George B. Moody PhysioNet Challenge

Logo

Quick links for this year's Challenge:

Please post questions and comments in the forum. However, if your question reveals information about your entry, then please email info at physionetchallenge.org. We may post parts of our reply publicly if we feel that all Challengers should benefit from it. We will not answer emails about the Challenge to any other address.

Session S31.3

Prediction of Spontaneous Termination of Atrial Fibrillation Using Time-Frequency Analysis of the Atrial Fibrillatory Wave

F Castells, C Mora, R Ruiz, JJ Rieta, J Millet, C Sanchez, S Morell

Universidad Polit├ęcnica de Valencia

Valencia, Spain

The prediction of the spontaneous termination of atrial fibrillation (AF) still remains an open issue. Any improved information that could be extracted from the surface ECG about the mechanisms of paroxysmal AF (PAF) and sustained AF may provide important advances in the treatment of PAF in order to prevent its perpetuation. In this event, the challenge is to discriminate PAF episodes from sustained AF episodes. A database including a learning set with 20 recordings and a test set with 30 recordings was provided. We hypothesized that the most relevant information about the mechanisms that triggers the AF is contained in the atrial fibrillatory wave (FW). Thus, a previous QRST cancellation stage was necessary, which was implemented using principal component analysis (PCA) concepts. The time- frequency analysis of the estimated FW was then carried out using the Choi-Williams transform. This processing was firstly realized for the recordings corresponding to the learning set. The analysis of the FW corresponding to PAF patients revealed two different fashions in the time- frequency domain: some patients showed a main frequency between 3.75Hz and 5.5Hz, which was very stable during the whole interval; in other patients it was not possible to identify a main frequency, but the FW seemed to be completely irregular and chaotic. On the other hand, the analysis of the FW corresponding to sustained AF patients showed main frequencies between 5.5Hz and 8Hz that evolved with slight frequency variations. These observations were employed in the classification algorithm for the test set, obtaining 16 sustained and 14 self-terminating AF episodes, with an initial result of 26 correct classifications out of 30. The most relevant parameter to discriminate among both groups was the main frequency. These results are coherent with existing AF theories, since the refractory cycle tends to decrease with the perpetuation of the arrhythmia. In addition, possible hypothesis for explaining PAF behaviours could be based on excitations of local foci for the case of stabilized frequencies and on the absence of reentrant self- perpetuating wavefronts in the case of patients where a main frequency could not be identified.


Supported by the National Institute of Biomedical Imaging and Bioengineering (NIBIB) under NIH grant number R01EB030362.

Back