Skip to content
Review Open access

A Comprehensive Review of Sequence and Generative Models in Motor Imagery (MI) Classification for Brain–Computer Interfaces (BCIs)

Aug 2026 · Italian National Conference on Sensors · Vol 26, pp. 5097 · 0 citations · 150 references
Medicine

TL;DR

This study performs an in-depth analysis on the strengths and limitations of traditional models such as RNNs and LSTMs as well as emergent models such as VAEs and transformers, which have demonstrated superior performance in extracting the intricate patterns of the EEG data with low latency.

Abstract

Motor imagery (MI) classification serves as the backbone to brain–computer interfaces (BCIs) by strengthening the communication bridge between the human brain and external peripheral devices. The past two decades have witnessed unprecedented success in MI-BCIs, with applications not only in medical fields but also in several other domains, such as gaming and robotic control. Initially, MI classification primarily relied on classical signal processing techniques that were heavily impacted by signal variations; however, recent trends in deep learning (DL), specifically in sequence-oriented, attention-based, hybrid, and generative architectures such as recurrent neural networks (RNNs), variational autoencoders (VAEs), generative adversarial networks (GANs), and transformers have significantly improved the efficiency and robustness of MI classification. This study presents a comprehensive review of these sequence-oriented, attention-based, hybrid, and generative architectures including RNNs, VAEs, GANs, and transformers, comparing their robustness across various public MI datasets, highlighting their challenges, such as inter-subject variation, low signal-to-noise ratio (SNR), and the obstacles in real-time signal classification. We perform an in-depth analysis on the strengths and limitations of traditional models such as RNNs and LSTMs as well as emergent models such as VAEs and transformers, which have demonstrated superior performance in extracting the intricate patterns of the EEG data with low latency. Moreover, we critically examine the future potential of such models in overcoming current bottlenecks, such as weak generalization on unseen data and high computational load. This study aims to assist researchers in attaining significant insights into the state-of-the-art sequence, attention-based, hybrid, and generative models used in MI classification, thus offering a direction for future innovation.

Read PDF

Similar papers

Review Open access Sep 2026

Cross-variability decoding for motor imagery EEG signals: a comprehensive review

A comprehensive taxonomy of MI EEG cross-variability decoding studies from 2020 to 2025 is presented, systematically organizing advances in deep learning and transfer learning and critically evaluate core algorithmic approaches, including Convolutional Neural Networks, transformers, feature alignment, domain adaptation...

Li-Jun Wang, Yue-Ying Zhou, Peng-Pai Wang et al. · 0 citations
Open access Sep 2026

Subject-Specific Deep Learning Model for Motor Imagery Direction Decoding

A novel deep learning framework for online decoding of binary directional MI signals from the dominant hand of 20 healthy subjects using EEGNet-based convolutional filters to extract temporal and spatial features and is outperforming the existing deep learning models.

Praveen K. Parashiva, Sagila Gangadharan Kutteri, A. Vinod · 0 citations
Open access Sep 2026

Subject-Independent Multi-Class Motor Imagery EEG Classification Using Hybrid CNN–BiLSTM with Channel Attention

A hybrid deep-learning architecture that combines convolutional neural networks, bidirectional long short-term memory, and squeeze-and-excitation feature-channel attention for subject-independent four-class MI-EEG classification provides a favorable accuracy–complexity trade-off for subject-independent multi-class BCI...

Saif Rasool · 0 citations
Open access Aug 2026

Subject-Specific BCI Frameworks for Motor Imagery Classification Based on BSS-Free and BSS-Equipped Pipelines

A localized architectural framework is introduced to evaluate whether a lightweight pipeline operating without BSS (No-BSS) is sufficiently efficient for real-time control when compared against two BSS-equipped pipelines utilizing Independent Component Analysis (ICA) and Empirical Mode Decomposition (EMD).

Nerita Ramsoonder, Rito Clifford Maswanganyi, Philani Khumalo · 0 citations
Sep 2026

Comprehensive deep representation-based Wasserstein domain discriminator for cross-subject motor imagery EEG classification.

A novel deep domain adaptation model, referred to as Comprehensive Deep Representation based Wasserstein Domain Discriminator (CDR-WDD), to simultaneously represent comprehensive deep features and adversarial learned based on the domain discriminator with stable learning gradient is proposed.

Tian-Jian Luo, Zi-Kun Cai, Xuan Cao · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.