Multi-modal pelvic fracture data generation method and system, medium, program and electronic terminal

Abstract

The invention provides a multi-mode pelvic fracture data generation method and system, a medium, a program and an electronic terminal, and the method comprises the steps: extracting a skeleton mask image from an original CT image, and carrying out the Euclidean distance transformation of the skeleton mask image, so as to obtain a distance transformation image; a watershed segmentation algorithm is utilized to generate a first watershed label graph with an initial fracture line along a mechanical weak area, and affine transformation of controlled Gaussian noise and biomechanical constraint is introduced, so that a second watershed label graph with form diversity and biomechanical authenticity is generated to represent a real pelvic fracture form; and meanwhile, the corresponding 6DOF parameters are output, so that the problem of scarcity of clinical data is effectively relieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical image processing, in particular to a multi-modal pelvic fracture data generation method, system, medium, program and electronic terminal. BACKGROUND

[0002] In the complex pelvic fracture scenario, a large amount of data is needed to establish a reliable algorithm and preoperative decision-making process. However, the number of complex pelvic fracture cases caused by severe trauma is limited. In addition, the CT / MRI images of each patient need to be labeled by a senior orthopedic specialist layer by layer. The time-consuming single-labeling process can take several hours, which is inefficient. In addition, medical privacy protection policies strictly limit the sharing and circulation of patient data, resulting in a severe shortage of relevant data samples (such as pelvic fracture morphology structure diagrams and the displacement path of bone fragments during the fracture process).

[0003] Traditional pelvic fracture generation methods (such as using a generative adversarial network (GAN) to generate relevant data samples) can synthesize images, but it is difficult to strictly constrain the anatomical rationality and spatial distribution of the fragments, and the generated results often deviate from the real clinical situation. SUMMARY

[0004] In view of the above-mentioned shortcomings of the prior art, the purpose of the present application is to provide a multi-modal pelvic fracture data generation method, system, medium, program and electronic terminal to solve the above-mentioned problems.

[0005] To achieve the above-mentioned purposes and other related purposes, the first aspect of the present application provides a multi-modal pelvic fracture data generation method, comprising: obtaining a bone mask image; wherein the bone mask image is a mask CT image containing only target pelvic bone information; performing Euclidean distance transformation on the bone mask image to obtain a distance transformation image representing the internal structure of the bone; performing a marker-based watershed segmentation algorithm on the distance transformation image to generate a first watershed label image with an initial fracture morphology; adding Gaussian noise to the first watershed label image to dynamically perturb the water level parameter and generate a bone fragment boundary that conforms to the clinical variability; performing affine transformation on the bone fragments in the first watershed label image with Gaussian noise to obtain a second watershed label image with a final fracture morphology and / or 6DOF displacement parameters of the affine transformation; wherein the affine transformation includes translation and / or rotation transformation limited by biomechanical rules.

[0006] In an embodiment of the first aspect of the present application, the effectiveness screening is performed on the initial bone fragments before the affine transformation is performed after the Gaussian noise is added to the initial bone fragments; wherein the effectiveness screening method comprises: calculating the number of voxels v contained in each bone fragment in the first watershed label graph; taking the bone fragment with the largest number of voxels v as the main bone fragment; sequentially judging whether the number of voxels v of each bone fragment is greater than a preset first threshold k1, if v≤k1, merging the bone fragment into the main bone fragment to form a new main bone fragment; if v>k1, retaining the bone fragment.

[0007] In an embodiment of the first aspect of the present application, the process of the affine transformation comprises: performing a rotation or translation transformation on the transformed bone fragment; judging whether the transformed bone fragment meets the biomechanical rule; if yes, accepting the rotation or translation transformation and stopping the transformation; if not, backtracking the rotation or translation transformation and re-performing a rotation or translation transformation; wherein, after backtracking the rotation or translation transformation and before performing the rotation or translation transformation again, judging whether the total number of rotation and / or translation transformations performed on the transformed bone fragment reaches n times, if yes, stopping the transformation and merging the transformed bone fragment into the bone fragment with the largest number of voxels; if not, allowing the next rotation or translation transformation to be performed.

[0008] In an embodiment of the first aspect of the present application, the biomechanical rule comprises: the voxel overlap rate of the transformed bone fragment and any other bone fragment does not exceed a second threshold k2, wherein k2∈(0,1).

[0009] In an embodiment of the first aspect of the present application, the biomechanical rule comprises: judging whether the transformed bone fragment contains an articular surface; when the transformed bone fragment contains an articular surface, the center of rotation is limited in the spherical domain (c, r) with the anatomical center c of the articular surface as the spherical center and the physiological radius r as the radius; wherein the anatomical center c is the area centroid obtained after surface fitting of the articular surface or the geometric center of the corresponding anatomical structure, and the r is a third threshold preset based on clinical statistical data; when the transformed bone fragment is a fragment without an articular surface, the contact surface centroid between the transformed bone fragment and any other bone fragment in contact with it is taken as the center of rotation.

