Enhancing feature extraction for VF detection using data mining techniques

A. Rosado-Munoz, G. Camps-Valls, J. Guerrero-Martinez, J.V. Frances-Villora, J. Munoz-Mari, A.J. Serrano-Lopez
Computers in Cardiology  
Previous studies developed by the authors proposed VF detection algorithms, including VT discrimination, based on time-frequency distributions. However, due to the large number of parameters extracted from the distributions, efficient schemes for parameter selection and significance estimation are needed. This study proposes a combined strategy of classical and modern techniques for the selection of parameters to develop improved VF detection algorithms. We show how exhaustive exploration of
more » ... e exploration of the input space using data mining techniques simplifies and improves the solution and reduces the computational cost of detection algorithms. Jointly with classical selection techniques (correlation, Wilks' Lambda, statistical significance), other approaches are used (PCA, SOM-Ward and CART). We show that better results are achieved using less number of parameters than previous VF detection algorithms. Data collection and feature extraction Data from 29 patient recordings were analyzed, each containing an average of 30 minutes of continuous ECG, of which 100 minutes contained VF. Data were processed to obtain 25 time-frequency parameters from the Pseudo Wigner-Ville (PWV) distribution calculated over 128 point 0276−6547/02 $17.00 © 2002 IEEE 209 Computers in Cardiology 2002;29:209−212. © A pruning strategy. Training the trees does not follow any stopping rule but an over-growing and then pruning back methodology.
doi:10.1109/cic.2002.1166744 fatcat:7mvf7twzi5ccfpste43nqm6lga