Deep-Learning Assisting Cerebral Palsy Patient Handgrip Task Translation

• Published in: Journal of physics. Conference series 2021-07, Vol.1962 (1), p.12047
• Main Authors: Zakaria, Fazrul Faiz, Warip, Mohd Nazri Mohd, Ehkan, Phaklen, Mustapa, Muslim, Ilyas, Mohd Zaizu
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Computer architecture; Deep learning; Electroencephalography; Human-computer interface; Machine learning; Neural networks; Signal processing; Spatial filtering; Training
• ISSN: 1742-6588; 1742-6596
• DOI: 10.1088/1742-6596/1962/1/012047

An electro-encephalography (EEG) brain-computer interface (BCI) can provide the brain and external environment with separate information sharing and control networks. EEG impulses, though, come from many electrodes, which produce different characteristics, and how the electrodes and features to enhance classification efficiency have been chosen has become an urgent concern. This paper explores the deep convolutional neural network architecture (CNN) hyper-parameters with separating temporal and spatial filters without any pre-processing or artificial extraction processes. It selects the raw EEG signal of electrode pairs over the cortical area as hybrid samples. Our proposed deep-learning model outperforms other neural network models previously applied to this dataset in training time (∼40%) and accuracy (∼6%). Besides, considerations such as optimum order for EEG channels do not limit our model, and it is patient-invariant. The impact of network architecture on decoder output and training time is further discussed.