Automatic methods for cortex-wide layer identification of electrophysiological signals reveals a cortical motif for the expression of neuronal rhythms

A 2019 conference paper presents two automatic methods for identifying cortical layers from electrophysiological recordings. It also reveals a layer-specific pattern in neuronal rhythms across frontal, parietal, and visual cortex.

The paper by André M. Bastos, Omar Costilla-Reyes, and Earl K. Miller was presented as a poster at the 2019 Conference on Cognitive Computational Neuroscience (CCN), held in Berlin on September 13–16, 2019. It comes from Dr. Costilla Reyes’s postdoctoral work in MIT Brain and Cognitive Sciences, where he applied machine learning to cortical dynamics in the Miller laboratory.

The cerebral cortex has six anatomical layers, and how they support cognition is a central question in neuroscience. Answering it depends on knowing which layer each recording contact sits in, and that must be determined from data recorded in a living animal. Established approaches draw on assumptions about the signal and on expert judgment, and anatomical confirmation after the experiment is available for only a few sites, so an automatic, quantitative method opens the way to comparing many more sites and areas on the same footing. The authors recorded local field potentials (LFPs), the population-level electrical activity measured at each electrode, in monkeys with multi-electrode probes spanning all layers of frontal, parietal, and visual cortex.

The authors introduce two fully automatic, quantitative methods for layer identification. The first converts the relative power across electrodes into a two-dimensional image and maximizes the similarity of those images across probes. The second uses ensemble machine learning, which combines several models, to classify LFP data into layer labels. Both methods detect consistent patterns. With layers assigned, the image-similarity method reveals a motif across cortex: slow delta/theta rhythms and fast gamma rhythms are strongest in the superficial layers, while alpha/beta rhythms are strongest in the deep layers. This layer-specific organization of rhythms across frontal, parietal, and visual cortex gives neuroscientists a concrete reference for relating neural oscillations to the layered circuitry that produces them.

The paper is paired with a companion study at the same conference, led by Costilla-Reyes, “Pattern recognition of deep and superficial layers of the macaque brain using large-scale local field potentials,” which focuses on the prefrontal cortex and visual area V4. Both apply machine learning to neural recordings, where, as in his doctoral work on floor-sensor data, the task is to find reliable patterns in recordings that change over time.