Skull mark point identification method and system based on bone seam boundary extraction and intersection identification

Abstract

A skull mark point recognition method based on bone seam boundary extraction and intersection recognition comprises the steps that firstly, a segmented skull three-dimensional CT tag image is obtained to serve as input; then, performing binarization processing and morphological expansion operation on each skeleton label, and extracting boundary voxels of each skeleton; thirdly, fusing and superposing all skeleton boundaries, and detecting voxel points which are overlapped at two or more skeleton boundaries as candidate bone seam intersections; then, classifying and identifying the candidate intersection points based on a predefined skeleton combination rule, and determining the types of the skull anatomical mark points corresponding to the candidate intersection points; and finally, a geometric feature optimization strategy is adopted for the position of each mark point, and the final space coordinate of each mark point is obtained through calculation. The invention further provides a skull mark point recognition system. The whole process does not depend on machine learning model training, the result interpretability is high, a large amount of labeled data is not needed, the positioning precision is high, and the method can be widely applied to scenes such as medical image three-dimensional reconstruction and surgical planning navigation.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of medical image processing and computer-aided surgery, and particularly relates to a skull landmark point recognition method and system based on suture boundary extraction and intersection identification, which is used for recognizing specific skull landmark points with anatomical significance from three-dimensional skull images. BACKGROUND

[0002] Skull anatomical landmark points refer to specific reference points on the skull with anatomical significance, such as suture intersection points, etc. In medical diagnosis, surgical planning, and anthropological research, accurate positioning of these landmark points plays an important role. Currently, the labeling of skull feature points in medical images mainly relies on manual completion: professional personnel manually label these key points on three-dimensionally reconstructed skull models or two-dimensional slices. Since manual labeling is influenced by subjective experience, there are differences between different operators, and the repeatability is poor. Moreover, the manual positioning process is time-consuming and laborious, and the efficiency is low.

[0003] Some existing automatic positioning methods are mainly based on simple threshold segmentation or edge detection of grayscale images, but it is often difficult to achieve ideal results in the face of complex skull structures. In particular, the suture lines between adjacent skull plates (i.e. the thin and narrow gaps formed at the junctions of different skull plates) are not always clearly visible in CT images, resulting in low recognition accuracy based on traditional image processing methods. At the same time, since the skull is composed of multiple irregular plates, the overall structure is complex, and existing methods have limitations in understanding the three-dimensional structural relationship and identifying suture intersection features, making it difficult to reliably identify all anatomical landmark points. Therefore, there is an urgent need for a new method that can efficiently and accurately automatically identify skull landmark points to reduce human error and improve work efficiency. SUMMARY

[0004] In order to overcome the shortcomings of the prior art, the present application provides a skull landmark point recognition method and system based on suture boundary extraction and intersection identification, which can automatically identify important anatomical landmark points of the skull on three-dimensional CT segmented images. This method uses a computer program to analyze the boundaries and junction features of each skull plate, achieving efficient and accurate positioning of the landmark points.

[0005] To achieve the above purpose, the present application adopts the following technical solutions:

[0006] A skull landmark point recognition method based on suture boundary extraction and intersection identification, comprising the following steps:

[0007] (1) Three-dimensional image acquisition: acquire three-dimensional CT segmented image data of the skull, wherein different gray scales or label values in the segmented image correspond to different skull plate regions, and preferably, the input image data can be pre-processed as necessary, such as denoising or format conversion, to ensure the accuracy and stability of subsequent processing.

[0008] (2) Boundary extraction: traversing the three-dimensional segmented image, a set of boundary points of each bone plate is extracted, specifically, scanning the three-dimensional voxel data, when a voxel and its adjacent voxel belong to different bone plate labels, the position of the voxel belongs to the boundary of the bone plate. In the present application, all voxels (or their corresponding spatial positions) that meet this condition are collected to form the boundary point set of each bone plate of the skull. Through this step, the initial boundary description of the outer surface of the skull bone plate and the adjacent bone plate junction can be obtained.

