Time variant spindle dynamics using statistical signal analysis

OBJECTIVES: A spindle reflects neurophysiological mechanisms of interactions between
inhibitory cells in the thalamic reticular nucleus (RE) and bursting thalamocortical (TC) relay neurons. A gradual cell recruitment in RE-TC-RE loops is linked to the waxing of spindles. The cause of spindle waning is less clear, but a depolarizing action by the thalamic IH current may be involved. The influence of these neurophysiological mechanisms on spindle morphology inthe scalp EEG is still to be clarified. The transient nature of spindling and the lack of knowledge of the precise mechanism behind spindles require that the analysis of scalp EEG spindles should be performed with a minimum of assumptions and a high temporal resolution. This paper presents a statistical approach for the analysis of scalp EEG to determine the dynamics of spindle power.

METHODS: The algorithm, developed in the Galaxy sleep system, has a very high timeresolution and deploys a minimum of assumptions. EEG is band-pass filtered (11.0-16.0 Hz) using a FIR-filter. The standard deviation of the signal is computed with a moving window of 0.2 second. The resulting power has a time resolution of the sample rate of the signal. Waxing and waning characteristics of a spindle are represented by the time-variant characteristics of the power. A pattern recognition algorithm detects all waxing/waning couplets. Various characteristics like peak power, total intensity, duration, symmetry, polarization/depolarization speed etc. are calculated for each waxing and waning couplet. In addition power dynamics in different spindle bands are calculated.

RESULTS: Sleep EEG data of previously published studies on sleep and memory were reanalyzed to compare the statistics of the waxing/waning dynamics with spindles detected heuristically by visual criteria. The statistical analysis without prior assumptions provided more details of a spindle that could not be detected by visual heuristics (even with automated detection). The power dynamics showed that the power within the slow band (10-13 Hz) increased during the waxing part whereas the power in the fast band (13-16Hz) increased during the waning part of the spindle.

CONCLUSION: Detecting all waxing/waning patterns without prior criteria like amplitude, duration etc. is useful. Subsequent analysis of waxing/waning parameters reveals more details missed by analyzing only heuristically detected spindles.