Predicting Clinical Outcomes Using Principal Component Analysis-Derived Knee Morphology
Faculty Mentor Information
Dr. Clare K. Fitzpatrick, Boise State University
Presentation Date
7-2025
Abstract
INTRODUCTION: Knee osteoarthritis (OA) is a degenerative joint disease that causes chronic knee pain and mobility loss, affecting hundreds of millions globally. OA-related joint changes could be streamlined with predictive tools. This study aims to predict OA outcomes using principal component analysis (PCA)-derived knee bone morphologies from magnetic resonance (MR) images.
METHODS: MR images were segmented using an auto-segmentation algorithm and registered to a low-density template knee. The dataset was split into training (80%) and testing (20%) sets. PCA was applied to the training set, and machine learning models were trained to classify and predict specific clinical outcomes based on the resulting principal components.
RESULTS: The first 21 principal components captured the majority of shape variation in the test set. Reconstruction accuracy was evaluated using average Euclidean distances between original and reconstructed nodes: femur (0.63 mm), tibia (0.76 mm), and patella (0.65 mm). Initial modeling results suggest strong potential for outcome prediction based on morphology.
DISCUSSION: These findings support the utility of PCA-derived morphology as a compact and accurate representation of knee shape that retains predictive value for joint mechanics and disease state.
SIGNIFICANCE: This approach may streamline OA diagnosis and prognosis, reducing clinician workload through automated, shape-based predictions.
Predicting Clinical Outcomes Using Principal Component Analysis-Derived Knee Morphology
INTRODUCTION: Knee osteoarthritis (OA) is a degenerative joint disease that causes chronic knee pain and mobility loss, affecting hundreds of millions globally. OA-related joint changes could be streamlined with predictive tools. This study aims to predict OA outcomes using principal component analysis (PCA)-derived knee bone morphologies from magnetic resonance (MR) images.
METHODS: MR images were segmented using an auto-segmentation algorithm and registered to a low-density template knee. The dataset was split into training (80%) and testing (20%) sets. PCA was applied to the training set, and machine learning models were trained to classify and predict specific clinical outcomes based on the resulting principal components.
RESULTS: The first 21 principal components captured the majority of shape variation in the test set. Reconstruction accuracy was evaluated using average Euclidean distances between original and reconstructed nodes: femur (0.63 mm), tibia (0.76 mm), and patella (0.65 mm). Initial modeling results suggest strong potential for outcome prediction based on morphology.
DISCUSSION: These findings support the utility of PCA-derived morphology as a compact and accurate representation of knee shape that retains predictive value for joint mechanics and disease state.
SIGNIFICANCE: This approach may streamline OA diagnosis and prognosis, reducing clinician workload through automated, shape-based predictions.