From MRI to Motion: Evaluating Pipelines for Participant-Specific Musculoskeletal Modelling in an ACL-Injured Pediatric Population

Abstract

Purpose: The development of participant-specific musculoskeletal (MSK) models has the potential to improve the accuracy and clinical relevance of biomechanical simulations. However, while prior research has explored aspects of segmentation and personalization in MSK modelling, the specific trade-offs between segmentation efficiency, anatomical fidelity, and their downstream impact on pediatric knee simulations remain underexplored. This thesis aims to address two key stages in the imaging-to-simulation pipeline for adolescent MSK modelling: (1) evaluating the efficiency and concurrent validity of two segmentation platforms, MITK and Elucis, and (2) assessing whether incorporating partial participant-specific femur and tibia geometries into OpenSim models meaningfully alters simulation outputs during a functional task.

Methods: Study 1 compared segmentation time and agreement between MITK (typical, 2D segmentation software) and Elucis (3D, virtual reality-based segmentation software) using MRI data from 33 pediatric participants. Inter-rater reliability and inter-software agreement were assessed using Dice Similarity Coefficients (DSC) for the femur, tibia, fibula, and posterior cruciate ligament (PCL) while segmentation time was compared using paired t-tests for the femur, tibia, fibula, anterior cruciate ligament (ACL) and PCL. Study 2 incorporated segmented partial femur and tibia bones into two OpenSim models (Rajagopal and Catelli) using rigid and non-rigid registration (iterative closest point (ICP) and coherent point drift (CPD)) followed by integration via the Shared Tools for Automatic Personalized Lower Extremity Modelling (STAPLE) pipeline. Models were scaled using the Automatic Scaling Tool (AST). Model outputs from a forward lunge task were compared across model types using statistical parametric mapping (SPM), and repeated measures ANOVA for three outcomes: knee joint center (KJC) location, knee flexion kinematics, and marker trajectory error.

Results: Elucis significantly reduced segmentation time (mean difference = 8.6 minutes; p < 0.001) while maintaining comparable or better DSC agreement between raters for the soft-tissue structure (e.g.., PCL; inter-rater DSC = 0.77 in Elucis vs. 0.57 in MITK). Femur, tibia, and fibula DSC scores remained high across both platforms (inter-software) and between raters (inter-rater; DSC ≥ 0.90). In MSK simulations, partial bone personalization led to inconsequential changes in KJC (< 0.006 mm) and non-significant differences in knee flexion waveforms (p > 0.05). Only one model comparison (Participant-specific Catelli vs. Generic Catelli) showed a statistically significant difference in marker root mean square (RMS) error (p < 0.001), but the absolute difference remained below 0.25 mm, far under biomechanical relevance thresholds.

Conclusion: This thesis demonstrates that Elucis offers a time-efficient alternative to traditional segmentation platforms, particularly for challenging soft-tissue structures like the PCL. However, incorporating partial participant-specific geometry alone did not substantially alter kinematic or marker-based outputs during a controlled functional task of the forward lunge. These findings highlight the importance of aligning the degree and type of model personalization with the specific research or clinical application. Full anatomical personalization or more complex movement tasks may be required to detect meaningful difference in simulation outcomes.