Spiral wave clustering using normalized compression distance

Cardiac fibrillatory dynamics are identified with spiral waves in mathematical modeling of cardiac electrical propagation. Automatic identification of spiral wave dynamics is essential for patient specific cardiac modeling. In our work we used normalized compression distance (NCD), an information theoretical distance measure, in order to cluster the simulated spiral waves as stable, meandering and break up. Different representation of the data was introduced to NCD in the form of raw time series, fast Fourier transform (FFT), feature summarization and symbolic quantization of the simulated electrograms. Clustering was done in an unsupervised way using spectral method. Clustering analysis was performed using diferent validation methods. Gap statistics was used to find optimal number of groups. Jaccard coefficient was used in order to evaluate accuracy of clustering. We had a perfect evaluation results from the raw data representation and Fourier transformation with a jaccard index of J, and a very good performance of feature summarization with a jaccard index of 0.98.