Binary Classification of Spinal Cord Injury Patients’ EEG Data Based on the Local Linear Embedding and Spectral Embedding Methods

Spinal cord injury (SCI) is a chronic disease that damages the spinal cord, leading to the loss of neuronal function, especially sensorimotor functions. Brain-Computer Interface (BCI) controlled rehabilitation systems offer a promising therapeutic treatment for individuals with SCI. Their treatment often involves a lengthy and demanding rehabilitation process. For this reason, we introduced an innovative approach that utilizes an electroencephalography (EEG)-based BCI rehabilitation system to assist SCI patients. Our study involved the analysis of low-frequency EEG signals from nine individuals with SCIs while attempting arm and hand movements. EEG analysis generally consists of preprocessing, feature extraction, and Machine Learning (ML) algorithms for classification. However, relying solely on traditional methods for each step may prove inadequate for real-time applications. Traditional approaches can sometimes be limited by the complexity and high dimensionality of the signals. To address these challenges, we employed Manifold Learning, which allows for a more effective representation of the temporal and spatial features of brain activity in a lower-dimensional space of EEG signals. Our study obtained certain results by trying Spectral Embedding and Local Linear Embedding methods as Manifold Learning algorithms. The classification was implemented using k-NN, Naive Bayes, and SVM methods. According to the average results, the k-NN algorithm gives the best accuracies for Local Linear Embedding methods obtained at 0.995, and the Spectral Embedding methods obtained at 0.996. While comparing the Manifold Learning methods, we achieved the highest success in Spectral Embedding.