Adaptive Data Analysis Methods for Biomedical Signal Processing Applications

Introduction

Physiological processes are complex phenomena and are characterized by certain biomedical signals revealing their nature and performance. Thus the health status of a physiological system can be evaluated through analysis of the corresponding biomedical signal. Manual analysis of biomedical signals has many limitations and is very subjective. The accuracy and reliability of manual diagnostic processes are limited by several factors including the constraint of humans in extracting and detecting certain features from signals. Thus, computer-based processing and analysis of biomedical signals is necessary since it enables quantitative measurements and thus accurate diagnosis can be provided. Biomedical Signal processing techniques employ the mathematical tools for extracting some key features from a recorded signal based on which a clinical decision is made. The biomedical signal processing applications includes various stages. After signal acquisition by sensors/electrodes, the raw data is pre-processed and filtered. This pre-processing is essential because the measured signals often contain some undesirable noise that is combined with relevant signal information. The next step is to extract features from the processed signal that represent the status and condition of physiological system under consideration. The final step is classification and diagnostics in which clinical decisions are made (Rangayyan et al. 2002). Several techniques for processing of biomedical signals are currently available and many more are being developed for effective analysis and processing of signals. The earlier signal processing techniques were limited to standard linear filters and frequency spectrum based processing. Standard linear filters based analysis techniques have limited utility. In real life, the measured biomedical signals are typically more complex in nature and are often composed of different modes. These modes carry worthy information regarding the originating system and must be studied carefully. Due to nonlinear and non-stationary nature of many biomedical signals, adaptive data analysis techniques become a prime choice. The standard Fourier transform is the most widely used mathematical tool in signal processing but is useful for stationary signals only due to prior selection of sinusoidal basis set (Li, et al. 1995). Wavelet transform is an advanced signal processing method but, its performance is limited by choice of mother wavelet (Unser, 1997; Sharma et al. 2010). A fully adaptive data decomposition technique is Empirical mode decomposition (EMD) which tend to break a signal into a several oscillatory functions termed as intrinsic mode functions (IMF’s) and has gained a lot attention in bio-medical signal processing (Huang,et al. 1998). The aim of EMD algorithm is to identify the primary components representing the data, where a component roughly corresponds to a signal having a narrow bandwidth in spectral domain. The EMD algorithm is inherently adaptive and because of its data driven characteristics it can be employed for analyzing non-stationary and non linear signals. However, the major concern with EMD is insufficient mathematical background. In reality, it is an experimental approach and is difficult to model due to its inherent non-linearity. Another adaptive and data driven analysis technique is Empirical wavelet transform (EWT). EWT is relatively a new technique and decompose a time series into its various modes using adaptively designed filter banks (Gilles, J. 2013). The EWT technique first computes frequency spectrum of signal followed by its segmentation and then extraction of mode using adaptively designed wavelets. EWT is relatively a new mathematical technique that is still being researched and applications for it are being found. Variational mode decomposition (VMD) is another adaptive method used to break a signal into separate sub-signals (Dragomiretskiy et al 2013). In the next section, these techniques are discussed with their contemporary applications for biomedical signals.