Fetal Heart Rate Deceleration Detection from the Discrete Cosine Transform Spectrum

Automated detection of decelerations in fetal heart rate (FHR) signals can be posed as a problem of signal detection in the presence of noise. We present an algorithm that adaptively selects the resolution of analysis and uses the discrete cosine transform (DCT) to describe the spectrum at short-term and longer-term scales. In so doing we generate near-orthogonal and scale-invariant features that are presented to a feedforward neural network for classification