# Signal detection in extracellular neural ensemble recordings using higher criticism

## TL;DR

This paper introduces a statistical method based on higher criticism for detecting neural signals in complex, noisy extracellular recordings without prior filtering, enhancing analysis of neural ensemble activity.

## Contribution

The study presents a novel, robust detection strategy using higher criticism that operates directly on raw neural data, improving signal detection in noisy, correlated recordings.

## Key findings

- Effective detection of small signals in noisy data
- No need for filtering or denoising prior to analysis
- Versatile across various experimental conditions

## Abstract

Information processing in the brain is conducted by a concerted action of multiple neural populations. Gaining insights in the organization and dynamics of such populations can best be studied with broadband intracranial recordings of so-called extracellular field potential, reflecting neuronal spiking as well as mesoscopic activities, such as waves, oscillations, intrinsic large deflections, and multiunit spiking activity. Such signals are critical for our understanding of how neuronal ensembles encode sensory information and how such information is integrated in the large networks underlying cognition. The aforementioned principles are now well accepted, yet the efficacy of extracting information out of the complex neural data, and their employment for improving our understanding of neural networks, critically depends on the mathematical processing steps ranging from simple detection of action potentials in noisy traces - to fitting advanced mathematical models to distinct patterns of the neural signal potentially underlying intra-processing of information, e.g. interneuronal interactions. Here, we present a robust strategy for detecting signals in broadband and noisy time series such as spikes, sharp waves and multi-unit activity data that is solely based on the intrinsic statistical distribution of the recorded data. By using so-called higher criticism - a second-level significance testing procedure comparing the fraction of observed significances to an expected fraction under the global null - we are able to detect small signals in correlated noisy time-series without prior filtering, denoising or data regression. Results demonstrate the efficiency and reliability of the method and versatility over a wide range of experimental conditions and suggest the appropriateness of higher criticism to characterize neuronal dynamics without prior manipulation of the data.