Deep Learning-Based Estimation of Ground Reaction Forces in Parkinsonian Gait Using an Optimized Set of IMU Data

Accurate gait analysis in Parkinson's disease 1 (PD) typically relies on laboratory-based systems to cap-2 ture biomechanical data, such as ground reaction forces 3 (GRFs). Estimating GRFs using inertial measurement units 4 (IMUs) provides a feasible alternative. However, this ap-5 proach remains challenging in pathological gait like PD 6 due to its high variability and complexity. Moreover, ex-7 isting monitoring approaches often require multiple body-8 mounted sensors, which limit practicality and reduce pa-9 tient compliance. To date, no study has investigated the 10 application of deep learning approaches to address this 11 challenge. This study proposes, for the first time, a deep 12 learning framework to estimate bilateral vertical GRFs 13 (vGRFs) in PD using an optimized set of wearable IMUs. A 14 hybrid CNN-BiLSTM model was trained separately on data 15 from 61 PD patients and 65 healthy controls (HC) using 13 16 IMUs. The model achieved high intra-subject accuracy (R 2 17 = 0.98) and strong inter-subject generalization (R 2 = 0.93 18 for HC, R 2 = 0.91 for PD). Sensor configuration was found 19 to significantly influence estimation accuracy, with optimal 20 sensor placement varying between PD patients and HC. 21 For PD patients, estimation accuracy dropped markedly 22 when reducing to a single IMU. The optimal configuration 23 for PD used four IMUs. We identified a minimal setup with only two IMUs still enabled robust estimation. This compact setup offers a practical and scalable solution. Overall, the proposed approach supports the development of wearable vGRF-based gait analysis systems for Parkinsonian gait and potentially other pathological conditions, enabling accessible clinical assessments, remote monitoring, and personalized rehabilitation.