[0010] In an embodiment of the first aspect of the present application, the biomechanical rule comprises: when the transformed bone fragment is a hip bone fragment, the articular surface of the transformed bone fragment is A, the femoral head contact surface point cloud is B, the nearest point x of the sampling point y on A on B is calculated; the average nearest neighbor distance between x and y is calculated Wherein, m is the number of sampling points on A; the (x, y) point set of d≤dmax is reserved, and the average value of the opposite normal angle of each point in the point set is calculated Wherein, a is the unit normal of A at y, and b is the unit normal of B at x obtained by local surface fitting; let The average value of the opposite normal angle of each point in the point set is calculated

[0011] To achieve the above object and other related objects, the second aspect of the present application provides a multi-modal pelvic fracture data generation system, comprising: an image acquisition module configured to acquire a bone mask image; wherein the bone mask image is a mask CT image containing only target pelvic bone information; a distance transformation module configured to perform Euclidean distance transformation on the bone mask image to obtain a distance transformation image representing the internal structure of the bone; a watershed segmentation module configured to perform a marker-based watershed segmentation algorithm on the distance transformation image to generate a first watershed label image with an initial fracture morphology; a noise adding module configured to add Gaussian noise to the first watershed label image and dynamically perturb the water level parameter to generate a bone fragment boundary conforming to clinical variability; and an affine transformation module configured to perform affine transformation on the bone fragments in the first watershed label image with Gaussian noise added to obtain a second watershed label image with a final fracture morphology and / or 6DOF displacement parameters of the affine transformation; wherein the affine transformation includes translation transformation and / or rotation transformation limited by biomechanical rules.

[0012] To achieve the above object and other related objects, the third aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method described in any of the preceding aspects.

[0013] To achieve the above object and other related objects, the fourth aspect of the present application provides a computer program product comprising computer program code, which, when executed on a computer, causes the computer to implement the method described in any of the preceding aspects.

[0014] To achieve the above object and other related objects, the fifth aspect of the present application provides an electronic terminal comprising a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to implement the method described in any of the preceding aspects.

[0015] As described above, the present application has the following beneficial effects:

[0016] The application takes the original CT image as the starting point, first extracts the bone mask image containing only the target pelvic skeleton from the original CT image, accurately limits the scope of subsequent fracture morphological transformation, then performs Euclidean distance transformation on the bone mask image to obtain a distance field calculated according to the physical distance of voxels, so that the thickness, inside-outside relationship and approximate center axis of the bone are explicitly described in numerical form, then performs label-based three-dimensional watershed segmentation on the distance field, generates a first watershed label image with an initial fracture morphology by controlling flooding and partitioning, so that the initial fracture line can be formed preferentially near the mechanical center axis of the skeleton, that is, it expands along the mechanical weak area of the skeleton, thereby more conforming to the real fracture biomechanical mechanism in morphology. In the first watershed label image, Gaussian noise is added to dynamically disturb the water level parameter to generate a bone fragment boundary that conforms to the clinical variability, and by introducing controlled randomness, a large number of diversified morphologically variable fracture morphologies with statistical significance can be output. Finally, affine transformation (translation transformation and / or rotation transformation limited by biomechanical rules) is performed on the bone fragments in the first watershed label image with Gaussian noise added, to obtain a second watershed label image with a final fracture morphology and / or 6DOF displacement parameters of the affine transformation, so that gaps or overlaps between the bone fragments are formed under the action of muscle traction and external force, making the pelvic fracture morphology more realistic, and effectively solving the problem of scarcity of real fracture data. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart showing a method for generating multi-modal pelvic fracture data according to an embodiment of the present application.

[0018] Figure 2 A flowchart showing the effectiveness screening of initial bone fragments according to an embodiment of the present application.

[0019] Figure 3 A flowchart showing the affine transformation according to an embodiment of the present application.

[0020] Figure 4 A mask CT image containing only target pelvic skeleton information according to an embodiment of the present application.

[0021] Figure 5 A first watershed label image with Gaussian disturbance added according to an embodiment of the present application.

[0022] Figure 6 A second watershed label image obtained after affine transformation according to an embodiment of the present application.

[0023] Figure 7 A structural diagram of a system for generating multi-modal pelvic fracture data according to an embodiment of the present application.

[0024] Figure 8 Fig. 1 shows a schematic diagram of an electronic terminal according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] The above and other advantages and features of the present application will become apparent from the following description of the embodiments, taken in conjunction with the accompanying drawings, which illustrate, by way of example, the principles of the application. This description is given for the sake of example and the details are not intended to limit the present application. The present application can be implemented in many different ways and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the scope of the application to those skilled in the art. Other embodiments can be employed, and logical substitutions and changes can be made without departing from the spirit of the present application. It is understood that the embodiments shown in the drawings are merely meant to be illustrative and that other embodiments can be used, and that logical substitutions and changes can be made without departing from the spirit of the present application. Embodiments and features of embodiments described hereinbelow can be combined with each other unless specifically stated otherwise.

[0026] In the embodiments of the present application, the terms "first", "second", and the like are used to distinguish between similar or identical items or components having substantially the same function and role. For example, first XX and second XX are merely used to distinguish between different XXs, and do not limit the order. Those skilled in the art can understand that the terms "first", "second", and the like do not limit the number and execution order, and the terms "first", "second", and the like do not necessarily mean different.

