Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
Background: Signal processing tools are required to efficiently analyze data collected in body-surface-potential map (BSPM) recordings. A limited number of such tools exist for studying persistent atrial fibrillation (persAF). We propose two novel, spatiotemporal indices for processing BSPM data and test their clinical applicability through a comparison with the recently proposed non-dipolar component index (NDI) for prediction of single-procedure catheter ablation (CA) success rate in persAF patients. Methods: BSPM recordings were obtained with a 252-lead vest in 13 persAF patients (8 men, 63 +/- 8 years, 11 +/- 13 months sustained AF duration) before undergoing CA. Each recording was divided into seven 1-min segments of high signal quality. Spatiotemporal ventricular activity (VA) cancellation was applied to each segment to isolate atrial activity (AA). The two novel indices, called error-ratio, normalized root-mean-square error (ERNRMSE) and error-ratio, mean-absolute error (ERABSE), were calculated. These indices quantify the capacity of a subset of BSPM vest electrodes to accurately represent the AA, and AA dominant frequency (DF), respectively, on all BSPM electrodes over time, compared to the optimal principal component analysis (PCA) representation. The NDI, quantifying the fraction of energy retained after removal of the three largest PCs, was also calculated. The two novel indices and the NDI were statistically compared between patient groups based on single-procedure clinical CA outcome. Finally, their predictive power for univariate CA outcome classification was assessed using receiver operating characteristic (ROC) analysis with cross-validation for a logistic regression classifier. Results: Patient clinical outcomes were recorded 6 months following procedures, and those who had an arrhythmia recurrence at least 2 months post-CA were defined as having a negative outcome. Clinical outcome information was available for 11 patients, 6 with arrhythmia recurrence. Therefore, a total of 77 1-min AA-BSPM segments were available for analysis. Significant differences were found in the values of the novel indices and NDI between patients with arrhythmia recurrence post-ablation and those without. ROC analysis showed the best CA outcome predictive performance for ERNRMSE (AUC = 0.77 +/- 0.08, sensitivity = 76.2%, specificity = 84.8%). Conclusion: Significant association was found between the novel indices and CA success or failure. The novel index ERNRMSE additionally shows good predictive power for single-procedure CA outcome.
,
Dario Floreano, Yegor Piskarev, Jun Shintake, Yi Sun, Matteo Righi