Magnetic resonance imaging-based biomarkers of knee osteoarthritis progression via deep learning
Thesis event information
Date and time of the thesis defence
Place of the thesis defence
Auditorium F202, Kontinkangas campus
Topic of the dissertation
Magnetic resonance imaging-based biomarkers of knee osteoarthritis progression via deep learning
Doctoral candidate
Master of Engineering and Technology Egor Panfilov
Faculty and unit
University of Oulu Graduate School, Faculty of Medicine, Research Unit of Health Sciences and Technology
Subject of study
Medical Physics and Technology
Opponent
Associate Professor Cem M. Deniz, New York University
Custos
Professor Simo Saarakkala, University of Oulu
Predicting knee osteoarthritis development and progression from magnetic resonance images using AI
Knee osteoarthritis is one of the leading causes of disability worldwide, especially in developed countries. It leads to structural joint degeneration, pain, stiffness, and reduced mobility. However, not everyone with this disease experiences it the same way. Predicting who is at higher risk of progression — and how quickly the condition will worsen — remains a major challenge in medicine today.
Magnetic Resonance Imaging (MRI) provides detailed images of joint tissues, but comprehensive manual analysis of this data is time-consuming and requires expert knowledge. This thesis aimed to develop artificial intelligence (AI)-based methods that can automatically extract imaging biomarkers from MRI scans and to evaluate their potential for predicting knee osteoarthritis progression.
The first part of the thesis focused on developing a deep learning method that automatically identifies articular cartilage tissue from MRI images and quantifies its health status. These quantitative biomarkers offer a more precise way to track how the disease affects the knee joint over time. The method was validated using several MRI protocols and performed comparably to expert-guided semi-automatic software tools.
The second part of the thesis addressed a broader question: how well can progression of knee osteoarthritis be predicted using whole-knee MRI scans — not just cartilage tissue — and which MRI protocols are the most informative? Using AI, imaging biomarkers were learned directly from the multi-protocol MRI data. The predictive performance of AI models was analyzed across multiple time horizons, highlighting the contribution of individual MRI protocols.
To conclude, this thesis demonstrates how AI enables efficient, large-scale MRI analysis in osteoarthritis research and supports the identification of patients at risk of disease progression. All the tools developed are openly shared to support future studies and clinical translation.
Magnetic Resonance Imaging (MRI) provides detailed images of joint tissues, but comprehensive manual analysis of this data is time-consuming and requires expert knowledge. This thesis aimed to develop artificial intelligence (AI)-based methods that can automatically extract imaging biomarkers from MRI scans and to evaluate their potential for predicting knee osteoarthritis progression.
The first part of the thesis focused on developing a deep learning method that automatically identifies articular cartilage tissue from MRI images and quantifies its health status. These quantitative biomarkers offer a more precise way to track how the disease affects the knee joint over time. The method was validated using several MRI protocols and performed comparably to expert-guided semi-automatic software tools.
The second part of the thesis addressed a broader question: how well can progression of knee osteoarthritis be predicted using whole-knee MRI scans — not just cartilage tissue — and which MRI protocols are the most informative? Using AI, imaging biomarkers were learned directly from the multi-protocol MRI data. The predictive performance of AI models was analyzed across multiple time horizons, highlighting the contribution of individual MRI protocols.
To conclude, this thesis demonstrates how AI enables efficient, large-scale MRI analysis in osteoarthritis research and supports the identification of patients at risk of disease progression. All the tools developed are openly shared to support future studies and clinical translation.
Last updated: 4.7.2025