3D Model Segmentation Research Articles

Reproducibility of the 3D-Models for Baby and Mother phantom using the platform 3D-Slicer

The evolution of segmentation over the years has offered unlimited potential, with extensive applications in the field of Anatomy. Female pelvis anatomy is reaping great benefits, especially in terms of surgical planning and guidance. Personalize mannequin design are the basic element in virtual garment simulation and visualization and mannequin construction from scanned point cloud and regenerates their shapes by inputting variant dimension., we will describe the segmentation of an existing mannequin and regenerating a new one through 3D Slicer platform, this work is the result of cooperative work between CIGT (Image Guided Therapy Center in the Faculty of Medicine in Nouakchott University-Mauritania) and the IUIBS (Institute for Biomedical and Healthcare Research- University of Las Palmas de Gran Canarias- Spain), involving the segmentation and generation of 3D models for both baby and mother using the 3D Slicer platform.

MLGCN: an ultra efficient graph convolutional neural model for 3D point cloud analysis.

With the rapid advancement of 3D acquisition technologies, 3D sensors such as LiDARs, 3D scanners, and RGB-D cameras have become increasingly accessible and cost-effective. These sensors generate 3D point cloud data that require efficient algorithms for tasks such as 3D model classification and segmentation. While deep learning techniques have proven effective in these areas, existing models often rely on complex architectures, leading to high computational costs that are impractical for real-time applications like augmented reality and robotics. In this work, we propose the Multi-level Graph Convolutional Neural Network (MLGCN), an ultra-efficient model for 3D point cloud analysis. The MLGCN model utilizes shallow Graph Neural Network (GNN) blocks to extract features at various spatial locality levels, leveraging precomputed KNN graphs shared across GCN blocks. This approach significantly reduces computational overhead and memory usage, making the model well-suited for deployment on low-memory and low-CPU devices. Despite its efficiency, MLGCN achieves competitive performance in object classification and part segmentation tasks, demonstrating results comparable to state-of-the-art models while requiring up to a thousand times fewer floating-point operations and significantly less storage. The contributions of this paper include the introduction of a lightweight, multi-branch graph-based network for 3D shape analysis, the demonstration of the model's efficiency in both computation and storage, and a thorough theoretical and experimental evaluation of the model's performance. We also conduct ablation studies to assess the impact of different branches within the model, providing valuable insights into the role of specific components.

Methodology for Activity Unit Segmentation of Design 3D Models Using PointNet Deep Learning Technique

This paper discusses the challenge of using 3D models created during the design stage in the construction stage due to the need for segmentation or remodeling to incorporate activity concepts. To address this challenge, the paper proposes a methodology that uses the PointNet deep learning technique to automatically classify and segment design model elements into activity units for the construction stage while preserving attribute information. The authors introduce a module that utilize the bounding box concept to simplify the segmentation process after importing the 3D model into the 4D system used in the construction stage. This eliminates the need for separate CAD software and allows for direct segmentation into activity units within the 4D system and simultaneous simulation. The proposed methodology was applied to two real bridge projects and demonstrated increased 3D model reusability in the construction stage. The paper concludes that this approach can improve the usability of 3D models for construction projects.

Interior structural change detection using a 3D model and LiDAR segmentation

Interior structural change detection using a 3D model and LiDAR segmentation

Segmentation of the 3D printed mirror image auricular model to ease sculpture of the costal cartilages in total auricular aesthetic reconstruction

BackgroundRecently, three-dimensional (3D) models have been used more frequently than the traditional two-dimensional (2D) models as an intraoperative guide to ease sculpture of costal cartilages in total auricular aesthetic reconstruction in cases of microtia. Usually, 3D imaging techniques are used to create compact ear models, however, there is insufficient clinical knowledge of using them to create segmented 3D auricular models for cartilage framework reconstruction. In this studyassessment of the advantages of segmentation of 3D models over the traditional compact 3D models in total auricular aesthetic reconstruction has been discussed. Patients and methodsIn the current study, 16 patients who underwent total auricular aesthetic reconstruction using 3D models were included, patients were divided into two groups (Group I, 8 patients had total auricular aesthetic reconstruction using compact 3D printed mirror image model). Whereas (Group II, 8 patients had total auricular aesthetic reconstruction using segmented 3D printed mirror image model into auricular subunits. Then, each subunit was used for reconstruction of the corresponding part of the auricle. In both groups the patient's unaffected ear was mirrored and used as a reference to reconstruct the affected side. ResultsThe results were evaluated in both groups. The overall operative and carving times were found to be less in Group II; Also, postoperative aesthetic outcome and patient satisfaction were favorable in this group. ConclusionUsing segmentation of 3D printed models in total auricular aesthetic reconstruction had less intraoperative time, better aesthetic results, and high postoperative patient satisfaction.

