Automated posture adjustment system for immobilized patients using EEG signals.

Abstract

This paper presents a Brain Computing Interface (BCI) system utilizing Electroencephalography (EEG) for human posture Identification. The proposed approach follows a structured five-step process, ensuring accurate and efficient classification. The dataset collected using the MindRove EEG device captures brain activity during four motor imagery tasks: Leftward, Rightward, Upward, and Zeroth. Pre-processing involved filtering, followed by feature extraction using a Convolutional Recurrent Denoising Autoencoder (CRDAE) model. After that Classification is performed using artificial intelligence (AI) models, including Gated Recurrent Unit (GRU) with Attention, Temporal Transformer (TT), Bidirectional Long Short-Term Memory with attention mechanisms (Bi-LSTM with AM), and proposed Graph Transformer All Attention (GTAA). The GTAA model demonstrates superior performance, achieving the highest classification accuracy among the evaluated models. Additionally, the proposed system validated against the BCI Competition IV 2a datasets and ten-fold subject cross-validation, demonstrating its reliability and efficiency for real-time BCI applications. This study underscores the potential of integrating advanced AI techniques with EEG signal measurement and instrumentation for practical implementations.