Calcaneal fracture reduction effect dynamic evaluation system based on three-dimensional image

Abstract

The present application relates to the medical image processing technical field, specifically to a kind of calcaneus fracture reduction effect dynamic evaluation system based on three-dimensional image.The system obtains the broken bone shape feature of each bone block based on the surface gauss curvature information of bone block, determines the broken bone corresponding to the main bone of fracture by the method of classification and direction correlation analysis, quantifies broken bone index, obtains skeletal region.Using damage eigenvalue, the skeletal region is divided into to-be-registered region and damage region, in combination with the information in damage region, the stress influence direction of to-be-registered region is determined, the extrusion feature and the offset feature of bone block in stress influence direction are obtained, and the initial registration point of bone block is screened out in combination with damage eigenvalue for registration and reduction effect evaluation.The present application classifies the bone block in three-dimensional model effectively, and the initial registration point is screened out for registration by analyzing the situation affected by fracture, so that the registration effect and the accuracy of reduction evaluation are guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, and in particular to a calcaneal fracture reduction effect dynamic evaluation system based on three-dimensional images. BACKGROUND

[0002] Calcaneal fracture is a common type of foot injury in clinical practice, and the quality of surgical reduction directly affects the postoperative functional recovery and quality of life of patients. Calcaneus is an important bone for weight-bearing in the foot, and its anatomical structure is complex, making it prone to comminuted fractures in high-energy injuries. Calcaneal fractures account for about 60% of foot fractures, and severe cases can lead to loss of foot function, significantly affecting the daily life and work ability of patients.

[0003] The existing technology mainly relies on medical images for qualitative observation or angle measurement in fracture reduction evaluation. In the dynamic evaluation of three-dimensional image reduction effect, the patient's position, scanning direction, and bone tissue state may be different in CT scans at different time points. Therefore, it is necessary to register the three-dimensional models formed at different time points and unify all three-dimensional models to a reference coordinate system, thereby realizing quantitative comparison of angle changes, bone displacement, and callus formation. The existing technology mainly uses the ICP algorithm to register the models. The ICP algorithm performs initial coarse registration by manually selecting key feature points in the ankle and foot to avoid falling into local optimal solutions in the subsequent iteration process. However, in the actual registration process, due to different bone fracture types leading to different anatomical shapes of the skeleton, it is difficult to determine accurate registration points, resulting in poor registration results and affecting the accuracy of dynamic evaluation of reduction effect. SUMMARY

[0004] To solve the technical problem of inaccurate registration points leading to poor registration results and affecting reduction effect evaluation in the three-dimensional model registration process before and after fracture surgery in the existing technology, the purpose of the present application is to provide a calcaneal fracture reduction effect dynamic evaluation system based on three-dimensional images, and the technical solution adopted is as follows:

[0005] The present application proposes a calcaneal fracture reduction effect dynamic evaluation system based on three-dimensional images, which comprises:

[0006] A preoperative ankle and foot three-dimensional model construction module is used to obtain a preoperative ankle and foot three-dimensional model containing multiple bone blocks.

[0007] The bone region division module is configured to obtain a bone fragment shape feature of each bone fragment according to data distribution prominence of Gaussian curvature of the upper surface of the bone fragment; obtain a main fracture bone and a bone fragment according to the bone fragment shape feature; perform traversal in a preset traversal direction with the main fracture bone as a center, and obtain a fracture trace feature of the bone fragment in each traversal direction according to a normal direction change rate of the surface of the main fracture bone in the traversal direction, and obtain a bone fragment index of each bone fragment by combining the bone fragment shape feature; and reclassify the bone fragment and the main fracture bone according to the bone fragment index, to obtain a plurality of bone regions.

[0008] The initial registration point acquisition module is configured to obtain an injury feature value according to the bone fragment index of the bone fragment in the bone region and the number of bone fragments; classify the bone region into a registration area and an injury area by using the injury feature value; obtain a stress influence direction of the registration area according to a traversal direction distribution of the bone fragments in the injury area, obtain a compression feature and a displacement feature in the stress influence direction according to a shape distribution of the bone fragments in the registration area, and obtain an initial registration coefficient of each bone fragment in the registration area by combining the injury feature value; and screen out an initial registration point by using the initial registration coefficient.

[0009] The reduction effect evaluation module is configured to register the preoperative ankle-foot three-dimensional model and the postoperative ankle-foot based on the initial registration point, and evaluate the reduction effect according to the registered model.

[0010] Further, the method for obtaining the bone fragment shape feature comprises the following steps.

[0011] The Gaussian curvatures of the surface voxel points are arranged according to sizes to obtain a Gaussian curvature sequence; and a surface mutation point is a surface voxel point corresponding to a maximum element in a difference sequence of the Gaussian curvature sequence.

[0012] If the surface mutation point is a maximum value in the Gaussian curvature sequence, the surface mutation point and an adjacent surface voxel point are subjected to normalization processing on a Gaussian curvature difference therebetween to obtain a sharpness, the Gaussian curvature of the surface mutation point is increased according to the sharpness to obtain a sharpness weight; otherwise, the Gaussian curvature is taken as the sharpness weight of the surface mutation point.

[0013] A surface mutation vector of the bone fragment centroid is obtained, which points to each surface mutation point; and a cosine similarity of each surface mutation vector and an average surface mutation vector is obtained.

[0014] A ratio of the sharpness weight and the cosine similarity is taken as a first bone fragment saliency value of the surface mutation point; a ratio of a Gaussian curvature mean value of the surface voxel point and a surface voxel point quantity is taken as a second bone fragment saliency value; and a bone fragment shape feature of the bone fragment is obtained according to a mean value of the second bone fragment saliency value and the first bone fragment saliency value.

