Author: Roohollah Milimonfared
Milimonfared, Roohollah, 2019 Development and implementation of an artificial intelligence system for assessing corrosion damage at stem taper of hip replacement implants: A retrieval study, Flinders University, College of Science and Engineering
Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.
Despite the clinical benefits of modularity in total hip replacement (THR) implants, modular
interfaces such as head-neck taper junction are susceptible to fretting wear and corrosion due to
relative micro-motions at the interface and also the presence of corrosive body fluid. This fact
induces a chain of host body responses that may ultimately result in revision surgery to retrieve the
failed implant. Through large-scale implant retrieval studies, the damage sustained by the implants
is assessed, and possible associations between several implant/patients factors and the
extent/location of the damage are investigated.
This PhD study aims to conduct the first large-scale retrieval study in Australia through exploring a
database of 2100 operation records available at Royal Adelaide Hospital and a retrieval library of
implants with approximately the same number of implants that had been retrieved since 1980s. The
database was filtered at multiple occasions to identify implants suitable for this study.
Visual scoring of damage at taper junctions is the sole method to quantify corrosion in large-scale
retrieval studies. In this work, an intelligent image analysis-based method was developed and
implemented that can objectively assess corrosion at the stem taper of retrieved hip implants,
according to the popular Goldberg’s scoring method. A Support Vector Machine classifier was used
that takes in vectors of global and local textural features and assigns scores to the corresponding
images. Bayesian optimisation fine-tuned the hyperparameters of six binary learners of this
classifier to minimise the cross-validation error and increase the accuracy level to 85%.
Moreover, the spatial distribution and the severity of corrosion damage onto the surface of the
metallic stem tapers were objectively explored. An ordinal logistic regression model was developed
to find the odds of receiving a higher score at eight distinct zones of stem tapers. A method to find
the order of damage severity across the eight zones was introduced based on an overall test of
statistical significance. The findings showed that corrosion at the stem tapers occurred more
commonly in the distal region in comparison with the proximal region. Also, the medial distal zone
was found to possess the most severe corrosion damage among all the studied eight zones.
In the last phase of the project, several multivariate analyses of patient and implant factors were
carried out to identify the challenges regarding the causal-explanatory statistical modelling
techniques that are currently used in the literature of retrieval studies. It was elaborated why this
group of techniques are not suitable for looking at multiple confounding variables. Predictive
analytics was recommended to be utilised in conjunction with the existing methods to enable
clinicians to predict the likelihood of implants failure for prospective recipients.
Keywords: Total hip arthroplasty, Metallic implants, Machine learning, Digital image processing, Texture analysis, Regression
Subject: Engineering thesis
Thesis type: Doctor of Philosophy
Completed: 2019
School: College of Science and Engineering
Supervisor: Reza Hashemi Oskouei