Computationally Efficient Machine Learning Techniques for TKR Assessment
Date of Award
Summer 8-24-2024
Document Type
Dissertation
Degree Name
Ph.D.
Organizational Unit
Daniel Felix Ritchie School of Engineering and Computer Science, Mechanical and Materials Engineering
First Advisor
Paul Rullkoetter
Second Advisor
Peter Laz
Third Advisor
Kevin Shelburne
Fourth Advisor
Chadd Clary
Copyright Statement / License for Reuse
All Rights Reserved.
Keywords
Total knee replacement (TKR), Machine learning, Statistics, Finite element analysis
Abstract
This dissertation introduces innovative machine learning methodologies for the evaluation of total knee replacements (TKRs). A validated lower limb simulator was used to generate a set of boundary conditions for telemetrically implanted set of patients. The boundary conditions of these nine patients was expanded through the application of principal component analysis. Combining these boundary conditions with surgical alignment parameters a comprehensive dataset of knee joint mechanics was developed. This dataset was crucial for the development of predictive models. Linear regression, multi-layer perceptron, bi-directional long short term (biLSTM), convolutional neural network, and transformer-based approaches were explored to forecast the outputs of the validated lower limb model. Ultimately the biLSTM was found to have the best accuracy prediction of the models. Exploration of the variable inputs necessary for accurate prediction as well as generalization of linear regression, biLSTM and transformer-based models showed that it may be possible to accurately predict joint biomechanics give only surgical inputs to the models. The linear regression model showed that it was able to generalize the database the best while the biLSTM still retained the best prediction once all input variables were used. These results underscore the importance of a diverse and comprehensive dataset, as well as the need for ongoing data collection to enhance the models' predictive capabilities. The implications of this research are far-reaching, offering the potential to revolutionize the field of orthopedics. The ability to predict TKR mechanics in real-time holds promise for improving intraoperative decision-making, optimizing postoperative rehabilitation protocols, and informing the design of next-generation prosthetic components. This work not only contributes to academic discourse but also has practical applications that could improve patient outcomes and quality of life.
Copyright Date
8-2024
Publication Statement
Copyright is held by the author. Permanently suppressed.
Rights Holder
Chase Maag
Provenance
Received from author
File Format
application/pdf
Language
English (eng)
Extent
121 pgs
File Size
6.7 MB
Recommended Citation
Maag, Chase, "Computationally Efficient Machine Learning Techniques for TKR Assessment" (2024). Electronic Theses and Dissertations. 2494.
https://digitalcommons.du.edu/etd/2494
