BRIEF REVIEW OF NON-INVASIVE MOTION TRAJECTORY PREDICTION BASED BRAIN-COMPUTER INTERFACES Attila Korik 1 , Nazmul Siddique 1 , Damien Coyle 1 1 Intelligent Systems Research Centre, University of Ulster, Derry, UK korik-a@email.ulster.ac.uk, nh.siddique@ulster.ac.uk, dh.coyle@ulster.ac.uk Abstract – This paper presents a brief overview of non- invasive motion trajectory prediction (MTP) for brain- computer interface (BCI). This method is an ideal solution for controlling an artificial limb or robot arm because the applied signal processing algorithm reconstruct the track of the imagined movement. Regularly used neurosensory technique for an online outside of the lab application is electroencephalography (EEG). To date only a limited number of motion trajectory prediction studies have been reported using non-invasive techniques. This review provides a brief summary of the state of the art in this area. I. INTRODUCTION A common three dimensional (3D) brain- computer interface (BCI) provides the capacity to control objects in real or virtual three dimensional spaces using only brain signals. At the beginning of the millennium, invasive techniques were the most prevalent for motion control BCIs in which case surgical procedure is required to implant sensor(s) in the brain. Over the last decade there have been several non-invasive techniques applied in this area. The most common techniques are electroence- phalography (EEG), magnetoencephalography (MEG) and near-infrared spectroscopy (NIRS). Two different approaches are common in 3D BCIs. Multiclass classification based sensorimotor rhythms (SMR) BCIs take advantages of modulations in SMR which is related to a real or imagined motor task. This approach calculates power distribution of mu (8-12Hz) and beta (18-26Hz) EEG frequency range over central and parietal cortex. Participants can learn to modulate these bands voluntarily [1]. A multiclass BCI is an ideal solution for those applications where the control is possible with some continuously regulated two-stage switches. Mouse and wheelchair control are typical applications but a multiclass SMR BCI cannot control efficiently an artificial arm. An alternative is Motion trajectory prediction (MTP). This method is an ideal solution for controlling an artificial limb or a robot arm because MTP can reconstruct the imagined movement from neural signals. The following sections provide insight into this specific approach. II. CHRONOLOGICAL AND CATEGORICAL REVIEW Figure 1. Chronological distribution of MTP BCI studies. The first studies in non-invasive MTP were published about six years ago. While multiclass SMR BCIs counts hundreds of publications, for non-invasive MTP only a number of studies present results. Figure 1. shows statistics of published EEG and MEG papers in this topic. The search was conducted via Google Scholar database [2], the most hits were result of specific combination of the following key words: brain-computer interface, BCI, movement, motion, finger, trajectory, prediction, velocity, EEG, MEG, non-invasive. TABLE I. Categorical distribution of MTP BCI papers. Tasks EEG MEG Both Conference, Journal paper Drawing 2 4 2 Hand movement or centre out task 9 2 - Complex motion 4 - - Finger motion 4 1 1 Book, Book Chapters 2 2 3 TABLE I presents a categorical distribution of those papers which are published results of MTP BCIs. The most frequently investigated task is hand movement but some publications report on finger movements or more complex motions such as filling of glass of water. III. METHODOLOGY MTP regularly uses linear regression (LR) algorithm to achieve maximal correlation between trajectory of estimated and real limb movement. Although the most studies