In many diseases of the nervous system, the cortical command centers remain intact, but cannot connect to the peripheral organs, such as in spinal cord injury and ALS. The ability to create neuronal prosthetics that can link cortical activity to a computer or robotic device has long been the stuff of science fiction, although recent advances have used motor cortex recordings in primates to drive cursors or other devices from these cortical recordings. The feasibility of using such invasive recordings in humans was described recently (Neurosurgery 55:27–38, 2004). EEG recordings can also be used to run such brain-computer interfaces with considerable training (Clin Neurophysiol 113: 767–791, 2002). These attempts have been greatly advanced by the recent demonstration that electrocorticographic recordings can drive such brain-computer interfaces in humans (J Neural Eng 1:63–71, 2004). Using the recordings over motor or frontal language sites in patients with subdural electrodes placed for seizure localization, Leuthardt and colleagues identified reliable signals (figure) that were used to control a computer cursor to move up or down on a screen to reach a target. Patients learned with striking accuracy to manipulate the cursor, using the electrocortico-graphic signal even from imagined movement or speech. This method is particularly promising as patients appear to have much more rapid control than with EEG and does not require direct penetration of the cortex to obtain a task-specific signal. Going beyond proof of concept, the authors “closed the loop” using a readily applicable technology with only a few minutes of training. Such a technique will lead to exciting advances both in braincomputer interface and in the study of cortical function.Figure: Examples of electrode placement and ECoG signals. (a) Intra-operative placement of a 64-electrode subdural array. (b) Post-operative lateral skull radiograph showing grid placement. (c) Raw ECoG signals from patient C during control of cursor movement. Black and red traces are from one of the electrodes that controlled cursor movement and are examples for the patient resting and imagining saying the word “move,” respectively. (d) Spectra for the corresponding conditions for the final run of online performance. Imagery is associated with decrease in μ (8–12 Hz) and β (18–26 Hz) frequency bands. (Courtesy, J Neural Eng 1:63–71, 2004.)RICHARD G. ELLENBOGEN, M.D. PRINCIPAL