[0015] Further, the method for obtaining the main fracture bone and the bone fragment according to the bone fragment shape feature comprises the following steps.

[0016] The bone pieces are clustered based on the K-means clustering algorithm by using the bone piece shape features, and three bone piece clusters are obtained, wherein the bone piece cluster with the maximum average bone piece shape feature value is the collection of bone chips, the second largest is the collection of broken main bones, and the smallest is the collection of normal bone pieces.

[0017] Further, the method for obtaining the fracture trace feature comprises:

[0018] The minimum enclosing cuboid of the broken main bone is obtained, and the center point of the broken main bone is traversed in the direction of each surface of the minimum enclosing cuboid; the traversal is stopped until the normal bone piece is contained in the traversal space.

[0019] The connected domain formed by the surface mutation points of the broken main bone is a surface mutation connected domain; the normal direction change rate variance of all voxel points on the surface mutation connected domain is obtained.

[0020] If the projection area of the surface mutation connected domain on one surface of the minimum enclosing cuboid is the largest and the distance is the closest, the surface mutation connected domain is taken as the reference connected domain of the cuboid surface; the mean value of the normal direction change rate variance of all reference connected domains is multiplied by the number of reference connected domains to obtain the fracture trace feature of each cuboid surface.

[0021] If the bone chip belongs to multiple traversal spaces, the fracture trace feature of the bone chip is the fracture trace feature of the cuboid surface corresponding to the minimum traversal space; otherwise, the fracture trace feature of the bone chip is the fracture trace feature of the cuboid surface corresponding to the traversal space.

[0022] Further, the method for obtaining the bone chip index comprises:

[0023] For each bone chip, the fracture trace feature is normalized to obtain an increase ratio, and the bone piece shape feature is increased according to the increase ratio to obtain the bone chip index of the bone chip.

[0024] Further, the method for dividing the bone region comprises:

[0025] For each broken main bone, the bone piece shape feature is increased according to the maximum increase ratio to obtain the bone chip index of the broken main bone.

[0026] Based on the bone piece centroid coordinates and the bone chip index, the K-means clustering algorithm is used to cluster the broken main bones and the bone chips to obtain multiple bone regions.

[0027] Further, the method for classifying the bone regions into a to-be-registered region and a damaged region by using the damage feature value comprises:

[0028] The product of the mean value of the bone fragmentation index in each bone region and the number of bone blocks is taken as a damage feature value; threshold segmentation is performed on the damage feature value, two types of bone regions are obtained, one type with the minimum damage feature value is taken as a registration region, and the other type is taken as a damage region.

[0029] Further, the method for obtaining the stress influence direction comprises:

[0030] Optionally, one of the registration regions is taken as a target registration region;

[0031] If the target registration region is adjacent to the damage region, the sum direction of the traversal directions of the traversal space corresponding to all the fragmented bones in the damage region in the fracture trace feature assignment process is taken as the stress influence direction of the target registration region.

[0032] If the target registration region is not adjacent to the damage region, the stress influence direction of the adjacent registration region is taken as a reference direction, and the sum direction of the reference directions passing through the target registration region is taken as the stress influence direction of the target registration region.

[0033] Further, the method for obtaining the extrusion feature and the offset feature comprises:

[0034] The contact surface between the adjacent bone blocks in the registration region is obtained; on the contact surface, the surface voxel points on the contact surface are traversed in columns perpendicularly to the stress influence direction, the average distance between the surface voxel points and the adjacent bone block voxel points in each column is obtained, the average distance sequence is obtained according to the traversal order, and the absolute value of the sum of the elements of the difference sequence of the average distance sequence is normalized to obtain the extrusion feature.

[0035] The first angle between the main axis of each bone block and the XY plane of the three-dimensional space is obtained, the second angle between the line connecting the center point of the registration region and the center of the calcaneus and the XY plane is obtained, and the offset feature is obtained according to the difference between the first angle and the second angle.

[0036] Further, the method for obtaining the initial registration point comprises:

[0037] The initial registration coefficient is threshold segmented to obtain two types of initial registration coefficients, the bone block corresponding to the maximum initial registration coefficient is selected as a registration bone block, and the registration point obtained by using the ICP registration algorithm on the registration bone block is taken as the initial registration point.

[0038] The present application has the following beneficial effects:

[0039] After obtaining the three-dimensional model of the ankle-foot part, considering that various bone blocks exist in the three-dimensional model before the operation due to the fracture factor, the model registration needs to be performed by taking the bone blocks less affected by the fracture as the reference. Therefore, the application first obtains the broken bone shape features of each bone block based on the surface Gaussian curvature information of the bone block, then determines the broken bone corresponding to the broken main bone through the classification and direction correlation analysis method, and quantifies the broken bone index to obtain the bone region. Further, the damage characteristic value can divide the bone region into a registration area and a damage area, i.e., the bone block set that can be used as the registration reference. By combining the information in the damage area, the stress influence direction of the registration area can be determined, and then the extrusion feature and the offset feature of the bone block in the stress influence direction are obtained, and the initial registration coefficient is further obtained by combining the damage characteristic value. That is, the initial registration coefficient quantified by the extrusion and offset features of the bone block in the registration area can effectively represent the degree of influence of the bone block by the fracture, and then the initial registration points of the bone block can be screened for registration and reduction effect evaluation. The application can accurately identify and classify the bone blocks in the three-dimensional model, screen the accurate initial registration points for registration by analyzing the influence of the fracture, and ensure the registration effect and the accuracy of the reduction evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, below, a brief introduction will be given to the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0041] Figure 1 A block diagram of a calcaneal fracture reduction effect dynamic evaluation system based on three-dimensional images provided by an embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, below, the specific implementation, structure, features and effects of a calcaneal fracture reduction effect dynamic evaluation system based on three-dimensional images according to the present application are described in detail in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0044] Specifically, the application provides a three-dimensional image-based calcaneal fracture reduction effect dynamic evaluation system.

