as P = {p i | i = 1, 2, ..., N};

[0111] The preoperative region of interest three-dimensional mesh data comprises normal data, denoted as N = {n i | i = 1, 2, ..., M};

[0112] The error optimization function of the weighted ICP algorithm is:

[0113]

[0114] Where ω i The weight of the most recent correspondence;

[0115] The error optimization function of the weighted ICP algorithm is: (Function 1), where w i The weight of the most recent correspondence;

[0116] The x(R,T) is obtained by calculating all the nearest correspondences using the linear least squares method, that is: (Function 2);

[0117] The preoperative region of interest 3D data is 3D mesh data, and the intraoperative region of interest 3D data is 3D point cloud data;

[0118] The specific steps are as follows:

[0119] (1) For each point in the intraoperative region of interest 3D point cloud data, the nearest corresponding grid in the preoperative region of interest 3D grid data is found to form the nearest correspondence;

[0120] (2) The points in the nearest correspondence are projected onto the three-dimensional mesh data of the preoperative region of interest along the mesh normal direction;

[0121] (3) The point projected onto the preoperative three-dimensional mesh data is located within the ω region of the weighted area based on the intraoperative stable region three-dimensional data. i >1, the point projected onto the preoperative 3D mesh data is located outside the weighted region based on the intraoperative stable region 3D data at ω. i -1;

[0122] (4) After each iteration completes the search for the nearest correspondence, calculate x(R,T) corresponding to the current nearest correspondence according to the function 2, and then use R and T to calculate the error after this iteration using the function 1. Stop iterating when the expected error threshold is met, and obtain the second spatial transformation matrix TR. B .

[0123] Specifically, the second spatial transformation matrix Let be the homogeneous transformation matrix, where, Let T be the rotation transformation matrix. B =[t 11 t 21 t 31 ] T It is a translation vector.

[0124] The stable region is any one or more of the nasal region, the eyebrow bone region and the frontal bone region. Since the facial region includes easily deformable regions such as the mouth and cheeks, while the nasal region, the eyebrow bone region and the frontal bone region are not prone to deformation, increasing the weight of this region during fine registration by selecting at least one of the nasal region, the eyebrow bone region and the frontal bone region can ensure that the obtained second spatial transformation matrix is more accurate, and further improve the accuracy of spatial registration for neurosurgical navigation.

[0125] The stable region is segmented by cumulative difference or a pre-trained stable region detection model as a region that does not deform in the plurality of three-dimensional data of the intraoperative region of interest.

[0126] <S700, obtain the spatial mapping relationship from the real intraoperative space to the preoperative medical image space>

[0127] S700: Obtain the spatial mapping relationship TR from the real intraoperative space to the preoperative medical image space according to the first spatial transformation matrix and the second spatial transformation matrix c ;

[0128] Specifically, the spatial mapping relationship from the real intraoperative space to the preoperative medical image space is: TR c =TR B ·TR A . It should be noted that, as can be understood by those skilled in the art, the spatial registration method provided by the present invention is not only applicable to neurosurgery, but also applicable to other surgical operations, such as otolaryngology surgery, orthopedics and other surgical operations.

[0129] It should be noted that, as can be understood by those skilled in the art, the obtained preoperative medical image is a three-dimensional medical image including a plurality of two-dimensional medical images, such as CT images or MRI images or images collected by other medical imaging equipment, which is not limited in the present invention.

[0130] With the above method, for neurosurgery, coarse registration is performed on the three-dimensional data of preoperative single-scale unique feature, multi-scale persistent region and the three-dimensional data of intraoperative single-scale unique feature, multi-scale persistent region of the facial region through the feature-based automatic registration method to obtain the first spatial transformation matrix; then fine registration weighted based on the three-dimensional data of the intraoperative stable region is performed on the three-dimensional data of the preoperative region of interest and the three-dimensional data of the intraoperative region of interest based on the first spatial transformation matrix to obtain the second spatial transformation matrix. According to the first spatial transformation matrix and the second spatial transformation matrix, the spatial mapping relationship from the real intraoperative space to the preoperative medical image space is obtained, which can automatically complete the spatial registration process of neurosurgery without marker points, and effectively improve the accuracy and efficiency of spatial registration for neurosurgical navigation.

[0131] like Figure 3 As shown, intraoperative 3D data of patients in different states during surgery are collected, and the intraoperative facial stable region that does not deform in multiple intraoperative 3D data is segmented by averaging or a pre-trained stable region detection model.

[0132] The above-described embodiments are preferred embodiments of the present invention and are only used to facilitate the illustration of the present invention. They are not intended to limit the present invention in any way. Any person skilled in the art who makes local modifications or alterations to the technical content disclosed in the present invention without departing from the scope of the technical features of the present invention shall still fall within the scope of the technical features of the present invention.

