کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Electroencephalogram-Based Spinal Cord Injury Signals Decoding Using Temporal-Spatial Dense Deep Convolutional Neural Network

نویسندگان :

Mohammadreza Abbasi Sardari¹ Hamed Mirzabagherian² Amir Abolfazl Suratgar³ Mohammad Bagher Menhaj⁴

1- صنعتی امیرکبیر(پلی تکنیک تهران) 2- صنعتی امیرکبیر(پلی تکنیک تهران) 3- صنعتی امیرکبیر(پلی تکنیک تهران) 4- صنعتی امیرکبیر(پلی تکنیک تهران)

کلمات کلیدی :

electroencephalogram،Spinal Cord Injury،convolutional neural network،Dense Module،Temporal-Spatial،Classification،Brain-Computer Interface

چکیده :

Brain-computer interfaces (BCIs) offer innovative solutions for restoring hand functionality in individuals with cervical spinal cord injuries (SCI). Accurate brain activity decoding through automated and efficient feature extraction techniques from raw EEG signals is essential for improving BCI performance while minimizing reliance on extensive preprocessing. In this study, we present a deep learning architecture—Temporal-Spatial Dense Deep Convolutional Neural Network (TSD-Net)—specifically designed to classify electroencephalogram (EEG) signals related to various hand movements in SCI patients. Utilizing Temporal-Spatial convolutions and designing a dense module, we successfully extracted long-range dependencies and complex patterns from the Temporal-Spatial features of the model, thereby enhancing the performance of previous models. Our findings showcase the models' capability to decode distinct movement patterns, achieving a classification accuracy of 77.89% for the challenging SCI EEG dataset sourced from subjects with various spinal cord injuries. Our end-to-end approach not only addresses the challenge of manual feature extraction from raw data but also significantly improves the performance of BCI systems. We performed a comparative analysis of the TSD-Net model against well-established methodologies in the field. The proposed model demonstrated superior accuracy compared to prominent models, underscoring their potential to aid the advancement of BCI technologies for individuals with SCI.