[0045] Please refer to Figure 1 It shows a three-dimensional image-based calcaneal fracture reduction effect dynamic evaluation system block diagram provided by an embodiment of the application, which comprises a preoperative ankle foot three-dimensional model construction module 101, a bone region division module 102, an initial registration point acquisition module 103 and a reduction effect evaluation module 104.

[0046] The preoperative ankle foot three-dimensional model construction module 101 is used to obtain a preoperative ankle foot three-dimensional model containing a plurality of bone blocks, and the purpose of the embodiment of the application is to find a bone block less affected by fracture in the preoperative ankle foot three-dimensional model and obtain a registration point. In the construction process of the preoperative ankle foot three-dimensional model, the basic information contained therein, i.e. the information of each bone block and the center point of the calcaneus, needs to be determined first. When constructing the ankle foot three-dimensional model, different bone blocks are usually treated as independent model blocks to help bone block matching and defect recognition. Herein, only the model construction method in the embodiment of the application is briefly described:

[0047] (1) A multi-slice spiral CT scanner (such as a GE Revolution 256-slice spiral CT scanner) is used to collect high-resolution ankle foot CT scan images at multiple time points (such as 1 month, 3 months and 6 months) before and after operation, and thin layer scanning (layer thickness ≤0.625 mm) is set, and the scanning range covers 1 / 3 of the tibia to the distal end of the foot;

[0048] (2) The DICOM format image is imported into 3D Slicer software for preprocessing:

[0049] (3) The collected CT images are denoised and resampled;

[0050] (4) Using the Segment Editor module, each bone block is treated as an independent Segment, and the bone part in the CT image is separated from the cartilage and soft tissue;

[0051] (5) The Marching Cubes algorithm is applied to generate an ankle foot three-dimensional model (STL format);

[0052] (6) All verified key feature points (such as the top point of the calcaneal tuberosity, the center of the posterior articular surface and the center of the calcaneus) in the preoperative ankle foot three-dimensional model are manually marked.

[0053] Each voxel point in the three-dimensional model has three-dimensional (X-axis, Y-axis, Z-axis) coordinate properties, and the CT value of the point is inherited from the original two-dimensional pixel.

[0054] In the three-dimensional model of the ankle foot, the bone fragments caused by fracture will be regarded as independent bone regions in the three-dimensional reconstruction. If the registration is directly performed on the bone fragments, the registration result will be poor. Therefore, the system proposed in the embodiment of the present application needs to analyze the properties of each bone fragment, evaluate the degree of influence of the fracture, and then select effective registration points. The bone region division module 102 mainly integrates and divides the bone fragments into multiple bone regions by analyzing the bone fracture properties of each bone fragment, and reduces the interference of the bone fragments in the fracture damage area on the selection of registration points through integrated analysis.

[0055] The bone region division module 102 considers that the broken bone and the fractured main bone have obvious corner points compared with the normal bone fragments. When the fracture occurs, the bone is forced to break and form multiple sections. At this time, the bone fragments are mostly asymmetric small blocks, and the shape is sharp and the edge is sharp. The normal bone fragments are not affected by the fracture and are smoother, and usually only the surface of the bone end point has a larger curvature area. The broken bone and the fractured main bone are both affected by the fracture of different sizes, and produce more prominent corner points. Therefore, the broken bone shape feature of each bone fragment can be obtained by the data distribution prominence of the Gaussian curvature of the bone fragment surface, that is, the more prominent the Gaussian region distribution of the surface of the bone fragment is, the more likely the corresponding bone fragment is the broken bone or the fractured main bone affected by the fracture, and the greater the broken bone shape feature is.

[0056] The specific calculation method of the Gaussian curvature is a technology known to those skilled in the art, and will not be repeated here.

[0057] Therefore, the fractured main bone and the broken bone of the bone fragment can be selected based on the broken bone shape feature. It should be noted that the selection of the fractured main bone and the broken bone of the bone fragment described in the embodiment of the present application does not mean that the bone fragments in the three-dimensional model are only these two, and in fact, it also includes normal bone fragments with smaller broken bone shape features or even no obvious broken bone features. Here, no further description is made.

[0058] The broken bone can be considered as a small piece affected by the fracture of the main bone, so it is necessary to match the broken bone with the main broken bone and analyze the fracture trace features when affected by the fracture. The bone region division module takes the main broken bone as the center, presets multiple traversal directions, i.e., traverses from the center point of the main broken bone to each traversal direction, and the broken bone passed in the traversal space can be considered as the broken bone caused by the stress in this traversal direction. Therefore, the normal change rate of the surface of the main broken bone in the traversal direction is further analyzed. In the traversal direction, the greater the fluctuation of the normal direction change rate of the surface of the main broken bone, the more irregular, rough and jagged the surface, and the more likely it is to be affected by external force to cause broken bone and form uneven and irregular bone block surface. Therefore, the fracture trace features of the broken bone in each traversal direction can be obtained based on this. Further, the fracture trace features of a certain broken bone are combined with the shape features of the broken bone, i.e., the greater the fracture trace features and the greater the shape features of the broken bone, the greater the influence of the fracture on the broken bone, and the greater the broken bone index. Based on the broken bone index, the broken bone can be reclassified with the main broken bone to obtain multiple bone regions. That is, each bone region represents a group of bone blocks, and the group of bone blocks has a clear belonging relationship. For example, the bone region containing the main broken bone and the broken bone at the same time indicates that the bone region is a fracture affected region, and the fracture has a certain influence but not much. If the bone region only contains the broken bone, it indicates that the bone region is a more obvious damage region and belongs to a larger fracture affected region, which should not be used as a registration reference in three-dimensional model registration.

