Low-complexity image processing for real-time detection of neonatal clonic seizures

In this paper, we consider a novel low-complexity image processing-based approach to the detection of neonatal clonic seizures. Our approach is based on the extraction, from a video recording of a newborn, of an average luminosity signal representative of the body movements. Since clonic seizures are characterized by periodic movements of parts of the body (e.g., the limbs), by evaluating the periodicity of the extracted average luminosity signal it is possible to estimate the presence of a seizure. The periodicity is detected, through a hybrid autocorrelation-Yin estimation technique, on a per-window basis, where a window is constituted by a sequence of consecutive video frames. While we first consider single windows, we extend our approach to a scenario with interlaced windows. The performance of the proposed algorithm is investigated, in terms of sensitivity and specificity, considering video recordings of newborns affected by neonatal seizures. Our results show that the use of interlaced windows guarantees both sensitivity and specificity values above 90%.