Modeling and classifying Parkinsonian tremors with nonlinear dynamics and Kalman optimization

Abstract

Parkinson Disease (PD) is a chronic neurodegenerative disorder due to the degeneration of dopamine-producing neurons in a brain region, crucial for motor function regulation. Rest tremor is a vital parameter and is essential to diagnose the disease and assess its prognosis. The frequency of rest tremor is a key characteristic to differentiate it from other types of tremors such as Essential tremors. This highlights the need for reliable and accurate methods to estimate the rest tremor frequency as accurate as possible. This paper proposes a novel method, Extended Lagrangian combined with Kalman techniques for rest tremor frequency extraction. The Extended Lagrangian mechanism addresses the irregular oscillatory behavior of rest tremor by incorporating energy dissipation via Rayleigh’s dissipation function, nonlinear stiffness, as well as external forces. This mechanism serves as a foundation and helps to track the oscillatory nature of tremors. The video recordings of PD tremor taken in clinical settings may contain noise and have significant impact on the rest frequency estimate. The proposed approach addresses this issue, thus improving the accuracy of these estimates. This algorithm is validated on a dataset of 60 video recordings from PD patients, annotated by movement disorder specialists. The rest tremor frequency, along with other key features are then passed to a classifier to determine the severity of PD. The model achieved an accuracy of 98% with 1D CNN-LSTM classifier. This approach could be used in remote health assessment for PD patients, providing increased convenience to patients and caregivers.