[0059] In three-dimensional reconstruction and registration analysis, the spatial relationship and structural features of the bones around the calcaneus need to be analyzed to achieve more accurate model segmentation, registration, and postoperative dynamic change evaluation. The calcaneus and multiple bones form joints around the calcaneus, which have stable relative position relationship and morphological performance under normal circumstances. To further optimize the selection of initial registration position points, the relative position change features and bone damage degree of different bone blocks affected by the fracture need to be analyzed. In the ankle and foot three-dimensional model, the calcaneus is the core bone block of the hind foot, and the calcaneus fracture will cause a cascade of spatial changes in the surrounding structure. In order to ensure the accuracy of model registration, it is necessary to select as many key feature points on the bone blocks that are not affected or weakly affected as possible as the initial registration position points. In the same bone region, the damage degree of the bone blocks is relatively close, and after selecting a bone region with low damage degree, the key position points in the current bone region can be evaluated by analyzing the damage degree and damage direction of the adjacent bone regions. Therefore, after the region division by the bone region division module 102, the initial registration point module 103 can further analyze the damage features in the bone region and select the initial registration points.

[0060] The more the number of bone pieces contained in a bone region and the greater the fragmentation index of the fragmented bone, the more likely the bone region is a significant damage region with more obvious damage characteristic values. Therefore, after the initial registration module 103 obtains the damage characteristic values, the bone regions are classified into a registration region and a damage region using the damage characteristic values. Because the damage region is a significant damage region obviously affected by the fracture, it should not be used as a reference for registration. However, because the damage region is a region greatly affected by the fracture, the force-affected direction of the registration region can be obtained according to the distribution of the bone piece traversal direction in the damage region.

[0061] When the calcaneus is fractured, the angle of the calcaneus changes, the originally well-coordinated articular surface is irregular or misaligned, causing local stress concentration, resulting in displacement of the joint space, and this displacement is a force-affected factor at the moment of fracture. In general, the contact surface between the bone pieces is relatively smooth. Therefore, the distribution of the gap between the bones can be used as one of the influence parameters for selecting the initial registration position points in the registration region. In addition, the arrangement of the bones in the ankle and foot will also be affected: taking the standing posture as an example, the calcaneus is located in the lower rear region of the ankle and foot, and after the calcaneus is fractured, other bones in the ankle and foot may appear to be collapsed and offset, that is, the arrangement angle of the bones is concave towards the calcaneus, and this concave is a gravity structure-affected factor after the fracture. In general, the vertically supporting bones (such as the talus) and the laterally supporting bones (such as the cuboid) maintain the corresponding support structure. Therefore, the direction of the bone pieces can be used as another parameter for selecting the registration point position.

[0062] Therefore, analyzing the bone piece morphology distribution under the force-affected direction in the registration region can determine the extrusion characteristics and the offset characteristics of the bone pieces in the registration region after the bone pieces are affected by the external force of the fracture. The greater the extrusion characteristics, the more obvious the impact of the bone pieces on the fracture, and the bone pieces have a displacement under the obvious trend of the external force. The greater the offset characteristics, the more the bone pieces are offset from the center of the calcaneus after being affected by the external force, and the greater the impact of the fracture. Therefore, the initial registration coefficient of each bone piece in the registration region can be obtained in combination with the damage characteristic values in the registration region. Based on the initial registration coefficient, the bone pieces with greater reference can be screened out in the registration region, and then the registration points in these bone pieces can be used as the initial registration points during the initial registration.

[0063] The reduction effect evaluation module 104 can register the preoperative ankle and foot three-dimensional model and the postoperative ankle and foot based on the initial registration points, and then evaluate the reduction effect according to the registered model.

[0064] It should be noted that the process of registration realized by the embodiment of the present application is realized by using the ICP registration algorithm, and the purpose of the initial coarse registration is to preliminarily align two point clouds, reduce the spatial difference, and avoid subsequent iterations from falling into local optimization. Registration is realized by including steps such as nearest point matching and precise registration iteration, and specific technical means familiar to those skilled in the art will not be described here. After obtaining the registered model, the change of the anatomical angle (such as the subtalar joint angle, the angle formed by the posterior articular surface of the calcaneus and the connecting line from the calcaneal sulcus to the anterior process) of the preoperative and postoperative model, and the bone block coaptation degree (such as the Dice coefficient) are extracted to evaluate the recovery degree. In the embodiment of the present application, only the evaluation process in one embodiment is described, and the specific evaluation process is a technical means familiar to those skilled in the art, which will not be described here.

[0065] The specific evaluation process is:

[0066] 1. The difference between the anatomical angle and the healthy angle;

[0067] a) The Gissane (the angle formed by the posterior articular surface of the calcaneus and the connecting line from the calcaneal sulcus to the anterior process) angle of the male group is (123.3±8.5) °, and the female group is (122.7±8.9) °;

[0068] b) The Böhler (subtalar joint angle) angle of the male group is (35.98±4.34) °, and the female group is (38.31±4.43) °;

[0069] The smaller the difference between the postoperative anatomical angle and the healthy angle compared with the difference between the preoperative anatomical angle and the healthy angle, the better the reduction effect.

