Minasyan G.R., Chatten J.B., Harner R.N. "Detection of epileptiform activity in unresponsive patients using ANN" Neural Networks, 2009. IJCNN 2009. International Joint Conference on Neural Networks, Atlanta, 2009. pp. 2117 – 2124

Abstract:

Advanced EEG analysis tools are needed for use on unconscious or comatose patients in hospital ICU and emergency departments. Their purpose is to monitor brain state and provide warnings, by analysis of EEG, that the patient's brain is in a dangerous state of ongoing epileptic seizure activity. In this study, our objective is limited to detection of Non-Convulsive Status Epilepticus (NCSE). The proposed NCSE detection techniques start with detection and classification of 1-sec EEG features (basic rhythms, paroxysmal events, sleep events, and artifacts). On the next step, 1-sec EEG features are accumulated over a period of 1-minute, and 10-element EEG State Vector (ESV) is computed. ESV vectors are passed to a multi-layer perceptron that classifies 1-min EEG epochs as: NCSE, SLOW, FAST, BURST-SUPPRESSION, or ARTIFACT. One minute epochs from 9 training and 12 test records were expertly scored into one of the 5 EEG states listed above. The ANN correctly classified 71% epochs of NCSE and 99% epochs of non-NCSE. These findings suggest the potential for accurate detection of NCSE in the ICU and emergency departments.