[0009] (3) Suture recognition: overlapping analysis is performed on the extracted bone plate boundary point set to identify the skull suture line features. The so-called overlapping analysis refers to comparing the boundary data of different bone plates to detect whether there are parts that are positionally consistent or adjacent; if the boundaries of two different bone plates are adjacent in three-dimensional space and form a face-to-face relationship, it is determined that this is the location of the suture; in other words, by identifying overlapping boundary points from two adjacent bone plates, the suture line segments formed by the intersection of these bone plates can be extracted; after this step, a set of suture lines is obtained, each suture line corresponding to the intersection of two specific skull bone plates.

[0010] (4) Intersection extraction: analyzing the suture line set to extract the intersection points between the suture lines as candidate skull landmark points. When three or more suture lines intersect in space, the intersection points often correspond to important anatomical landmark points, for example, the intersection of multiple bone plate junctions (such as the intersection of the coronal suture and the sagittal suture) is a typical skull landmark point. In this step, the algorithm traverses the suture line set to find any suture line pairs / groups that are close to each other and intersect. For the detected intersection positions, their spatial coordinates are extracted and recorded as candidate landmark points. In addition, to improve robustness, a fine analysis (such as increasing neighborhood search) can be performed on the region near the intersection point to ensure the accuracy of the intersection point positioning.

[0011] (5) Classification and screening: classifying and screening the candidate landmark points according to the bone plate combinations they involve, and outputting the final list of skull anatomical landmark points. Classification refers to assigning a classification identifier to the landmark point based on the number of bone plates involved in the intersection near the landmark point and their anatomical positions (for example, points involving the intersection of three bone plates are classified as one type, points involving the intersection of four bone plates are classified as another type, etc.). At the same time, some non-typical or unstable intersection points can be screened out based on anatomical knowledge or predetermined rules, for example, if a certain intersection point is formed by a very small bone fragment, it may not be a major anatomical landmark point. After classification and screening, the remaining landmark points are the key anatomical landmark points of the skull detected by the present application. These landmark points will be labeled with additional information (such as name or number) and can be stored or transmitted for use by doctors in diagnosis or surgical planning.

[0012] A skull landmark point recognition system based on suture boundary extraction and intersection identification, the system comprises:

[0013] a skeleton segmentation module for obtaining a skull three-dimensional image and segmenting the image to obtain a label region of each skeleton;

[0014] a suture boundary extraction module for performing morphological dilation and difference operation on each of the skeleton label region to extract the suture boundary at the junction of the skeleton and the adjacent skeleton;

[0015] an intersection detection module for fusing the suture boundaries of all skeletons to construct an accumulation matrix to mark the voxel positions where at least two of the skeleton boundaries coincide to obtain a candidate landmark point set;

[0016] a landmark point classification module for comparing and matching the skeleton label combination involved in each voxel in the candidate landmark point set with a pre-stored anatomical landmark point skeleton combination template to identify the landmark point category to which each candidate voxel belongs;

[0017] a coordinate determination module for calculating the representative coordinates of the landmark point of each landmark point category according to the spatial distribution of the candidate voxels when there are multiple candidate voxels corresponding to each landmark point category, and outputting the three-dimensional coordinates of all recognized skull landmark points.

[0018] Compared with the prior art, the beneficial effects of the present application are mainly as follows: first, it avoids the tedious manual annotation process, greatly improving the efficiency of skull landmark point positioning; second, the automatic identification strategy based on suture boundary overlap and intersection analysis reduces human subjective error, and the positioning result has higher consistency and objectivity; in addition, the present application makes full use of the anatomical structure information of three-dimensional CT segmentation data, and can accurately identify the key points formed by the intersection of sutures, and has a more comprehensive understanding of the complex three-dimensional structure of the skull. In summary, the recognition method provided by the present application can more accurately and efficiently obtain the anatomical landmark points of the skull, and has important significance for related medical image analysis and clinical application. DETAILED DESCRIPTION

[0019] The specific embodiments of the present application will be further described below. It should be understood that the embodiments described herein are only for the purpose of explaining the present application and not limiting the scope of the present application.

