Towards Development of a 3-State Self-Paced Brain-Computer Interface

Ali Bashashati, Rabab K. Ward, Gary E. BirchComput. Intell. Neurosci 2007 · OpenReview

TL;DR

This paper introduces and evaluates two novel designs for a 3-state self-paced brain-computer interface (BCI) capable of distinguishing between right-hand extensions, left-hand extensions, and a "no-control" (idle) state. This advancement offers more control options for users compared to existing 2-state systems while maintaining comparable performance and a crucial ability to ignore idle brain activity.

What Problem Does This Paper Address?

Traditional brain-computer interfaces (BCIs) often operate synchronously, meaning they only process brain signals during predefined time windows. This can be restrictive for users. Self-paced (asynchronous) BCIs, on the other hand, are always active and allow users to initiate control whenever they intend to. The existing 2-state asynchronous BCIs, like the Low-Frequency Asynchronous Switch Design (LF-ASD), can detect a single type of intended control (e.g., a finger flexion) but struggle to differentiate between multiple distinct commands or to reliably ignore periods of user inactivity (the "no-control" or NC state). This limitation restricts the complexity of commands a user can issue. This paper aims to bridge this gap by developing a 3-state self-paced BCI that can differentiate between two distinct motor commands (right vs. left hand extensions) and robustly support the NC state, offering a more nuanced and practical control solution.

Methods

The proposed 3-state BCI is built upon a two-stage detection process:

  1. Detector 1 (DET1): This stage determines whether any movement-related brain activity is present, distinguishing between an "intentional control" (IC) state and the "no-control" (NC) state. Two designs for DET1 were explored:
  2. DET1-LF-1NN: This design utilizes features from the 0-4 Hz band of bipolar EEG channels and employs a template matching algorithm similar to the original LF-ASD. It uses a 1-nearest neighbor (1-NN) classifier.
  3. DET1-PSD-LDA: This design extracts power spectral density (PSD) features from bipolar EEG channels. It uses stepwise linear discriminant analysis (LDA) to select the most informative features and an LDA classifier.
  4. Detector 2 (DET2): If DET1 detects movement, DET2 then classifies the movement as either a right-hand or a left-hand extension. The paper focuses on DET2-PSD-LDA, which uses PSD features and stepwise LDA to identify subject-specific ERD/ERS frequency bands and channels for discrimination. This detector is similar to DET1-PSD-LDA but omits the moving average and debounce blocks.

The overall 3-state BCI system combines these detectors. Two main configurations were evaluated: <DET1-LF-1NN + DET2-PSD-LDA> and <DET1-PSD-LDA + DET2-PSD-LDA>.

Datasets

The study utilized EEG data collected from four able-bodied individuals. The data was recorded using 15 monopolar electrodes over the motor cortex and supplementary motor area, along with EOG for artifact detection. Participants performed a cue-based task where they were instructed to extend their right or left hand when a visual cue indicated. Approximately 150 trials for each movement type were collected per subject across two sessions. The data was split into 80% for training and 20% for testing, with a cross-validation procedure repeated five times. No new benchmark datasets were introduced; the study used data collected specifically for this research.

Key Insights and Ideas

  • The Power of Differentiation: Moving beyond a simple "move/don't move" detection to differentiating between specific motor commands (right vs. left hand extensions) significantly enhances the potential for richer control in BCIs.
  • Robust NC Support is Crucial: The ability of a BCI to reliably ignore periods of inactivity (NC state) is paramount for practical applications. The proposed 3-state system explicitly designs for this, unlike some prior attempts that might have masked NC states with "unknown" outputs.
  • Hand Extensions as a Promising Motor Task: The choice of right- and left-hand extensions as the control tasks appears to yield more discriminative EEG patterns compared to simpler movements like finger flexions, potentially due to involving more body parts and requiring more effort.
  • PSD Features Show Promise: The DET1-PSD-LDA design, utilizing power spectral density features and stepwise LDA, demonstrated competitive or superior performance compared to the LF-ASD-based approach, suggesting the effectiveness of spectral analysis for movement detection.