[0027] It should be noted that in the embodiments of the present application, the words "exemplary" or "for example" mean an example, an illustration, or an illustration. Any embodiment or design scheme described as "exemplary" or "for example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the use of the words "exemplary" or "for example" is intended to present the relevant concept in a specific manner.

[0028] In the embodiments of the present application, "at least one" means one or more, and "multiple" means two or more. The association relationship between the associated objects is described, which means that there can be three relationships, for example, A and / or B, which means that A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after it. "At least one of the following" or similar expressions means any combination of these items, including any combination of single item or multiple items. For example, at least one of a, b, or c can mean a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0029] As shown in Figure 1 The first aspect of the present application provides a method for generating multi-modal pelvic fracture data, comprising:

[0030] S1: Obtain a bone mask image; wherein, the bone mask image is a masked CT image containing only information about the target pelvic bones.

[0031] The original 3D CT image includes not only the target pelvic bones, but also abdominal organs, muscles, subcutaneous fat, air, the scan bed, and other bony areas unrelated to the target. For example... Figure 4 As shown, the masked CT image is an image obtained by masking the original CT image, used to display the target pelvic bone region separately. The masking process is as follows: the original CT image is multiplied by a binary mask voxel by voxel. For example, the target pelvic bone region is labeled as 1, and non-target regions (such as the aforementioned abdominal organs, muscles, subcutaneous fat, air, scanning bed, and other non-target bony regions) are labeled as 0, so that the grayscale of non-target voxels is masked (set to 0 or uniformly set to background value), and only the original HU grayscale of the voxels corresponding to the target pelvic bones is retained, thereby obtaining a masked CT image that only displays the target pelvic bones; at the same time, the corresponding binary mask map is output as the index region for subsequent distance transformation, watershed segmentation, and affine transformation.

[0032] It should be understood that the CT images used to obtain skeletal mask images described in this invention include images of the pelvis of a healthy individual with intact anatomy and no fractures or defects.

[0033] Preferably, the process for acquiring the CT image after masking is as follows:

[0034] S11: Standardize the CT values ​​(HU values) of the original 3D CT images to eliminate intensity differences and enhance tissue contrast.

[0035] S12: Within the pre-selected pelvic region (ROI), a histogram-based adaptive dual-threshold algorithm is used for bone tissue segmentation: First, a high-threshold region is selected as a seed point for strong bone tissue. Then, a three-dimensional region growing algorithm is used to merge medium-strength bone voxels that are spatially connected to the seed point and meet the preset low-threshold condition to generate the initial bone segmentation region.

[0036] S13: Perform three-dimensional connected domain analysis on the initial skeletal region: First, remove noisy regions with too few voxels. Then, based on the anatomical location of each connected domain and its spatial matching degree with the standard pelvic atlas, select the connected domains belonging to the pelvis. Based on morphological prior knowledge, exclude non-target skeletal structures such as the femur and lumbar vertebrae.

[0037] S14: Apply three-dimensional morphological opening and closing operations to eliminate minor noise and connect tiny fractures, while filling internal holes to obtain a skeletal mask with a complete structure and smooth boundaries.

[0038] S15: Detect whether there is a metal implant artifact in the image, if there is an artifact, identify and mask it, or re-segment the image after pre-processing with a metal artifact correction algorithm if necessary, to ensure the quality of the final bone mask.

[0039] The mask post-CT image and the corresponding binary mask obtained through the above steps S11-S15 can ensure that the cortical bone structure is continuous and unbroken, the overall topological relationship is complete, and the interference of common artifacts such as metal implants is effectively suppressed. The output result can achieve a high standard in terms of shape accuracy and noise control, thereby providing clear, stable and structurally complete input data for subsequent distance transformation, watershed segmentation and affine transformation, and ensuring the accuracy and reliability of the subsequent processing flow.

[0040] S2: Perform Euclidean distance transformation on the bone mask image to obtain a distance transformation image representing the internal structure of the bone.

[0041] Euclidean distance transformation is an algorithm for converting a binary image into a grayscale image. In the generated Euclidean distance transformation image, the value of each foreground pixel (within the target pelvic bone region) is defined as the Euclidean distance from that point to the nearest background pixel (bone boundary). Therefore, the distance value of the points on the bone boundary is zero, and the distance value of the points closer to the geometric center of the bone is larger, which is displayed as brighter pixels in the image.

[0042] The watershed algorithm divides the image as a terrain surface, and its ideal input is a gradient image that clearly reflects the natural boundaries of the object. The distance transformation image provides terrain gradient information for the watershed algorithm: after Euclidean distance transformation, the geometric center of the bone appears as a bright ridge (peak) in the image, and the area near the bone boundary appears as a dark hill foot. Since the bone is subjected to external impact, its fracture usually begins in the internal area where stress is concentrated and tends to expand along the mechanically weak path. The skeleton geometric axis highlighted by the distance transformation image often corresponds to the high stress area and the potential fracture path in structural mechanics. Therefore, the ridges divided in the distance transformation image are consistent with the potential weak lines in the biomechanics of the bone.

[0043] S3: Perform a label-based watershed segmentation algorithm on the distance transformation image to generate a first watershed label image with an initial bone fracture shape.

