SBIR Phase I: Information Theoretic Learning and Application to Fetal ECG

点击翻译按钮获取中文摘要

This Small Business Innovation Research (SBIR)Phase I project focuses on the development and evaluation of a new class of algorithms for blind source separation (BSS) and independent component analysis (ICA) based on a recently proposed information theoretic learning (ITL) criterion. The algorithms yield several practical criteria to adapt universal mappers, either under unsupervised or supervised paradigms. The ITL criterion can dramatically improve upon systems trained with mean square error. NeuroDimension will develop new algorithms to choose the segments for separation, address BSS of noisy mixtures, and extend the ITL criterion to convolutive mixtures. The firm further proposes to validate these methods via the fetal heart rate monitoring problem, which requires the separation of the maternal and fetal ECGs, a blind source separation problem. The ITL criterion of minimum cross entropy can exploit the fact that the ECGs are statistically independent. The expectation is that the new information theoretic learning will extract a much cleaner ECG because it is exploiting all the information about the signal statistics, not only the second order statistics (as MSE does). Finally the ITL criterion will be compared with the conventional interference cancellation algorithms in real data obtained from the University of Florida College of Medicine. The project has the potential to develop a new piece of clinical instrumentation, a fetal heart monitor, for which there is a demonstrated market. The firm utilizes a new approach to information signal process that may be able to identify the elusive fetal heart signal in a practical, real-time manner.