Key Results

  • 2-State Performance (DET1 alone): When used as a 2-state BCI (detecting any movement vs. no movement), the system achieved an average detection rate (TPIC) of 58.1% at a fixed false positive rate (FP) of 1%. This is a significant improvement over previous 2-state LF-ASD systems, which reported around 41% TP at 1% FP.
  • 3-State Performance:
    • The <DET1-LF-1NN + DET2-PSD-LDA> design achieved an average true positive rate of 36.0% for right-hand extensions (TPR) and 33.9% for left-hand extensions (TPL) at a 1% FP rate.
  • The <DET1-PSD-LDA + DET2-PSD-LDA> design achieved an average TPR of 37.5% and TPL of 42.8% at a 1% FP rate.
  • Subject Variability: Performance varied across subjects, with Subject 4 showing the best differentiation between right and left movements.
  • Comparison: The 3-state system's performance was comparable to existing 2-state systems when considered as a 2-state BCI, but with the added benefit of multiple control states.

Limitations and Future Work

  • Limited Subject Pool: The study was conducted with only four able-bodied individuals. The authors emphasize the need to confirm these findings on a larger subject pool, particularly individuals with motor disabilities (the target population).
  • No Real-time Feedback: Participants did not receive real-time feedback during the experiment. While this simulates a more natural control mode, it might have contributed to lower performance in some subjects. Future work should explore the impact of online feedback.
  • No Subject Prescreening or Prior Training: The absence of prescreening and prior training for subjects could have influenced performance. Future studies should incorporate these elements.
  • False Positive Rate: While a 1% FP rate is used, the authors acknowledge that even this rate can lead to an average of one false activation every six seconds, which is still too high for practical applications.
  • Future Directions: The authors suggest exploring more efficient feature extraction and classification methods, implementing subject training protocols, providing online feedback, and verifying results on a larger subject pool to improve performance and address the limitations. They also note that combining complementary designs might yield further gains.

Summary updated

Abstract

Most existing brain-computer interfaces (BCIs) detect specific mental activity in a so-called synchronous paradigm. Unlike synchronous systems which are operational at specific system-defined periods, self-paced (asynchronous) interfaces have the advantage of being operational at all times. The low-frequency asynchronous switch design (LF-ASD) is a 2-state self-paced BCI that detects the presence of a specific finger movement in the ongoing EEG. Recent evaluations of the 2-state LF-ASD show an average true positive rate of 41&#37; at the fixed false positive rate of 1&#37;. This paper proposes two designs for a 3-state self-paced BCI that is capable of handling idle brain state. The two proposed designs aim at detecting right- and left-hand extensions from the ongoing EEG. They are formed of two consecutive detectors. The first detects the presence of a right- or a left-hand movement and the second classifies the detected movement as a right or a left one. In an offline analysis of the EEG data collected from four able-bodied individuals, the 3-state brain-computer interface shows a comparable performance with a 2-state system and significant performance improvement if used as a 2-state BCI, that is, in detecting the presence of a right- or a left-hand movement (regardless of the type of movement). It has an average true positive rate of 37.5&#37; and 42.8&#37; (at false positives rate of 1&#37;) in detecting right- and left-hand extensions, respectively, in the context of a 3-state self-paced BCI and average detection rate of 58.1&#37; (at false positive rate of 1&#37;) in the context of a 2-state self-paced BCI.
BibTeX
@inproceedings{bashashati2007towards,
  author = {Ali Bashashati and Rabab K. Ward and Gary E. Birch},
  title = {{Towards Development of a 3-State Self-Paced Brain-Computer Interface}},
  booktitle = {Comput. Intell. Neurosci},
  year = {2007},
  url = {https://openreview.net/forum?id=zTiPF1HL_Vw},
}
Figure