OrthoFusion: A Super-Resolution Algorithm to Fuse Orthogonal CT Volumes

OrthoFusion: A Super-Resolution Algorithm to Fuse Orthogonal CT Volumes | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article OrthoFusion: A Super-Resolution Algorithm to Fuse Orthogonal CT Volumes Rebecca E. Abbott, Alain Nishimwe, Hadi Wiputra, Ryan E. Breighner, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4117386/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 09 Jan, 2025 Read the published version in Scientific Reports → Version 1 posted 11 You are reading this latest preprint version Abstract OrthoFusion, an intuitive super-resolution algorithm, is presented in this study to enhance the spatial resolution of clinical CT volumes. The efficacy of OrthoFusion is evaluated, relative to high-resolution CT volumes (ground truth), by assessing image volume and derived bone morphological similarity, as well as its performance in specific applications in 2D-3D registration tasks. Results demonstrate that OrthoFusion significantly reduced segmentation time, while improving structural similarity of bone images and relative accuracy of derived bone model geometries. Moreover, it proved beneficial in the context of biplane videoradiography, enhancing the similarity of digitally reconstructed radiographs to radiographic images and improving the accuracy of relative bony kinematics. OrthoFusion's simplicity, ease of implementation, and generalizability make it a valuable tool for researchers and clinicians seeking high spatial resolution from existing clinical CT data. This study opens new avenues for retrospectively utilizing clinical images for research and advanced clinical purposes, while reducing the need for additional scans, mitigating associated costs and radiation exposure. Biological sciences/Biological techniques/Imaging/X ray tomography Health sciences/Medical research Super Resolution Bone Models Computed Tomography Image Fusion Spatial Resolution Enhancement Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 INTRODUCTION Three-dimensional visualization and creation of bone models are essential for many applications in medical imaging, such as diagnosis of fracture or disease, pre-operative planning, implant design, trainee and patient education, and motion analyses requiring digitally reconstructed radiographs (DRRs). Computed tomography (CT) is a common imaging modality that uses x-rays to produce cross-sectional images of the body, offering detailed volumetric renderings of anatomy and enabling precise examination of internal structures. Although CT is the gold standard for morphologic bone imaging and thus generation of bone models, CT acquisition incurs radiation exposure, cost, and time. Thus, there are concerted efforts across disciplines to reduce radiation exposure to patients to be ‘as low as reasonably achievable’ (ALARA) for the clinical indication. Despite these risks, clinical CT orders are still increasing. 1 Clinical CT scans are routinely archived in the patient’s electronic medical records (EMR). Typical clinical scans in EMR are highly anisotropic for a variety of reasons, but primarily driven by methods to reduce both radiation exposure to patients and image file size to account for storage limitations. Imaging protocols export 2 or 3 orthogonal volumes (axial, sagittal, and/or coronal) with high in-plane resolution (e.g., 0.3x0.3 mm), but with greater slice thickness and spacing, resulting in a 10-fold courser through-plane resolution (e.g., 3 mm). Unfortunately, high spatial resolution and cortical bone contrast are necessary for visualization of the segmentation process necessary for bone model construction. Therefore, spatial resolution of many CT volumes in EMR is inadequate for multiplanar reconstruction and bone modelling applications. A method to increase the spatial resolution of clinical CT volumes obtained from EMR would open new opportunities for researchers and clinicians to create multi-planar reconstructions and accurate bone models from the wealth of existing data already stored in EMRs. Super-Resolution is a broad term that refers to a family of techniques to combine multiple low-resolution volumes to produce higher resolution volumes. 2 – 4 Various super-resolution techniques have been employed to improve medical imaging quality and resolution, including iterative reconstruction 2 , deconvolution 3 , deep learning-based approaches 5 , and registration and fusion 6 . ‘Registration and fusion’ involves spatially aligning multiple volumes (registration) and combining the information to a single volume that incorporates data from multiple views (fusion). Despite these technological advancements, super-resolution tools in medical imaging are often tailored to a specific research task and are not readily available to researchers or clinicians, requiring advanced programming skills to implement. Additionally, many require large datasets for learning or are not generalizable to other applications. The objective of this study was to design, implement, and evaluate an intuitive super-resolution algorithm to increase the spatial resolution of low-resolution clinical CT volumes. This effort was motivated by the fact that some participants in our cervical kinematics studies already had recently acquired clinical CT imaging in their EMR. To avoid an additional CT scan and associated radiation exposure, an ethical alternative was desired. Our application requires a subject-specific bone model created from a CT volume to build the DRR for shape-matching via biplane videoradiography. Biplane videoradiography is an advanced imaging technique that captures dynamic radiographs from two views simultaneously so that 3-D bone kinematics can quantified. Our super-resolution solution is generalizable beyond our specific application and will provide an opportunity to other researchers and clinicians to use clinical CTs from EMR to create high-resolution 3-D visualizations and bone models. METHODS OrthoFusion Super-Resolution Algorithm The OrthoFusion super-resolution algorithm registers and fuses 2 or 3 orthogonal low-resolution CT volumes to create a single volume with high spatial resolution. A graphical representation of the algorithm is provided in Fig. 1 , starting from the axial, coronal, and sagittal “Clinical” volumes. First, each low-resolution clinical volume is linearly interpolated to the same high-resolution isotropic grid (0.2 x 0.2 x 0.2 mm) such that an intensity value is assigned to each voxel (registration). Second, the voxel-by-voxel average is computed (fusion). By including multiple orthogonal volumes (axial, coronal, and sagittal), all available information is utilized, capturing features that would otherwise be missed. Image Acquisition and Processing To evaluate the effectiveness of the OrthoFusion algorithm, we acquired high-resolution CT volumes to serve as the ground truth. CT imaging (B60S; Siemens Biograph PET/CT, Knoxville, USA) of the cervical spine was acquired from 4 participants (3 female/1 male, age 21 to 40 years old) as part of a larger study approved by the University of Minnesota Institutional Review Board. Participants provided informed signed consent and all methods were performed in accordance with all relevant guidelines and regulations. The CT scans were reconstructed into 3 orthogonal high-resolution volumes (axial, sagittal, and coronal), with in-plane resolutions of approximately 0.2 x 0.2 and a through-plane resolution of 0.6 mm. Image Processing Four CT cases were compared within each participant: 1) high-resolution (HR), 2) clinical (Clin), 3) resliced (RS), and 4) super-resolution (OrthoFusion) volumes. Image processing was performed using custom code in MeVisLab 3.7.2 (MeVisLab Medical Solutions AG). The study design is illustrated in Fig. 2 . An axial clinical (Clin) volume with a through-plane resolution of 3 mm was simulated by down-sampling the axial HR volume using Gaussian interpolation. A resampled (RS) volume was created by then up-sampling the axial Clin volume back to 0.6 mm through-plane resolution using Lanczos interpolation, a commonly used technique. 7 The super-resolution (OrthoFusion) volume was constructed using the OrthoFusion algorithm described above, fusing all 3 orthogonal volumes and resulting in an isotopic resolution of 0.2 mm. General Procedure Bone Segmentation Following image acquisition and processing, subject-specific bone models of the C4 and C5 vertebrae were segmented from each CT case in Mimics (v23, Materialize, Leuven, Belgium) by RA – an experienced segmentation expert. Initial segmentation was performed using a semi-automated CT bone segmentation tool (CT Bone Wizard) that utilizes initial thresholding and a continuity-based algorithm. Additional manual refinement was performed as needed. The segmented CT volumes were then used to create 3-D geometries for morphological comparisons and Digitally Reconstructed Radiographs (DRRs) for kinematic analysis. Segmentation Time The approximate time and tools (semi-automated vs. manual) required for segmenting each cervical vertebra were recorded for each case. Image Similarity Image quality assessment was performed comparing each case to the respective HR volume using the structural similarity (SSIM) index and peak signal to noise ratio ( PSNR ). 