Supervised Machine Learning Based Noninvasive Prediction of Atrial Flutter Mechanism from P-to-P Interval Variability under Imbalanced Dataset Conditions

Table 11

Related works on discrimination of AFL mechanism (PWM: P-wave morphology, CL: cycle length, and NA: not applicable).

Author	Year	Ratio macro/focal (ECGs)	Technique	Classifier	Parameter (significant)	Performance
Brown et al.	2007	27/14	PWM, CL	NA (autocorrelation performed)	P < 160 ms P/CL < 45%	P: Sensitivity: 90%; Specificity: 90%; P/CL: Sensitivity: 86%; Specificity: 98%
Chang et al.	2011	51/17	PWM CL	NA (empirical-based study)	V6 > 0.9 mV, CL > 265 ms V6 < 0.9 mV, CL > 290 ms	Accuracy focal: 93%; Accuracy macro: 88%; Accuracy focal: 100%; Accuracy macro: 100%
Luongo et al.	2020	Original: 11/9; Augmented: 1256	Recurrance quantification	Decision tree (DT), KNN, and radial basis neural network (rbNN)	RQA-based features	Hit rate: 67.7%
Proposed model		Original: 41/5; Augmented: 15 (focal)	Consecutive P-P intervals (within R-R interval)	LDA, LOG, and SVM	Sum of all consecutive intervals (p < 0.05)	Accuracy: 76.88%; Specificity: 49.50; Sensitivity: 90.24%