Signal Processing

The physiological data collected with biomagnetic (MCG and infant MEG), EEG and fMRI systems, are analyzed using both linear and nonlinear signal processing techniques. Depending on the information of interest to be extracted from the recorded signals, diverse type of signal processing techniques can be performed.

Since 2004, our group has extensively employed Independent Component Analysis (ICA) for the processing of biomedical data of different origin, with the purpose of separating physiological signals and denoising the signals of interest. Signal denoising is also achieved by means of techniques that utilize wavelets decomposition.

More recently, our interest has focused on the development of methods to assess the non linear dynamics of complex physiological systems such as the heart or the brain, and to perform an automatic pattern recognition between clinical and non clinical data sets.

Independent Component Analysis

Independent component analysis (ICA) is a statistical Blind Source Separation (BSS) technique that models a set of input data in terms of statistically independent variables. ICA algorithms aim at separating mutually independent source signals from their linear instantaneous mixtures, without any a priori information about the spatial mixing.

During the last decade, ICA has been successfully used for signal extraction tasks in sound and image processing and in telecommunications. More recently, ICA has been employed in the field of biomedical signal processing, with the primary application of noise reduction. We use ICA for the separation of physiological signals recorded with multi-channel devices, such as biomagnetic, EEG and fMRI systems. This is possible because one major requirement for ICA application i.e. that the number of input signals be larger than or equal to the number of expected signals, is fulfilled when using data sets recorded with multi-channel devices.

Publications (selected):


Nonlinear dynamics

We estimate linear and nonlinear parameters (such as Short Term Variability (STV), Approximated Entropy (ApE), Sample Entropy (SE) and Multiscale Entropy (MSE)) in order to describe more comprehensively the behavior of complex physiological systems.

We are developing a method that, by means of dynamic estimators of chaos, such as correlation dimensions and Lyapunov exponents, might permit to assess the dynamics of physiological systems that exhibit a complex behavior. Furthermore, by means of those estimators we intend to characterize the evolution of systems such as the heart and the brain during their development in the life span.



Automatic pattern recognition systems

We are developing a system for the automatic detection of functional patterns embedded in physiological data sets recorded with multi-channel systems. This system aims at classifying, in a observer-independent manner, clinical and non clinical populations. Our system is based on the conjoint use of a multilayer perceptron (MLP) neural network that is trained with estimates of linear and nonlinear system parameters, such as ICA features and entropy estimators.




Write your email address to receive BIND Newsletters:


Click on the date to see the associated event:
October 2017
  • No events.