Fast and robust femur segmentation from computed tomography images for patient-specific hip fracture risk screening

ABSTRACT

Osteoporosis is a common bone disease that increases the risk of bone fracture. Hip-fracture risk screening methods based on flnite element analysis depend on segmented computed tomography (CT) images; however, current femur segmentation methods require manual delineations of large data sets. Here we propose a deep neural network for fully automated, accurate, and fast segmentation of the proximal femur from CT. Evaluation on a set of 1147 proximal femurs with ground truth segmentations demonstrates that our method is apt for hip-fracture risk screening, bringing us one step closer to a clinically viable option for screening at-risk patients for hip-fracture susceptibility.

KEYWORDS:

• Computed tomography
• femur
• segmentation
• convolutional neural networks
• osteoporosis

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. The Hounsfield unit is a relative quantitative measurement of radio density used by radiologists in the interpretation of CT images.

2. This sample set was used by (Pauchard et al. 2016) to evaluate their model.

3. This sub-cohort from AGES-RS, including all fracture cases, was used by (Enns-Bray et al. 2019).

4. MIScnn is an open-source Python library and intuitive API for medical image segmentation pipelines (Müller and Kramer 2019).

5. On-the-fly data augmentation eliminates the need for excessive storage of augmented images by performing the augmentation prior to each optimisation iteration.

Additional information

Funding

The work was supported by the Icelandic Centre for Research []; The Strategic Focus Area “Personalized Health and Related Technologies” of the ETH Domain [018-430, 2018-325]

Notes on contributors

Pall Asgeir Bjornsson

Pall Asgeir Bjornsson received his M.Sc. degree in Electrical and Computer Engineering from the University of Iceland in 2020. He is currently working for the University in the field of image processing and deep neural networks, as well as the field of geothermal heat transfer. His primary research interests are the use of AI in image processing, financial markets, and decision making.

Alexander Baker

Alexander Baker is a doctoral student within the Laboratory for Orthopaedic Technologies at ETH Zurich. He possesses a B.Sc in Mechanical Engineering from UCLA and a M.Sc. in Mechanical Engineering from UC Berkeley. He has done previous work in the field of injury biomechanics, including assisting in the development of a finite element model of an American football helmet, and improving the methodology to construct injury risk curves. His doctoral research focuses on computational modeling of hip fracture using finite element analysis and applying those models to large population cohorts in order to better understand and predict hip fracture.

Ingmar Fleps

Dr. Ingmar Fleps is a mechanical engineer and material scientist with a Ph.D. and postgraduate training in biomechanics. His work focuses on medical image processing, experimental modelling of biomechanical systems, and computational modelling based on medical images to improve our understanding of the mechanical interaction and failure processes of human tissues, as well as tissues and implants. Dr. Fleps is a postdoctoral associate at Boston University where he currently works on mechanical modelling of the spine.

Yves Pauchard

Dr. Yves Pauchard is an electrical engineer with graduate and post-graduate training in medical imaging and analysis. His research interests include software development, image analysis and computational modelling in musculoskeletal applications with a focus on bone. He is the project lead of MITK-GEM https://simtk.org/projects/mitk-gem, an image analysis application to generate finite element models from medical images. Dr Pauchard holds a faculty position in Electrical and Software Engineering at the University of Calgary. In his role, he teaches courses in software development, data analysis and machine learning.

Halldor Palsson

Dr. Halldór Pálsson finished a CS degree in Mechanical Engineering from the University of Iceland in 1993. He then finished a M.Sc. degree in the same subject from the Technical University of Denmark in 1995. This degree was followed by a Ph.D. degree at the Technical University of Denmark in the year 2000, where the subject was ”Methods for planning and operating decentralized combined heat and power plants”. After that Halldór worked on a research project involving mathematical modeling of high energy electric arc FeSi production, until acquiring a position as an Associate Professor at the University of Iceland in 2005. Halldór has been a full professor at the University of Iceland since 2016, working on subjects regarding geothermal energy utilization, fluid mechanics, biomechanics, and mathematical modeling in general.