[0020] A skull landmark point recognition method based on suture boundary extraction and intersection identification, the method comprising the following steps:

[0021] (1) obtaining three-dimensional medical image data of the skull, and performing skull skeleton segmentation processing on the image to obtain a three-dimensional skull segmentation image labeled with multiple skeleton labels;

[0022] (2) performing morphological boundary extraction on each bone label region in the segmented image to obtain a set of suture boundary voxels at the junction of the bone and adjacent bones;

[0023] The suture boundary extraction is performed by executing a three-dimensional dilation operation with a voxel width on each bone label region once, and performing an exclusive OR operation between the dilation result and the original bone region to extract the boundary voxels of the bone region, wherein the influence of the background region is shielded, so that the extracted boundary only contains the junction line voxels between bones;

[0024] (3) fusing the sets of suture boundary voxels of all bones to identify intersection voxels containing at least two bone boundaries as candidate landmark positions;

[0025] (4) for each candidate landmark position, obtaining the set of bone labels involved, and matching the set with a pre-stored anatomical landmark bone combination template to identify the skull landmark category corresponding to the candidate landmark;

[0026] A skull anatomical landmark bone combination correspondence relationship is established in advance, and the names of important skull landmarks and the sets of bone labels involved are stored in a dictionary; when the set of bone labels involved in the position of a candidate landmark is completely consistent with the bone combination of a landmark in the dictionary, it is determined that the candidate position is the corresponding anatomical landmark.

[0027] When there are multiple candidate points for a certain landmark category, a single representative coordinate is obtained by calculating the geometric centroid of the candidate point set; wherein, for the nasion landmark, when multiple candidate points are distributed in a linear manner along the skull surface, the candidate point with the highest Z-axis direction coordinate value is selected as the final recognition result of the nasion.

[0028] The detection of the candidate landmark position uses a cumulative matrix fusion method to map and accumulate the boundary voxels of each bone into a three-dimensional cumulative matrix with the same size as the segmented image to mark the overlapping regions of different bone boundaries, thereby identifying the junction voxels containing at least two bones.

[0029] (5) when a certain skull landmark category corresponds to multiple candidate points, the multiple candidate points are integrated to determine the unique landmark coordinate of the category, and the spatial coordinates of each skull landmark are output.

[0030] A skull landmark recognition system based on suture boundary extraction and intersection identification, the system comprising:

[0031] A bone segmentation module for obtaining a skull three-dimensional image and segmenting the image to obtain label regions of each bone;

[0032] a suture boundary extraction module configured to perform morphological dilation and difference operation on each of the bone label regions to extract suture boundaries at the junctions of adjacent bones;

[0033] a junction detection module configured to fuse the suture boundaries of all bones to construct an accumulation matrix to mark the voxel positions where at least two bone boundaries coincide, thereby obtaining a candidate landmark set;

[0034] a landmark classification module configured to compare the bone label combination involved by each voxel in the candidate landmark set with a pre-stored anatomical landmark bone combination template to match and identify the landmark category to which each candidate voxel belongs;

[0035] a coordinate determination module configured to calculate a representative coordinate of a landmark category when multiple candidate voxels belong to the category according to the spatial distribution of the candidate voxels, and output the three-dimensional coordinates of all identified skull landmarks.