[0070] 2. The bone block coaptation degree evaluates the overlap rate of the postoperative bone block and the preoperative model: the Dice coefficient and the Hausdorff distance: calculate the overlap degree and the maximum surface distance of the registered model; the larger the Dice coefficient and the Hausdorff distance of the postoperative model and the preoperative model, the better the reduction effect.

[0071] It should be noted that the embodiment of the present application can also regularly perform CT examination on the calcaneus of the patient at 1 month, 3 months and 6 months after the operation, repeat the modeling and registration process, and obtain three-dimensional models at consecutive time points for observing the dynamic trend of bone healing, bone block stability and remodeling. Other evaluation periods can also be set in other embodiments of the present application, which will not be limited and described here.

[0072] In summary, after obtaining the ankle-foot three-dimensional model, the bone fragment shape features of each bone block are obtained based on the surface Gaussian curvature information of the bone block, then the broken bone corresponding to the main bone is determined by the classification and direction correlation analysis method, and the bone fragment index is quantified to obtain the bone region. The bone region can be divided into a registration region and a damage region by using the damage feature value, and the stress influence direction of the registration region can be determined by combining the information in the damage region, and then the extrusion feature and the offset feature of the bone block in the stress influence direction are obtained, and the initial registration coefficient is further obtained by combining the damage feature value. The initial registration points of the bone block are screened out for registration and reduction effect evaluation. The accurate initial registration points are screened out for registration by effectively identifying and classifying the bone blocks in the three-dimensional model, and the registration effect and the accuracy of the reduction evaluation are ensured.

[0073] Preferably, in an embodiment of the present application, the method for obtaining the bone fragment shape feature comprises:

[0074] The surface voxel points are arranged according to the size of the Gaussian curvature to obtain a Gaussian curvature sequence, and the surface voxel point corresponding to the maximum element in the difference sequence of the Gaussian curvature sequence is a surface mutation point. That is, the surface mutation point can be regarded as an angle point formed by the jagged texture on the surface of the bone block, because the surface mutation point is the surface voxel point corresponding to the maximum element in the difference sequence of the Gaussian curvature sequence, which means that the point has a significantly prominent Gaussian curvature compared with other surface voxel points. It should be noted that the Gaussian curvature is arranged in ascending order in the embodiment of the present application, and the difference sequence is the value obtained by subtracting the previous element from the next element in the Gaussian curvature sequence, that is, each element in the difference sequence corresponds to two surface voxel points. After screening out the maximum element in the difference sequence, the surface voxel point with larger Gaussian curvature is selected as the surface mutation point from the two corresponding surface voxel points. Moreover, there can be multiple maximum elements in the difference sequence, that is, the surface mutation point can contain multiple points.

[0075] If the surface mutation point is the maximum value in the Gaussian curvature sequence, it means that the surface mutation point is an obvious bone block angle point, and the sharpness at this position can be evaluated by the Gaussian curvature difference between the surface mutation point and the adjacent surface voxel points. Therefore, the absolute value of the difference between the Gaussian curvature of the surface mutation point and each adjacent surface voxel point is calculated, the mean value of all the calculated difference absolute values is obtained, and then the sharpness is obtained by normalization processing. According to the increase of the Gaussian curvature of the surface mutation point according to the sharpness, the sharpness weight is obtained; otherwise, if the surface mutation point is not the maximum value in the Gaussian curvature sequence, the Gaussian curvature is taken as the sharpness weight of the surface mutation point.

[0076] It should be noted that the normalization algorithm in the embodiment of the present application can select the range standardization, and the value range of the sharpness before normalization is counted, and the sharpness is obtained based on the maximum value and the minimum value. Because the value range of the sharpness after normalization is between 0 and 1, the positive integer 1 can be added to the sharpness to obtain the magnification, that is, the value range of the magnification is between 1 and 2, and the magnification is directly multiplied by the Gaussian curvature to obtain the sharpness weight.

[0077] It should be noted that because the Gaussian curvature sequence is arranged in size order, the maximum value is the maximum value in the sequence.

[0078] The surface mutation vector of the centroid of the bone block points to each surface mutation point, and the cosine similarity between each surface mutation vector and the average surface mutation vector is obtained. The smaller the cosine similarity, the more the surface mutation vector of the corresponding surface mutation point deviates from the overall vector.

[0079] Because the greater the sharpness weight and the smaller the cosine similarity, the more the surface mutation point belongs to the obvious protruding, irregular and jagged corner point in the bone block. Therefore, the ratio of the sharpness weight to the cosine similarity is taken as the first bone fragment significant value of the surface mutation point. Further, the greater the Gaussian curvature and the smaller the number of surface voxel points, the smaller and more irregular the bone block, and the more likely it is to be a bone fragment generated by a fracture, so the ratio of the average Gaussian curvature of the surface voxel point to the number of surface voxel points is taken as the second bone fragment significant value. The bone fragment shape feature is obtained according to the average of the second bone fragment significant value and the first bone fragment significant value.

[0080] In the embodiment of the present application, after the average of the first bone fragment significant value is calculated, the bone fragment shape feature is obtained by multiplying the second bone fragment significant value.

[0081] Preferably, in the embodiment of the present application, the broken main bone and the bone fragment are obtained according to the bone fragment shape feature, comprising:

[0082] The bone block is clustered based on the K-means clustering algorithm by using the bone fragment shape feature to obtain three types of bone block clusters. The bone block cluster with the maximum average value of the bone fragment shape feature is the set of bone fragments, the second largest is the set of broken main bones, and the smallest is the set of normal bone blocks. That is, three types of clustering clusters are obtained by setting K equal to 3 for clustering. The specific clustering algorithm is a routine technical means for those skilled in the art, which is not described here.