[0044] Specifically, when the watershed segmentation algorithm is applied to the aforementioned distance transformation map, each local maximum point (i.e., the geometric axis of the bone) in the distance transformation map is defined as a marker point of the catchment basin. This simulates the process of water overflowing from these marker points. When water bodies overflowing from different marker points meet, their confluence boundary forms a watershed, which is also the initial position of the fracture line. Since the geometric axis of the bone often corresponds to a high-stress distribution area in structural mechanics, and is the potential path most prone to fracture under external force, this segmentation method allows the initial fracture line (the boundary line of bone fragments) to preferentially form near the mechanical axis of the bone, that is, to extend along the mechanically weak areas of the bone, thus morphologically more consistent with the actual biomechanical mechanism of fractures.

[0045] S4: Add Gaussian noise to the first watershed label map to dynamically perturb the water level parameters in order to generate bone fragment boundaries that conform to clinical variability.

[0046] like Figure 5 The image shown is the first watershed label image after adding Gaussian perturbation. A Gaussian random perturbation with a mean of 0 and a standard deviation of σ (e.g., σ∈[2,5]) is superimposed on each voxel value in the input image (i.e., the aforementioned distance transformation image) D_original(x, y, z) to obtain a new input image D_noisy(x, y, z) = D_original(x, y, z) + N(0, σ). This allows the boundaries (fracture lines) extracted by the watershed algorithm to produce pixel-level random shifts, thereby generating morphological micro-variations without changing the overall number of fragments. For example, the fracture line may slightly fluctuate, becoming more tortuous or coarser; the boundary contours of the fragments may no longer be smooth, exhibiting subtle convex-concave changes. This method is suitable for generating subdivided variants under the same fracture type, thereby simulating subtle changes caused by differences in bone microstructure or imaging noise.

[0047] Furthermore, based on D_noisy(x, y, z), the marker positions or threshold parameters of the inundation process in the watershed algorithm are randomly fine-tuned (this embodiment is not used in conjunction with the bone fragment validity screening described below). Thus, by interfering with the decision-making process of the watershed algorithm itself, more significant structural changes can be induced. Its effect goes beyond simple boundary fine-tuning; it can alter the merging and splitting behavior of fragments, thereby simulating macroscopic morphological variations caused by different magnitudes or directions of force. For example, it can cause some small regions that should have merged to become independent new fragments due to threshold changes, simulating a more severe comminuted fracture morphology; or it can cause some fragments that should have separated to merge due to changes in marker positions, generating simpler fracture patterns. This method can generate fracture morphologies of different severity and fragmentation patterns, greatly enhancing the diversity of generated data and better covering the complex situations encountered in clinical practice.

[0048] The two foregoing embodiments with Gaussian noise added effectively solve the problem of the scarcity of real fracture data by introducing controlled randomness. The large number of diversified statistically significant morphological variations in fracture morphology output can be used in the training data set of the deep learning model, so that the model learns a general representation of fracture features rather than being limited to a few specific morphologies, thereby potentially improving its generalization ability in real clinical scenarios.

[0049] S41: As shown in the first aspect of the present application, in an embodiment, after the initial bone fragments are added with Gaussian noise, before performing affine transformation, the effectiveness of the initial bone fragments is screened; wherein the effectiveness screening method comprises: calculating the number of voxels v contained in each bone fragment in the first watershed label graph. The bone fragment with the largest number of voxels v is taken as the main bone block. The number of voxels v of each bone fragment is sequentially judged whether it is greater than a preset first threshold k1, if v≤k1, the bone fragment is merged into the main bone block to form a new main bone block; if v>k1, the bone fragment is retained. Figure 2 It should be understood that since the watershed algorithm is prone to produce very small bone fragments (usually containing only a few or tens of voxels) due to image noise or minor details, merging such bone fragments into the main bone block (so that the bone fragment and the main bone block become a whole bone fragment) can eliminate the false fragments and burrs produced by the watershed, making the input morphology more coherent and the boundary more stable. This is because, very small bone fragments are prone to produce outliers and ill-conditioned matrices in subsequent nearest neighbor distance, normal estimation and affine fitting, and after merging, the variance of the mean distance and the angle statistics can be significantly reduced, reducing calculation jitter and error accumulation. Moreover, small and isolated fragments can cause misjudgment of contact consistency and collision detection, and after merging, the inspection can be focused on the main force surface with anatomical significance, making the biomechanical rule determination more reliable.

[0050] Further, in an embodiment of the first aspect of the present application, the larger bone fragments that meet the voxel number standard after the foregoing effectiveness screening are further screened, that is, only a preset proportion of larger bone fragments are retained, and the remaining larger bone fragments are merged into the main bone block as the foregoing small bone fragments.

[0051]

[0052] ​It should be understood that after the aforementioned effectiveness screening eliminates small fragments, only the larger bone fragments that meet the preset proportion are retained, and the rest are merged into the main bone block, aiming to actively reduce the number of fragments under the premise of not violating the anatomical and mechanical constraints (contact consistency, topological closure, boundary continuity), and to systematically include the case of reducing the number of fragments in the data generation process. From the same anatomical sample, a variety of fracture morphologies from few fragments to many fragments are derived, which conforms to the real distribution that not all cases in the clinic are highly fragmented. In this way, the sample space can be expanded and the diversity can be improved, avoiding the over-fragmented fracture morphology in the data set, and enhancing the generalization ability of the model to different fragmentation levels.