8 The structural similarity map includes each pixel in the image, based on its relationship to other pixels in a local radius. The measures were calculated on masked CT volumes to isolate the bone, and on the whole CT volumes. The mean and standard deviation of the local SSIM values were compared between cases. The PSNR was computed on the whole volumes. Morphological Similarity The morphological similarity of the subject specific 3-D bone geometries was assessed using CloudCompare (v2.10). The point-by-point 3-D minimum Euclidean distance between the case geometries and reference HR geometry was computed to determine 1) bias: the mean signed error, 2) precision: the standard deviation of the signed error, and 3) MAE: the mean absolute error. Application Specific Procedure Each of the 4 cases (HR, Clin, RS, and SR) for each of the 4 participants was taken through the following pipeline shown in Fig. 3 for extracting segmental kinematics from biplane videoradiography. Briefly, the process includes simultaneous collection of dynamic radiographs from two views during cervical motion tasks, shape-matching the DRRs to the biplane radiographs, and computing the relative kinematics between the C4 and C5 vertebrae. The error associated with our traditional biplane videoradiography pipeline for the cervical spine kinematics is ≤ 0.49 mm and ≤ 1.80⁰. 9 The performance of each case was compared to our traditional high-resolution CT at multiple phases in pipeline. Biplane Videoradiography Dynamic radiographs of the cervical spine were acquired using a custom biplane videoradiography system (Imaging Systems & Services, Inc., Painesville, OH, USA) at 30 Hz with approximate imaging parameters of 160 mA, 70 kV, and 0.16 mSv/trial. Each participant performed a 3-second trial of neck flexion-extension, lateral bending, and axial rotation. Post-processing of the radiographic images included undistortion and filtering (DSX Suite, C-Motion Inc., Germantown, MD, USA) and calibration (XMALab, Brown University, RI, USA). Shape-matching Shape-matching is the process by which 3-D kinematics are extracted by registering the DRRs (created from the projections of the segmented CT volumes) to the biplane radiographs. The dynamic 3-D positions and orientations of the C4 and C5 vertebrae were resolved using Autoscoper (Brown University, RI, USA), an open-source shape-matching software 10 , 11 . Both the DRRs and radiographs are filtered (Gaussian, sobel, etc.) to enhance features and edges. Auto-registration is performed using a particle swarm optimization technique with the inverse of the normalized cross-correlation ( NCC ) as the cost function. Manual refinement was performed when the optimization algorithm was unable to find an adequate match. The global 3-D kinematics of each bone was exported. Shape-Matching Similarity The match of the DRR to radiographs during the shape-matching process is quantified using the NCC . The NCC is a pixel-by-pixel correlation of the projection of the 3-D DRR to the 2-D radiographs. The inverse of the NCC is used as the optimization criteria for auto-registration by Autoscoper. A lower value for NCC value indicates a better fit. Because the NCC may be affected by radiograph image parameters, the NCC error metric was normalized frame-by-frame by the corresponding HR NCC for each participant and trial. $$NCC error = \frac{\left(NC{C}_{case}-NC{C}_{HR}\right)}{NC{C}_{HR}}$$ The mean and standard deviation of the NCC error was computed across all frames of each trial. Relative Kinematics Relative kinematics between vertebrae were computed using KinematicsToolbox v.4.7.2. 12 Local coordinate systems were created for each vertebra by digitizing anatomic landmarks such that the origin is in the center of the body, x-axis points anterior, and y-axis points left, z-axis points superior. 9 The rotations and displacements of C4 (relative to C5) were computed for each of the trials, cases, and participants. Relative Kinematics Accuracy The relative kinematics accuracy is represented by two scalar values: the orientation error and position error. The orientation error is the distance between quaternion rotations 13 of the case, p, and HR, q, defined as: $$Ѳ = 2 co{s}^{-1}\left(\right|\left|\right)$$ $$ = {p}_{1}{q}_{1} + {p}_{2}{q}_{2} + {p}_{3}{q}_{3} + {p}_{4}{q}_{4}$$ The position error was computed as the Euclidean distance between the relative C4-C5 displacements of the case and HR. The mean and variability (standard deviation) of the orientation and position error were computed across each trial. Statistics ANOVA was used to determine groupwise differences for each outcome measure, followed by appropriate post-hoc comparisons. Normality was assessed and vertebral levels were treated as independent. A one-way repeated measures ANOVA was conducted for measures that are not impacted by trial, i.e.., image similarity and bone morphology. A two-way ANOVA was used to examine the effect of both cases and trial for the shape-matching and relative kinematics measures. RESULTS The results are reported in Fig. 4 , with representative examples in Fig. 5 . Image Similarity The SSIM of the masked volumes takes both the internal structure and the segmentation of the bone into account. The mean masked SSIM of the SR volume (0.696 ± 0.024) was statistically greater than the Clin (0.589 ± 0.047; p = 0.001) and RS (0.602 ± 0.037; p = 0.001) volumes. Similarly, the SR volume had significantly less SSIM variability (0.2 ± 0.014) than the Clin (0.313 ± 0.009; p < 0.001) and RS (0.290 ± 0.011; p < 0.001) masked volumes. The RS volume also had less variability than the Clin volume (p = 0.001). The SSIM of the whole CT volumes represents the structural similarity of the non-segmented CT volumes. The mean SSIM of the whole RS volume (0.766 ± 0.033) was statistically higher than SR (0.747 ± 0.031, p = 0.018) and the Clin volume (0.753 ± 0.031, p < 0.001). The SSIM variability was not statistically different between the cases. The PSNR of the whole volumes were also not statistically different. Segmentation Time Segmentation of each vertebra was successfully completed with the automated “CT Bone Wizard” tool in Mimics (v23) for the HR and OrthoFusion (SR) volumes with minimal need for manual refinement. This resulted in a segmentation time of approximately 15 minutes per vertebra. Manual segmentation was required for the Clin and RS vertebrae due to poor distinction between bones (primarily facet joints). This resulted in a segmentation time of approximately 2.5 hours per vertebra. Morphological Similarity Significant differences between cases were found for bias, precision, and MAE distances of the 3-D bone geometry. Bias: The SR bone models (-0.01 ± 0.12 mm) had significantly less bias than the Clin (0.47 ± 0.39 mm; p = 0.017) and RS (-0.24 ± 0.15 mm; p = 0.031) bone models. Precision: The SR (0.38 ± 0.13 mm) bone models were significantly more precise than the RS (0.71 ± 0.19 mm; p = 0.003) or Clin (1.18 ± 0.44 mm; p = 0.005) bone models. MAE: The MAE of the SR bone models (0.29 ± 0.06 mm) were significantly less than the Clin (0.85 ± 0.31 mm; p = 0.002) and RS (0.54 ± 0.12; p = 0.001) bone models. Shape-Matching Similarity The normalized cross-correlation coefficient ( NCC ) is the measure of pixel-by-pixel similarity of the shape-matched DRR and biplane radiographs. The NCC error is the normalized difference between the case and HR NCC values at each frame of the trial. Within-subjects effects of both trial and case were found to be significant for the normalized NCC error measures. A significant trial effect was observed such that the lateral bending trials had a significantly lower (better matched) mean NCC error than the axial rotation (p = 0.006) and flexion-extension (p = 0.001) trials. A significant case effect was observed such that the mean NCC error for the SR cases (0.106 ± 0.162) were significantly lower than the Clin (0.261 ± 0.119; p = 0.002) and RS (0.194 ± 0.105; p = 0.024) cases. A significant trial effect was observed such that the LB trials had significantly less variability in NCC error than the FE trials (p = 0.042). A significant case effect was observed such that the NCC error variability for the SR cases (0.060 ± 0.014) was significantly lower than that of the Clin (0.162 ± 0.023; p = 0.002) or RS (0.109 ± 0.010; p = 0.005). Relative Kinematics Accuracy The relative kinematics of the C4 to C5 vertebrae were compared between the HR and cases to determine kinematics accuracy. Significant within-subjects effects of case, but not