Optimal length of R–R interval segment window for Lorenz plot detection of paroxysmal atrial fibrillation by machine learning

Table 7 Comparisons of AF detection performance in the PhysioNet database

AF detection method	Segment window length, beats	Classification performance
AF detection method	Segment window length, beats	Sensitivity	Specificity	PPV	NPV	Accuracy	PLR	NLR	AUC
LP image CNN (this study)^a	10	0.975	0.928	0.741	0.994	0.936	13.55	0.027	0.987
	20	0.980	0.958	0.830	0.996	0.962	23.24	0.021	0.989
	50	0.975	0.971	0.877	0.995	0.972	33.71	0.025	0.983
	85	0.984	0.978	0.905	0.997	0.979	45.49	0.016	0.987
	100	0.984	0.978	0.905	0.997	0.979	45.19	0.016	0.986
	200	0.992	0.976	0.897	0.998	0.979	41.08	0.008	0.982
	500	0.998	0.963	0.85	1.000	0.969	26.94	0.002	0.980
RdR map [9]	32	0.944	0.926	–	–	–	–	–	0.978
	64	0.958	0.943	–	–	–	–	–	0.986
	128	0.959	0.954	–	–	–	–	–	0.989
CoSen [13]	41	–	–	–	–	0.911	–	–	–
Poincare plot [13]	82	–	–	–	–	0.912	–	–	–
SVM [13]	65	–	–	–	–	0.909	–	–	–
Normalized fuzzy entropy [11]	12	0.956	0.925	0.812	0.971	0.890	–	–	0.927
	30	0.967	0.952	0.852	0.973	0.914	–	–	0.953
	60	0.985	0.968	0.878	0.987	0.935	–	–	0.968

AUC area under the curve, CoSen coefficient of sample entropy, NLR negative likelihood ratio, NPV negative-predictive value, PLR positive likelihood ratio, PPV positive-predictive value, SVM support vector machine
^aAF images were annotated with the non-strict criteria

ISSN: 1475-925X