[0083] Preferably, in the embodiment of the present application, the method for obtaining the fracture trace feature comprises:

[0084] The minimum surrounding cuboid of the fractured main bone is obtained, and the center point of the fractured main bone is traversed in the direction of each surface of the minimum surrounding cuboid as a unit. The traversal is stopped until a normal bone block is contained in the space. That is, six directions are set for traversal in the embodiment of the present application, and in the traversal process, each surface of the cuboid is taken as a scanning surface, and the step is 1 for step-by-step traversal in the three-dimensional space.

[0085] The connected domain formed by the surface mutation points of the fractured main bone is a surface mutation connected domain. It should be noted that the connected domain obtained by directly detecting the surface mutation points may not be complete, and therefore, in the embodiment of the present application, a morphological closing operation is further used to fill the pores in the initially obtained connected domain to obtain the surface mutation connected domain.

[0086] The variance of the normal direction change rate of all voxel points on the surface mutation connected domain is obtained. In the embodiment of the present application, for a voxel point, the normal is obtained, and the average of the included angle between the normal and the normals of adjacent voxel points is taken as the normal change rate. The greater the variance of the change rate, the more uneven the current surface connected domain, and the greater the fracture trace of the fracture surface of the fractured main bone.

[0087] If the surface mutation connected domain has the maximum projection area and the closest distance on one surface of the minimum surrounding cuboid, the surface mutation connected domain is taken as the reference connected domain of the surface of the cuboid. That is, these reference connected domains are the fracture surfaces of the bone fracture in the direction of the surface of the cuboid. Therefore, the mean value of the normal direction change rate variance of all reference connected domains corresponding to the surface of the cuboid is normalized by the product of the number of reference connected domains to obtain the fracture trace feature of each surface of the cuboid. That is, the greater the normal direction change rate variance and the greater the number of reference connected domains, the more significant the fracture trace in the direction of the surface of the cuboid, and the greater the fracture trace feature.

[0088] If the broken bone belongs to multiple traversal spaces, the fracture trace feature of the broken bone is the fracture trace feature of the surface of the cuboid corresponding to the minimum traversal space. Because the smaller the traversal space, the closer the broken bone to the fractured main bone, and the more the broken bone belongs to the direction of the fracture. Otherwise, if the broken bone belongs to only one traversal space, the fracture trace feature of the broken bone is the fracture trace feature of the surface of the cuboid corresponding to the traversal space.

[0089] Preferably, in the embodiment of the present application, the greater the fracture trace feature and the greater the shape feature of the broken bone, the more the broken bone belongs to the small bone piece caused by the fracture, and therefore, for each broken bone, the fracture trace feature is normalized to obtain an increase ratio, and the shape feature of the broken bone is increased according to the increase ratio to obtain a broken bone index of the broken bone. It should be noted that the increase method of the increase ratio is the same as the increase method of the sharpness weight increase ratio, which is not described herein.

[0090] Preferably, in the embodiments of the present application, the method for dividing the bone region comprises:

[0091] For the bone pieces with higher bone fragment indexes, because they are the bone fragments with more obvious characteristics generated by the broken main bone, i.e., they are the bone pieces with more obvious fracture influence as the broken main bone, the bone pieces with higher bone fragment indexes should be divided into the same region as the broken main bone in the bone region division; the bone pieces with lower bone fragment indexes are divided into the same bone region. The bone pieces with more obvious fracture influence are reduced to interfere with the registration points of the complete bone pieces.

[0092] Therefore, for each broken main bone, the bone fragment shape characteristics are enlarged according to the maximum enlargement ratio to obtain the bone fragment index of the broken main bone. The specific enlargement method is not described again. After the enlargement, the broken main bone has a larger bone fragment index, so the K-means clustering algorithm can be used to cluster the broken main bone and the bone fragments based on the bone piece centroid coordinates and the bone fragment index to obtain a plurality of bone regions. The K value of the clustering algorithm can be obtained by the elbow method, and finally a plurality of bone regions are obtained. The specific clustering algorithm is not described again.

[0093] Preferably, in the embodiments of the present application, the method for classifying the bone regions into the registration region and the damage region by using the damage feature value comprises:

[0094] The product of the average value of the bone fragment index and the number of bone pieces in each bone region is taken as the damage feature value. That is, the larger the bone fragment index and the more the bone pieces in a bone region, the greater the damage feature value, indicating that the fracture influence is greater and more bone fragments are generated. The damage feature value is used for threshold segmentation to obtain two types of bone regions. The type with the smallest damage feature value is taken as the registration region, and the other type is taken as the damage region. It should be noted that the threshold segmentation can adopt the Otsu threshold method, or the two types of regions can be obtained by sorting according to the size and then selecting the point with the largest difference between adjacent elements as the segmentation point. The specific method is not described again.

[0095] Preferably, in the embodiments of the present application, the method for obtaining the stress influence direction comprises:

[0096] Optionally, one registration region is taken as a target registration region;

[0097] If the target registration region and the damage region are adjacent, because they are adjacent, the damage region is a region with significant fracture influence, so the traversal direction in the damage region can be taken as a reference. The sum direction of the traversal directions of all the bone fragments in the damage region in the fracture trace feature assignment process is taken as the stress influence direction of the target registration region;

[0098] If the target to-be-registered region is not adjacent to the damaged region, the adjacent to-be-registered region whose stress influence direction has been determined can be taken as a reference, the stress influence direction of the adjacent to-be-registered region is taken as a to-be-referenced direction, and the sum direction of the to-be-referenced directions passing through the target to-be-registered region is taken as the stress influence direction of the target to-be-registered region.

