Imu-based hip adduction and flexion moment estimation for osteoarthritis and healthy individuals during normal and modified gait

Abstract Excessive cumulative hip joint loading and harmful compensatory gait patterns during daily activities often remain undetected outside the biomechanics laboratory, limiting timely adjustment of rehabilitation strategies and potentially accelerating disease progression in patients with hip osteoarthritis (HOA). Wearable inertial measurement units (IMUs) could enable continuous monitoring of hip joint loading, but accurate estimation must generalize across individuals, pathological gait patterns, walking environments, and gait modifications. We recruited 84 participants (34 with osteoarthritis, 19 healthy older adults, and 31 healthy young adults) who performed normal and modified walking trials (fast and slow walking, toe-in and toe-out gait, wide and narrow step widths, increased lateral trunk sway, and dual-task walking) on an instrumented treadmill or overground force plates while wearing IMUs and optical motion capture markers. We developed a novel, personalized deep learning framework to estimate continuous hip adduction moment (HAM) and hip flexion moment (HFM) waveforms from input IMU signals at the trunk, pelvis, thigh, and shank. Personalization using only baseline walking data achieved RMSEs of 0.48 and 0.52 %BW $$\cdot $$ BH with waveform correlations of 0.96 and 0.94 for HAM and HFM, respectively. During independent overground evaluation, the personalized model achieved RMSEs of 0.92 and 0.77 %BW $$\cdot $$ BH with correlations of 0.91 and 0.93. Predicted peak moments and stance-phase moment impulses preserved gait-dependent differences across the evaluated gait modifications. By enabling individualized, low-burden hip-loading assessment across diverse gait patterns, this framework may help identify harmful compensatory strategies earlier, inform timely rehabilitation adjustments, broaden access to biomechanical monitoring outside specialized laboratories, and provide a technical basis for future wearable biofeedback and personalized rehabilitation assessment.