Feature-Optimized Time-Series Neural Network for Predicting Rehabilitation Progress in Tremor Disorders

In recent years, the growing population and limited time resources have increased the likelihood of human errors in treating motor disorders such as tremor-related conditions, often leading to ineffective treatment and a waste of resources. Leveraging artificial intelligence for the diagnosis and prediction of treatment progress can significantly enhance the speed and accuracy of clinical decision-making. In this study, a time-series neural network is employed to predict the recovery trajectory of patients with tremor disorders. Features extracted from tremor signals over multiple rehabilitation sessions were evaluated using a t-test, and statistically significant features were selected. The neural network was then designed with optimized delay values and neuron counts to estimate future feature values based on current measurements. Given the strong dependence of prediction accuracy on the number of delays and neurons, as well as the improvement in performance when incorporating longer historical data, optimal network design is identified as one of the main challenges addressed in this research.

Keywords:

Tremor Disorders , Essential Tremor , Time-Series Neural Network , Signal Processing , Feature Selection.