[0053] S5: performing affine transformation on the bone fragments in the first watershed label image after adding Gaussian noise to obtain a second watershed label image with a final fracture morphology and / or 6DOF displacement parameters of the affine transformation; wherein the affine transformation includes translation transformation and / or rotation transformation limited by biomechanical rules.

[0054] The foregoing steps S3-S4 divide the complete pelvis skeleton into multiple fragments and generate diversified fracture lines, but at this time each fragment is still closely attached together and has not moved relative to each other, that is, there is no gap or overlap between the bone fragments to form a real pelvis fracture morphology visible in the clinic. Based on this, step S5 performs affine transformation including translation transformation and rotation transformation on each bone fragment, so that it can simulate the displacement of the bone fragments under the action of muscle traction and external force in a real pelvis fracture scenario.

[0055] 6DOF (Six Degrees of Freedom) completely describes all possible movements of a rigid body in three-dimensional space, and the 6DOF displacement parameters include translation freedom and rotation freedom. The translation freedom is the linear motion of the object along the X, Y, and Z coordinate axes, and the rotation freedom is the rotational motion of the object around the X, Y, and Z coordinate axes. Through the combination of rotation and translation, the angularity, displacement, and overlap of the fragments can be simulated, and the clinical fracture characteristics such as shortening deformity, separation displacement, angular displacement, etc. can be perfectly reproduced. However, not all mathematically possible 6DOF transformations are biologically reasonable, so the transformation must be limited by biomechanical rules. Biomechanical rules can be set by those skilled in the art according to widely available biomechanical rules.

[0056] As Figure 3As shown, in an embodiment of the first aspect of the present application, the affine transformation process includes: performing a rotation or translation transformation on the transformed bone fragment; determining whether the transformed bone fragment meets the biomechanical rule; if yes, accepting the rotation or translation transformation and stopping the transformation; if not, backtracking the rotation or translation transformation and re-performing a rotation or translation transformation; wherein, after backtracking the rotation or translation transformation and before performing the rotation or translation transformation again, determining whether the total number of rotation and / or translation transformations performed on the transformed bone fragment reaches n times, if yes, stopping the transformation and merging the transformed bone fragment into the bone fragment with the largest number of voxels; if not, allowing the next rotation or translation transformation to be performed.

[0057] It should be understood that the case of "whether the total number of rotation and / or translation transformations performed on the transformed bone fragment reaches n times" in the above embodiment includes: n times of translation transformation has been performed, or n times of rotation transformation has been performed, or the total number of translation and rotation transformations performed is n times. The backtracking of the rotation or translation transformation refers to undoing the current transformation and restoring the bone fragment to the state before the transformation. This embodiment sets the maximum number of iterations n, so that the affine transformation process has a clear calculation cutoff boundary, effectively preventing the algorithm from falling into an invalid loop or consuming excessive computing power due to repeated attempts to meet the displacement that does not satisfy the biomechanical rule. At the same time, whether the biomechanical rule is met is used as a judgment threshold in the iteration process, which ensures that the generated training data has high biomechanical rationality, avoids inputting a large amount of noise data that does not meet the physical law into the model, and significantly improves the training efficiency and performance of the deep learning model. Moreover, the small bone fragments that still cannot reach a stable position after iteration are merged into the main bone block, simulating the clinical phenomenon that small bone fragments often remain relatively continuous with the original main bone and heal quickly in a real trauma environment due to lack of significant displacement, accurately restoring the clinical prognosis characteristics of such fractures that do not require special treatment, and making the generated data spectrum more close to the clinical reality.

[0058] In an embodiment of the first aspect of the present application, the affine transformation method includes: the rotation angle of each rotation transformation α ∈ (0°, k4], the translation distance of each translation transformation p ∈ (0, k5], k4 is a preset fourth threshold value, and k5 is a preset fifth threshold value.

[0059] Since real bone displacement is a continuous process, limiting the magnitude of single-step transformation makes each frame of displacement state output by the algorithm only a slight change from the previous moment, which simulates the mechanical behavior of the gradual displacement of the bone after being stressed. During the iterative transformation process of step S5, the algorithm needs to try different displacements multiple times to find a form that not only satisfies other complex rules (such as no collision and joint surface matching) but also looks real. If the step of rotation or translation is allowed to be too large at a time, it may directly skip the reasonable solution space, causing the optimization process to oscillate, diverge or fail to converge. However, by limiting the transformation magnitude of single iteration, the affine transformation process is limited to a gradual and controllable fine-tuning process, avoiding jumping changes and ensuring the smoothness and stability of the optimization process, thereby improving the robustness of the algorithm.

[0060] Preferably, the following several special biomechanical rule setting methods disclosed in the present application can achieve better technical effects, and one or more of them can be selected for use on the basis of the foregoing scheme (for example, Figure 6 The second watershed label map obtained after the affine transformation that satisfies the following three biomechanical rules (voxel coincidence rate judgment and processing rule, joint surface judgment and processing rule, and processing rule for hip acetabulum fragments) is shown:

[0061] In an embodiment of the first aspect of the present application, the biomechanical rule includes that the voxel coincidence rate of the transformed bone fragment and any other bone fragment does not exceed a second threshold k2, where k2∈(0, 1).

