Implementation Of Online Brain Computer Interface Using Motor Imagery In LabVIEW
Authors
Binderup, Asbjørn ; Garg, Anurag
Term
10. term
Publication year
2007
Pages
122
Abstract
Formålet var at undersøge, om træning med feedback kan forbedre en persons præstation i et online Brain-Computer Interface (BCI), et system der gør det muligt at styre en computer via hjernesignaler. Vi udviklede et online system, BCILab, der håndterede dataindsamling, signalbehandling, udtræk af kendetegn (features), klassifikation og lagring af parametre. Data blev indsamlet via en TCP-forbindelse til programmet ACQUIRE. Under signalbehandlingen brugte vi en spatial filtermetode (common average reference) for at mindske støj og detrending for at fjerne langsomme driftskomponenter, så signalerne blev mere stabile. Vi beregnede effekten ved 8, 10 og 12 Hz ud fra autoregressive modeller for kanalerne C3 og C4 og brugte disse kendetegn i en bayesiansk klassifikator (en statistisk metode). Træningsdata til klassifikatoren blev indsamlet i én screeningssession og flere feedbacksessioner, hvor deltageren blev bedt om at forestille sig en venstre eller højre bevægelse afhængigt af målet. Vi analyserede præstationen i hver session og på tværs af dage. Da der kun var to deltagere, betragtes undersøgelsen som et casestudie. I begge tilfælde sås forbedring over dage: for deltager 1 fra 48%±15% til 67%±10% over 3 dage og for deltager 2 fra 50%±0% til 56%±11% over 4 dage.
This study examined whether training with feedback improves a person's performance in an online Brain-Computer Interface (BCI), a system that allows users to control a computer using brain signals. We developed an online system called BCILab that handled data acquisition, signal processing, feature extraction, classification, and parameter storage. Data were collected via a TCP connection to the ACQUIRE program. For signal processing, we applied a common average reference spatial filter to reduce noise and detrending to remove slow drifts, aiming for higher signal-to-noise ratio and more stable signals. We extracted features by computing power at 8, 10, and 12 Hz from autoregressive models for channels C3 and C4, and used these features in a Bayesian classifier (a statistical method). Training data for the classifier were gathered in one screening session and multiple feedback sessions, during which the participant imagined a left or right movement according to the target. We analyzed performance within sessions and across days. Because there were only two participants, this is a case study. In both cases, accuracy improved across days: for subject 1 from 48%±15% to 67%±10% over 3 days, and for subject 2 from 50%±0% to 56%±11% over 4 days.
[This abstract was generated with the help of AI]
Keywords