INVESTIGADORES
PETERSON Victoria
congresos y reuniones científicas
Título:
Movement decoding from subdural electrocorticography and subthalamic local field potentials in Parkinson?s disease
Autor/es:
TIMON MERK; VICTORIA PETERSON; MARK RICHARSON; WOLF-JULIAN NEUMANN
Reunión:
Conferencia; Bernstein Conference 2020; 2020
Resumen:
The basal ganglia are known to be involved in the planning, execution and control of movement vigor often measured as grip force. Machine learning based decoding of grip force may augment clinical brain computer interfaces to support motor function in patients with neurological disorders. Although this has been primarily researched using cortical signals, recent publications have shown promising results using local field potentials (LFP) from the subthalamic nucleus (STN) in patients with deep brain stimulation (DBS) electrodes [1]. In order to determine ideal biomarkers for a brain computer interface (BCI) in the application of adaptive DBS [2], the present study aims at comparing the performance of subdural electrocorticography [ECoG] and subthalamic LFP signals for movement decoding. Therefore, ECoG signals were recorded simultaneously with STN-LFP signals from 11 Parkinson?s disease patients performing a force gripping task during awake functional neurosurgery. Neural sources, signal features and machine learning methods are compared in a systematic manner. Multiple band power features are computed and used to train Linear Models (LM), Neural Networks (NN) and ensemble based methods (XGBOOST) within a Bayesian Optimization procedure to decodegrip force. The results are also compared against the Source Power Comodulation (SPoC) method, the state-of-the-art spatial filtering and band-power feature extraction method for continuous target decoding. Figure 1 shows the prediction performance of each method considered. We find that ensemble methods perform best in the cross-validated test set. In addition, our results consistently show that there is a significant performance advantage of ECoG over STN signals. Although the implantation of ECoG strip electrodes is currently not practiced in clinical routine, our results show that ECoG promises significant advantages for BCI based decoding for next-generation neurostimulation approaches, such as intelligent adaptive DBS for Parkinson?s disease [2].