Sleep spindle detection through amplitude–frequency normal modelling

Abstract

h i g h l i g h t sAn automated model-based spindle detection algorithm is proposed. It models the amplitude-frequency spindle distribution with a bivariate normal distribution. It automatically adapts to each individual subject and derivation. It was tested in seven healthy children and six adult patients suffering from different pathologies, and performs similarly or better than sleep experts. Normal modelling enhances spindle detection quality compared to fixed amplitude and frequency thresholds. t r a c tManual scoring of sleep spindles can be very time-consuming, and achieving accurate manual scoring on a long-term recording requires high and sustained levels of vigilance, which makes it a highly demanding task with the associated risk of decreased diagnosis accuracy. Although automatic spindle detection would be attractive, most available algorithms are sensitive to variations in spindle amplitude and frequency that occur between both subjects and derivations, reducing their effectiveness.We propose here an algorithm that models the amplitude-frequency spindle distribution with a bivariate normal distribution (one normal model per derivation). Subsequently, spindles are detected when their amplitude-frequency characteristics are included within a given tolerance interval of the corresponding model. As a consequence, spindle detection is not directly based on amplitude and frequency thresholds, but instead on a spindle distribution model that is automatically adapted to each individual subject and derivation.The algorithm was first assessed against the scoring of one sleep scoring expert on EEG samples from seven healthy children. Afterward, a second study compared performance of two additional experts versus the algorithm on a dataset of six EEG samples from adult patients suffering from different pathologies, to submit the method to more challenging and clinically realistic conditions. Smaller and shorter spindles were more difficult to evaluate, as false positives and false negatives showed lower amplitude and smaller length than true positives. In both studies, normal modelling enhanced performance compared to fixed amplitude and frequency thresholds. Normal modelling is therefore attractive, as it enhances spindle detection quality.

Automatic detection of sleep spindles using Teager energy and spectral edge frequency

“…We believe that together with using the metrics proposed in [10] it is important to report the fraction of spindles detected in each stage of sleep as well to evaluate the performance of any spindle detection algorithm. 1 190 10 3 135 42 0 2 101 0 0 77 24 0 3 91 5 8 75 3 0 4 104 32 8 54 10 0 5 164 4 0 110 50 0 6 146 0 0 97 49 0 Total 796 51 19 548 178 0 Comparison of results against [10] and [13] (evaluated using the same dataset) shows superior sensitivity and similar specificty values. These are shown in Table III. V. CONCLUSION Identification of sleep spindles is an integral part of sleep staging.…”

“…It is a transient waveform with waxing-waning morphology and exhibits strong presence in stage 2 of NREM sleep (N2), although it may be present in N3 stage with a lower frequency of occurence. According to the American Academy of Sleep Medicine, a sleep spindle is defined as "a train of distinct waves with frequency [11][12][13][14][15][16] with a duration ≥ 0.5 seconds" [1]. An example of typical sleep spindles in stage 2 of NREM sleep is shown in Fig.…”

“…[11] used power thresholding and Hilbert-Huang transform to detect spindles in healthy children, using two EEG channels, and reported sensitivitiy and specificity of 88.2% and 89.7% respectively. [12] reported 96.2% sensitivity using maximum frequency in spindle range, Teager energy and harmonic decomposition of signals while [13] used amplitude-frequency normal modelling and reported 78.5% and 75.1% sensitivity for children and adult spindle detection respectively.…”

EEG Beta Wave Trains Are Not the Second Harmonic of Mu Wave Trains in Parkinson’s Disease Patients

“…The idea of our method of analyzing EEG is in that we consider EEG signal as a composition of so-called wave trains. In contract to papers devoted to detecting wave trains of one or two specific types, such as alpha spindles [1] and sleep spindles [2,3,4,5,6,7], we analyze any kind of wave trains in a wide frequency area. The developed method differs from analogous method for detailed analysis of time-frequency dynamics of EEG [8] in that the statistical analysis of samples of wave trains and a new method for visualizing the results of the analysis are proposed.…”

Evaluating the use of line length for automatic sleep spindle detection

“…A data-driven probabilistic approach was used by Babadi et al [11] that showed high detection accuracy. For detection of spindles in children [12] presented a method using Hilbert-Huang transform while [13] used amplitude-frequency normal modelling to detect spindles in both children and adults. In another method, Schönwald et al [14] evaluated the use of matching pursuit (MP) for spindle detection and achieved good results.…”