A Deep Learning Approach to Mesh Segmentation

In the shape analysis community, decomposing a 3D shape into meaningful parts has become a topic of interest. 3D model segmentation is largely used in tasks such as shape deformation, shape partial matching, skeleton extraction, shape correspondence, shape annotation and texture mapping. Numerous approaches have attempted to provide better segmentation solutions; however, the majority of the previous techniques used handcrafted features, which are usually focused on a particular attribute of 3D objects and so are difficult to generalize. In this paper, we propose a three-stage approach for using Multi-view recurrent neural network to automatically segment a 3D shape into visually meaningful sub-meshes. The first stage involves normalizing and scaling a 3D model to fit within the unit sphere and rendering the object into different views. Contrasting viewpoints, on the other hand, might not have been associated, and a 3D region could correlate into totally distinct outcomes depending on the viewpoint. To address this, we ran each view through (shared weights) CNN and Bolster block in order to create a probability boundary map. The Bolster block simulates the area relationships between different views, which helps to improve and refine the data. In stage two, the feature maps generated in the previous step are correlated using a Recurrent Neural network to obtain compatible fine detail responses for each view. Finally, a layer that is fully connected is used to return coherent edges, which are then back project to 3D objects to produce the final segmentation. Experiments on the Princeton Segmentation Benchmark dataset show that our proposed method is effective for mesh segmentation tasks.

Development of a 3D simulator for training the mouse in utero electroporation.

In utero electroporation (IUE) requires high-level training in microinjection through the mouse uterine wall into the lateral ventricle of the mouse brain. Training for IUE is currently being performed in live mice as no artificial models allow simulations yet. This study aimed to develop an anatomically realistic 3D printed simulator to train IUE in mice. To this end, we created embryo models containing lateral ventricles. We coupled them to uterus models in six steps: (1) computed tomography imaging, (2) 3D model segmentation, (3) 3D model refinement, (4) mold creation to cast the actual model, (5) 3D mold printing, and (6) mold casting the molds with a mix of soft silicones to ensure the hardness and consistency of the uterus and embryo. The results showed that the simulator assembly successfully recreated the IUE. The compression test did not differ in the mechanical properties of the real embryo or in the required load for uterus displacement. Furthermore, more than 90% of the users approved the simulator as an introduction to IUE and considered that the simulator could help reduce the number of animals for training. Despite current limitations, our 3D simulator enabled a realistic experience for initial approximations to the IUE and is a real alternative for implementing the 3Rs. We are currently working on refining the model.

Limited Hip Flexion and Internal Rotation Resulting From Early Hip Impingement Conflict on Anterior Metaphysis of Patients With Untreated Severe SCFE Using 3D Modelling.

Slipped capital femoral epiphysis (SCFE) is the most common hip disorder in adolescent patients that can result in complex 3 dimensional (3D)-deformity and hip preservation surgery (eg, in situ pinning or proximal femoral osteotomy) is often performed. But there is little information about location of impingement.Purpose/Questions: The purpose of this study was to evaluate (1) impingement-free hip flexion and internal rotation (IR), (2) frequency of impingement in early flexion (30 to 60 degrees), and (3) location of acetabular and femoral impingement in IR in 90 degrees of flexion (IRF-90 degrees) and in maximal flexion for patients with untreated severe SCFE using preoperative 3D-computed tomography (CT) for impingement simulation. A retrospective study involving 3D-CT scans of 18 patients (21 hips) with untreated severe SCFE (slip angle>60 degrees) was performed. Preoperative CT scans were used for bone segmentation of preoperative patient-specific 3D models. Three patients (15%) had bilateral SCFE. Mean age was 13±2 (10 to 16) years and 67% were male patients (86% unstable slip, 81% chronic slip). The contralateral hips of 15 patients with unilateral SCFE were evaluated (control group). Validated software was used for 3D impingement simulation (equidistant method). (1) Impingement-free flexion (46±32 degrees) and IRF-90 degrees (-17±18 degrees) were significantly ( P <0.001) decreased in untreated severe SCFE patients compared with contralateral side (122±9 and 36±11 degrees).(2) Frequency of impingement was significantly ( P <0.001) higher in 30 and 60 degrees flexion (48% and 71%) of patients with severe SCFE compared with control group (0%).(3) Acetabular impingement conflict was located anterior-superior (SCFE patients), mostly 12 o'clock (50%) in IRF-90 degrees (70% on 2 o'clock for maximal flexion). Femoral impingement was located on anterior-superior to anterior-inferior femoral metaphysis (between 2 and 6 o'clock, 40% on 3 o'clock and 40% on 5 o'clock) in IRF-90 degrees and on anterior metaphysis (40% on 3 o'clock) in maximal flexion and frequency was significantly ( P <0.001) different compared with control group. Severe SCFE patients have limited hip flexion and IR due to early hip impingement using patient-specific preoperative 3D models. Because of the large variety of hip motion, individual evaluation is recommended to plan the osseous correction for severe SCFE patients. Level III.

