Consciousness encompasses a range of cognitive and physiological states, including wakefulness, sleep, and altered states, as well as pathological conditions such as disorders of consciousness (DoC). In recent decades, quantitative EEG analysis has been used in adults to distinguish between wakefulness and sleep states and to study brain activity patterns associated with different DoC. However, such applications in pediatric subjects remain underexplored due to significant age-related changes in brain activity and limited data availability.The aim of this work was to develop a Support Vector Machine (SVM) classifier capable of distinguishing between wakefulness and sleep states in healthy pediatric subjects using a combination of linear and nonlinear EEG features. In addition, the feasibility of extending this classifier to pediatric patients with DoC was preliminarily evaluated.In 89 healthy pediatric subjects, 4 linear features (power spectral values in the canonical frequency bands: delta, theta, alpha, and beta) and 2 nonlinear features (spectral exponent and Higuchi fractal dimension) were computed from EEG data. SVM classifiers with different kernels (linear, polynomial and radial basis function) were trained and tested. The best classifier achieved an accuracy of 95% in distinguishing wakefulness from sleep states. Ignoring nonlinear features would reduce the accuracy by approximately 4%. In the pediatric DoC cohort, the SVM classifier showed that patients transitioning to a minimally conscious state had a gradual increase in the percentage of wakefulness-labeled EEG sweeps, while those who remained unresponsive had consistently low percentages.Clinical relevance- The proposed quantitative EEG analysis could potentially anticipate/assist the clinician's diagnosis of the change in the state of consciousness (from unresponsive wakefulness syndrome to minimally conscious state) in pediatric DoC patients.
State of consciousness classification in pediatric subjects using SVM with both linear and nonlinear EEG features
Colussi F.;Favaro J.;Ancona C.;Masiero S.;Sparacino G.;Toldo I.;Sartori S.;Rubega M.
2025
Abstract
Consciousness encompasses a range of cognitive and physiological states, including wakefulness, sleep, and altered states, as well as pathological conditions such as disorders of consciousness (DoC). In recent decades, quantitative EEG analysis has been used in adults to distinguish between wakefulness and sleep states and to study brain activity patterns associated with different DoC. However, such applications in pediatric subjects remain underexplored due to significant age-related changes in brain activity and limited data availability.The aim of this work was to develop a Support Vector Machine (SVM) classifier capable of distinguishing between wakefulness and sleep states in healthy pediatric subjects using a combination of linear and nonlinear EEG features. In addition, the feasibility of extending this classifier to pediatric patients with DoC was preliminarily evaluated.In 89 healthy pediatric subjects, 4 linear features (power spectral values in the canonical frequency bands: delta, theta, alpha, and beta) and 2 nonlinear features (spectral exponent and Higuchi fractal dimension) were computed from EEG data. SVM classifiers with different kernels (linear, polynomial and radial basis function) were trained and tested. The best classifier achieved an accuracy of 95% in distinguishing wakefulness from sleep states. Ignoring nonlinear features would reduce the accuracy by approximately 4%. In the pediatric DoC cohort, the SVM classifier showed that patients transitioning to a minimally conscious state had a gradual increase in the percentage of wakefulness-labeled EEG sweeps, while those who remained unresponsive had consistently low percentages.Clinical relevance- The proposed quantitative EEG analysis could potentially anticipate/assist the clinician's diagnosis of the change in the state of consciousness (from unresponsive wakefulness syndrome to minimally conscious state) in pediatric DoC patients.
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
