Recurrence quantification analysis for characterizing atrial electrogram fractionation in human chronic atrial fibrillation

Background – Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and radiofrequency catheter ablation is the most accepted interventional procedure for its treatment. During AF ablation, two types of atrial electrograms (AEGs) are usually identified: (i) normal AEGs, with organized discrete activations and; (ii) complex fractionated atrial electrograms (CFAEs), with continuous complex activations. CFAEs have been used as targets for chronic AF ablation, but this strategy has shown suboptimal outcomes due to, among other factors, poor understanding of AEGs dynamics during AF. In this work, we employed recurrence quantification analysis (RQA) for characterizing AEG complexity during chronic AF using the clinical labels provided by the broadly used commercial system, the CARTO (Biosense Webster). Methods – 797 AEGs were collected from 18 chronic AF patients undergoing ablation. Automated CFAE classification (Normal AEG or CFAE) was performed in all cases using the CARTO criteria. CARTO calculates the Interval Confidence level (ICL), Average Complex Interval (ACI) and the Shortest Complex Interval (SCI). The AEGs were considered CFAEs if ICL≥4, ACI≤82 ms and SCI≤58 ms. Nine RQA attributes were calculated from the AEGs: DET; Lmax; ENTR; RR; ZIPrate; LAM; TT; Vmax; and L. Eight four linear discriminant analyses (LDA) were performed considering all possible combinations of three RQA attributes, and were compared with the LDA created using the three CARTO attributes. Results – A total of 307 (39% of total 797) AEGs were classified as CFAEs by CARTO. As expected, the LDA using the three CARTO attributes achieved a high hit rate (93% overall; 98% for CFAEs; 89% for Normal AEGs). The LDA using RQA attributes with best classification was achieved with DET, ZIPrate and Vmax (70% overall hit rate; 51% for CFAEs; 82% for Normal AEGs). These RQA attributes were also effective in significantly discriminating Normal AEGs and CFAEs (Normal AEGs vs CFAE, respectively [Mean±SD]. DET: 0.98±0.03 vs 0.97±0.02; ZIPrate: 0.20±0.08 vs 0.26±0.05; Vmax: 505±731 vs 275±166; p<0.0001 for all cases). Conclusion – This work takes a first step towards the characterization of AEGs using RQA complexity measures, in which we have identified the best RQA-based three-dimensional space attribute and compared it with the CARTO criteria. The attained difference between the classification performances motivates the analysis of AEGs during AF using unsupervised methods with either the CARTO criteria, or an alternative paradigm based on RQA, or even the combination of those methods.