Semantic segmentation of 3D car parts using UAV-based images

Environment understanding in real-world scenarios has gained an increased interest in research and industry. The advances in data capture and processing allow a high-detailed reconstruction from a set of multi-view images by generating meshes and point clouds. Likewise, deep learning architectures along with the broad availability of image datasets bring new opportunities for the segmentation of 3D models into several classes. Among the areas that can benefit from 3D semantic segmentation is the automotive industry. However, there is a lack of labeled 3D models that can be useful for training and use as ground truth in deep learning-based methods. In this work, we propose an automatic procedure for the generation and semantic segmentation of 3D cars that were obtained from the photogrammetric processing of UAV-based imagery. Therefore, sixteen car parts are identified in the point cloud. To this end, a convolutional neural network based on the U-Net architecture combined with an Inception V3 encoder was trained in a publicly available dataset of car parts. Then, the trained model is applied to the UAV-based images and these are mapped on the photogrammetric point clouds. According to the preliminary image-based segmentation, an optimization method is developed to get a full labeled point cloud, taking advantage of the geometric and spatial features of the 3D model. The results demonstrate the method’s capabilities for the semantic segmentation of car models. Moreover, the proposed methodology has the potential to be extended or adapted to other applications that benefit from 3D segmented models.

Three-Dimensional Magnetic Resonance Imaging Bone Models of the Hip Joint Using Deep Learning: Dynamic Simulation of Hip Impingement for Diagnosis of Intra- and Extra-articular Hip Impingement.

Background:Dynamic 3-dimensional (3D) simulation of hip impingement enables better understanding of complex hip deformities in young adult patients with femoroacetabular impingement (FAI). Deep learning algorithms may improve magnetic resonance imaging (MRI) segmentation.Purpose:(1) To evaluate the accuracy of 3D models created using convolutional neural networks (CNNs) for fully automatic MRI bone segmentation of the hip joint, (2) to correlate hip range of motion (ROM) between manual and automatic segmentation, and (3) to compare location of hip impingement in 3D models created using automatic bone segmentation in patients with FAI.Study Design:Cohort study (diagnosis); Level of evidence, 3.Methods:The authors retrospectively reviewed 31 hip MRI scans from 26 symptomatic patients (mean age, 27 years) with hip pain due to FAI. All patients had matched computed tomography (CT) and MRI scans of the pelvis and the knee. CT- and MRI-based osseous 3D models of the hip joint of the same patients were compared (MRI: T1 volumetric interpolated breath-hold examination high-resolution sequence; 0.8 mm3 isovoxel). CNNs were used to develop fully automatic bone segmentation of the hip joint, and the 3D models created using this method were compared with manual segmentation of CT- and MRI-based 3D models. Impingement-free ROM and location of hip impingement were calculated using previously validated collision detection software.Results:The difference between the CT- and MRI-based 3D models was <1 mm, and the difference between fully automatic and manual segmentation of MRI-based 3D models was <1 mm. The correlation of automatic and manual MRI-based 3D models was excellent and significant for impingement-free ROM (r = 0.995; P < .001), flexion (r = 0.953; P < .001), and internal rotation at 90° of flexion (r = 0.982; P < .001). The correlation for impingement-free flexion between automatic MRI-based 3D models and CT-based 3D models was 0.953 (P < .001). The location of impingement was not significantly different between manual and automatic segmentation of MRI-based 3D models, and the location of extra-articular hip impingement was not different between CT- and MRI-based 3D models.Conclusion:CNN can potentially be used in clinical practice to provide rapid and accurate 3D MRI hip joint models for young patients. The created models can be used for simulation of impingement during diagnosis of intra- and extra-articular hip impingement to enable radiation-free and patient-specific surgical planning for hip arthroscopy and open hip preservation surgery.