[0099] Preferably, in the embodiments of the present application, the method for obtaining the extrusion feature and the offset feature comprises:

[0100] The contact surface between the adjacent bone blocks in the to-be-registered region is obtained. In the embodiments of the present application, the contact surface between the adjacent bone blocks can be directly determined by using the nearest neighbor and normal consistency, that is, the surface closest to the to-be-registered region and the surface with the most consistent normal direction in the three-dimensional space are taken as the contact surface.

[0101] On the contact surface, the surface voxel points on the contact surface are traversed in columns perpendicularly to the stress influence direction, and the average distance between the surface voxel points in each column and the adjacent bone block is obtained. The average distance sequence is obtained according to the traversal order, and the absolute value of the sum of the elements of the difference sequence of the average distance sequence is normalized to obtain the extrusion feature. That is, if the elements of the difference sequence are all of the same sign, that is, all positive or all negative, it indicates that the voxel points at different positions have a consistent distance change trend, for example, all tend to be close to the adjacent bone block or all tend to be far away from the adjacent bone block, and therefore the greater the absolute value of the sum of the elements, the greater the extrusion feature under the influence of the external force. The hyperbolic tangent function mapping method can be selected as the normalization method.

[0102] It should be noted that the method for obtaining the average distance between the surface voxel points in each column and the adjacent bone block is that the distance between each surface voxel point in a column and the adjacent bone block is obtained, and then the average distance is obtained by averaging the distances of all the surface voxel points in the column.

[0103] The first angle between the principal axis of each bone block and the XY plane of the three-dimensional space is obtained, and the second angle between the line connecting the center point of the to-be-registered region and the center of the calcaneus and the XY plane is obtained. Since both angles are obtained with the XY plane as a reference, the smaller the difference between the two angles, the closer the principal axis direction of the bone to the direction of the region and the center of the calcaneus, that is, the greater the probability of collapse offset of the bone, that is, the greater the offset feature. Therefore, the offset feature can be obtained according to the difference between the first angle and the second angle. In the embodiments of the present application, the reciprocal of the difference between the two angles is taken as the offset feature.

[0104] It should be noted that because the extrusion feature, the offset feature of a bone block and the damage feature value of the corresponding to-be-registered region are all negatively correlated with the registration reference degree, the reciprocal of the product of the three features is taken as the initial registration coefficient in the embodiments of the present application. That is, the greater the initial registration coefficient, the more the registration point on the bone block should be taken as the initial registration point.

[0105] Preferably, in the embodiments of the present application, the method for obtaining the initial registration points comprises:

[0106] The initial registration coefficients are threshold segmented to obtain two types of initial registration coefficients, and the bone block corresponding to the largest type of initial registration coefficients is selected as the registration bone block. The registration points obtained on the registration bone block by using the ICP registration algorithm are used as the initial registration points. It should be noted that the threshold segmentation herein is the same as the threshold segmentation method of the bone region described above, and will not be described here.

[0107] It should be noted that, in order to ensure the accuracy of the initial registration, at least three initial registration points should be selected, that is, at least three registration bone blocks are selected. When the selected registration bone blocks do not meet the three conditions, the bone blocks with larger initial registration coefficients can be selected based on the size of the initial registration coefficients until the number condition is met.

[0108] It should be noted that: the above-mentioned embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or may be advantageous.

[0109] Each of the embodiments in the specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments.

Claims

1. A system for dynamic evaluation of the reduction effect of calcaneal fracture based on three-dimensional images, characterized in that, The system comprises: A preoperative ankle and foot three-dimensional model construction module is configured to obtain a preoperative ankle and foot three-dimensional model comprising a plurality of bone blocks; A bone region division module is configured to obtain a broken bone shape feature of each bone block according to a data distribution prominence of a Gaussian curvature of an upper surface of the bone block; obtain a main broken bone and a broken bone according to the broken bone shape feature; perform traversal in a preset traversal direction with the main broken bone as a center; obtain a broken trace feature of the broken bone in each traversal direction according to a normal direction change rate of a surface of the main broken bone in the traversal direction; obtain a broken bone index of each broken bone by combining the broken bone shape feature and the broken trace feature; and reclassify the broken bone and the main broken bone according to the broken bone index to obtain a plurality of bone regions; An initial registration point acquisition module is configured to obtain a damage feature value according to the broken bone index of the broken bone and the number of bone blocks in the bone region; classify the bone region into a registration area to be registered and a damage area by using the damage feature value; obtain a stress influence direction of the registration area to be registered according to a traversal direction distribution of the bone blocks in the damage area; obtain a compression feature and a displacement feature in the stress influence direction according to a shape distribution of the bone blocks in the registration area to be registered; and obtain an initial registration coefficient of each bone block in the registration area to be registered by combining the damage feature value; and screen out an initial registration point by using the initial registration coefficient; A reduction effect evaluation module is configured to register the preoperative ankle and foot three-dimensional model and a postoperative ankle and foot based on the initial registration point; and evaluate a reduction effect according to the registered model.

