Author(s): Yasuhiko Nakanishi 1, Takufumi Yanagisawa 2,3,4, Duk Shin 1,*, Ryohei Fukuma 3, Chao Chen 1, Hiroyuki Kambara 1, Natsue Yoshimura 1, Masayuki Hirata 2, Toshiki Yoshimine 2, Yasuharu Koike 1
A number of prominent brain-machine interface studies have arisen, in which electroencephalography (EEG), magnetoencephalography (MEG), electrocorticography (ECoG), and intracortical microelectrode have been applied to neuroprosthesis control, neurorehabilitation and novel communication tools for paralyzed or "locked-in" patients suffering from neuromuscular disorders. Since EEG and MEG are non-invasive and have high temporal resolution, they have been used in various paradigms, such as online control of a computer cursor -, direction inference of hand movements -, operation of a spelling device , and neurofeedback for rehabilitation -. Although a large proportion of these non-invasive methods succeeded in classification of movement direction or intention, prediction of time-varying trajectories is likely difficult due to insufficient spatial resolution and low signal-to-noise ratio in such methods.
Signal recording with intracortical microelectrodes is a powerful tool to realize precise trajectory prediction or accurate device control. Using motor cortical signals in animals, studies have shown successful prediction of hand trajectories - and grasp types and velocity , control of a computer cursor  or a robot arm -, and controlled stimulation to a paralyzed arm . These techniques have also been applied in humans to control a cursor  and a virtual keyboard and virtual hand . However, though intracortical electrodes can provide rich information for BMI control, they face limitations such as signal degradation due to glial scarring and potential displacement from the recording site .
Conversely, ECoG is less invasive than microelectrodes and can offer higher spatial resolutions than EEG and MEG. Researchers have been applying ECoG in humans for several years now and in numerous applications. The classification of hand movement directions or grasp types -, one-, two-, or three-dimensional cursor control , -, and prediction of finger flexion  are just some examples of ECoG applications in human patients. Studies concerning the prediction of three-dimensional (3D) trajectory or muscle activities from primate ECoG have shown outstanding results -. Investigations on the prediction of 3D arm trajectory using ECoG in humans, however, are lacking, despite the potential to provide significant improvement in neuroprosthesis and neurorehabilitation technology. The inadequate quality of ECoG signals recorded from patients is one potential obstacle in predicting 3D trajectories. Specifically, (1) paralyzed or elderly patients may find it difficult to perform a long series of repeating trials and stably replicate the same motion for each trial, (2) ECoG signals in patients can include pathological activity, depending on the condition, and (3) the electrode sites on the cortex and the recording lengths can differ, depending on the treatment.
The aim of this study was to predict 3D arm trajectories from ECoG time series in human patients as a basis for a neuroprosthesis. Patients diagnosed with thalamic hemorrhage, ruptured spinal dural arteriovenous fistula (dAVF) and intractable epilepsy executed rotating tasks with three objects on a table. We simultaneously recorded arm trajectories and...