Computer-Aided Diagnosis of Children with Cerebral Palsy under Deep Learning Convolutional Neural Network Image Segmentation Model Combined with Three-Dimensional Cranial Magnetic Resonance Imaging.

In this paper, we analyzed the application value and effect of deep learn-based image segmentation model of convolutional neural network (CNN) algorithm combined with 3D brain magnetic resonance imaging (MRI) in diagnosis of cerebral palsy in children. 3D brain model was segmented based on CNN algorithm to obtain the segmented MRI images of brain tissue, and the validity was verified. Then, 70 children with cerebral palsy were rolled into the observation group (n = 35), which received MRI for diagnosis after segmentation of brain tissue, and control group (n = 35), which were diagnosed by computed tomography (CT). The diagnosis results of the two groups were compared. The validity experiment verified that the image segmentation method based on CNN algorithm can obtain effective style graphics. In clinical trials, the diagnostic accuracy of 88.6% in the observation group was evidently superior to that of 80% in the control group (P < 0.05). In the observation group, one patient was diagnosed as normal, four patients had white matter lesions, 17 patients had corpus callosum lesions, and five patients had basal ganglia softening foci. In the control group, two patients were diagnosed as normal, two patients had white matter lesions, 19 patients had corpus callosum lesions, and four patients had basal ganglia softening foci. No notable difference was found between the two groups (P > 0.05). According to the research results, in the diagnosis of cerebral palsy in children, the image segmentation of brain 3D model based on CNN to obtain the MRI image of segmented brain tissue can effectively improve the detection accuracy. Moreover, the specific symptoms can be diagnosed clearly. It can provide the corresponding diagnostic basis for clinical diagnosis and treatment and was worthy of clinical promotion.

Deep learning-based optimal segmentation of 3D printed product for surface quality improvement and support structure reduction

Large-sized product cannot be printed as one piece by a 3D printer because of the volume limitation of most 3D printers. Some products with the complex structure and high surface quality should also not be printed into one piece to meet requirement of the printing quality. For increasing the surface quality and reducing support structure of 3D printed models, this paper proposes a 3D model segmentation method based on deep learning. Sub-graphs are generated by pre-segmenting 3D triangular mesh models to extract printing features. A data structure is proposed to design training data sets based on the sub-graphs with printing features of the original 3D model including surface quality, support structure and normal curvature. After training a Stacked Auto-encoder using the training set, a 3D model is pre-segmented to build an application set by the sub-graph data structure. The application set is applied by the trained deep-learning system to generate hidden features. An Affinity Propagation clustering method is introduced in combining hidden features and geometric information of the application set to segment a product model into several parts. In the case study, samples of 3D models are segmented by the proposed method, and then printed using a 3D printer for validating the performance.

3D mesh model segmentation based on skeleton extraction

ABSTRACT In order to solve the problem of low segmentation accuracy of 3D mesh models, a non-deep learning method based on skeleton extraction is proposed. First, the skeleton of the model is extracted. Then, the skeleton feature points are extracted by analysing the relationship between the skeleton data. The feature point with the smallest distance from the centroid in the same neighbourhood is selected as the segmentation point. Finally, based on the improved region-based algorithm, the segmented skeleton region is obtained with the segmentation points as the seeds, and the corresponding original data is segmented through inverse mapping. The experimental results show that compared with the existing non-deep learning methods, the proposed method can obtain better semantic separation results and segmentation boundaries.

Reliability and accuracy of automatic segmentation of mandibular 3D models on linear measurements.

Evaluate if automatic segmentation of mandibular three-dimensional (3D) models is reliable and accurate. Eight dry mandibles with eight silica markers were scanned in the i-CAT Classic device (Imaging Sciences International). Automatic segmentation was performed using nine standard preset thresholds in the Dolphin software (Dolphin Imaging & Management Solutions). Three observers individually made twice eight linear measurements on the mandibular 3D models. Another observer made physical measurements, twice as well, on the dry mandibles. Reliability and accuracy were evaluated with intraclass correlation coefficients (ICCs), Dahlberg's formula, Bland-Altman analyses, and changing bias with regression analyses. Inter-observer and intra-observer ICCs and Dahlberg's error were ≥ 0.75 and ≤ 1.0mm, respectively, for all measurements. Inter-observer agreement between mandibular 3D models and physical measurements ranged from -0.37 to 0.91mm. Linear measurements made on mandibular 3D models obtained using standard preset thresholds are reliable and accurate. However, additional studies are necessary to confirm this hypothesis for clinical applications. Since the 3D models are useful for diagnostics and surgical planning, it is necessary to determinate whether the linear measurements made on 3D models obtained by automatic segmentation are sufficiently reliable and accurate.

