The Oscillatory ReConstruction Algorithm (ORCA) adaptively identifies frequency bands to improve spectral decomposition in human and rodent neural recordings

Neural oscillations are routinely analyzed using methods that measure activity in canonical frequency bands (e.g. alpha, 8-12 Hz), though the frequency of neural signals is not fixed and varies within and across individuals based on numerous factors including neuroanatomy, behavioral demands, and species. Further, band-limited activity is an often assumed, typically unmeasured model of neural activity and band definitions vary considerably across studies. These factors together mask individual differences and can lead to noisy spectral estimates and interpretational problems when linking electrophysiology to behavior. We developed the Oscillatory ReConstruction Algorithm (ORCA), an unsupervised method to measure the spectral characteristics of neural signals in adaptively identified bands which incorporates two new methods for frequency band identification. ORCA uses the instantaneous power, phase, and frequency of activity in each band to reconstruct the signal and directly quantify spectral decomposition performance using each of four different models. To reduce researcher bias, ORCA provides spectral estimates derived from the best model and requires minimal hyperparameterization. Analyzing human scalp EEG data during eyes open and eyes-closed resting conditions, we first identify variability in the frequency content of neural signals across subjects and electrodes. We demonstrate that ORCA significantly improves spectral decomposition compared to conventional methods and captures the well-known increase in low-frequency activity during eyes closure in electrode- and subject-specific frequency bands. We further illustrate the utility of our method in rodent CA1 recordings. ORCA is a novel analytic tool that will allow researchers to investigate how non-stationary neural oscillations vary across behaviors, brain regions, individuals, and species.