trial, were found for the kinematics measures. The SR case produced relative kinematics with the least orientation and displacement error. The mean orientation error for the SR case (1.7 ± 1.5 deg) was significantly lower than the Clin (8.0 ± 5.8 deg; p = 0.019) and approaching significance for RS (3.8 ± 2.2 deg; p = 0.053) case. Similarly, the variability of the orientation error for the SR case (1.1 ± 1.2 deg) was significantly lower than the Clin (4.5 ± 2.3 deg; p = 0.004) and approaching significance for RS (3.1 ± 2.4 deg; p = 0.061) case. The mean displacement error for the SR case (0.5 ± 0.2 mm) was significantly lower than that of the Clin (2.5 ± 0.5 mm) and RS (1.2 ± 0.3 mm) cases. The variability of displacement error for the SR case (0.4 ± 0.49 mm) was significantly lower than the Clin (1.4 ± 1.1 mm; p = 0.014) and RS (0.8 ± 0.8 mm; p = 0.026) cases. DISCUSSION We designed, implemented, and evaluated OrthoFusion super-resolution technique to increase the spatial resolution of previously acquired low-resolution CT volumes. OrthoFusion significantly improved segmentation time, the structural similarity of the bone images, and the accuracy of bone model geometry. Specific to our biplane videoradiography application, OrthoFusion improved the DRR to radiograph similarity during shape-matching and the accuracy of the relative C4-C5 kinematics. Ultimately, these improvements enable us to use previously acquired clinical CT scans from participant EMR instead of acquiring additional CT imaging as part of our studies, reducing radiation exposure, time, and cost. Super-resolution has utility well beyond our specific biplane videoradiography application. The abundance of imaging in electronic medical records is an untapped opportunity for retrospective studies that were previously not possible due to poor through-plane resolution. Super-resolution can benefit applications requiring accurate and high resolution multiplanar reconstruction, segmentation, DRRs, and 3-D geometries. To our knowledge, currently available imaging software applications do not provide a tool that increases spatial resolution from multiple low-resolution volumes. Many super-resolution techniques have been reported in the medical imaging literature, often with the goal of exceeding hardware resolution and decreasing noise 2 However, these algorithms are complex and challenging to implement for a typical researcher or clinician. Machine learning algorithms offer another approach, however they require training on large datasets which are not always feasible. 5 , 14 , 15 Morphology morphing algorithms in conjunction with biplane slot scanners offer a low-dose approach to generate bone models, but are limited based on target specific geometries and do not yield a volumetric imaging dataset. 16 OrthoFusion is an easily implemented technique that is generalizable to any CT imaging target and does not require training. Further development of the OrthoFusion algorithm by integrating filtering and iterative algorithms will provide new possibilities for the use of the large dataset of clinical imaging in EMR. There are several limitations that should be considered in the interpretation of this study. First, we applied OrthoFusion super-resolution to our specific application. The benefits of improving spatial resolution for bone segmentation are generalizable, but performance should be assessed considering the requirements of new applications. Second, the processing and analysis involved in our biplane videoradiography pipeline requires significant user interaction and the cases were visibly identifiable, making blinding impossible. Finally, OrthoFusion uses a simple voxel-by-voxel averaging algorithm that creates a blurring effect. The addition of anisotropic filtering 17 or a spatial weighted average approach 18 are likely to enhance the image quality. This study demonstrates the utility of super-resolution to increase the spatial resolution of previously acquired low-resolution CT volumes. The OrthoFusion technique enabled us to use CT imaging from participant EMR for our biplane videoradiography studies, thus reducing exposure to radiation and costs associated with acquiring imaging. This straight-forward technique is easy to implement, generalizable to other applications, and allows researchers and clinicians that require high spatial resolution CT to utilize the abundance of clinical imaging in EMR. Declarations Funding Disclosures: This work was funded by the following NIH grants: R03HD09771, TL1R002493, UL1TR002494, and F32AR082276 Additionally, this work was supported by the Minnesota Partnership for Biotechnology and Medical Genomics (MHP IF #14.02). Acknowledgements The authors would like to thank Craig Kage for his assistance with data collection. Author Contributions Conceptualization, study design: R.E.A., A.M.E. Methodology: R.E.A., A.N., H.W., R.E.B. Formal analysis: R.E.A., A.N., A.M.E., Writing original draft: R.E.A. Review and editing: all authors. Funding acquisition, project management: A.M.E. All authors have read and agreed to the published version of the manuscript. Data Availability The code and datasets analyzed during this study are available in the Data Repository for U of M (https://conservancy.umn.edu/drum: specific link TBD). Competing Interests The authors declare no competing interests. References Smith-Bindman, R. et al. Trends in Use of Medical Imaging in US Health Care Systems and in Ontario, Canada, 2000–2016. JAMA 322, 843–856 (2019). Hakimi, W. E. & Wesarg, S. Accurate super-resolution reconstruction for CT and MR images. in Proceedings of the 26th IEEE International Symposium on Computer-Based Medical Systems 445–448 (2013). doi: 10.1109/CBMS.2013.6627837 . Odet, C., Peyrin, F. & Goutte, R. Improved resolution of medical 3-D x-ray computed-tomographic images. in Visual Communications and Image Processing ’90: Fifth in a Series vol. 1360 658–664 (SPIE, 1990). Isaac, J. S. & Kulkarni, R. Super resolution techniques for medical image processing. in 2015 International Conference on Technologies for Sustainable Development (ICTSD) 1–6 (2015). doi: 10.1109/ICTSD.2015.7095900 . Jiang, C., Zhang, Q., Fan, R. & Hu, Z. Super-resolution CT Image Reconstruction Based on Dictionary Learning and Sparse Representation. Sci Rep 8, 8799 (2018). Greenspan, H. Super-Resolution in Medical Imaging. The Computer Journal 52, 43–63 (2009). Moraes, T., Amorim, P., Da Silva, J. V. & Pedrini, H. Medical image interpolation based on 3D Lanczos filtering. Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization 8, 294–300 (2020). Renieblas, G. P., Nogués, A. T., González, A. M., Gómez-Leon, N. & Del Castillo, E. G. Structural similarity index family for image quality assessment in radiological images. J Med Imaging (Bellingham) 4, 035501 (2017). Kage, C. C. et al. Validation of an automated shape-matching algorithm for biplane radiographic spine osteokinematics and radiostereometric analysis error quantification. PLOS ONE 15, e0228594 (2020). Miranda, D. L. et al. Static and Dynamic Error of a Biplanar Videoradiography System Using Marker-Based and Markerless Tracking Techniques. J Biomech Eng 133, 121002-121002-8 (2011). Akhbari, B., Morton, A. M., Moore, D. C. & Crisco, J. J. Biplanar Videoradiography to Study the Wrist and Distal Radioulnar Joints. J Vis Exp 10.3791 /62102 (2021) doi:10.3791/62102. Rebekah L. Lawrence. KinematicsToolbox: A Comprehensive Software Program for Analyzing and Visualizing Radiographic Motion Capture Data. (2023) doi: https://doi.org/10.7936/6RXS-103624 . Huynh, D. Q. Metrics for 3D Rotations: Comparison and Analysis. J Math Imaging Vis 35, 155–164 (2009). Jia, Y., Gholipour, A., He, Z. & Warfield, S. K. A New Sparse Representation Framework for Reconstruction of an Isotropic High Spatial Resolution MR Volume From Orthogonal Anisotropic Resolution Scans. IEEE Transactions on Medical Imaging 36, 1182–1193 (2017). Danker Khoo, J. J., Lim, K. H. & Sien Phang, J. T. A Review on Deep Learning Super Resolution Techniques. in 2020 IEEE 8th Conference on Systems, Process and Control (ICSPC) 134–139 (2020). doi: 10.1109/ICSPC50992.2020.9305806 . Dubousset, J. et al. A new 2D and 3D imaging approach to musculoskeletal physiology and pathology with low-dose radiation and the standing position: the EOS system. Bull Acad Natl Med 189, 287–297; discussion 297–300 (2005). Kurt, B., Nabiyev, V. V. & Turhan, K. Medical images enhancement by using anisotropic filter and CLAHE. in 2012 International Symposium on Innovations in Intelligent Systems and Applications 1–4 (2012). doi: 10.1109/INISTA.2012.6246971 . Bätz, M., Eichenseer, A. & Kaup, A. Multi-image super-resolution using a dual weighting scheme based on Voronoi tessellation. in 2016 IEEE International Conference on Image Processing (ICIP) 2822–2826 (2016). doi: 10.1109/ICIP.2016.7532874 . Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 09 Jan, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 23 Sep, 2024 Reviews received at journal 19 Sep, 2024 Reviews received at journal 09 Sep, 2024 Reviewers agreed at journal 09 Sep, 2024 Reviewers agreed at journal 31 Aug, 2024 Reviewers agreed at journal 04 Jun, 2024 Reviewers invited by journal 15 Apr, 2024 Editor assigned by journal 15 Apr, 2024 Editor invited by journal 29 Mar, 2024 Submission checks completed at journal 29 Mar, 2024 First submitted to journal 17 Mar, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4117386","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":286588109,"identity":"12e83582-545c-49fd-8d94-6c99296c2307","order_by":0,"name":"Rebecca E. Abbott","email":"","orcid":"","institution":"University of Minnesota","correspondingAuthor":false,"prefix":"","firstName":"Rebecca","middleName":"E.","lastName":"Abbott","suffix":""},{"id":286588110,"identity":"f475d8bc-9b70-4a3d-bd6b-cadf711ad206","order_by":1,"name":"Alain Nishimwe","email":"","orcid":"","institution":"University of Minnesota","correspondingAuthor":false,"prefix":"","firstName":"Alain","middleName":"","lastName":"Nishimwe","suffix":""},{"id":286588111,"identity":"7eef5ff3-4926-42fc-a0f6-620ce83d771e","order_by":2,"name":"Hadi Wiputra","email":"","orcid":"","institution":"University of Minnesota","correspondingAuthor":false,"prefix":"","firstName":"Hadi","middleName":"","lastName":"Wiputra","suffix":""},{"id":286588112,"identity":"0d163542-7f99-40f7-a10e-eeab335e3e67","order_by":3,"name":"Ryan E. Breighner","email":"","orcid":"","institution":"Hospital for Special Surgery","correspondingAuthor":false,"prefix":"","firstName":"Ryan","middleName":"E.","lastName":"Breighner","suffix":""},{"id":286588114,"identity":"009d08b9-055c-4ddf-9bc1-c7315ccaca3a","order_by":4,"name":"Arin M. Ellingson","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvklEQVRIiWNgGAWjYDACduaGAwwVNsz8EC4zEVqYGYFazqSxSzaQooWBse0wv8EBYrXwMzM2Hi5gY5Y2vpGd9oChwjqxgZAWyWbGhsMzeNiMzW7kbjdgOJNOWIvBYaAWHgmeZKCWbRJAFxLWYg/WYiBRv3kGSMs/IrQYMIO0JBgwG0iAtDQQoUUCbMuBBGaJM2+3SSQcSzcmqIW/vfnwZ95//5n524G2fKixliWoBRUkkKZ8FIyCUTAKRgEuAADBnzwnXDtsVQAAAABJRU5ErkJggg==","orcid":"","institution":"University of Minnesota","correspondingAuthor":true,"prefix":"","firstName":"Arin","middleName":"M.","lastName":"Ellingson","suffix":""}],"badges":[],"createdAt":"2024-03-17 14:14:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4117386/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4117386/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-85516-y","type":"published","date":"2025-01-09T15:57:04+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":54036059,"identity":"dca06031-98c2-43e3-8cf7-6a8f4af99835","added_by":"auto","created_at":"2024-04-03 17:00:53","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":141258,"visible":true,"origin":"","legend":"\u003cp\u003eOrthoFusion Super Resolution Algorithm. Orthogonal clinical volumes with low through-plane resolution (top row) are linearly interpolated onto an isotropic high-resolution 3-D grid (bottom row). A voxel-by-voxel average is computed to produce the super resolution volume (right).\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4117386/v1/c2ee3f6afa06a447270885b7.png"},{"id":54036058,"identity":"efdb83f8-2de1-4c01-99ff-f86dc91c2eb7","added_by":"auto","created_at":"2024-04-03 17:00:53","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":155830,"visible":true,"origin":"","legend":"\u003cp\u003eStudy design. Four cases were extracted to evaluate the OrthoFusion algorithm – High Resolution (HR), Clinical (Clin), Resliced (RS), and Super Resolution (SR). The axial HR volume was considered the ‘ground truth’. First, HR CT volumes were acquired and reconstructed into 3 orthogonal volumes (axial, coronal, sagittal) with in-plane resolutions of ~0.2 x 0.2 mm and a through-plane resolutions of 0.6 mm (top row). Next, Clin volumes were simulated by down-sampling the HR volumes using Gaussian interpolation to a 3 mm through-plane resolution (middle row). The OrthoFusion algorithm (box) was applied, as depicted in Figure 1. The RS case volume was derived from the axial Clin volume by interpolating back to a 0.6 mm through-plane resolution using Lanczos interpolation.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-4117386/v1/28b7d4696429969b010fe358.png"},{"id":54037083,"identity":"c3de7e0a-ece7-4e89-a1a8-57e5b0ab5f62","added_by":"auto","created_at":"2024-04-03 17:08:53","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":220473,"visible":true,"origin":"","legend":"\u003cp\u003eThe biplane videoradiography procedure for measuring intersegmental kinematics and performance measures across the pipeline. To capture the dynamic motion of individual vertebrae, biplane videoradiography is used in conjunction with corresponding bone model(s). This technique uses two synchronized radiographic units to simultaneously capture planar x-ray movies from two perspectives. In parallel, we create subject-specific bone models segmented from a CT scan. The process of shape-matching involves software that simulates a radiographic projection of the bone model – called a digitally reconstructed radiograph or DRR – onto the dynamic radiographic images. An optimization in each frame finds the closest match for the two projections to place and orient the vertebra in 3D space. The OrthoFusion Super Resolution approach was evaluated at each step along this process by quantifying key performance measures.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-4117386/v1/563c9d0c01e025746bc2c8b7.png"},{"id":54036060,"identity":"31f56835-de46-42ea-98fd-c6aa617ab5cb","added_by":"auto","created_at":"2024-04-03 17:00:53","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":72761,"visible":true,"origin":"","legend":"\u003cp\u003eComparisons between the simulated Clinical (Clin), Resliced (RS), and Super Resolution (SR) volumes referenced to the “ground truth” High Resolution (HR) volume in a) image similarity measures, b) bone model morphological similarity, c) shape-matching similarity, and d) relative kinematics accuracy. Error bars represent standard deviation. Asterisk indicates a significant difference of p \u0026lt; 0.05 between groups. SSIM = Structural Similarity Index Measure, NCC = normalized cross-correlation coefficient.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-4117386/v1/164f3570279a6c5ae2f7be6e.png"},{"id":54036062,"identity":"14e620d1-228f-4e13-8dfe-1ed86c0f92a8","added_by":"auto","created_at":"2024-04-03 17:00:53","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":439880,"visible":true,"origin":"","legend":"\u003cp\u003eSummary Figure. Representative data illustrating the performance measures across the 4 case volumes, with the High Resolution case as the ground truth. Multiplanar Reconstructions emphasize the separation of the facet joints. Segmentation time was substantially less for the High Resolution and Super Resolution cases. Image similarity was improved with the Super Resolution approach, depicted from local SSIM maps where white indicates greater similarity. These improved performances over the Clinical and Resliced volumes yielded a greater morphological similarity and ultimately, for our applications purpose, the Super Resolution Digitally Reconstructed Radiograph (DRR) exhibited better performance for shape-matching similarity and relative kinematics accuracy.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-4117386/v1/74c3f83e69ad89d657aad043.png"},{"id":73693813,"identity":"e5e6e5d8-4a6c-4e08-85ec-eac9b229b5f5","added_by":"auto","created_at":"2025-01-13 16:07:26","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1652545,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4117386/v1/597ee847-67d9-4e09-b9c4-7d1f71a66caa.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"OrthoFusion: A Super-Resolution Algorithm to Fuse Orthogonal CT Volumes","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eThree-dimensional visualization and creation of bone models are essential for many applications in medical imaging, such as diagnosis of fracture or disease, pre-operative planning, implant design, trainee and patient education, and motion analyses requiring digitally reconstructed radiographs (DRRs). Computed tomography (CT) is a common imaging modality that uses x-rays to produce cross-sectional images of the body, offering detailed volumetric renderings of anatomy and enabling precise examination of internal structures. Although CT is the gold standard for morphologic bone imaging and thus generation of bone models, CT acquisition incurs radiation exposure, cost, and time. Thus, there are concerted efforts across disciplines to reduce radiation exposure to patients to be \u0026lsquo;as low as reasonably achievable\u0026rsquo; (ALARA) for the clinical indication. Despite these risks, clinical CT orders are still increasing.