Application of LSTM-based models in Impaired Finger Dexterity of Parkinson’s Disease

This study proposes a novel multi-class classification model for the early detection and severity level classification of Parkinson’s Disease (PD) using data from a 6 Degrees of Freedom (6DoF) sensor. Recognizing that movement dysfunction, particularly impaired finger dexterity, is a hallmark of PD, this study focuses on analyzing finger movement patterns during a simple keyboard-tapping test. Data collected from wearable sensors are first transformed into the frequency domain using Fourier transformation. Gaussian filtering is then applied to reduce noise and emphasize salient features. These processed spectral features are then used to train a Long Short-Term Memory (LSTM) network for classification. Evaluated on a dataset of 1,553 records, the model achieved a promising accuracy of 74%, with a 90% sensitivity for detecting early-stage PD (symptom level 1) and a 9.7% false-negative rate. This study highlights the potential of combining wearable sensor technology with LSTM-based analysis to develop an accessible, non-invasive, and scalable solution for early PD screening and severity assessment, ultimately aiding clinical decision-making.