guest ::
| Home > Publications database > Phase based Methods for Data Fusion |
| Home > Publications database > Phase based Methods for Data Fusion |
| Contribution to a conference proceedings | FZJ-2012-00825 |
2011
Abstract: Introduction Data fusion is the method to merge signals from different sources or sensors in order to, e.g., study the relationship among events or objects, reconstruct the laws of motion of a dynamical system, identify specific events or states of a dynamical system or even suppress noise and artifacts. Phase based methods are proposed for data fusion in physiological signals. Methods In physiology the far most activities obey a cyclic dynamic; therefore data fusion based on phase synchronization analysis is a promising approach in order to reveal the mutual interaction between the different processes, e.g. cardiac and respiratory activity. Phase calculation requires proper band pass filtering as a prerequisite, the cut-off frequencies being defined by the power spectral densities of the different signals. Applying this method the data fusion results e.g. in a histogram of phase differences whose deviation from a uniform distribution can be used to quantify the degree of phase synchronization. Another approach uses recurring events within one signal to mark trial epochs within the other signal in order to apply Cross Trial Phase Statistics (CTPS) [1]. Deviations from uniform distribution sharply indicate intermittent interactions between the signals also in those cases where amplitude based methods fail to reveal any cross correlation at all. In long term recordings during daily activity the occurrence of movement artifacts is very likely. In order to preserve the information of the undisturbed epochs accelerometer based movement detection can be used for artifact identification. The former mentioned CTPS allows for doing this in an automated manner. Results and Conclusion Phase analysis based methods are well suited for data fusion in physiological long term recordings. The analysis of phase difference histograms or Cross Trial Phase Statistics (CTPS) result in quantitative measures for synchronization or intermittent interaction of the merged signal and therefore reduces the huge amount of raw data to single numbers. This information reduction is an essential issue in long term recordings. CTPS also allows for automated artifact identification which is an important demand in home monitoring. [1] Dammers J, et al., (2008). IEEE transactions on bio-medical engineering 55:2353-62
|
The record appears in these collections: |