The individual adjustment method of sleep spindle analysis: Methodological improvements and roots in the fingerprint paradigm

https://doi.org/10.1016/j.jneumeth.2008.11.006Get rights and content

Abstract

Evidence supports the robustness and stability of individual differences in non-rapid eye movement (NREM) sleep electroencephalogram (EEG) spectra with a special emphasis on the 9–16 Hz range corresponding to sleep spindle activity. These differences cast doubt on the universal validity of sleep spindle analysis methods based on strict amplitude and frequency criteria or a set of templates of natural spindles. We aim to improve sleep spindle analysis by the individual adjustments of frequency and amplitude criteria, the use of a minimum set of a priori knowledge, and by clear dissections of slow- and fast sleep spindles as well as to transcend the concept of visual inspection as being the ultimate test of the method's validity. We defined spindles as those segments of the NREM sleep EEG which contribute to the two peak regions within the 9–16 Hz EEG spectra. These segments behaved as slow- and fast sleep spindles in terms of topography and sleep cycle effects, while age correlated negatively with the occurrence of fast type events only. Automatic detections covered 92.9% of visual spindle detections (A&VD). More than half of the automatic detections (58.41%) were exclusively automatic detections (EADs). The spectra of EAD correlated significantly and positively with the spectra of A&VD as well as with the average (AVG) spectra. However, both EAD and A&VD had higher individual-specific spindle spectra than AVG had. Results suggest that the individual adjustment method (IAM) detects EEG segments possessing the individual-specific spindle spectra with higher sensitivity than visual scoring does.

Introduction

Sleep spindles are usually defined as groups of 12–15 Hz sinusoidal electroencephalogram (EEG) waves occurring mainly during stage 2 non-rapid eye movement (NREM) sleep but occasionally appearing in stages 3 and 4 sleep as well (De Gennaro and Ferrara, 2003). There is growing neurophysiological knowledge regarding the nature of neural mechanisms generating sleep spindles, suggesting the role of hyperpolarization-rebound sequences in thalamocortical relay cells triggered, grouped and synchronized by cortico-cortical networks (Steriade, 2003). Moreover, there is a high-degree of interindividual difference in sleep spindle features accompanied by a remarkable intraindividual (night-to-night) stability (Silverstein and Levy, 1976, Gaillard and Blois, 1981, Werth et al., 1997, Tan et al., 2000, De Gennaro et al., 2005). The NREM sleep EEG power spectra at the 8 to 16 Hz frequency covering alpha and spindle activities is characterized by an individual profile, which is stable over time, resistant to experimental perturbations and strongly influenced by genetic factors (De Gennaro et al., 2008). A distinction of slower and faster sleep spindles based on frequency and topography was given by Gibbs and Gibbs (1950) and confirmed by studies using modern techniques like low-resolution electromagnetic tomography (Anderer et al., 2001), magnetoencephalography (Urakami, 2008), electrocorticography (Nakamura et al., 2003) and functional magnetic resonance imaging (Schabus et al., 2007). The frequency of slow spindles mostly corresponds to the alpha frequency range and detailed EEG studies suggest the possibility that slow spindles are anterior peaks of alpha activity during NREM sleep (De Gennaro and Ferrara, 2003). Given these evidences it is quite surprising that most of the methods of sleep spindle analysis are ultimately still based on, validated by, and tied to visual detection of spindles performed by experienced human scorers. By accepting visual scoring as the final test of automatic sleep spindle analysis one is implicitly assuming that human pattern recognition capacities are still superior to computer-based methods of spindle detection or that modern neurophysiological knowledge did not influence the definition of sleep spindles. Since we do not agree with these assumptions, we developed an improved method of sleep spindle analysis, which is a modified version of our previously published one (Bódizs et al., 2005). Our main starting points in the development of our method were the followings:

  • 1.

    Sleep spindles are groups of waves in the 9–16 Hz range lasting at least 0.5 s and appearing in NREM sleep EEG records (De Gennaro and Ferrara, 2003, De Gennaro et al., 2005).

  • 2.

    The exact spectral content of sleep spindles is individual-specific. In humans it emerges as an individually stable trait-like feature characterized most often by two distinct spectral peaks with different topography and sleep cycle dynamics (Werth et al., 1997, De Gennaro et al., 2005, Buckelmüller et al., 2006). There is an exceptionally low channel-by-channel, cycle-by-cycle and night-by-night variation in the individual-specific frequency bands delimiting the slow- and fast sleep spindles (Werth et al., 1997).

