Date of Award
1-1-2019
Document Type
Masters Thesis
Degree Name
M.S.
Organizational Unit
Daniel Felix Ritchie School of Engineering and Computer Science, Mechanical and Materials Engineering
First Advisor
Paul J. Rullkoetter, Ph.D.
Keywords
Computer vision, Convolutional neural networks, Deep learning, Machine learning, Orthopaedics
Abstract
The reemergence of deep learning in recent years has led to its successful application in a wide variety of fields. As a subfield of machine learning, deep learning offers an array of powerful algorithms for data-driven applications. Orthopaedics stands to benefit from the potential of deep learning for advancements in the field. This thesis investigated applications of deep learning for the field of orthopaedics through the development of three distinct projects.
First, algorithms were developed for the automatic segmentation of the structures in the knee from MRI. The resulting algorithms can be used to accurately segment full MRI scans in a matter of seconds. Reconstructed structures from predicted segmentation maps yielded on average submillimeter geometric errors when compared to geometries from ground truth segmentation maps on a test set. The resulting frameworks can further be applied to develop algorithms for automatic segmentation of other anatomies and modalities in the future.
Next, neural networks (NNs) were developed and evaluated for the prediction of muscle and joint reaction forces of patients performing activities of daily living (ADLs) in a gait lab environment. The performance of these models demonstrates the potential of NNs to supplement traditional gait lab data collection and has implications for the development of new gait lab workflows with less hardware and time requirements. Additionally, the models performed activity classification using standard gait lab data with near-perfect accuracy.
Lastly, a deep learning-based computer vision system was developed for the detection and 6-degree of freedom (6-DoF) pose estimation of two surgical tracking tools routinely used in total knee replacement (TKR). The resulting model demonstrated competitive object detection capabilities and translation error as little as a few centimeters for the pose estimation task. A preliminary evaluation of the system shows promise for its applications in skill assessment and operations research.
The development of these three projects represents a significant step towards the adoption of deep learning methodologies by the field of orthopaedics and shows potential for future additional applications.
Publication Statement
Copyright is held by the author. User is responsible for all copyright compliance.
Rights Holder
William Stewart Burton
Provenance
Received from ProQuest
File Format
application/pdf
Language
en
File Size
131 p.
Recommended Citation
Burton, William Stewart II, "Applied Deep Learning in Orthopaedics" (2019). Electronic Theses and Dissertations. 1568.
https://digitalcommons.du.edu/etd/1568
Copyright date
2019
Discipline
Biomedical engineering, Medical imaging, Biomechanics