[0062] In a real biomechanical environment, since the bone is not an absolutely ideal rigid body and is affected by the elasticity and deformation of the surrounding soft tissue (such as periosteum and ligament), there may indeed be a small and local embedding or interpenetration between fragments, especially in high-energy trauma. By setting a second threshold k2 to control the degree of penetration between fragments, a reasonable balance is established between calculation simulation and biomechanical reality, thereby generating a high-fidelity fracture morphology that conforms to physical laws and is close to clinical observation. For example, when k2 = 0.05, the voxel coincidence rate of the transformed bone fragment and any other bone fragment is allowed to be no more than 5%. If the aforementioned coincidence rate exceeds the second threshold k2, the transformation is determined to be invalid. In this way, not only is the search space of the algorithm limited to a biomechanically reasonable range, allowing for the small embedding commonly seen in clinical practice, but also the generation of serious and unrealistic physical penetration is prevented.

[0063] In an embodiment of the first aspect of the present application, the biomechanical rule comprises: judging whether the transformed bone fragment contains an articular surface; when the transformed bone fragment contains an articular surface, the center of rotation is defined within a spherical region (c, r) with the anatomical center c of the articular surface as the spherical center and a physiological radius r as the radius; wherein the anatomical center c is the area centroid of the articular surface after surface fitting or the geometric center of the corresponding anatomical structure, and the r is a third threshold value preset based on clinical statistical data; when the transformed bone fragment is a fragment that does not contain an articular surface, the center of rotation is the centroid of the contact surface between the transformed bone fragment and any other bone fragment in contact with it.

[0064] Since the joint is the pivot of movement, its movement is strictly subject to the physiological constraints of the joint capsule, ligaments and muscle insertion points, therefore, strictly defining the center of rotation of such fragments within a spherical region (c, r) with the anatomical center as the spherical center and a small physiological radius r as the range can simulate the soft tissue constraint effect, ensuring that the joint fragment can only perform a small "wobble" or "slide" that conforms to its physiological function, and cannot rotate out of its anatomical position, thus accurately reproducing the typical displacement characteristics of joint fractures. For ordinary fragments that do not contain articular surfaces (such as fracture fragments on the pelvic ring), since their movement is mainly affected by mechanical contact with adjacent bone fragments and the pulling of surrounding muscles, therefore, by setting their center of rotation as the centroid of the contact surface, the natural pivot point of mechanics around the contact point is simulated, thus generating a translation and rotation displacement that is more in line with mechanical common sense, avoiding the unrealistic movement caused by randomly selecting the center of rotation.

[0065] In an embodiment of the first aspect of the present application, the biomechanical rule comprises: when the transformed bone fragment is an acetabular fragment, setting the articular surface after affine transformation as A, the femoral head contact surface point cloud as B, and the nearest point of a sampling point y on A to B as x; calculating the average nearest neighbor distance d between x and y wherein m is the number of sampling points on A; retaining the (x, y) point set with d≤dmax, and calculating the opposite normal angle θ of each point in the point set wherein a is the unit normal of A at y, and β is the unit normal of B at x after local surface fitting; let the average value of be within the preset interval [θmin, θmax].

[0066] In a real physiological environment, the acetabulum and the femoral head must maintain a certain contact and congruence relationship to ensure joint function, so the method first calculates the average nearest neighbor distance d between the transformed acetabular joint surface A and the femoral head point cloud B, and restricts it to be less than the maximum threshold dmax. This ensures that the acetabular fragments after fracture do not separate from the femoral head (i.e. dislocation), simulating the stabilizing effect of soft tissues such as joint capsule, ligament, etc. on the joint. At the same time, by calculating the included angle of the normal vectors of the two curved surfaces at the contact point and restricting the average value to be in a reasonable range, it is ensured that even after fracture, the contact surface of the acetabulum and the femoral head can still match in curvature, simulating the smooth mechanical interaction between the articular cartilage surfaces, avoiding unnatural hard collision or stress concentration. Therefore, this method ensures that each case generated containing acetabular fracture has a joint relationship that is clinically possible and has a certain functional significance (such as subluxation, poor congruence), rather than an extreme case such as complete dislocation that does not need to discuss reduction scheme, thereby accurately simulating and restricting the joint function compatibility after hip fracture, ensuring that the generated fracture displacement form is highly realistic in anatomy and biomechanics, so that the trained model can learn the complex relationship between joint function and fracture displacement.

[0067] As shown in Figure 7 , the second aspect of the present application provides a multi-modal pelvic fracture data generation system, comprising: an image acquisition module for acquiring a bone mask image; wherein the bone mask image is a mask CT image containing only target pelvic bone information; a distance transformation module for performing Euclidean distance transformation on the bone mask image to obtain a distance transformation image representing the internal structure of the bone; a watershed segmentation module for performing a marker-based watershed segmentation algorithm on the distance transformation image to generate a first watershed label image with an initial fracture morphology; a noise adding module for adding Gaussian noise to the first watershed label image and dynamically perturbing the water level parameter to generate a bone fragment boundary conforming to clinical variability; an affine transformation module for performing affine transformation on the bone fragments in the first watershed label image after adding Gaussian noise to obtain a second watershed label image with a final fracture morphology and / or 6DOF displacement parameters of the affine transformation; wherein the affine transformation includes translation transformation and / or rotation transformation limited by biomechanical rules.