[0036] The skull landmark recognition method of the embodiment first reads the three-dimensional CT skull segmentation image data using an image processing module. For example, the image can be a volume data obtained in advance by a medical image segmentation algorithm, and each voxel is assigned a corresponding bone plate label. After the image is read, the data can be simply smoothed to remove noise points, and the voxel intensity range or resolution can be adjusted as needed to optimize the effect of subsequent boundary detection. Then, enter the boundary extraction step. The embodiment uses a three-dimensional traversal method to extract the bone plate boundary: check the voxels of the volume data one by one, and compare the labels with their six adjacent (front, back, up, down, left and right) voxels. If the label value of a voxel is different from any adjacent voxel, it is considered that the voxel is on the surface boundary of the bone plate. Collect the coordinates of all voxels that meet this condition into a set, and obtain the boundary point set of all bone plates in the current image. By this method, the contact surface between the internal bone plates of the skull and the edge of the outer surface of the skull can be captured. In implementation, a boundary point list can be maintained for each bone plate, and the boundary points are usually stored in the form of three-dimensional coordinates or indexes. After obtaining the bone plate boundary, perform suture recognition processing. The suture at the junction of two adjacent bone plates is manifested in the data as two groups of boundary points with different labels that are closely distributed. For example, assuming that bone plate A and bone plate B are adjacent, their respective boundary point clouds will present a corresponding relationship in the contact area. The feature extraction module discovers this correspondence by comparing the boundary sets of different bone plates. When a boundary point of A and a boundary point of B are close enough in space and the two positions are connected to form a continuous column (along a certain local surface), then this series of corresponding boundary points constitutes a suture line. The algorithm can use neighborhood search or iterative clustering methods to merge the boundary points belonging to the same pair of bone plates and connected to each other into a continuous line segment. For example, a boundary point can be started from, its adjacent boundary points belonging to another bone plate are found, and the direction consistent trend is repeatedly expanded until it cannot continue, thereby extracting the longest suture line segment. For the whole skull, this process will extract several suture lines, each line corresponding to a specific two bone plate combination (such as the coronal suture between the parietal bone and the frontal bone). Next, the intersection point analysis module performs intersection point extraction. In the suture line set, the spatial relationship between two suture lines is checked. If it is found that two suture lines intersect or approach in the three-dimensional space trajectory, the intersection point position is determined. Usually, the distance threshold can be used to judge the intersection point: when the nearest point distance of two suture lines is lower than the preset threshold, the average position is recorded as an intersection point. For the case involving multiple suture lines (for example, four suture lines are connected to each other in a certain area to form a complex intersection), there may be multiple close intersection points in a region. In implementation, these close intersection points can be further clustered and merged to represent a single landmark to improve positioning stability. By traversing all suture line combinations and performing the above calculation, a series of candidate landmark points can be obtained, each candidate point being described by its three-dimensional coordinates and an associated suture line (or bone plate) list.Then, the classification determining module analyzes and screens the candidate landmarks. The classification is mainly based on the number of bone plates involved and the combination type. For example, if a landmark is formed by the intersection of three bone plates, it can be marked as a "three-bone intersection point" type; if it is formed by the intersection of four bone plates, it can be marked as a "four-bone intersection point" type. For landmarks that have a special name in anatomy (such as "Lambda" corresponding to the intersection of the parietal bone and the occipital bone, etc.), they can also be identified and named according to their bone plate combinations. In the screening process, some rules can be applied: eliminate those false intersection points that may be generated due to data noise (for example, points formed by the intersection of only very short suture line segments); if multiple candidate points are very close, only a representative one can be retained. The screened landmarks will be finally determined as the anatomical landmarks of the skull and given clear identification. Finally, the detected skull landmarks are output and visualized. Through the three-dimensional visualization module, the positions of these landmarks can be highlighted on the reconstructed three-dimensional model of the skull. For example, a small sphere or cross mark can be drawn at each landmark, and a red landmark can be used. In this way, users (such as doctors) can intuitively observe the distribution of key points on the skull. If necessary, the application can also output the coordinate values of the landmarks or the index positions in the image for quantitative analysis or subsequent processing.

[0037] In order to facilitate the understanding of the above process, a pseudo code can be designed to outline the implementation logic. Through the pseudo code, the method flow of the application can be understood as follows: first, extract the bone plate boundaries in the volume data, then identify the sutures between the bone plates, then find the intersection points of these sutures and screen the final landmarks. In specific implementation, various optimizations can be made to the above process. For example, parallel computing is used to accelerate voxel checking during boundary extraction, a more robust path search algorithm is used for suture line extraction, and spatial indexing is introduced to accelerate the proximity query during intersection detection. These optimizations are within the protection scope of the application. It should be emphasized that the method of the application relies on high-precision skull segmentation images as input. If the segmentation quality of the input data is poor, the bone sutures may be interrupted or incorrectly labeled. For this purpose, appropriate error correction or completion mechanisms can be added to the algorithm, such as connecting the broken suture line segments according to anatomical knowledge, thereby improving the integrity and accuracy of landmark identification.

