Neural Network Modeling of Ambulatory Systolic Blood Pressure for Hypertension Diagnosis

A neural network model is used for estimating Ambulatory Systolic Blood Pressure (ASBP) variations from corporal acceleration and heart rate measurements. The temporal correlation of the estimation residual, modeled by a first order autoregressive (AR) process, is used for training the neural network in a maximum likelihood framework, which yields a better estimation performance. As data are collected at irregular time intervals, the first order AR model is modified for taking into account this irregularity. The results are compared by those of a neural network trained using an ordinary least square method.