Automatic identification of Slow Biphasic Complexes in EEG: an effective tool to detect Encephalitis

Objective: the presence of slow biphasic complexes in EEG was found to be associated to different neural diseases, in particular inflammatory. This paper proposes a method to identify automatically the complexes and provides preliminary tests on paediatric patients with encephalitis. Approach: a prototype waveform was computed aligning and averaging complexes manually identified during years of investigation. A wide range of amplitudes and durations was noticed. The proposed automatic method is based on the cross-correlation of the tested EEG with scaled versions of the prototype waveform. The waves with high correlation with the prototype are identified as complexes if their amplitude and duration are reasonable and if they do not appear as pseudo-periodic oscillations. Main results: the algorithm was tested on a dataset of 128 EEGs from healthy controls and patients (for which the follow-up was also available). Experts assigned to each trace a severity score with 5 levels considering both electrophysiological and clinical manifestations. The number and amplitude of complexes are markers of encephalitis, they show statistically significant differences across EEGs with different severity scores and correlate with the condition of patients during the follow-up (median correlation about 80%). Significance: inflammatory brain pathologies can have serious sequelae. Quickness of diagnosis and targeted therapeutic interventions are essential to avoid or limit the damages. The proposed automated method is feasible for the fast, accurate and non-invasive diagnosis of encephalitis and the follow-up of patients.