UC San Diego

Neural coding is primarily concerned with characterizing the relationship between stimulus and neuronal responses and is classified to stimuli encoding and brain response decoding. Although there are existing models for neural coding, most are not sophisticated enough to describe the relationship of population of neural responses to natural stimuli such as human speech or bird songs. In this study we propose utilizing composite receptive fields (CRF) as a new tool for neural coding. CRFs are quadratic receptive fields which are built from mutual information between stimuli and related brain responses. Here we create a pool of 3080 CRFs from a population of 154 cells recorded from the brain auditory region of European starling songbirds. Following, this pool of CRFs is used to build a spatial-temporal map for the population of cells along the brain coronal plane in respect to bird song stimuli. This map has revealed novel information about the relationship between neural responses and related stimuli such as: 1) Natural sound stimuli can be encoded by populations of neurons. 2) The number of cells needed to encode the stimuli can be quantified. 3) Stimuli encoding mechanisms of the brain appeared to be uniform and independent of cells’ topology and their locations. 5) From this map, connectivity between cells as well as their response plasticity to diverse stimuli were observed. 6) CRFs were used as intermediate tools to reconstruct stimuli and predict brain responses. These results have confirmed that quadratic receptive fields can be a novel candidate for population neural coding. Testing neural coding by CRFs was originally performed on cells recorded from the brain coronal depth plane. We expanded this coding method and evaluated CRFs mapping on cells recorded from two novel in-house fabricated electrodes: a surface electrode and a combination of surface and depth electrodes. The CRFs extracted from cells recorded by these electrodes can be employed to create a 2D and 3D spatial-temporal map which is useful to explore neural information distribution and their perception mechanisms from deep brain to cortical surface. Furthermore, the CRF neural coding method and the brain implants described in this study have potentials to be used in BCI prosthetics.