A digitized approach for amplitude-integrated electroencephalogram transformation towards a standardized procedure

The transformation procedure of deriving amplitude-integrated (a)EEG from raw EEG has been well described in a purely analog prototype, however the inherent specifications within the prototype are not established or disclosed. In this paper, we aim at providing an accessible and digitalized aEEG algorithm that is evaluated quantitatively and validated in clinical practice. The algorithms, especially the filter design and envelope detection methods, applied in the transformation procedure, are investigated. The effectiveness and feasibility of the filters namely, symmetric and asymmetric filters followed with four different envelope detection methods (low-pass filtering, squaring and low-pass filtering, Hilbert transform, and moving average) are evaluated on a clinical dataset collected at Children's Hospital of Fudan University which involves 30 infants. Compared to the outputs of the commercial available (a)EEG device NicoletOne, the Spearman rank correlations (SRs) of the upper/lower tracings of aEEG using the asymmetric filter and squaring and low-pass filtering-based envelope detection method can reach over 0.97. Meanwhile, the SRs of the upper and the lower margin amplitude of aEEG can achieve 0.98 and 0.97, respectively. Furthermore, the accuracy of the obtained aEEG tracings in identifying the aEEG background activities can achieve 100%. To our knowledge, this is the first work to present a digital procedure to transform the EEG into aEEG by assessing the impact of different filters and envelope detection methods. With the high performance of the proposed approach, this work can promote the standardization of aEEG transformation procedure and the exploration of the sophisticated automatic aEEG interpretation algorithms.

Automated summary

This paper investigates the digital transformation procedure of deriving amplitude-integrated (a)EEG from raw EEG and evaluates the impact of different filters and envelope detection methods. The results demonstrate that the proposed approach achieved excellent performance in a clinical dataset, promoting the standardization of aEEG transformation procedure and exploration of automatic interpretation algorithms.