Signal Analysis for Data-Driven Neurocritical Monitoring

Abstract

Intracranial hypertension is a major driver of secondary injury after traumatic brain injury, motivating continuous intracranial pressure (ICP) monitoring and interpretable risk estimation. This paper analyses ICP recordings from the CHARIS database (13 patients, 50 Hz) using wavelet denoising, short-time spectral measures, and empirical distribution analysis. After Haar wavelet denoising, the signal is segmented with overlapping Hann windows, and compact short-time descriptors are computed, including entropy, mean frequency, and band power. A probability-based risk indicator is derived from the empirical distribution of ICP values relative to clinically used thresholds. In this cohort, shorttime band power shows a strong association with the probability of elevated pressure, making it a promising lightweight candidate feature for rapid screening and monitoring. The proposed ICP-only workflow is interpretable, computationally simple, and can complement autoregulation indices when multisignal channels are unavailable or unreliable.