2. The system for dynamic evaluation of the reduction effect of calcaneal fracture based on three-dimensional images according to claim 1, characterized in that, The method for obtaining the broken bone shape feature comprises: Arranging the Gaussian curvatures of the surface voxel points according to sizes to obtain a Gaussian curvature sequence; and a surface mutation point is a surface voxel point corresponding to a maximum element in a difference sequence of the Gaussian curvature sequence; If the surface mutation point is a maximum value in the Gaussian curvature sequence, performing normalization processing on a mean value of a Gaussian curvature difference between the surface mutation point and an adjacent surface voxel point to obtain a sharpness, increasing the Gaussian curvature of the surface mutation point according to the sharpness to obtain a sharpness weight; otherwise, taking the Gaussian curvature as the sharpness weight of the surface mutation point; Obtaining a surface mutation vector of the bone block centroid pointing to each surface mutation point, and obtaining a cosine similarity of each surface mutation vector and an average surface mutation vector; Taking a ratio of the sharpness weight and the cosine similarity as a first broken bone saliency value of the surface mutation point; taking a ratio of a mean value of the Gaussian curvature of the surface voxel point and a number of the surface voxel points as a second broken bone saliency value; and obtaining the broken bone shape feature of the bone block according to a mean value of the second broken bone saliency value and the first broken bone saliency value.

3. The system for dynamic evaluation of the reduction effect of calcaneal fracture based on three-dimensional images according to claim 2, characterized in that, The method for obtaining the broken bone shape feature comprises: Clustering the bone blocks based on a K-means clustering algorithm by using the broken bone shape feature to obtain three types of bone block clusters, wherein a bone block cluster with a maximum average value of the broken bone shape feature is a collection of broken bones, a second maximum is a collection of main broken bones, and a minimum is a collection of normal bone blocks.

4. The system for dynamic evaluation of the reduction effect of calcaneal fracture based on three-dimensional images according to claim 3, characterized in that, The method for obtaining the broken trace feature comprises: Obtaining a minimum enclosing cuboid of the main broken bone, taking each surface of the minimum enclosing cuboid as a unit, and performing traversal in a direction of the six surfaces from the center point of the main broken bone; and stopping the traversal until the normal bone block is contained in the traversal space. The connected domain formed by the surface mutation points of the broken main bone is a surface mutation connected domain; a normal direction change rate variance of all voxel points on the surface mutation connected domain is obtained; If the surface mutation connected domain has the maximum projection area on a surface of the smallest enclosing cuboid and the closest distance, the surface mutation connected domain is taken as a reference connected domain of the surface of the cuboid; a mean value of the normal direction change rate variances of all the reference connected domains is multiplied by a number of the reference connected domains to obtain a fracture trace feature of each surface of the cuboid; If the broken bone belongs to multiple traversal spaces, the fracture trace feature of the broken bone is a fracture trace feature of a surface of a cuboid corresponding to a minimum traversal space; otherwise, the fracture trace feature of the broken bone is a fracture trace feature of a surface of a cuboid corresponding to a traversal space.

5. The system for dynamic evaluation of the reduction effect of calcaneal fracture based on three-dimensional images according to claim 1, characterized in that, The method for obtaining the broken bone index comprises: For each broken bone, the fracture trace feature is normalized to obtain an enlargement ratio, and the broken bone shape feature is enlarged according to the enlargement ratio to obtain a broken bone index of the broken bone.

6. The system for dynamic evaluation of the reduction effect of calcaneal fracture based on three-dimensional images according to claim 5, characterized in that, The method for dividing the bone region comprises: For each broken main bone, the broken bone shape feature is enlarged according to the maximum enlargement ratio to obtain a broken bone index of the broken main bone; Based on the bone block centroid coordinates and the broken bone index, the K-means clustering algorithm is used to cluster the broken main bones and the broken bones to obtain a plurality of bone regions.

7. The system for dynamic evaluation of the reduction effect of calcaneal fracture based on three-dimensional images according to claim 1, characterized in that, The method for classifying the bone regions into a registration region and a damage region based on the damage feature value comprises: The product of the mean value of the broken bone index and the number of bone blocks in each bone region is taken as a damage feature value; the damage feature value is used for threshold segmentation to obtain two types of bone regions, one type with the minimum damage feature value is taken as the registration region, and the other type is taken as the damage region.

8. The system for dynamic evaluation of the reduction effect of calcaneal fracture based on three-dimensional images according to claim 4, characterized in that, The method for obtaining the stress influence direction comprises: An optional registration region is taken as a target registration region; If the target registration region and the damage region are adjacent, a sum direction of the traversal directions of all the broken bones in the damage region in the fracture trace feature assignment process is taken as the stress influence direction of the target registration region; If the target registration region is not adjacent to the damage region, a stress influence direction of an adjacent registration region is taken as a reference direction, and a sum direction of the reference directions passing through the target registration region is taken as the stress influence direction of the target registration region.

9. The system for dynamic evaluation of the reduction effect of calcaneal fracture based on three-dimensional images according to claim 1, characterized in that, The method for obtaining the extrusion feature and the offset feature comprises: A contact surface between adjacent bone blocks in the registration region is obtained; on the contact surface, a surface voxel point on each column is obtained by traversing the surface voxel points on the contact surface perpendicularly to the stress influence direction; an average distance sequence is obtained according to the traversal order; an absolute value of an element sum of a difference sequence of the average distance sequence is normalized to obtain the extrusion feature; A first included angle between a principal axis of each bone block and an XY plane of a three-dimensional space is obtained; a second included angle between a line connecting a center point of the registration region and a center of the calcaneus and the XY plane is obtained; the offset feature is obtained according to a difference between the first included angle and the second included angle.

10. The system for dynamic evaluation of the reduction effect of calcaneal fracture based on three-dimensional images according to claim 1, characterized in that, The method for obtaining the initial registration point comprises: The initial registration coefficients are threshold segmented to obtain two types of initial registration coefficients, and a bone block corresponding to a largest type of initial registration coefficients is selected as a registration bone block, and a registration point obtained by using an ICP registration algorithm on the registration bone block is used as an initial registration point.

Images