RISE-iEEG: Robust to Inter-Subject Electrodes Implantation Variability iEEG Classifier

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Intracranial electroencephalography (iEEG) offers high-resolution neural data for clinical use and brain-computer interfaces (BCIs).
• Inter-subject variability in electrode placement hinders the development of generalized neural decoders.
• Existing models struggle to account for anatomical differences in electrode implantation sites.

Purpose of the Study:

• To introduce a novel deep learning model, RISE-iEEG, designed to be robust to inter-subject variability in iEEG electrode implantation.
• To develop a generalized neural decoder that does not require precise electrode coordinates for each participant.
• To improve the accuracy and generalizability of neural decoding from iEEG data.

Main Methods:

• Developed RISE-iEEG (Robust to Inter-Subject Electrode Implantation Variability iEEG Classifier), a deep neural network.
• Incorporated a participant-specific projection network to map individual neural data onto a common low-dimensional space.
• Evaluated RISE-iEEG on the Music Reconstruction and AJILE12 datasets, comparing its performance against established models like HTNet and EEGNet.

Main Results:

• RISE-iEEG demonstrated superior performance compared to HTNet and EEGNet, achieving an average F1 score of 0.83, approximately 7% higher.
• The model successfully compensated for electrode implantation variability without needing electrode coordinates.
• Analysis of projection network weights identified the Superior Temporal and Postcentral lobes as key encoding regions for the tested datasets.

Conclusions:

• RISE-iEEG offers a robust and generalizable solution for neural decoding from iEEG data, effectively handling inter-subject variability.
• The model enhances decoding accuracy while preserving interpretability, making it valuable for clinical applications and BCIs.
• The findings highlight the potential of adaptive projection networks in overcoming anatomical differences for cross-subject neural decoding.