\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eClinical CT scans are routinely archived in the patient\u0026rsquo;s electronic medical records (EMR). Typical clinical scans in EMR are highly anisotropic for a variety of reasons, but primarily driven by methods to reduce both radiation exposure to patients and image file size to account for storage limitations. Imaging protocols export 2 or 3 orthogonal volumes (axial, sagittal, and/or coronal) with high in-plane resolution (e.g., 0.3x0.3 mm), but with greater slice thickness and spacing, resulting in a 10-fold courser through-plane resolution (e.g., 3 mm). Unfortunately, high spatial resolution and cortical bone contrast are necessary for visualization of the segmentation process necessary for bone model construction. Therefore, spatial resolution of many CT volumes in EMR is inadequate for multiplanar reconstruction and bone modelling applications.\u003c/p\u003e \u003cp\u003eA method to increase the spatial resolution of clinical CT volumes obtained from EMR would open new opportunities for researchers and clinicians to create multi-planar reconstructions and accurate bone models from the wealth of existing data already stored in EMRs. Super-Resolution is a broad term that refers to a family of techniques to combine multiple low-resolution volumes to produce higher resolution volumes.\u003csup\u003e\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e Various super-resolution techniques have been employed to improve medical imaging quality and resolution, including iterative reconstruction\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e, deconvolution\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e, deep learning-based approaches\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e, and registration and fusion\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. \u0026lsquo;Registration and fusion\u0026rsquo; involves spatially aligning multiple volumes (registration) and combining the information to a single volume that incorporates data from multiple views (fusion). Despite these technological advancements, super-resolution tools in medical imaging are often tailored to a specific research task and are not readily available to researchers or clinicians, requiring advanced programming skills to implement. Additionally, many require large datasets for learning or are not generalizable to other applications.\u003c/p\u003e \u003cp\u003eThe objective of this study was to design, implement, and evaluate an intuitive super-resolution algorithm to increase the spatial resolution of low-resolution clinical CT volumes. This effort was motivated by the fact that some participants in our cervical kinematics studies already had recently acquired clinical CT imaging in their EMR. To avoid an additional CT scan and associated radiation exposure, an ethical alternative was desired. Our application requires a subject-specific bone model created from a CT volume to build the DRR for shape-matching via biplane videoradiography. Biplane videoradiography is an advanced imaging technique that captures dynamic radiographs from two views simultaneously so that 3-D bone kinematics can quantified. Our super-resolution solution is generalizable beyond our specific application and will provide an opportunity to other researchers and clinicians to use clinical CTs from EMR to create high-resolution 3-D visualizations and bone models.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eOrthoFusion Super-Resolution Algorithm\u003c/h2\u003e \u003cp\u003eThe OrthoFusion super-resolution algorithm registers and fuses 2 or 3 orthogonal low-resolution CT volumes to create a single volume with high spatial resolution. A graphical representation of the algorithm is provided in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, starting from the axial, coronal, and sagittal \u0026ldquo;Clinical\u0026rdquo; volumes. First, each low-resolution clinical volume is linearly interpolated to the same high-resolution isotropic grid (0.2 x 0.2 x 0.2 mm) such that an intensity value is assigned to each voxel (registration). Second, the voxel-by-voxel average is computed (fusion). By including multiple orthogonal volumes (axial, coronal, and sagittal), all available information is utilized, capturing features that would otherwise be missed.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eImage Acquisition and Processing\u003c/h2\u003e \u003cp\u003eTo evaluate the effectiveness of the OrthoFusion algorithm, we acquired high-resolution CT volumes to serve as the ground truth. CT imaging (B60S; Siemens Biograph PET/CT, Knoxville, USA) of the cervical spine was acquired from 4 participants (3 female/1 male, age 21 to 40 years old) as part of a larger study approved by the University of Minnesota Institutional Review Board. Participants provided informed signed consent and all methods were performed in accordance with all relevant guidelines and regulations. The CT scans were reconstructed into 3 orthogonal high-resolution volumes (axial, sagittal, and coronal), with in-plane resolutions of approximately 0.2 x 0.2 and a through-plane resolution of 0.6 mm.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eImage Processing\u003c/h2\u003e \u003cp\u003eFour CT cases were compared within each participant: 1) high-resolution (HR), 2) clinical (Clin), 3) resliced (RS), and 4) super-resolution (OrthoFusion) volumes. Image processing was performed using custom code in MeVisLab 3.7.2 (MeVisLab Medical Solutions AG). The study design is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. An axial clinical (Clin) volume with a through-plane resolution of 3 mm was simulated by down-sampling the axial HR volume using Gaussian interpolation. A resampled (RS) volume was created by then up-sampling the axial Clin volume back to 0.6 mm through-plane resolution using Lanczos interpolation, a commonly used technique.\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e The super-resolution (OrthoFusion) volume was constructed using the OrthoFusion algorithm described above, fusing all 3 orthogonal volumes and resulting in an isotopic resolution of 0.2 mm.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eGeneral Procedure\u003c/h2\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003eBone Segmentation\u003c/h2\u003e \u003cp\u003eFollowing image acquisition and processing, subject-specific bone models of the C4 and C5 vertebrae were segmented from each CT case in Mimics (v23, Materialize, Leuven, Belgium) by RA \u0026ndash; an experienced segmentation expert. Initial segmentation was performed using a semi-automated CT bone segmentation tool (CT Bone Wizard) that utilizes initial thresholding and a continuity-based algorithm. Additional manual refinement was performed as needed. The segmented CT volumes were then used to create 3-D geometries for morphological comparisons and Digitally Reconstructed Radiographs (DRRs) for kinematic analysis.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eSegmentation Time\u003c/h2\u003e \u003cp\u003eThe approximate time and tools (semi-automated vs. manual) required for segmenting each cervical vertebra were recorded for each case.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eImage Similarity\u003c/h2\u003e \u003cp\u003eImage quality assessment was performed comparing each case to the respective HR volume using the structural similarity (SSIM) index and peak signal to noise ratio (\u003cem\u003ePSNR\u003c/em\u003e).\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e The structural similarity map includes each pixel in the image, based on its relationship to other pixels in a local radius. The measures were calculated on masked CT volumes to isolate the bone, and on the whole CT volumes. The mean and standard deviation of the local \u003cem\u003eSSIM\u003c/em\u003e values were compared between cases. The \u003cem\u003ePSNR\u003c/em\u003e was computed on the whole volumes.\u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003eMorphological Similarity\u003c/h2\u003e \u003cp\u003eThe morphological similarity of the subject specific 3-D bone geometries was assessed using CloudCompare (v2.10). The point-by-point 3-D minimum Euclidean distance between the case geometries and reference HR geometry was computed to determine 1) bias: the mean signed error, 2) precision: the standard deviation of the signed error, and 3) MAE: the mean absolute error.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eApplication Specific Procedure\u003c/h2\u003e \u003cp\u003eEach of the 4 cases (HR, Clin, RS, and SR) for each of the 4 participants was taken through the following pipeline shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e for extracting segmental kinematics from biplane videoradiography. Briefly, the process includes simultaneous collection of dynamic radiographs from two views during cervical motion tasks, shape-matching the DRRs to the biplane radiographs, and computing the relative kinematics between the C4 and C5 vertebrae. The error associated with our traditional biplane videoradiography pipeline for the cervical spine kinematics is \u0026le;\u0026thinsp;0.49 mm and \u0026le;\u0026thinsp;1.80⁰.