Transformer-based long-term predictor of subthalamic beta activity in Parkinson’s disease

Abstract

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) is a mainstay treatment for patients with Parkinson’s disease (PD). The adaptive DBS approach (aDBS) modulates stimulation, based on the power in the beta range ([12–30] Hz) of STN local field potentials, aiming to follow the patient’s clinical state. Control of aDBS relies on identifying the correct thresholds of pathological beta power. Currently, in-person reprogramming sessions, due to changes in beta power distribution over time, are needed to ensure clinical efficacy. Here we present LAURA, a Transformer-based framework predicting the nonlinear evolution of subthalamic beta power up to 6 days in advance, based on the analysis of chronic recordings. High prediction accuracy (>90%) was achieved in four PD patients with chronic DBS over months of recordings, independently of stimulation parameters. Our study paves the way for remote monitoring strategies and the implementation of new algorithms for personalized auto-tuning aDBS devices.