Inferring Arm Movement Direction from EEG Signals Using Explainable Deep Learning.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Decoding human movement from brain signals is crucial for developing naturalistic brain-computer interfaces (BCIs).
• Traditional machine learning methods have been used, but deep learning applications in this area remain limited.
• Electroencephalography (EEG) offers a non-invasive method for capturing brain activity.

Purpose of the Study:

• To evaluate a convolutional neural network (CNN) for decoding movement direction from EEG signals during a reaching task.
• To investigate the interpretability of the CNN model using explanation techniques.
• To assess the feasibility of CNN-based EEG decoding for non-invasive BCIs.

Main Methods:

• Collected EEG data from twenty healthy participants during a delayed center-out reaching task.
• Utilized EEGNet, a CNN architecture, to classify movement direction in three scenarios: fine-direction, coarse-direction, and proximity.
• Applied DeepLIFT and occlusion tests for spatio-temporal EEG feature analysis and model interpretability.

Main Results:

• The CNN achieved above-chance decoding accuracies: 0.45 (five endpoints), 0.64 (three endpoints), and 0.70 (two endpoints) on average.
• Explainability analyses indicated that movement direction information is primarily encoded during the preparation phase.
• Key brain regions involved in decoding were identified as parietal and parietal-occipital areas.

Conclusions:

• CNN-based EEG decoding is a feasible and interpretable method for analyzing reaching movements.
• The findings provide valuable insights into visuomotor planning mechanisms.
• This research supports the advancement of non-invasive brain-computer interfaces.