A novel paradigm for fast training data generation in asynchronous movement-based BCIs

Abstract

Introduction: Movement-based brain-computer interfaces (BCIs) utilize brain activity generated during executed or attempted movement to provide control over applications. By relying on natural movement processes, these BCIs offer a more intuitive control compared to other BCI systems. However, non-invasive movement-based BCIs utilizing electroencephalographic (EEG) signals usually require large amounts of training data to achieve suitable accuracy in the detection of movement intent. Additionally, patients with movement impairments require cue-based paradigms to indicate the start of a movement-related task. Such paradigms tend to introduce long delays between trials, thereby extending training times. To address this, we propose a novel experimental paradigm that enables the collection of 300 cued movement trials in 18 min. Methods: By obtaining measurements from ten participants, we demonstrate that the data produced by this paradigm exhibits characteristics similar to those observed during self-paced movement. Results and discussion: We also show that classifiers trained on this data can be used to accurately detect executed movements with an average true positive rate of 31.8% at a maximum rate of 1.0 false positives per minute.