Macroscopic EEG Reveals Discriminative Low-Frequency Oscillations in Plan-to-Grasp Visuomotor Tasks

TL;DR

This study develops a vision-based, plan-to-grasp paradigm and uses Filter-Bank Common Spatial Pattern (FBCSP) with SVM to decode grasp types from noninvasive EEG. By temporally isolating planning from execution, the authors demonstrate that low-frequency oscillations in the $0.5-8$ Hz range encode grasp information during planning and persist into execution, outperforming traditional MRCP methods. Broadband features across bands further enhance performance, with theta-dominated planning achieving up to about $75 ext{--}78$ ext{%} accuracy for grasp-type discrimination and beta/gamma activity contributing during execution for grasp vs no-grasp classification. These findings support the temporal macroscopic emergence of visuomotor networks and point to efficient, noninvasive BMI strategies that decode planning signals for naturalistic grasp control, potentially benefiting individuals with severe motor impairments.

Abstract

The vision-based grasping brain network integrates visual perception with cognitive and motor processes for visuomotor tasks. While invasive recordings have successfully decoded localized neural activity related to grasp type planning and execution, macroscopic neural activation patterns captured by noninvasive electroencephalography (EEG) remain far less understood. We introduce a novel vision-based grasping platform to investigate grasp-type-specific (precision, power, no-grasp) neural activity across large-scale brain networks using EEG neuroimaging. The platform isolates grasp-specific planning from its associated execution phases in naturalistic visuomotor tasks, where the Filter-Bank Common Spatial Pattern (FBCSP) technique was designed to extract discriminative frequency-specific features within each phase. Support vector machine (SVM) classification discriminated binary (precision vs. power, grasp vs. no-grasp) and multiclass (precision vs. power vs. no-grasp) scenarios for each phase, and were compared against traditional Movement-Related Cortical Potential (MRCP) methods. Low-frequency oscillations (0.5-8 Hz) carry grasp-related information established during planning and maintained throughout execution, with consistent classification performance across both phases (75.3-77.8\%) for precision vs. power discrimination, compared to 61.1\% using MRCP. Higher-frequency activity (12-40 Hz) showed phase-dependent results with 93.3\% accuracy for grasp vs. no-grasp classification but 61.2\% for precision vs. power discrimination. Feature importance using SVM coefficients identified discriminative features within frontoparietal networks during planning and motor networks during execution. This work demonstrated the role of low-frequency oscillations in decoding grasp type during planning using noninvasive EEG.

Figures (8)

• Figure 1: Experimental setup and data collection protocol for vision-based grasping platform. (a) Depicts a participant wearing the 16-channel EEG recording system during the experimental protocol. The top view of the turntable rotates to present one of three conditions at random: bottle, pen, or empty. (b) Single-trial structure and outline of EEG recording session, beginning with the planning phase, then the reach-to-grasp execution phase following audio cue.
• Figure 2: Overview of FBCSP multiclass classification pipeline using one-vs-rest strategy with three trained binary classifiers for final class prediction.
• Figure 3: Classification accuracies for cross-validation analysis across different experimental conditions and EEG frequency bands. Results show average classification performance ($\pm$STD) across all subjects using 10-fold CV on the training set for three binary conditions: pen vs. bottle (top), pen vs. empty (middle), and bottle vs. empty (bottom). Blue circles represent planning phase data, green circles represent movement phase data, with individual frequency bands (delta, theta, alpha, beta, gamma) and broadband features (concatenated across all bands) shown on the x-axis. Statistical significance between task phases are indicated by asterisks (*$p < 0.05$, ***$p < 0.001$; NS = not significant).
• Figure 4: Temporal evolution of 10-fold CV classification accuracy across frequency bands during plan-to-grasp tasks. Classification performance is shown over the plan-to-grasp task within three frequency bands (theta, beta, Broadband) and three binary conditions (pen vs. bottle, pen vs. empty, bottle vs. empty), averaged across all subjects. The number of extracted features remains constant at each time point, with classification performance varying based on the cumulative temporal information available for feature extraction. The vertical line at movement onset (t = 3s) separates the planning phase (blue shaded region) from the movement phase (green shaded region).
• Figure 5: Most significant CSP patterns and corresponding broadband SVM coefficients for representative Subject 8. The left panel shows the most significant CSP patterns in the theta band (CSP4 and CSP2) and their corresponding broadband SVM coefficients during the planning phase, while the right panel displays the most significant CSP patterns in the beta band (CSP9 and CSP1) and their corresponding broadband SVM coefficients during the movement phase. Topographic maps illustrate the spatial distribution of CSP patterns with positive (red) and negative (blue) weights across electrode locations. Bar plots show the magnitude of broadband SVM coefficients across different CSP indices (CSP1-CSP16), with colored bars representing individual CSP components and dashed horizontal lines indicating mean values across each frequency band.