Multispectral aerial imagery-based 3D digitisation, segmentation and annotation of large scale urban areas of significant cultural value

• Urban areas 3D digitisation using multispectral aerial imagery • Experimental low-cost multispectral camera for drone development • Supervised machine learning methods for 3D mesh segmentation using multispectral data • Performance evaluation of the proposed methods on a large scale case study • Web-based visualisation and annotation tool for segmentation fine-tuning Disaster risk management of movable and immovable cultural heritage is a highly significant research topic. In this work, we present a pipeline for 3D digitisation, segmentation and annotation of large scale urban areas in order to produce data that can be exploited in disaster management simulators (e.g fire spreading, crowd movement, firefighting training, evacuation planning, etc.). We have selected the old town of Xanthi (Greece) as a challenging case study. We developed a custom multispectral camera to be carried by a commercial drone. Using the structure from motion / multiview stereo (SFM/MVS) approach, we produced a 3D model of the urban area covering 0.5 k m 2 that is followed by a multilayer texture map which carries information from visible and near-infrared regions of the electromagnetic spectrum. We developed a set of machine learning approaches based on logistic regression, support vector machines and artificial neural networks that allow 3D model segmentation by exploiting not only morphological and structural features but also the multispectral behaviour of different material surfaces. We objectively evaluate the performance of the proposed segmentation approaches on six significant material-based classes (cobbled-roads granite kilns, building walls, ceramic roof-tiles, low-vegetation, high-vegetation and metal surfaces) that are used in simulating fire propagation and crowd movement. The experiments revealed that the segmentation accuracy can be enhanced by taking into consideration surface material multispectral properties as well as morphological features. A Web-based multi-user annotation tool complements our proposed pipeline by enabling further 3D model segmentation, fine tuning and semantics annotation (e.g. usage-based building classification and evacuation priorities, escape paths and gathering points).

Direct 3D model extraction method for color volume images.

BACKGROUND: There is a great demand for the extraction of organ models from three-dimensional (3D) medical images in clinical medicine diagnosis and treatment.OBJECTIVE: We aimed to aid doctors in seeing the real shape of human organs more clearly and vividly.METHODS: The method uses the minimum eigenvectors of Laplacian matrix to automatically calculate a group of basic matting components that can properly define the volume image. These matting components can then be used to build foreground images with the help of a few user marks.RESULTS: We propose a direct 3D model segmentation method for volume images. This is a process of extracting foreground objects from volume images and estimating the opacity of the voxels covered by the objects.CONCLUSIONS: The results of segmentation experiments on different parts of human body prove the applicability of this method.

MRI-based 3D models of the hip joint enables radiation-free computer-assisted planning of periacetabular osteotomy for treatment of hip dysplasia using deep learning for automatic segmentation

IntroductionBoth Hip Dysplasia(DDH) and Femoro-acetabular-Impingement(FAI) are complex three-dimensional hip pathologies causing hip pain and osteoarthritis in young patients. 3D-MRI-based models were used for radiation-free computer-assisted surgical planning. Automatic segmentation of MRI-based 3D-models are preferred because manual segmentation is time-consuming.To investigate(1) the difference and(2) the correlation for femoral head coverage(FHC) between automatic MR-based and manual CT-based 3D-models and (3) feasibility of preoperative planning in symptomatic patients with hip diseases. MethodsWe performed an IRB-approved comparative, retrospective study of 31 hips(26 symptomatic patients with hip dysplasia or FAI). 3D MRI sequences and CT scans of the hip were acquired. Preoperative MRI included axial-oblique T1 VIBE sequence(0.8 mm3 isovoxel) of the hip joint. Manual segmentation of MRI and CT scans were performed. Automatic segmentation of MRI-based 3D-models was performed using deep learning. Results(1)The difference between automatic and manual segmentation of MRI-based 3D hip joint models was below 1 mm(proximal femur 0.2 ± 0.1 mm and acetabulum 0.3 ± 0.5 mm). Dice coefficients of the proximal femur and the acetabulum were 98 % and 97 %, respectively. (2)The correlation for total FHC was excellent and significant(r = 0.975, p < 0.001) between automatic MRI-based and manual CT-based 3D-models. Correlation for total FHC (r = 0.979, p < 0.001) between automatic and manual MR-based 3D models was excellent.(3)Preoperative planning and simulation of periacetabular osteotomy was feasible in all patients(100 %) with hip dysplasia or acetabular retroversion. ConclusionsAutomatic segmentation of MRI-based 3D-models using deep learning is as accurate as CT-based 3D-models for patients with hip diseases of childbearing age. This allows radiation-free and patient-specific preoperative simulation and surgical planning of periacetabular osteotomy for patients with DDH.

