Multiparameter physiological signal reconstruction using NARX Neural Networks

Constant monitoring of a variety of physiological signals is vitally important in numerous clinical care settings. This signals are not perfect, however, and can be corrupted or lost. The loss of a signal can be devastating to the patient, as the physician may lose key information to understanding disease processes, or worse, be unaware of the patient´s status in either surgery or the ICU. This study uses a NARX-type Artificial Neural Network to reconstruct portions of physiological signals that have become corrupted. The effectiveness of this network was tested using signals and guidelines from the Computing in Cardiology/Physionet 2010 challenge, “Mind the Gap.” The NARX network performs quite well under these conditions, comparing favorably with other top entrants in the Physionet competition. Additionally, it is noted that the accuracy of the signal reconstructions also depends on which channel was corrupted. This work has important implications in many areas ranging from sports medicine and sleep studies to surgery and the ICU.