Deep Learning-Based Gait Analysis from Wearable Sensors for Early Parkinson's Disease Detection

Parkinson's Disease (PD) is a progressive neurodegenerative disorder that impacts a person's motor control, causing symptoms like tremors, muscle stiffness, and problems with balance. Early detection is vital for starting treatment and slowing the disease's progression. However, traditional clinical exams often miss the subtle initial signs, leading to delayed diagnosis. Recent advances in wearable technology provide a non-invasive way to monitor gait patterns, which often contain early clues of PD-related motor decline. This study presents a deep learning framework that uses inertial measurement units (IMUs) and pressure insoles to analyze spatiotemporal gait dynamics. The framework uses Convolutional Neural Networks (CNNs) to capture the spatial features of gait and Long Short-Term Memory (LSTM) networks to learn and represent their temporal dynamics. To improve its robustness and ability to generalize, our pipeline includes data augmentation, dimensionality reduction, and attention mechanisms. It also features an Explainable AI (XAI) component, using Grad-CAM and SHAP to highlight key gait features that are most relevant to PD onset. We validated the framework on public benchmark datasets and custom gait recordings. Our results show high classification accuracy that surpasses traditional statistical and machine learning methods. The proposed approach connects wearable sensing, advanced AI, and clinical decision support, providing a scalable, accurate, and interpretable solution for the early detection of PD+.