A hybrid biological/in silico neural network based brain-machine interface

Abstract

Brain-machine interfaces (BMIs) aim to improve the lives of individuals with neurological disease or injury, by opening new information transfer channels between brain tissue and prosthetic actuators. In a majority of the BMI work, the data acquired from the motor cortex neurons are decoded into user's intended prosthetic actions by some "optimized" input-output mathematical model. Although this approach is quite sound, the information processing principles used are fundamentally di erent from those of natural neural circuits. In this thesis, we propose a novel, neurally-inspired design approach; the BMI controller consists of spiking model neurons and receives simulated synaptic inputs from extracellularly recorded neurons. The controller therefore forms a hybrid biological/in silico neural network with the neuronal circuits of the user's brain. In order to ful ll the challenging real-time requirements of the present design approach, we rst developed the Bioinspired Model Development Environment (BMDE). The BMDE, implemented on a hard real-time system, signi cantly facilitates BMI model development processes with powerful online data visualization tools while satisfying the strict timing constraints of the proposed design approach. Using the BMDE, we realized a novel, adaptive BMI controller which consists of in silico striatal medium spiny neurons, each receiving simulated synaptic inputs from extracellularly recorded motor cortex neurons. By implementing a reward-modulated spike timing-dependent plasticity rule and a winner-takes-all mechanism, the BMI controller, based on real-time closed-loop simulations, achieves perfect target reach accuracy for a two target reaching task in one dimensional space. Using this design approach and the BMDE, new generation BMI controllers that better mimic brain circuits can be developed. Moreover, by investigating the interactions between biological and in silico neural networks during neuroprosthetic control tryouts new neuroscienti c insights concerning motor control and learning can be obtained.