\u003csup\u003e9\u003c/sup\u003e The performance of each case was compared to our traditional high-resolution CT at multiple phases in pipeline.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eBiplane Videoradiography\u003c/h2\u003e \u003cp\u003eDynamic radiographs of the cervical spine were acquired using a custom biplane videoradiography system (Imaging Systems \u0026amp; Services, Inc., Painesville, OH, USA) at 30 Hz with approximate imaging parameters of 160 mA, 70 kV, and 0.16 mSv/trial. Each participant performed a 3-second trial of neck flexion-extension, lateral bending, and axial rotation. Post-processing of the radiographic images included undistortion and filtering (DSX Suite, C-Motion Inc., Germantown, MD, USA) and calibration (XMALab, Brown University, RI, USA).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eShape-matching\u003c/h2\u003e \u003cp\u003eShape-matching is the process by which 3-D kinematics are extracted by registering the DRRs (created from the projections of the segmented CT volumes) to the biplane radiographs. The dynamic 3-D positions and orientations of the C4 and C5 vertebrae were resolved using Autoscoper (Brown University, RI, USA), an open-source shape-matching software\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e,\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. Both the DRRs and radiographs are filtered (Gaussian, sobel, etc.) to enhance features and edges. Auto-registration is performed using a particle swarm optimization technique with the inverse of the normalized cross-correlation (\u003cem\u003eNCC\u003c/em\u003e) as the cost function. Manual refinement was performed when the optimization algorithm was unable to find an adequate match. The global 3-D kinematics of each bone was exported.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eShape-Matching Similarity\u003c/h2\u003e \u003cp\u003eThe match of the DRR to radiographs during the shape-matching process is quantified using the \u003cem\u003eNCC\u003c/em\u003e. The \u003cem\u003eNCC\u003c/em\u003e is a pixel-by-pixel correlation of the projection of the 3-D DRR to the 2-D radiographs. The inverse of the \u003cem\u003eNCC\u003c/em\u003e is used as the optimization criteria for auto-registration by Autoscoper. A lower value for \u003cem\u003eNCC\u003c/em\u003e value indicates a better fit. Because the \u003cem\u003eNCC\u003c/em\u003e may be affected by radiograph image parameters, the \u003cem\u003eNCC\u003c/em\u003e error metric was normalized frame-by-frame by the corresponding HR \u003cem\u003eNCC\u003c/em\u003e for each participant and trial.\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$NCC error = \\frac{\\left(NC{C}_{case}-NC{C}_{HR}\\right)}{NC{C}_{HR}}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe mean and standard deviation of the \u003cem\u003eNCC\u003c/em\u003e error was computed across all frames of each trial.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eRelative Kinematics\u003c/h2\u003e \u003cp\u003eRelative kinematics between vertebrae were computed using KinematicsToolbox v.4.7.2.\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e Local coordinate systems were created for each vertebra by digitizing anatomic landmarks such that the origin is in the center of the body, x-axis points anterior, and y-axis points left, z-axis points superior.\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e The rotations and displacements of C4 (relative to C5) were computed for each of the trials, cases, and participants.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eRelative Kinematics Accuracy\u003c/h2\u003e \u003cp\u003eThe relative kinematics accuracy is represented by two scalar values: the orientation error and position error. The orientation error is the distance between quaternion rotations\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e of the case, p, and HR, q, defined as:\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$Ѳ = 2 co{s}^{-1}\\left(\\right|\u0026lt;p,q\u0026gt;\\left|\\right)$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\u0026lt;p,q\u0026gt; = {p}_{1}{q}_{1} + {p}_{2}{q}_{2} + {p}_{3}{q}_{3} + {p}_{4}{q}_{4}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe position error was computed as the Euclidean distance between the relative C4-C5 displacements of the case and HR. The mean and variability (standard deviation) of the orientation and position error were computed across each trial.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eStatistics\u003c/h2\u003e \u003cp\u003eANOVA was used to determine groupwise differences for each outcome measure, followed by appropriate post-hoc comparisons. Normality was assessed and vertebral levels were treated as independent. A one-way repeated measures ANOVA was conducted for measures that are not impacted by trial, i.e.., image similarity and bone morphology. A two-way ANOVA was used to examine the effect of both cases and trial for the shape-matching and relative kinematics measures.\u003c/p\u003e \u003c/div\u003e"},{"header":"RESULTS","content":"\u003cp\u003e \u003c/p\u003e \u003cp\u003eThe results are reported in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, with representative examples in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eImage Similarity\u003c/h2\u003e \u003cp\u003eThe \u003cem\u003eSSIM\u003c/em\u003e of the masked volumes takes both the internal structure and the segmentation of the bone into account. The mean masked \u003cem\u003eSSIM\u003c/em\u003e of the SR volume (0.696\u0026thinsp;\u0026plusmn;\u0026thinsp;0.024) was statistically greater than the Clin (0.589\u0026thinsp;\u0026plusmn;\u0026thinsp;0.047; p\u0026thinsp;=\u0026thinsp;0.001) and RS (0.602\u0026thinsp;\u0026plusmn;\u0026thinsp;0.037; p\u0026thinsp;=\u0026thinsp;0.001) volumes. Similarly, the SR volume had significantly less \u003cem\u003eSSIM\u003c/em\u003e variability (0.2\u0026thinsp;\u0026plusmn;\u0026thinsp;0.014) than the Clin (0.313\u0026thinsp;\u0026plusmn;\u0026thinsp;0.009; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and RS (0.290\u0026thinsp;\u0026plusmn;\u0026thinsp;0.011; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) masked volumes. The RS volume also had less variability than the Clin volume (p\u0026thinsp;=\u0026thinsp;0.001).\u003c/p\u003e \u003cp\u003eThe \u003cem\u003eSSIM\u003c/em\u003e of the whole CT volumes represents the structural similarity of the non-segmented CT volumes. The mean \u003cem\u003eSSIM\u003c/em\u003e of the whole RS volume (0.766\u0026thinsp;\u0026plusmn;\u0026thinsp;0.033) was statistically higher than SR (0.747\u0026thinsp;\u0026plusmn;\u0026thinsp;0.031, p\u0026thinsp;=\u0026thinsp;0.018) and the Clin volume (0.753\u0026thinsp;\u0026plusmn;\u0026thinsp;0.031, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The \u003cem\u003eSSIM\u003c/em\u003e variability was not statistically different between the cases. The \u003cem\u003ePSNR\u003c/em\u003e of the whole volumes were also not statistically different.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eSegmentation Time\u003c/h2\u003e \u003cp\u003eSegmentation of each vertebra was successfully completed with the automated \u0026ldquo;CT Bone Wizard\u0026rdquo; tool in Mimics (v23) for the HR and OrthoFusion (SR) volumes with minimal need for manual refinement. This resulted in a segmentation time of approximately 15 minutes per vertebra. Manual segmentation was required for the Clin and RS vertebrae due to poor distinction between bones (primarily facet joints). This resulted in a segmentation time of approximately 2.5 hours per vertebra.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eMorphological Similarity\u003c/h2\u003e \u003cp\u003eSignificant differences between cases were found for bias, precision, and MAE distances of the 3-D bone geometry. Bias: The SR bone models (-0.01\u0026thinsp;\u0026plusmn;\u0026thinsp;0.12 mm) had significantly less bias than the Clin (0.47\u0026thinsp;\u0026plusmn;\u0026thinsp;0.39 mm; p\u0026thinsp;=\u0026thinsp;0.017) and RS (-0.24\u0026thinsp;\u0026plusmn;\u0026thinsp;0.15 mm; p\u0026thinsp;=\u0026thinsp;0.031) bone models. Precision: The SR (0.38\u0026thinsp;\u0026plusmn;\u0026thinsp;0.13 mm) bone models were significantly more precise than the RS (0.71\u0026thinsp;\u0026plusmn;\u0026thinsp;0.19 mm; p\u0026thinsp;=\u0026thinsp;0.003) or Clin (1.18\u0026thinsp;\u0026plusmn;\u0026thinsp;0.44 mm; p\u0026thinsp;=\u0026thinsp;0.005) bone models. MAE: The MAE of the SR bone models (0.29\u0026thinsp;\u0026plusmn;\u0026thinsp;0.06 mm) were significantly less than the Clin (0.85\u0026thinsp;\u0026plusmn;\u0026thinsp;0.31 mm; p\u0026thinsp;=\u0026thinsp;0.002) and RS (0.54\u0026thinsp;\u0026plusmn;\u0026thinsp;0.12; p\u0026thinsp;=\u0026thinsp;0.001) bone models.