The final phase of general anaesthesia, emergence, requires precise timing for extubation. In paediatric patients, suboptimal timing carries a heightened risk of severe complications. Currently, there is no formal characterisation of the time-frequency signatures in paediatric EEG that correlate with safe extubation readiness. This gap represents an open signal processing problem: the identification of discriminative, temporally-localised features within a non-stationary time series to classify neurological state, and eventually predicting the optimal time for extubation.
We propose a novel time-frequency analysis framework to identify discriminative precursors in EEG signals. We analyzed a dataset of 64 patients (ages 2-18) by computing the relative spectral power in key frequency bands over time. This revealed a consistent, temporally-ordered sequence of spectral peaks preceding successful extubation.
The intervals between these peak events were highly consistent across the cohort. Leveraging this discovered structure, we trained a model to predict the optimal extubation time based on expert-annotated protocols. Using a leave-one-out cross-validation scheme, our model achieved a mean absolute error of 81 seconds (std: 79 seconds) from the expert-defined optimal time.
This work establishes a robust, data-driven framework for interpreting EEG during emergence. We identify a consistent sequence of time-frequency events that serve as precursors to extubation readiness. Our predictive model demonstrates the viability of translating this knowledge into a clinical tool, with the potential to standardise and improve patient safety in paediatric anaesthesia.