Stimuli and feature extraction methods for EEG-based brain-machine interfaces: a systematic comparison.

• Ano de defesa: 2017
• Autor(a) principal: Villalpando, Mayra Bittencourt
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Brain-Machine Interface (BMI); Eletroencefalografia; Eletroencephalography (EEG); Interface homem-computador; Neurociências; P300; Potenciais evocados; Processamento de sinais biomédicos; SSVEP
• Link de acesso: http://www.teses.usp.br/teses/disponiveis/3/3152/tde-19032018-090128/

A brain-machine interface (BMI) is a system that allows the communication between the central nervous system (CNS) and an external device (Wolpaw et al. 2002). Applications of BMIs include the control of external prostheses, cursors and spellers, to name a few. The BMIs developed by various research groups differ in their characteristics (e.g. continuous or discrete, synchronous or asynchronous, degrees of freedom, others) and, in spite of several initiatives towards standardization and guidelines, the cross comparison across studies remains a challenge (Brunner et al. 2015; Thompson et al. 2014). Here, we used a 64-channel EEG equipment to acquire data from 19 healthy participants during three different tasks (SSVEP, P300 and hybrid) that allowed four choices to the user and required no previous neurofeedback training. We systematically compared the offline performance of the three tasks on the following parameters: a) accuracy, b) information transfer rate, c) illiteracy/inefficiency, and d) individual preferences. Additionally, we selected the best performing channels per task and evaluated the accuracy as a function of the number of electrodes. Our results demonstrate that the SSVEP task outperforms the other tasks in accuracy, ITR and illiteracy/inefficiency, reaching an average ITR** of 52,8 bits/min and a maximum ITR** of 104,2 bits/min. Additionally, all participants achieved an accuracy level above 70% (illiteracy/inefficiency threshold) in both SSVEP and P300 tasks. Furthermore, the average accuracy of all tasks did not deteriorate if a reduced set with only the 8 best performing electrodes were used. These results are relevant for the development of online BMIs, including aspects related to usability, user satisfaction and portability.