Stephen J. Ferguson

Dr. Stephen Ferguson is a Full Professor of Biomechanics at the Institute for Biomechanics of the ETH Zurich in Switzerland. He has a B.A.Sc. in Mechanical Engineering from the University of Toronto (91) and a M.Sc. (1994) and Ph.D. (2000) from Queen's University. The focus of his research is the study of the mechanical and biological mechanisms of musculoskeletal disorders and injuries and the development and application of innovative technologies for their treatment. The challenge presented by an ageing population is of primary importance in his work, with research programs on bone fragility, joint arthroplasty, disc degeneration and postural decay. He and his group study new biomaterials, material fabrication technologies, implant concepts and diagnostic methods and develop the technical means for their application in the clinic. Dr. Ferguson is the author of over 250 scientific papers, 7 book chapters and 4 patent applications. He and his collaborators have received the European Spine Journal GRAMMER prize for best scientific work, the CTI Medtech Award for the translational project “BoneWelding”. He is on the editorial board of the journals Clinical Biomechanics, Journal of Biomechanics and European Cells & Materials and is a former president of the European Society of Biomechanics.

Sigurdur Sigurdsson

Sigurdur Sigurdsson is currently the CEO of the Icelandic Heart Association Radiology and a lecturer at the Faculty of Medicine, University of Iceland. Sigurdsson’s research involves radiology with focus on neuroimaging of the aging brain in cohort studies. Sigurdsson has published over 140 peer-reviewed papers in scientific journals. Sigurdsson received his MSc from South Bank University, London UK in clinical Magnetic Resonance Imaging in year 2001 and BSc in Radiography from the Technical University of Iceland in 1994.

Vilmundur Gudnason

Dr. Vilmundur Gudnason is the director of the Icelandic Heart Association and a professor at the Faculty of Medicine, University of Iceland. He obtained his medical degree from the University of Iceland and a PhD in molecular genetics from University College London. He is the Principal Investigator for the Age Gene/Environment Susceptibility Reykjavik (AGES-Reykjavik) study, based on the 50-year-long Reykjavik study, and for the REFINE Reykjavik study of younger generations. The AGES-Reykjavik study is a study on aging with extremely deep phenotyping, including using imaging with CT and MRI and detailed follow-up hospital data, such as myocardial infarction and fractures. Professor Gudnason (https://orcid.org/0000-0001-5696-0084) has published widely in various high impact journals and has been on the list of Thompson Reuter/Clarivate Highly Cited Researchers 2015, 2016, 2017 and 2018. ‪Vilmundur Gudnason - ‪Google Scholar.

Benedikt Helgason

Dr. Benedikt Helgason is a civil engineer with post-graduate training in biomechanics. His research focuses on the development of modeling techniques for simulating bone pathologies and treatments. He is the PI on several projects in the field of healthy aging, e.g. the Future Health Technologies program funded by the National Research Foundation of Singapore, and a project on in silico clinical trials on personalized preventive measures for managing hip fractures in the elderly, funded by the Swiss National Science Foundation. Dr. Helgason holds a position as a senior scientist at the ETH-Zurich, Zurich, Switzerland.

Lotta Maria Ellingsen

Dr. Lotta Maria Ellingsen is an Associate Professor and Vice Chair of the faculty of Electrical and Computer Engineering at the University of Iceland. She received her M.S.E. (2004) and Ph.D. (2008) degrees in Electrical and Computer Engineering from the Johns Hopkins University, Baltimore, USA. Her research interests are in the field of medical image processing and analysis with emphasis on 3D medical image registration and segmentation, with applications to brains, bones, and statistical atlases. She has particular expertise in deformable 3D-3D image registration of magnetic resonance images of the human brain for different application purposes. Her current research focuses on developing pipelines for automatic segmentation and labeling of brain anatomy for systematic analysis of brain dysmorphology to better characterize neurodegenerative diseases for personalized medicine.