phd abstract

Cardiac system complexity analysis in a macroscopic and microscopic scale

I. G. Chouvarda

Aristotle University of Thessaloniki (A.U.Th.)

The Medical School,

Dept of Radiology, Medical Physics and Informatics

Lab of Medical Informatics

http://www.didaktorika.gr/eadd/handle/10442/14010

The PhD thesis concerns cardiac signals processing by use of time-frequency transforms towards solving specific problems of the domain.

The purpose of this work is, besides the possible contribution to the solution of the specific problems, to investigate different ways that time-frequency transforms could be employed, to explore the possibilities of combining time-frequency features in a relevant manner, in order to result in quantitative and meaningful outcomes.

Three different approaches were developed, for different type of problem and cardiac signals each. Each problem considers a different viewpoint of the cardiac system, starting from macroscopic ECG recordings and ending to the microscopic domain of 1-D propagation along a fiber.

Specifically, regarding the 1D propagation, a method for the estimation between the recording electrode and the active fiber has been developed, based on experimental data from propagation on cardiac fibers. The source models and their relation to the extracellular recordings was modelled, and the model parameters were estimated based on an adapted simulated annealing method. It was found that model parameter optimisation had better results in distance estimation, when wavelet features were used for optimisation, than time domain features.

Following, in 2D propagation, the focus was on the problem of reconstruction of the activation map, which is nontrivial in cases of arrhythmogenic tissue. In this study, simulated data were used. A technique was developed, based on wavelet filtering, and compared with other methods, such as deconvolution methods. It was found that wavelet based method was more robust with sparser electrode grids.

As far as macroscopic recordings of the cardiac system are concerned, clinical ECG data were used and the problem under investigation was the classification of cases with respect to the success of thrombolysis, based on the features of Wigner-Ville transform of the ECG. A linear classifier was applied, with iterative feature selection and bootstrapping for statistical robustness. The method classified correctly the dataset. Furthermore, the evolution of the features with time was related to the differences in patient's medical condition after thrombolysis.

The aim was the development of new algorithms and methodologies addressing specific problems. The work introduced innovative points concerning the analysis of electrograms and the overall methodology.