A WaveNet Deep Learning Framework for Real-Time FoG Prediction in Patients with Parkinson’s Disease

Freezing of gait (FoG) is among the most disabling motor symptoms experienced by individuals with movement disorders, particularly Parkinson’s disease (PD). FoG episodes can severely limit mobility, heighten the risk of falls, disrupt daily routines, and profoundly diminish quality of life. Early and reliable detection of FoG is therefore crucial to enable timely intervention and more effective disease management. However, traditional detection methods often fall short in accuracy due to the complex and variable nature of gait patterns. In this study, we propose a novel WaveNet-based deep learning framework that effectively captures long-range temporal dependencies in gait signals, addressing key limitations of conventional models. Unlike standard recurrent or convolutional networks, WaveNet’s dilated causal convolutions allow for efficient modeling of sequential accelerometer data with minimal computational cost, making it well suited for real-time applications. To assess its performance, we benchmarked the proposed WaveNet model against established deep learning architectures, including Deep Convolutional long short-term memory and InceptionTime, using two comprehensive datasets, tDCS-FoG and DeFoG, containing annotated accelerometer recordings collected in both controlled laboratory and real-world home environments. To address the significant class imbalance present in FoG datasets, we employed the Adaptive Synthetic Sampling (ADASYN) technique alongside jerk-based features, which improves the detection of subtle gait changes. Our experimental results show that the proposed WaveNet model achieves a precision of 0.99 and a specificity of 0.98, outperforming previously reported precision levels, which ranged between 77% and 91% on comparable benchmarks. With its low inference latency and lightweight architecture, the proposed framework is well positioned for integration into wearable systems to enable continuous, real-time FoG monitoring. This advancement holds substantial promise for clinical and home use, ultimately contributing to safer mobility and an improved quality of life for individuals living with PD.