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eShape-Matching Similarity\u003c/h2\u003e \u003cp\u003eThe normalized cross-correlation coefficient (\u003cem\u003eNCC\u003c/em\u003e) is the measure of pixel-by-pixel similarity of the shape-matched DRR and biplane radiographs. The \u003cem\u003eNCC\u003c/em\u003e error is the normalized difference between the case and HR \u003cem\u003eNCC\u003c/em\u003e values at each frame of the trial. Within-subjects effects of both trial and case were found to be significant for the normalized \u003cem\u003eNCC\u003c/em\u003e error measures. A significant trial effect was observed such that the lateral bending trials had a significantly lower (better matched) mean \u003cem\u003eNCC\u003c/em\u003e error than the axial rotation (p\u0026thinsp;=\u0026thinsp;0.006) and flexion-extension (p\u0026thinsp;=\u0026thinsp;0.001) trials. A significant case effect was observed such that the mean \u003cem\u003eNCC\u003c/em\u003e error for the SR cases (0.106\u0026thinsp;\u0026plusmn;\u0026thinsp;0.162) were significantly lower than the Clin (0.261\u0026thinsp;\u0026plusmn;\u0026thinsp;0.119; p\u0026thinsp;=\u0026thinsp;0.002) and RS (0.194\u0026thinsp;\u0026plusmn;\u0026thinsp;0.105; p\u0026thinsp;=\u0026thinsp;0.024) cases. A significant trial effect was observed such that the LB trials had significantly less variability in \u003cem\u003eNCC\u003c/em\u003e error than the FE trials (p\u0026thinsp;=\u0026thinsp;0.042). A significant case effect was observed such that the \u003cem\u003eNCC\u003c/em\u003e error variability for the SR cases (0.060\u0026thinsp;\u0026plusmn;\u0026thinsp;0.014) was significantly lower than that of the Clin (0.162\u0026thinsp;\u0026plusmn;\u0026thinsp;0.023; p\u0026thinsp;=\u0026thinsp;0.002) or RS (0.109\u0026thinsp;\u0026plusmn;\u0026thinsp;0.010; p\u0026thinsp;=\u0026thinsp;0.005).\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003eRelative Kinematics Accuracy\u003c/h2\u003e \u003cp\u003eThe relative kinematics of the C4 to C5 vertebrae were compared between the HR and cases to determine kinematics accuracy. Significant within-subjects effects of case, but not trial, were found for the kinematics measures. The SR case produced relative kinematics with the least orientation and displacement error. The mean orientation error for the SR case (1.7\u0026thinsp;\u0026plusmn;\u0026thinsp;1.5 deg) was significantly lower than the Clin (8.0\u0026thinsp;\u0026plusmn;\u0026thinsp;5.8 deg; p\u0026thinsp;=\u0026thinsp;0.019) and approaching significance for RS (3.8\u0026thinsp;\u0026plusmn;\u0026thinsp;2.2 deg; p\u0026thinsp;=\u0026thinsp;0.053) case. Similarly, the variability of the orientation error for the SR case (1.1\u0026thinsp;\u0026plusmn;\u0026thinsp;1.2 deg) was significantly lower than the Clin (4.5\u0026thinsp;\u0026plusmn;\u0026thinsp;2.3 deg; p\u0026thinsp;=\u0026thinsp;0.004) and approaching significance for RS (3.1\u0026thinsp;\u0026plusmn;\u0026thinsp;2.4 deg; p\u0026thinsp;=\u0026thinsp;0.061) case. The mean displacement error for the SR case (0.5\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2 mm) was significantly lower than that of the Clin (2.5\u0026thinsp;\u0026plusmn;\u0026thinsp;0.5 mm) and RS (1.2\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3 mm) cases. The variability of displacement error for the SR case (0.4\u0026thinsp;\u0026plusmn;\u0026thinsp;0.49 mm) was significantly lower than the Clin (1.4\u0026thinsp;\u0026plusmn;\u0026thinsp;1.1 mm; p\u0026thinsp;=\u0026thinsp;0.014) and RS (0.8\u0026thinsp;\u0026plusmn;\u0026thinsp;0.8 mm; p\u0026thinsp;=\u0026thinsp;0.026) cases.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eWe designed, implemented, and evaluated OrthoFusion super-resolution technique to increase the spatial resolution of previously acquired low-resolution CT volumes. OrthoFusion significantly improved segmentation time, the structural similarity of the bone images, and the accuracy of bone model geometry. Specific to our biplane videoradiography application, OrthoFusion improved the DRR to radiograph similarity during shape-matching and the accuracy of the relative C4-C5 kinematics. Ultimately, these improvements enable us to use previously acquired clinical CT scans from participant EMR instead of acquiring additional CT imaging as part of our studies, reducing radiation exposure, time, and cost.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eSuper-resolution has utility well beyond our specific biplane videoradiography application. The abundance of imaging in electronic medical records is an untapped opportunity for retrospective studies that were previously not possible due to poor through-plane resolution. Super-resolution can benefit applications requiring accurate and high resolution multiplanar reconstruction, segmentation, DRRs, and 3-D geometries. To our knowledge, currently available imaging software applications do not provide a tool that increases spatial resolution from multiple low-resolution volumes. Many super-resolution techniques have been reported in the medical imaging literature, often with the goal of exceeding hardware resolution and decreasing noise\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e However, these algorithms are complex and challenging to implement for a typical researcher or clinician. Machine learning algorithms offer another approach, however they require training on large datasets which are not always feasible.\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e Morphology morphing algorithms in conjunction with biplane slot scanners offer a low-dose approach to generate bone models, but are limited based on target specific geometries and do not yield a volumetric imaging dataset.\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e OrthoFusion is an easily implemented technique that is generalizable to any CT imaging target and does not require training. Further development of the OrthoFusion algorithm by integrating filtering and iterative algorithms will provide new possibilities for the use of the large dataset of clinical imaging in EMR.\u003c/p\u003e \u003cp\u003eThere are several limitations that should be considered in the interpretation of this study. First, we applied OrthoFusion super-resolution to our specific application. The benefits of improving spatial resolution for bone segmentation are generalizable, but performance should be assessed considering the requirements of new applications. Second, the processing and analysis involved in our biplane videoradiography pipeline requires significant user interaction and the cases were visibly identifiable, making blinding impossible. Finally, OrthoFusion uses a simple voxel-by-voxel averaging algorithm that creates a blurring effect. The addition of anisotropic filtering\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e or a spatial weighted average approach\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e are likely to enhance the image quality.\u003c/p\u003e \u003cp\u003eThis study demonstrates the utility of super-resolution to increase the spatial resolution of previously acquired low-resolution CT volumes. The OrthoFusion technique enabled us to use CT imaging from participant EMR for our biplane videoradiography studies, thus reducing exposure to radiation and costs associated with acquiring imaging. This straight-forward technique is easy to implement, generalizable to other applications, and allows researchers and clinicians that require high spatial resolution CT to utilize the abundance of clinical imaging in EMR.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding Disclosures:\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was funded by the following NIH grants: R03HD09771,\u0026nbsp;TL1R002493, UL1TR002494, and F32AR082276 Additionally, this work was supported by the Minnesota Partnership for Biotechnology and Medical Genomics (MHP IF #14.02).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors would like to thank Craig Kage for his assistance with data collection.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConceptualization, study design: R.E.A., A.M.E. Methodology: R.E.A., A.N., H.W., R.E.B. Formal analysis: R.E.A., A.N., A.M.E., Writing original draft: R.E.A. Review and editing: all authors. Funding acquisition, project management: A.M.E. All authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe code and datasets analyzed during this study are available in the Data Repository for U of M (https://conservancy.umn.edu/drum: specific link TBD).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003cbr\u003e\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSmith-Bindman, R. \u003cem\u003eet al.