[Paper Review] Boosted Convolutional Neural Networks for Motor Imagery EEG Decoding with Multiwavelet-based Time-Frequency Conditional Granger Causality Analysis

Yang Li, Meng-Ying Lei+5 more|arXiv (Cornell University)|Oct 22, 2018

EEG and Brain-Computer Interfaces44 references4 citations

TL;DR

This paper proposes a boosted convolutional neural network (ConvNet) framework for motor imagery EEG decoding by leveraging multiwavelet-based time-frequency conditional Granger causality (TF-CGC) to extract discriminative spatio-temporal connectivity features. The method achieves 12.15% higher average accuracy and 74.02% lower inter-subject standard deviation than the competition winner on BCI Competition IV Dataset IIa.

ABSTRACT

Decoding EEG signals of different mental states is a challenging task for brain-computer interfaces (BCIs) due to nonstationarity of perceptual decision processes. This paper presents a novel boosted convolutional neural networks (ConvNets) decoding scheme for motor imagery (MI) EEG signals assisted by the multiwavelet-based time-frequency (TF) causality analysis. Specifically, multiwavelet basis functions are first combined with Geweke spectral measure to obtain high-resolution TF-conditional Granger causality (CGC) representations, where a regularized orthogonal forward regression (ROFR) algorithm is adopted to detect a parsimonious model with good generalization performance. The causality images for network input preserving time, frequency and location information of connectivity are then designed based on the TF-CGC distributions of alpha band multichannel EEG signals. Further constructed boosted ConvNets by using spatio-temporal convolutions as well as advances in deep learning including cropping and boosting methods, to extract discriminative causality features and classify MI tasks. Our proposed approach outperforms the competition winner algorithm with 12.15% increase in average accuracy and 74.02% decrease in associated inter subject standard deviation for the same binary classification on BCI competition-IV dataset-IIa. Experiment results indicate that the boosted ConvNets with causality images works well in decoding MI-EEG signals and provides a promising framework for developing MI-BCI systems.

Motivation & Objective

• Address the challenge of nonstationary perceptual decision processes in motor imagery (MI) EEG decoding for brain-computer interfaces (BCIs).
• Improve classification accuracy and robustness across subjects in MI-EEG signal decoding.
• Develop a data-driven, causality-aware feature representation that preserves time, frequency, and spatial connectivity information.
• Integrate advanced deep learning techniques—specifically spatio-temporal convolutions, cropping, and boosting—into a unified framework for EEG decoding.
• Provide a scalable and generalizable framework for next-generation MI-BCI systems with enhanced inter-subject performance consistency.

Proposed method

• Employ multiwavelet basis functions combined with Geweke spectral measure to compute high-resolution time-frequency conditional Granger causality (TF-CGC) representations.
• Apply a regularized orthogonal forward regression (ROFR) algorithm to identify a parsimonious and generalizable TF-CGC model.
• Construct causality images from alpha-band multichannel EEG signals that encode time, frequency, and spatial connectivity patterns for use as input to deep networks.
• Design a boosted ConvNet architecture using spatio-temporal convolutions to extract hierarchical discriminative features from causality images.
• Incorporate data augmentation and boosting techniques to enhance model generalization and performance on limited EEG data.
• Optimize the network through end-to-end training with cross-entropy loss and adaptive learning rate scheduling.

Experimental results

Research questions

• RQ1 Can multiwavelet-based time-frequency conditional Granger causality provide a more informative and discriminative representation of motor imagery EEG signals compared to traditional spectral or time-domain features?
• RQ2 To what extent can causality images derived from TF-CGC enhance the performance of deep learning models in decoding binary motor imagery tasks?
• RQ3 How does the proposed boosted ConvNet framework compare to state-of-the-art methods in terms of accuracy and inter-subject variability on benchmark EEG datasets?
• RQ4 Does the integration of spatio-temporal convolutions and boosting techniques improve feature learning and classification robustness in EEG decoding?
• RQ5 Can the proposed method achieve superior generalization across subjects while maintaining high decoding accuracy?

Key findings

• The proposed method achieved a 12.15% increase in average classification accuracy compared to the competition winner on BCI Competition IV Dataset IIa.
• Inter-subject standard deviation was reduced by 74.02% compared to the competition winner, indicating significantly improved robustness and consistency across subjects.
• The causality images effectively preserved time, frequency, and spatial connectivity information, enabling the model to learn discriminative patterns from EEG connectivity dynamics.
• The boosted ConvNet architecture demonstrated superior feature extraction capability, outperforming standard ConvNets and traditional methods in both accuracy and generalization.
• The use of ROFR for model selection ensured a parsimonious yet effective TF-CGC representation, minimizing overfitting and improving interpretability.
• The integration of multiwavelet-based TF-CGC with deep learning provides a promising, scalable framework for real-time and robust MI-BCI systems.