Multiple Correlated Component Analysis for Identifying the Bilateral Location of Target in Visual Search Tasks

李沛洋, Yin Tian, Haiyong Zhang, Peiyang Li, Yang LiIEEE Access 2019 · DOI

Introduction

Brain-computer interfaces (BCIs) represent a transformative technology for human-machine interaction, traditionally relying on paradigms like the P300 or Steady-State Visual Evoked Potential (SSVEP). However, researchers are increasingly exploring the N2pc component—an electroencephalogram (EEG) signal appearing approximately 250ms post-stimulus over the brain hemisphere contralateral to the focus of visual attention.

Despite its potential for intuitive control, the N2pc is notoriously difficult to detect in single trials due to its low signal-to-noise ratio and high inter-hemispheric synchronicity. Historically, researchers have relied on averaging many trials to extract a clean signal, which slows down BCI responsiveness. This study introduces Multiple Correlated Component Analysis (MCORCA), a novel signal-processing framework designed to achieve high-accuracy, single-trial classification of visual attention.

Methodology

The core of the proposed approach is the MCORCA framework, which optimizes spatial filters to extract the N2pc component more effectively than standard methods. While standard Correlated Component Analysis (CORCA) typically seeks a single linear combination of EEG channels to maximize correlation across trials, MCORCA utilizes an optimal subset of these correlated components.

Workflow of the proposed MCORCA method for target identification.

The workflow involves:

  1. Spatial Filtering: Using CORCA to create spatial filters that weight different electrodes to "stretch" the amplitude difference between the brain's hemispheres.
  2. Feature Extraction: Calculating the correlation between a single-trial test sample and "template" signals for both Left Visual Field (LVF) and Right Visual Field (RVF) targets.
  3. Component Selection: The researchers discovered that using exactly five pairs of electrodes (components) provided the best balance between information retention and noise reduction.

Experimental Setup

The study evaluated 30 healthy subjects (ages 19–21) using a visual search task. Participants were tasked with finding a target circle ('O') hidden among distractor arcs ('C') across 400 trials. EEG data was captured using a 64-channel Neuroscan system, with analysis specifically focusing on posterior-occipital electrodes (such as PO7 and PO8) where N2pc activity is most prominent.

Experimental paradigm showing the visual search task sequence.

Results and Key Insights

MCORCA demonstrated superior performance compared to the traditional Support Vector Machine with Recursive Feature Elimination (SVM-RFE) approach:

  • Classification Accuracy: The method achieved a mean single-trial accuracy of 82% ± 3.70%, significantly outperforming the 75% achieved by SVM-RFE.
  • Information Transfer Rate (ITR): The ITR reached 35 ± 3.31 bits/minute, a substantial jump over the 22 bits/minute baseline.
  • Sample Efficiency: Even with a very small number of training trials (as few as 10 per class), MCORCA maintained a significant performance lead.

The researchers found that the method performed best in the 200–300ms window post-stimulus. This perfectly aligns with the known physiological timing of the N2pc, confirming that the algorithm is accurately capturing the intended neural process.

Discussion and Future Work

The success of MCORCA lies in its ability to bypass hemispheric synchronicity and amplify the faint N2pc signal. While these results were achieved in an offline study, they provide a strong foundation for more responsive, real-time BCIs.

Future research will focus on transitioning this framework to "online" systems and reducing user fatigue during calibration. The authors suggest that transfer learning—utilizing data from previous subjects to pre-train models—could significantly decrease the time required for new users to calibrate the system. Additionally, tailoring the analysis window to individual variations in N2pc latency may further push the boundaries of classification accuracy.

Summary updated

BibTeX
@article{2019multiple,
  author = {李沛洋 and Yin Tian and Haiyong Zhang and Peiyang Li and Yang Li},
  title = {{Multiple Correlated Component Analysis for Identifying the Bilateral Location of Target in Visual Search Tasks}},
  journal = {IEEE Access},
  year = {2019},
  doi = {10.1109/access.2019.2929185},
}
Figure