Optimal feedback control of a neuromotor interface

No effective treatments currently exist for the permanent disabilities caused by damage to the motor system by spinal cord injuries, among which upper-limb disorders can be especially devastating. The emerging field of neuromotor prosthetics aims to restore function to paralysed patients by using signals derived directly from the brain to control computers, wheelchairs and other external devices. During normal movement, the discharge of motor cortex neurons is modulated to provide the appropriate drive to muscles. A Brain-Controlled Interface decodes the activity of these neurons to determine control signals that move assistive devices. However, performance of these interfaces must be improved before this approach can yield useful clinical applications. In particular, variability in the activity of individual neurons causes control errors that are only corrected slowly through visual feedback. In recent years, a sophisticated body of theoretical work has described strategies that the brain uses to minimise the consequences of this variability and thus optimise the accuracy of natural movements. This has led to a new understanding of the importance of ?feedback loops? that relay output signals back into the brain?s motor control system. This project aims to apply these principles to design improved interfaces that exploit similar mechanisms for minimising the influence of variability on relevant output signals.

A combined experimental approach will use non-invasive techniques to study operation of neuromotor interfaces by normal human subjects, and related tasks performed by macaque monkeys. The monkey experiments will use electrodes that are implanted in motor areas of the brain to control a computer cursor. Implant surgeries will be performed in sterile conditions under general anaesthesia as is routinely used for human surgery and monkeys will be given post-operative analgesics. Electrodes implanted in the brain do not cause pain, as evidenced by the thousands of human patients who have benefited from Deep Brain Stimulation electrodes for Parkinson?s Disease. The use of permanently implanted electrodes minimises the amount of restraint that is required for each animal to participate in these tests. Some sections of the experiment will be performed during natural movements in the animal?s home environment using a miniature wearable electronic device.

Experimental results will be applied to the development of new methods to decode brain activity, which will then be compared against existing approaches. If successful, improved Brain-Controlled Interface techniques, in combination with new electrode technologies being developed in related projects, could restore function to a large population of patients with spinal cord injuries.

Brain-Controlled Interfaces (BCIs) use signals derived directly from the brain to control external devices and restore function to paralysed patients. The most successful BCI implementations use implanted electrodes to record action potentials from large numbers of neurons, nevertheless the poor accuracy of control is a major obstacle which must be overcome before these proof-of-principle demonstrations can be translated into viable clinical applications. The traditional BCI approach is to decode movements based on the assumption that trajectories are represented in the firing rates of motor cortical neurons and much research effort is devoted to developing decoding algorithms of increasing sophistication. However, the trajectory-coding assumption has recently been questioned by a new theoretical framework for optimal motor control that points instead to the existence of flexible feedback controllers acting on internal state estimates to minimise noise in task-relevant dimensions. It is my belief that the poor performance of current BCI approaches arises from this fundamental misconception about how the brain controls normal movements, and that an understanding of how optimal feedback control is implemented by the motor system will lead to a a new strategy for developing successful BCIs.

I propose a novel experimental approach combining BCI tasks performed by monkeys and related myoelectric-controlled interface tasks performed by human subjects. Analysis of movement variability and the responses to unexpected perturbations under different goal conditions will reveal how feedback control is implemented at the neural level within the motor system. Results will be interpreted in relation to constraints on optimality imposed by the underlying functional architecture of the motor system. Output fields of individual neurons will be determined by recording cell and muscle activity during unrestrained behaviour with an electronic implant. Local synaptic connectivity between neurons will be assessed by the degree of synchronisation in their spiking activity. I hypothesize these features will help predict combinations of control signals that are optimal for minimising errors along different task dimensions. This information will be used to construct improved BCI decoders for exploiting fast error correction mechanisms mediated by intrinsic motor cortical circuitry. By comparing their performance with decoders based on trajectory-coding assumptions, I will test the hypothesis that an optimal feedback approach can lead to more accurate BCI control than existing strategies. This will provide a next generation of BCIs that will have practical value to spinal cord injured patients.