A high-performing neural decoder that reconstructs spiking data from local field potentials to analyze neural recordings for developing neuro-prosthetic devices.
Motor impairment is common in neurological conditions such as cerebral palsy, Parkinson's, stroke, spinal cord injury, multiple sclerosis, and other neurodegenerative disorders. In the United States, more than 20 million, or 8.5 percent of the population, have some form of disability that limits motor function. Moreover, two million people suffer from severe motor impairment and complete paralysis. These patients cannot perform everyday tasks (eating, walking, talking, etc.), often leading to obesity and cardiovascular disease due to sedentary lifestyles. Several research groups are now developing brain-computer interfaces (BCIs) that evaluate brain signaling with the hopes of helping those with severe paralysis communicate and perform tasks by thinking using neuroprosthetics. However, limited products are on the market due to the inherent complexities of acquiring, analyzing, and utilizing data obtained data.
Emory researchers are developing novel latent variable models (LVMs) that use field potentials as inputs and are trained to reconstruct spiking data at the output. LVMs are essential tools in understanding the patterns associated with neural activity. To date, LVMs have primarily been trained on spiking activity, as this tends to yield the highest accuracy predictions of behavior. However, this process can be unreliable as implanted neural recording hardware may experience signal degradation due to immune responses of surrounding tissue or unstable placement to nearby neurons. The inventors’ method combines the stability benefits of local field potential (LFP) with the high-accuracy decoding provided by the spiking data. These models can be used to understand neural signals and improve neuro-prosthetic devices.