[0068] It should be understood that the specific process of each module performing the corresponding steps described above has been described in detail in the method embodiment described above, and for the sake of brevity, it will not be repeated here.

[0069] It should also be appreciated that the division of modules in the embodiments of the present application is illustrative only, and is merely a logical function division, and another division manner can be used in actual implementation. In addition, each functional module in each embodiment of the present application can be integrated in one processor, or can be physically separated, or two or more modules can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software functional module.

[0070] The third aspect of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method of any one of the preceding aspects.

[0071] The fourth aspect of the present application provides a computer program product, which includes computer program codes, and when the computer program codes are run on a computer, the computer implements the method of any one of the preceding aspects.

[0072] As shown in Figure 8 The fifth aspect of the present application provides an electronic terminal, which includes a memory, a processor and a computer program stored in the memory, and the processor executes the computer program to implement the method of any one of the preceding aspects. The electronic terminal includes at least one processor 101, a memory 102, at least one network interface 103 and a user interface 105. Each component in the apparatus is coupled together through a bus system 104. It can be understood that the bus system 104 is used to realize the connection communication between the components. In addition to including a data bus, the bus system 104 also includes a power bus, a control bus and a status signal bus.

[0073] The user interface 105 can include a display, a keyboard, a mouse, a trackball, a click gun, a key, a button, a touchpad or a touch screen, etc.

[0074] It can be understood that the memory 102 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. The non-volatile memory can be a read-only memory (ROM, Read Only Memory), a programmable read-only memory (PROM, Programmable Read-Only Memory), which is used as an external cache. By way of example but not limitation, many forms of RAM can be used, such as static random access memory (SRAM, Static Random Access Memory), synchronous static random access memory (SSRAM, Synchronous Static Random Access Memory). The memory described in the embodiments of the present application is intended to include but not limited to these and any other suitable categories of memory.

[0075] The memory 102 in the embodiment of the present application is configured to store various types of data to support the operation of the electronic terminal 100. Examples of the data include any executable program for operating on the electronic terminal 100, such as an operating system 1021 and an application program 1022. The operating system 1021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application program 1022 can include various application programs, such as a media player, a browser, etc., for implementing various application services. The method provided by the embodiment of the present application can be included in the application program 1022.

[0076] The method disclosed by the above embodiment of the present application can be applied to the processor 101 or implemented by the processor 101. The processor 101 can be an integrated circuit chip having a signal processing capability. In the implementation process, each step of the above method can be completed by an integrated logic circuit or an instruction in the form of software in the processor 101. The above processor 101 can be a general-purpose processor, a digital signal processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. The processor 101 can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the present application. The general-purpose processor 101 can be a microprocessor or any conventional processor, etc. In combination with the steps of the accessory optimization method provided by the embodiment of the present application, the above-mentioned steps can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module can be located in a storage medium, which is located in the memory. The processor reads the information in the memory and combines the hardware to complete the steps of the above method.

[0077] In the exemplary embodiment, the electronic terminal 100 can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), etc., for executing the above method.

[0078] As used in this description, the terms "component," "module," "system," and the like are intended to refer to a computer-related entity, either hardware, firmware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to being, a process running on a processor, a processor, an object, an executable, a thread of execution, a program, and / or a computer. By way of illustration, both an application running on a computing device and the computing device can be a component. One or more components can reside within a process and / or thread of execution and a component can be localized, partially localized, or distributed across two or more computers. Also, these components can execute from various computer readable media having various data structures stored thereon. The components can communicate by way of local and / or remote processes such as in accordance with a signal having one or more data packets (e.g., data from one component interacting with another component in a local system, distributed system, and / or across a network such as the Internet with other systems via the signal).

[0079] Those of skill in the art would understand that the various illustrative logical blocks and steps described in connection with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. The choice of whether to implement the described functionality in hardware or software depends on the particular application and design constraints imposed on the overall system. Skilled artisans can implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

[0080] Those of skill in the art would understand that, for the described convenience and brevity, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiment, which will not be described here.

[0081] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative, for example, the division of units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be through some interface, indirect coupling or communication connection between devices or units, which can be electrical, mechanical or other forms.

[0082] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e., may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0083] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically, or two or more units can be integrated into one unit.

[0084] In the above embodiments, the functions of each functional unit can be implemented by software, hardware, firmware or any combination thereof, in whole or in part. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions (programs). When the computer program instructions (programs) are loaded and executed on a computer, the flow or function according to the embodiments of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. Computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another, for example, computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be magnetic media (such as floppy disk, hard disk, magnetic tape), optical media (such as high-density digital video disc (digital video disc, DVD), or semiconductor media (such as solid state disk (solid state disk, SSD) and the like.

[0085] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0086] The above merely illustrates the principles and effects of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0087] In summary, the present application effectively overcomes the shortcomings in the prior art and has a high industrial utilization value.

