Code: A general model based on Riemannian manifold for stable decoding continuous hand movement trajectory from ECoG signals in non-human primate

This repository contains source code relevant of the manuscript:
A general model based on Riemannian manifold for stable decoding continuous hand movement trajectory from ECoG signals in non-human primate.
Reza Eyvazpour, Behraz Farrokhi, and Abbas Erfanian

Summary:
This study proposes a novel method for decoding continuous 3D hand trajectories from electrocorticographic (ECoG) signals for brain-computer interface (BCI) applications, addressing the challenge of inter-session variability. The approach combines Riemannian geometry-based feature extraction with LSTM-BiLSTM networks to enable transfer learning across sessions. Using ECoG data from five monkeys performing reaching tasks, spatial covariance matrices were computed from ten frequency band powers and mapped onto a Riemannian manifold to extract session-invariant features. These features, along with spectral data, were used to train deep learning models (LSTM-BiLSTM). The proposed method outperformed baseline models using only spectral features, showing enhanced generalization across sessions and supporting the utility of geometric-temporal modeling in BCI development.

For more details and code usage, please refer to the description file.