Spectro-Temporal Fusion of High-Gamma and Low-Frequency ECoG Signals for Intracranial Finger Movement Decoding
Lists
Only you see this
Abstract
Electrocorticography (ECoG) provides a minimally invasive approach for acquiring high-fidelity intracranial brain signals. The capability to decode continuous finger movements from ECoG holds substantial promise for developing dexterous neuroprostheses, which can restore activities of daily living for individuals with motor disabilities. Despite this potential, the inherent complexity and distributed nature of finger somatotopy present a significant algorithmic challenge. To this end, we propose a novel decoding framework, driven by the specifically engineered HiLoFuseNet architecture, to effectively fuse the spectro-temporal dynamics of high-gamma and lowfrequency ECoG signals for continuous finger movement trajectory prediction. HiLoFuseNet achieves state-of-the-art decoding performance on the public BCI Competition IV and Stanford datasets. The average Pearson correlation coefficients between true and predicted finger movement trajectories were 0.631 and 0.534, respectively, representing improvements of 5.0% (p > 0.05) and 11.9% (p = 0.0049) over the previously best reported results. Furthermore, the model significantly outperforms all reimplemented baselines. Beyond superior performance, the model successfully captures the distinct neurophysiological behavior of high-gamma and low-frequency ECoG activities during finger movements, all while maintaining a concise and efficient profile in both feature extraction and network architecture. Collectively, this study provides a performant and neurophysiologically sound framework for continuous finger movement decoding from ECoG signals, establishing a solid foundation for the development of motor brain-computer interfaces.
BibTeX
@misc{sun2025spectro,
author = {Qiang Sun and E. Calvo and Bob Van Dyck and Yuan Yang and Jiayuan He and Marc M. Van Hulle},
title = {{Spectro-Temporal Fusion of High-Gamma and Low-Frequency ECoG Signals for Intracranial Finger Movement Decoding}},
year = {2025},
doi = {10.36227/techrxiv.176592065.57408366/v1},
}Loading related content...