METHODS: CFAE from several atrial sites, recorded for a duration of 16 s, were acquired from 10 patients with persistent and 9 patients with paroxysmal AF. These signals were appraised using non-overlapping windows of 1-, 2- and 4-s durations. The resulting data sets were analyzed with Recurrence Plots (RP) and Recurrence Quantification Analysis (RQA). The data was also quantified via entropy measures.
RESULTS: RQA exhibited unique plots for persistent versus paroxysmal AF. Similar patterns were observed to be repeated throughout the RPs. Trends were consistent for signal segments of 1 and 2 s as well as 4 s in duration. This was suggestive that the underlying signal generation process is also repetitive, and that repetitiveness can be detected even in 1-s sequences. The results also showed that most entropy metrics exhibited higher measurement values (closer to equilibrium) for persistent AF data. It was also found that Determinism (DET), Trapping Time (TT), and Modified Multiscale Entropy (MMSE), extracted from signals that were acquired from locations at the posterior atrial free wall, are highly discriminative of persistent versus paroxysmal AF data.
CONCLUSIONS: Short data sequences are sufficient to provide information to discern persistent versus paroxysmal AF data with a significant difference, and can be useful to detect repeating patterns of atrial activation.
METHODS: We first tested ten traditional machine learning algorithms, and then the three-best performing algorithms (three types of SVM) were used in the rest of the study. To improve the performance of these algorithms, a data preprocessing with normalization was carried out. Moreover, a genetic algorithm and particle swarm optimization, coupled with stratified 10-fold cross-validation, were used twice: for optimization of classifier parameters and for parallel selection of features.
RESULTS: The presented approach enhanced the performance of all traditional machine learning algorithms used in this study. We also introduced a new optimization technique called N2Genetic optimizer (a new genetic training). Our experiments demonstrated that N2Genetic-nuSVM provided the accuracy of 93.08% and F1-score of 91.51% when predicting CAD outcomes among the patients included in a well-known Z-Alizadeh Sani dataset. These results are competitive and comparable to the best results in the field.
CONCLUSIONS: We showed that machine-learning techniques optimized by the proposed approach, can lead to highly accurate models intended for both clinical and research use.