Based on these statements we defined sleep spindles as those segments of the NREM sleep EEG which last at least 0.5 s and contribute to one of the two individual-specific spectral peaks observed in the 9–16 Hz range. By accepting this definition our aim was to:

  • 1.

    define the individual-specific spectral peaks in the 9–16 Hz range;

  • 2.

    calculate the individual- and derivation-specific amplitude criteria for the two spindle types separately;

  • 3.

    perform an adequate bandpass-filtering and to obtain the precise envelope curves in the individual-specific frequency ranges;

  • 4.

    obtain not only the density, but also the mean amplitude and mean duration of the two sleep spindle types for each subject and each EEG derivation;

  • 5.

    validate our method by testing previously established relationships:

    • (a)

      topographical difference between slow- and fast sleep spindles (anterior versus posterior predominance of slow- and fast sleep spindles, respectively);

    • (b)

      declining trend of slow spindling and increasing trend of fast spindling over consecutive sleep cycles;

    • (c)

      drop of sleep spindles with ageing and its interaction with sleep cycle effects;

  • 6.

    compare the output of our automatic sleep spindle detection technique with the visual procedure performed by a trained expert.

Our previously published method (Bódizs et al., 2005) was modified in accordance with our sleep spindle definition and with new developments in the field. As regarding methodological improvements we introduced the zero-padding of EEG segments prior to fast Fourier transformation (FFT), as this was shown to be a reliable method of estimating the dominant spindle frequencies (Huupponen et al., 2006). Moreover, bandpass-filtering was based on Gauss-filters instead of Butterworth ones. And lastly we did not introduce any ad hoc correction in the amplitude criteria, but calculated a precise envelope of the filtered signals.

We hypothesized that the individual adjustment method (IAM) of sleep spindle analysis, which is an operationalization and application of our sleep spindle definition on human sleep EEG records:

  • 1.

    results in spindle detections, which behave as slow- and fast sleep spindles in terms of topography, sleep cycle effects and age;

  • 2.

    results in sleep spindles with an individual-specific spectral content paralleling the individual fingerprints of sleep EEG spectra, but being also articular in terms of the individual-specific spectral peaks;

  • 3.

    is much more sensitive than visual detection performed by human experts (detect more spindles than humans), but the extra-spindles detected by the IAM:

    • (a)

      share the individual-specific spectral content unlike those visually detected spindles which are not covered by IAM;

    • (b)

      are characterized by an amplitude spectrum exceeding the average amplitude spectrum of the whole EEG segment (spindle + no-spindle) in terms of individual-specific spindle-peaks unlike those visually detected spindles which are not covered by IAM.

Section snippets

Subjects and procedures

Polysomnographic recordings of 46 adult subjects (29 men and 17 women; mean age: 32.46 years; age range: 17–55) participating in two different research projects were used as a database for the current research. The research protocols were approved by the Ethical Committee of the Institute of Behavioural Sciences Semmelweis University Budapest. All subjects signed informed consent for the participation in the studies. According to a semi-structured interview subjects were healthy and free of any

The frequency of slow- and fast sleep spindles

Based on the procedure described above (Section 2.2) we determined the individualized frequency limits for slow- and fast sleep spindles in NREM periods 1–4. Results are presented in Table 1. In general slow spindles were slower than 12 Hz, while fast spindles were faster than 12.5 Hz. However, this cannot be considered a rule, as there were subjects with slow spindle frequencies between 12.15 and 13.26 Hz or subjects with fast spindle frequencies between 11.76 and 12.84 Hz for example. In the

Discussion

The accurate analysis of sleep spindling has remarkable practical and theoretical importance. This is underlined by the reported cognitive (Bódizs et al., 2005, Bódizs et al., 2008, Clemens et al., 2005, Clemens et al., 2006, Schabus et al., 2007, Schabus et al., 2008, Fogel et al., 2007) and clinical neuropsychiatric (Donnet et al., 1992, Ferrarelli et al., 2007, Petit et al., 2004, Limoges et al., 2005, Gürses et al., 2005) correlates of different sleep spindle measures. Given the high

Acknowledgements

This work was supported by the National Office for Research and Technology (NKFP-1B/020/04) and the National Research Fund (OTKA TS-049785 and OTKA-48927). The first author is supported by the János Bolyai Research Fellowship of the Hungarian Academy of Sciences.

References (30)

Cited by (102)

  • Analysis of slow and fast sleep spindle properties in Parkinson's disease – A comparative EEG study

    2022, International Journal of Psychophysiology
    Citation Excerpt :

    The artefactual segments were manually annotated on a 4 second basis. Automated SS analysis was performed on artefact-free segments of NREM sleep with the Individual Adjustment Method (IAM) (Bódizs et al., 2009; Ujma et al., 2015) using Fercio's EEG Plus Polysomnography program as well. SS density (number of SS/min), duration (s), middle amplitude (μV), and oscillatory frequency (Hz) of slow and fast spindles were analysed according to each EEG leads, than separately over the Fp1, Fp2, F3, F4, Fz, F7, and F8 leads for slow-, and over the C3, C4, Cz, P3, P4 and Pz leads for fast spindles using an average-reference montage.

View all citing articles on Scopus
View full text