Claims

1. A feature-based automatic registration method for surgical navigation, characterized in that, The steps include the following: S100, acquires three-dimensional data of the skin surface before surgery; S200, obtains three-dimensional data of the preoperative region of interest; S300, acquire 3D data of intraoperative region of interest: process the 3D data acquired by S200 to obtain 3D data of intraoperative region of interest; S400, the preoperative three-dimensional data of the region of interest and the intraoperative three-dimensional data of the region of interest are processed to obtain preoperative three-dimensional data of regions with unique single-scale features and continuous multi-scale features, as well as intraoperative three-dimensional data of regions with unique single-scale features and continuous multi-scale features. Includes the following sub-steps: S401, Based on the scale of the three-dimensional data of the preoperative and intraoperative regions of interest, a set of discrete feature descriptor scales within a continuous interval is obtained, and a set of continuous and discrete feature descriptor scales is obtained by using a feature descriptor neighborhood radius calculation method based on global metric. Among them, the feature descriptor neighborhood radius ρ based on global metric is used: ； where A M is the total surface area of the three-dimensional data M, and a is a parameter that controls the size of the neighborhood radius; S402, based on the radius of the feature descriptor scale obtained in S401, calculate each single-scale feature descriptor of the three-dimensional data of the preoperative and intraoperative regions of interest. S403, Calculate the mean μ and standard deviation σ of each single-scale feature descriptor method feature descriptor of the three-dimensional data of the preoperative region of interest based on distance metric; S404, based on the mean μ and standard deviation σ of each single-scale feature descriptor, obtain the single-scale feature unique descriptor outside the interval μ±α·σ and the corresponding single-scale feature unique region three-dimensional data. S405 performs intersection processing on the feature descriptors that are unique in each single scale to obtain feature descriptors that are unique in single scale and persistent in multi-scale, as well as the corresponding three-dimensional data of the preoperative and intraoperative regions that are unique in single scale and persistent in multi-scale. S500, obtain the first spatial transformation matrix TR A The method employs a feature-based automatic registration approach to coarsely register the preoperative 3D data with unique single-scale features and persistent multi-scale features, and the intraoperative 3D data with unique single-scale features and persistent multi-scale features, to obtain a first spatial transformation matrix. This includes the following sub-steps: S501, performing feature detection on the preoperative 3D data with unique single-scale features and persistent multi-scale features, and the intraoperative 3D data with unique single-scale features and persistent multi-scale features, to obtain key 3D data with unique single-scale features and persistent multi-scale features, and key 3D point cloud data with unique single-scale features and persistent multi-scale features, respectively; S502, performing feature detection on the key 3D data with unique single-scale features and persistent multi-scale features, and key 3D point cloud data with unique single-scale features and persistent multi-scale features, respectively. S503: Feature description is performed on key 3D data of the continuation region to obtain feature descriptors for the preoperative single-scale and multi-scale continuous key 3D data and the intraoperative single-scale and multi-scale continuous key 3D data; S504: Feature matching is performed based on the feature descriptors of the preoperative and intraoperative single-scale and multi-scale continuous key 3D data to obtain feature matching relationships; S505: Error matching is removed from the feature matching relationships to obtain optimized feature matching relationships; S506: Based on the optimized feature matching relationships, least squares or singular value decomposition is used to obtain a first spatial transformation matrix, completing the coarse registration process. S600, obtain the second spatial transformation matrix TR B The second spatial transformation matrix is ​​obtained by performing fine registration of the preoperative region of interest 3D data and the intraoperative region of interest 3D data based on the weighted intraoperative stable region 3D data, using the first spatial transformation matrix. The preoperative three-dimensional data of the region of interest is used as the target data, denoted as... ; The three-dimensional data of the intraoperative region of interest is used as the source data, denoted as... ; The preoperative region of interest 3D data includes normal data, denoted as... ; S600 includes the following sub-steps: S601, Process the three-dimensional data of the intraoperative region of interest to obtain three-dimensional data of the intraoperative stable region; S602, Based on the first spatial transformation matrix, perform fine registration on the preoperative region of interest 3D data and the intraoperative region of interest 3D data weighted based on the intraoperative stable region 3D data to obtain the second spatial transformation matrix and complete the fine registration process; The stable region is any part or more of the nose region, brow bone region, and frontal bone region; the stable region is segmented by accumulating differences or a pre-trained stable region detection model to identify the region that does not deform in the three-dimensional data of multiple intraoperative regions of interest. The fine registration method based on the weighted three-dimensional data of the intraoperative stable region adopts the weighted ICP algorithm. The three-dimensional data of the preoperative region of interest is three-dimensional mesh data, and the three-dimensional data of the intraoperative region of interest is three-dimensional point cloud data. Specifically, it includes: (1) For each point in the intraoperative region of interest 3D point cloud data, the nearest grid in the preoperative region of interest 3D grid data is found to form the nearest correspondence; (2) The points in the nearest correspondence are projected onto the three-dimensional mesh data of the preoperative region of interest along the mesh normal direction; (3) The point projected onto the preoperative region of interest 3D mesh data is located within the ω region weighted by the intraoperative stable region 3D data. i >1, the point projected onto the preoperative region of interest 3D mesh data is located outside the weighted region based on the intraoperative stable region 3D data at ω. i -1;ω i The weight of the most recent correspondence; (4) After each iteration completes the search for the nearest correspondence, calculate x(R,T) corresponding to the current nearest correspondence according to function 2. Then use the rotation transformation matrix R and the translation matrix T to calculate the error after this iteration using function 1. Stop the iteration when the expected error threshold is met, and obtain the second spatial transformation matrix TR. B ; Wherein, function 1 represents the error optimization function of the weighted ICP algorithm. ; Here, function 2 represents the result obtained by calculating using the linear least squares method based on all the nearest correspondences. , ; S700, obtains the spatial mapping relationship between the intraoperative real space and the preoperative medical imaging space. c ; TR c =TR B TR A .

2. The feature-based automatic registration method for surgical navigation according to claim 1, characterized in that, S100 includes the following sub-steps: S101, preprocess the acquired preoperative medical images to obtain preprocessed preoperative medical images; S102, threshold segmentation is used to segment the preprocessed preoperative medical images to obtain image data of the preoperative skin surface; S103, Reconstruct the preoperative skin surface image data to obtain preoperative three-dimensional skin surface data.

3. The feature-based automatic registration method for surgical navigation according to claim 1, characterized in that, S200 further includes: processing the preoperative three-dimensional data of the skin surface to obtain preoperative three-dimensional data of the region of interest.

4. A storage medium, characterized in that, The storage medium stores a program that runs the surgical navigation automatic registration method as described in any one of claims 1 to 3.

Images