On-line neuromorphic biomedical waveform analysis

Abstract

Addresses the real-time and on-line processing of clinical patient waveform patterns so as to improve critical-care maintenance in anesthesiology. The authors demonstrate that patient waveform pattern interpretation is achieved via a neuromorphic approach. This is primarily attributed to a neural networks ability to adaptively learn from examples. The authors utilized four distinct data sets (each containing a hundred patterns) encompassing patients vital clinical breathing patterns (i.e. hypocapnia, hypoventilation, and curare cleft) to train their network. Each pattern featured a base, ascending, plateau, and descending lines. The test and training data sets were obtained from actual strip-chart recordings. The Neuromorphic System (Neuro-Sys) was trained to correctly classify all of twenty-one unique clinical breathing patterns within a ten percent error-tolerance.