[0038] The scheme of the embodiment realizes the recognition of the skull landmark points through in-depth analysis of the three-dimensional CT skull segmentation image. The core is to find the sutures and their intersection points by using the spatial relationship of the bone plate boundary, and abandon the traditional method which purely relies on the gray intensity, so that the recognition is more reliable. The embodiment proves that the method can accurately locate important landmarks such as the inion, the bregma (the anterior fontanel) and the like, and has a broad application prospect in the fields of computer-aided surgery and craniofacial reconstruction. Those skilled in the art can understand that improvements or modifications can be made to the specific implementation modes without departing from the principles of the present application, but these improvements or modifications all fall within the protection scope of the claims of the present application.

Claims

1. A method for recognizing skull landmarks based on suture boundary extraction and intersection identification, characterized in that, The method includes the following steps: (1) Obtain three-dimensional medical image data of the skull and perform segmentation processing on the image of each bone of the skull to obtain a three-dimensional skull segmentation image labeled with multiple bone tags. (2) Perform morphological boundary extraction on each bone label region in the segmented image to obtain the set of voxels at the bone suture boundary at the junction of the bone and the adjacent bone. (3) Merge the set of voxels of the suture boundaries of all bones and identify the intersection voxels containing at least two kinds of bone boundaries as candidate landmark locations. (4) For each candidate landmark location, obtain the set of bone labels involved, and match the set with the pre-stored anatomical landmark bone combination template to identify the skull landmark category corresponding to the candidate landmark. (5) When a certain skull landmark category corresponds to multiple candidate points, integrate the multiple candidate points to determine the unique landmark coordinates of the category, and output the spatial coordinates of each skull landmark.

2. The method for recognizing cranial landmarks based on suture boundary extraction and intersection recognition as described in claim 1, characterized in that, In step (2), the bone seam boundary extraction is performed by performing a three-dimensional dilation operation of the voxel width on each bone label region and performing an XOR operation with the original bone region to extract the boundary voxel of the bone region. The influence of the background region is masked so that the extracted boundary only contains the boundary line voxel between bones.

3. The method for recognizing cranial landmarks based on suture boundary extraction and intersection recognition as described in claim 1 or 2, characterized in that, In step (4), a correspondence between the skeletal combinations of anatomical landmarks of the skull is established in advance, and the names of important skull landmarks and the sets of skeletal labels they involve are stored in a dictionary. When the set of skeletal labels involved in the location of a candidate landmark is completely consistent with the skeletal combination of a certain landmark in the dictionary, the candidate location is determined to be the corresponding anatomical landmark.

4. The method for recognizing cranial landmarks based on suture boundary extraction and intersection recognition as described in claim 3, characterized in that, When there are multiple candidate points for a certain marker category, a single representative coordinate is obtained by calculating the geometric centroid of the candidate point set; where, for the nasal root point marker, when there are multiple candidate points distributed linearly along the skull surface, the candidate point with the highest Z-axis coordinate value is selected as the final identification result of the nasal root point.

5. The method for recognizing cranial landmarks based on suture boundary extraction and intersection recognition as described in claim 3, characterized in that, The detection of the candidate marker positions adopts the cumulative matrix fusion method, which maps and accumulates the boundary voxels of each bone into a three-dimensional cumulative matrix of the same size as the segmented image, so as to mark the overlapping areas of different bone boundaries, thereby identifying the boundary voxels containing at least two kinds of bones.

6. A system implementing the cranial landmark recognition method based on suture boundary extraction and intersection recognition as described in claim 1, characterized in that, The system includes: The skeleton segmentation module is used to acquire a three-dimensional image of the skull and segment the image to obtain the label regions of each bone. The bone suture boundary extraction module is used to perform morphological dilation and difference operations on each of the bone label regions to extract the bone suture boundaries at the junction of the bone and the adjacent bones. The intersection detection module is used to fuse the suture boundaries of all bones, construct an accumulation matrix to mark the voxel positions where at least two types of bone boundaries overlap, and obtain a set of candidate marker points; The marker classification module is used to compare and match the bone tag combination involved in each voxel in the candidate marker set with the pre-stored anatomical marker bone combination template to identify the marker category to which each candidate voxel belongs. The coordinate determination module is used to calculate the representative coordinates of the marker points of that category based on the spatial distribution of the candidate voxels when each marker point category corresponds to multiple candidate voxels, and output the three-dimensional coordinates of all identified skull marker points.