Concurrent Validity of Two Inertial Measurement Unit Pipelines for Estimating Lumbar and Thoracic Kinematics During Lifting Tasks

Lumbosacral and thoracolumbar kinematics are key risk factors for lifting-related low back pain, yet their measurement is typically restricted to motion capture laboratories. Inertial measurement units (IMUs) offer the potential to quantify spine kinematics in more naturalistic settings, but the validity of IMU-based processing pipelines relative to optical motion capture (OMC) remains unclear. Nine healthy participants performed stoop, squat, free, and asymmetric lifting tasks while IMU and OMC data were simultaneously collected to evaluate the concurrent validity of two IMU pipelines: the proprietary MVN Analyze pipeline and an OpenSense pipeline using a validated OpenSim biomechanical model for lifting. Joint angles from both pipelines were compared against OMC-derived joint angles calculated using the same validated OpenSim model with one-way repeated-measures statistical parametric mapping (SPM) (α = 0.05), Bland–Altman analysis with Limits of Agreement (LoA) and 95% Confidence Intervals (CIs), and Concordance Correlation Coefficients (CCCs) with 95% CIs. Xsens MVN Analyze consistently overestimated flexion-extension at both spinal levels across all lift types (lumbosacral: RMSE ≤ 9.8◦, bias ≤ −14.5◦, LoA ≤ ±10◦; thoracolumbar: RMSE ≤ 5.4◦, bias ≤ −8.3◦, LoA ≤ ±5◦), with SPM confirming significant differences during the lifting and lowering phases of all lifting cycles. In contrast, processing Xsens data with OpenSense using the same biomechanical model as the OMC data yielded excellent agreement with OMC (RMSE ≤ 2.9◦, bias ≤ 3◦, LoA ≤ ±10◦). CCC was poor to moderate, specifically in lateral bending and axial rotation planes, likely reflecting limited between-participant ROM variability. These results suggest that discrepancies are driven primarily by biomechanical model differences rather than sensor or sensor fusion limitations. Ultimately, when paired with an appropriate biomechanical model, XSens sensors show promise for practical field-based assessment of lifting biomechanics, potentially requiring only sensors at the chest and pelvis.