\u003c/em\u003e Trends in Use of Medical Imaging in US Health Care Systems and in Ontario, Canada, 2000\u0026ndash;2016. JAMA 322, 843\u0026ndash;856 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHakimi, W. E. \u0026amp; Wesarg, S. Accurate super-resolution reconstruction for CT and MR images. in \u003cem\u003eProceedings of the 26th IEEE International Symposium on Computer-Based Medical Systems\u003c/em\u003e 445\u0026ndash;448 (2013). doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/CBMS.2013.6627837\u003c/span\u003e\u003cspan address=\"10.1109/CBMS.2013.6627837\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOdet, C., Peyrin, F. \u0026amp; Goutte, R. Improved resolution of medical 3-D x-ray computed-tomographic images. in \u003cem\u003eVisual Communications and Image Processing \u0026rsquo;90: Fifth in a Series\u003c/em\u003e vol. 1360 658\u0026ndash;664 (SPIE, 1990).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIsaac, J. S. \u0026amp; Kulkarni, R. Super resolution techniques for medical image processing. in 2015 \u003cem\u003eInternational Conference on Technologies for Sustainable Development (ICTSD)\u003c/em\u003e 1\u0026ndash;6 (2015). doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ICTSD.2015.7095900\u003c/span\u003e\u003cspan address=\"10.1109/ICTSD.2015.7095900\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiang, C., Zhang, Q., Fan, R. \u0026amp; Hu, Z. Super-resolution CT Image Reconstruction Based on Dictionary Learning and Sparse Representation. Sci Rep 8, 8799 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGreenspan, H. Super-Resolution in Medical Imaging. The Computer Journal 52, 43\u0026ndash;63 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoraes, T., Amorim, P., Da Silva, J. V. \u0026amp; Pedrini, H. Medical image interpolation based on 3D Lanczos filtering. Computer Methods in Biomechanics and Biomedical Engineering: Imaging \u0026amp; Visualization 8, 294\u0026ndash;300 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRenieblas, G. P., Nogu\u0026eacute;s, A. T., Gonz\u0026aacute;lez, A. M., G\u0026oacute;mez-Leon, N. \u0026amp; Del Castillo, E. G. Structural similarity index family for image quality assessment in radiological images. J Med Imaging (Bellingham) 4, 035501 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKage, C. C. \u003cem\u003eet al.\u003c/em\u003e Validation of an automated shape-matching algorithm for biplane radiographic spine osteokinematics and radiostereometric analysis error quantification. PLOS ONE 15, e0228594 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiranda, D. L. \u003cem\u003eet al.\u003c/em\u003e Static and Dynamic Error of a Biplanar Videoradiography System Using Marker-Based and Markerless Tracking Techniques. J Biomech Eng 133, 121002-121002-8 (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAkhbari, B., Morton, A. M., Moore, D. C. \u0026amp; Crisco, J. J. Biplanar Videoradiography to Study the Wrist and Distal Radioulnar Joints. J Vis Exp 10.3791\u003cdiv class=\"ExternalRefDOI\"\u003e/62102\u003c/div\u003e (2021) doi:10.3791/62102.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRebekah L. Lawrence. KinematicsToolbox: A Comprehensive Software Program for Analyzing and Visualizing Radiographic Motion Capture Data. (2023) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7936/6RXS-103624\u003c/span\u003e\u003cspan address=\"10.7936/6RXS-103624\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuynh, D. Q. Metrics for 3D Rotations: Comparison and Analysis. J Math Imaging Vis 35, 155\u0026ndash;164 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJia, Y., Gholipour, A., He, Z. \u0026amp; Warfield, S. K. A New Sparse Representation Framework for Reconstruction of an Isotropic High Spatial Resolution MR Volume From Orthogonal Anisotropic Resolution Scans. IEEE Transactions on Medical Imaging 36, 1182\u0026ndash;1193 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDanker Khoo, J. J., Lim, K. H. \u0026amp; Sien Phang, J. T. A Review on Deep Learning Super Resolution Techniques. in 2020 \u003cem\u003eIEEE 8th Conference on Systems, Process and Control (ICSPC)\u003c/em\u003e 134\u0026ndash;139 (2020). doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ICSPC50992.2020.9305806\u003c/span\u003e\u003cspan address=\"10.1109/ICSPC50992.2020.9305806\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDubousset, J. \u003cem\u003eet al.\u003c/em\u003e A new 2D and 3D imaging approach to musculoskeletal physiology and pathology with low-dose radiation and the standing position: the EOS system. Bull Acad Natl Med 189, 287\u0026ndash;297; discussion 297\u0026ndash;300 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKurt, B., Nabiyev, V. V. \u0026amp; Turhan, K. Medical images enhancement by using anisotropic filter and CLAHE. in 2012 \u003cem\u003eInternational Symposium on Innovations in Intelligent Systems and Applications\u003c/em\u003e 1\u0026ndash;4 (2012). doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/INISTA.2012.6246971\u003c/span\u003e\u003cspan address=\"10.1109/INISTA.2012.6246971\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eB\u0026auml;tz, M., Eichenseer, A. \u0026amp; Kaup, A. Multi-image super-resolution using a dual weighting scheme based on Voronoi tessellation. in 2016 \u003cem\u003eIEEE International Conference on Image Processing (ICIP)\u003c/em\u003e 2822\u0026ndash;2826 (2016). doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ICIP.2016.7532874\u003c/span\u003e\u003cspan address=\"10.1109/ICIP.2016.7532874\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Super Resolution, Bone Models, Computed Tomography, Image Fusion, Spatial Resolution Enhancement ","lastPublishedDoi":"10.21203/rs.3.rs-4117386/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4117386/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eOrthoFusion, an intuitive super-resolution algorithm, is presented in this study to enhance the spatial resolution of clinical CT volumes. The efficacy of OrthoFusion is evaluated, relative to high-resolution CT volumes (ground truth), by assessing image volume and derived bone morphological similarity, as well as its performance in specific applications in 2D-3D registration tasks. Results demonstrate that OrthoFusion significantly reduced segmentation time, while improving structural similarity of bone images and relative accuracy of derived bone model geometries. Moreover, it proved beneficial in the context of biplane videoradiography, enhancing the similarity of digitally reconstructed radiographs to radiographic images and improving the accuracy of relative bony kinematics. OrthoFusion's simplicity, ease of implementation, and generalizability make it a valuable tool for researchers and clinicians seeking high spatial resolution from existing clinical CT data. This study opens new avenues for retrospectively utilizing clinical images for research and advanced clinical purposes, while reducing the need for additional scans, mitigating associated costs and radiation exposure.\u003c/p\u003e","manuscriptTitle":"OrthoFusion: A Super-Resolution Algorithm to Fuse Orthogonal CT Volumes","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-04-03 17:00:48","doi":"10.21203/rs.3.rs-4117386/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-09-23T08:53:08+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-09-20T01:15:28+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-09-10T02:48:39+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"319066145354481923268642393844696446971","date":"2024-09-10T00:15:09+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"8641706376611534052554576621346719619","date":"2024-08-31T16:03:02+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"150899674387357306588442871098253214679","date":"2024-06-04T17:39:45+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-04-15T16:06:25+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-04-15T16:03:46+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-03-29T10:33:23+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-03-29T10:30:14+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-03-17T14:05:35+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d7b1b748-f6b2-4104-8424-74095bb56af4","owner":[],"postedDate":"April 3rd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":30164118,"name":"Biological sciences/Biological techniques/Imaging/X ray tomography"},{"id":30164119,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2025-01-13T15:59:22+00:00","versionOfRecord":{"articleIdentity":"rs-4117386","link":"https://doi.org/10.1038/s41598-025-85516-y","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-01-09 15:57:04","publishedOnDateReadable":"January 9th, 2025"},"versionCreatedAt":"2024-04-03 17:00:48","video":"","vorDoi":"10.1038/s41598-025-85516-y","vorDoiUrl":"https://doi.org/10.1038/s41598-025-85516-y","workflowStages":[]},"version":"v1","identity":"rs-4117386","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4117386","identity":"rs-4117386","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}