Nonparametric Regression Method for ECG Signal Pre-Processing Under Heteroscedasticity

• Activity Number: 188
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Donghui Zhang*+ and Cun-Hui Zhang
• Companies: Sanofi Aventis and Rutgers University
• Keywords: singal pre-processing ; wavelet method ; heteroscedasticity

Abstract

Electrocardiographic (ECG) analysis plays an important role in safety assessment during new drug development and in clinical diagnosis. The pre-processing of ECG analysis constitutes of low-frequency baseline wander (BW) correction and high-frequency artifact noise reduction from the raw ECG. Methodology for pre-processing when ECG signal is stationary (signal with homogenous noise) has been developed using wavelet method. However, methodology for signal pre-processing under heteroscedasticity is needed. This talk will present some development in this area. Briefly, we assume the signal variance is piecewise constant with unknown number of pieces m. We then develop method for estimate m and the variance within each piece. Afterwards, we normalize the signal to have homogenous noise so the pre-processing method for stationary signal would apply.