Nitish V. Thakor

Host: Dr. Terrence J. Sejnowski

Brain computer/machine interface (BCI/BMI) technology is presently receiving popular attention. Brain interface is created using noninvasive scalp electrodes and then BCI/BMI demonstrated by decoding EEG or event related potential signals. Alternately, neural interface is also created using microelectrode arrays implanted in the cortex for BCI or brain control of robotic mechanisms. First, I will present noninvasive BMI to control a prosthetic hand. I will present the BMI interface with and without feedback and compare visual and haptic feedback. Next, I will describe our recent work directed towards dexterous control of a multi-fingered prosthetic hand. In a collaborative experimental study, we obtained recordings from the M1 motor cortex of rhesus monkey during movements of individual fingers. We used a two step decoding process, a gating classifier to identify finger motion and a maximum likelihood classifier to decode individual finger movements. Both trained and blind decoding were evaluated. We decoded single finger movement with fewer than 25 neurons (99% accuracy) and multi-finger movements with 30 neurons (95% accuracy). Further, in a more challenging step, we decoded six 2-finger movement sets. The decoding accuracy was 90% with 100 neurons.
Blind decoding of single and multiple fingers points to possible neural control of dexterous multi-fingered prosthesis. I will present asynchronous decoding and actuation of fingers in a multi-fingered prosthetic hand. This research raises questions for discussion on how dexterous movement of ever increasing complexity might be encoded in the brain and what novel recording and decoding strategies may be needed to achieve further complexities in dexterous manipulation.