Three-dimensional accuracy of virtual planning in orthognathic surgery

This study aimed to assess the accuracy of virtual surgical planning (VSP) performed by Dolphin Imaging software (version 11.9; Dolphin Imaging and Management Solutions, Chatsworth, Calif). Ten people requiring bimaxillary surgery and genioplasty were followed up prospectively. All patients had preoperative cone-beam computed tomography, plaster models, and photographs allowing for VSP. Interocclusal intermediate surgical splints were produced using a 3-dimensional (3D) printer. Postoperative images were acquired 15days after surgery using cone-beam computed tomography. ITK-Snap (version 3.6; Cognitica, Philadelphia, Pa) allowed the segmentation of reliable 3D models. Geomagic Qualify 2013 (3D Systems, Rock Hill, SC) and MeshValmet (version 3.0) were used to identify the differences between VSP and actual surgical results through the root mean square values and the 3D translational displacement (3-axes) of the 3D centroid of each model. Discrepancies between the VSP and the actual result were found at the mandible (P=0.013) and the chin (P=0.013) when considering the root mean square values. In addition, 3D centroid differences were found in the transverse and sagittal direction of the right ramus (P=0.034 and P=0.005, respectively) and the sagittal aspect of the left ramus (P=0.025). Considering 2mm as a threshold of clinical relevance, almost all the bone fragments (maxilla, proximal, and distal mandibular segments) were accurately corrected by surgery, although not in the chin. On the basis of the obtained values, it is possible to consider the Dolphin Imaging software as clinically acceptable for performing virtual orthognathic surgical planning.

Segmentation of tibia femoral bone using graph cut method and 3D rendering for FEA

Development of a 3D model of tibia femoral bone is very important for diagnosis and treatment. Knee pain is most commonly observed problem in old age people and it is a very complex joint. Hence analysis of knee joint is very essential. Analysis can be done using 3D model and finite element model. Physical and mechanical properties of knee joint can be analyzed using Finite Element Analysis (FEA) technique. To develop 3D model segmentation process plays an important role which extracts the Region of Interest (ROI) or foreground from background image. This paper presents the graph cut method to segment tibia femoral bone from MRI/CT knee images. Also proposed algorithm illustrates the removal noise using median filter, graph cut method is used for segmenting tibia and femur bones. For volume rendering 3D model of a tibia femoral bone is created using segmented images. Finally finite element meshing is done on 3D model of tibia femoral bone to develop meshed model. This meshed model can be used for analysis and pre surgical planning of knee joint.

Computer-assisted intra-operative verification of surgical outcome for the treatment of syndesmotic injuries through contralateral side comparison.

Fracture reduction and fixation of syndesmotic injuries is a common procedure in trauma surgery. An intra-operative evaluation of the surgical outcome is challenging due to high inter-individual anatomical variation. A comparison to the contralateral uninjured ankle would be highly beneficial but would also incur additional radiation and time consumption. In this work, we pioneer automatic contralateral side comparison while avoiding an additional 3D scan. We reconstruct an accurate 3D surface of the uninjured ankle joint from three low-dose 2D fluoroscopic projections. Through CNN complemented 3D shape model segmentation, we create a reference model of the injured ankle while addressing the issues of metal artifacts and initialization. Following 2D-3D multiple bone reconstruction, a final reference contour can be created and matched to the uninjured ankle for contralateral side comparison without any user interaction. The accuracy and robustness of individual workflow steps were assessed using 81 C-arm datasets, with 2D and 3D images available for injured and uninjured ankles. Furthermore, the entire workflow was tested on eleven clinical cases. These experiments showed an overall average Hausdorff distance of [Formula: see text]mm measured at clinical evaluation level. Reference contours of the contralateral side reconstructed from three projection images can assist surgeons in optimizing reduction results, reducing the duration of radiation exposure and potentially improving postoperative outcomes in the long term.