[0088] The above embodiments only exemplarily illustrate the principles and effects of the present application, but are not used to limit the present application. Any person skilled in the art can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes completed by those skilled in the art without departing from the spirit and technical thought disclosed by the present application should be covered by the claims of the present application.

Claims

1. A method of generating multi-modal pelvic fracture data, the method comprising: The method comprises the following steps: obtaining a bone mask image, wherein the bone mask image is a mask CT image containing only target pelvic bone information; performing Euclidean distance transform on the bone mask image to obtain a distance transform image representing the internal structure of the bone; performing a marker-based watershed segmentation algorithm on the distance transform image to generate a first watershed label image with an initial fracture morphology; adding Gaussian noise to the first watershed label image to dynamically disturb the water level parameter and generate a bone fragment boundary conforming to clinical variability; performing affine transformation on the initial bone fragments after adding Gaussian noise to obtain a second watershed label image with a final fracture morphology and / or 6DOF displacement parameters of the affine transformation; wherein the affine transformation includes translation transformation and / or rotation transformation limited by biomechanical rules.

2. The method of generating multi-modal pelvic fracture data of claim 1, wherein, Before adding Gaussian noise to the initial bone fragments and performing affine transformation, the effectiveness of the initial bone fragments is screened; wherein the effectiveness screening method comprises: calculating the number of voxels v contained in each bone fragment in the first watershed label image; taking the bone fragment with the largest number of voxels v as the main bone block; sequentially determining whether the number of voxels v of each bone fragment is greater than a preset first threshold k1, if v≤k1, the bone fragment is merged into the main bone block to form a new main bone block; if v>k1, the bone fragment is retained.

3. The method of generating multi-modal pelvic fracture data of claim 1, wherein, The process of the affine transformation comprises: performing a rotation or translation transformation on the transformed bone fragment; determining whether the transformed bone fragment conforms to the biomechanical rules; if it conforms, the rotation or translation transformation is accepted and the transformation is stopped; if it does not conform, the rotation or translation transformation is rolled back and a rotation or translation transformation is performed again; wherein after rolling back the rotation or translation transformation and before performing the rotation or translation transformation again, it is determined whether the total number of rotation and / or translation transformations performed on the transformed bone fragment has reached n times, if it has, the transformation is stopped and the transformed bone fragment is merged into the bone fragment with the largest number of voxels; if it has not, the next rotation or translation transformation is allowed to be performed.

4. The method of generating multi-modal pelvic fracture data according to any one of claims 1, wherein, The biomechanical rules include: the voxel overlap rate of the transformed bone fragment and any other bone fragment does not exceed a second threshold k2, wherein k2∈(0,1).

5. The method of generating multi-modal pelvic fracture data according to any one of claims 1, wherein, The biomechanical rules include: determining whether the transformed bone fragment contains an articular surface; when the transformed bone fragment contains an articular surface, the center of rotation is limited within a spherical domain (c,r) with the anatomical center c of the articular surface as the spherical center and the physiological radius r as the radius; wherein the anatomical center c is the area centroid obtained by surface fitting of the articular surface or the geometric center of the corresponding anatomical structure, and the r is a third threshold preset based on clinical statistical data; when the transformed bone fragment is a fragment that does not contain an articular surface, the contact surface centroid between the transformed bone fragment and any other bone fragment in contact with it is taken as the center of rotation.

6. The method of generating multi-modal pelvic fracture data of any one of claims 1, wherein, The biomechanical rules include: When the bone fragment is transformed into the acetabular fragment, the articular surface of the transformed bone fragment is A, the femoral head contact surface point cloud is B, and the nearest point of the sampling point y on A to B is x; Computing the average nearest neighbor distance of x and y where m is the number of sample points on A; The set of (x, y) points with d < dmax is retained, and the opposite normal angle of each point in the set is calculated where a is the unit normal of A at y, and β is the unit normal of B at x obtained by local surface fitting Let the average value of the angle θ be within a preset interval [θmin, θmax].

7. A system for generating multi-modal pelvic fracture data, the system comprising: The application relates to a computer program product and a processor. An image acquisition module is configured to acquire a bone mask image; wherein the bone mask image is a mask CT image containing only target pelvic bone information; A distance transformation module is configured to perform Euclidean distance transformation on the bone mask image to obtain a distance transformation image representing the internal structure of the bone; A watershed segmentation module is configured to perform a marker-based watershed segmentation algorithm on the distance transformation image to generate a first watershed label image with an initial fracture morphology; A noise adding module is configured to add Gaussian noise to the first watershed label image to dynamically disturb the water level parameter and generate a bone fragment boundary conforming to clinical variability; An affine transformation module is configured to perform affine transformation on the bone fragment in the first watershed label image with added Gaussian noise to obtain a second watershed label image with a final fracture morphology and / or 6DOF displacement parameters of the affine transformation; wherein the affine transformation includes translation transformation and / or rotation transformation limited by biomechanical rules.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the method in any one of claims 1-6.

9. A computer program product, characterised in that, The computer program product includes computer program code, and when the computer program code is run on a computer, the computer program code causes the computer to implement the method in any one of claims 1-6.

10. An electronic terminal comprising a memory, a processor and a computer program stored on the memory, characterized in that, The processor executes the computer program to implement the method in any one of claims 1-6.

Images