Band Correlation Histogram to Improve Classification of Acute Encephalopathy in Infants

We propose a method to determine whether the early onset of acute encephalopathy causes severe sequela by analyzing the frequency of electroencephalogram waves. Even though sequela can severely damage the brains of infants, no prevalent method can automatically diagnose acute encephalopathy in them. We solve this problem by designing a discriminative feature that delivers impressive classification performance. Based on knowledge of the diagnosis, our method applies a bandpass filter, randomly selects pairs of waves over a short period, and computes a band correlation histogram from a distribution of their correlation coefficients. The results of experiments show that the band correlation histogram is superior to the prevalent method in the classification of a dataset of patients with acute encephalopathy.