An Improved Sliding Window Area Method for T Wave Detection

Table 5

Comparable detection results of T wave offset in the QT database.

Methods	Annotations	Se (%)	P+ (%)	(ms)
Improved SWA	3542	98.5	98.5	1.21 ± 25.82
Traditional SWA [31]	3542	95.5	95.5	−1.12 ± 21.19
Wavelet-based [13, 14]	3542	99.77	97.79	−1.6 ± 18.1
Low-pass differentiation-based [20]	3542	99.00	97.74	13.5 ± 27.0
Hidden Markov model-based [21, 22]	3542	NA	NA	−5 ± 14
Partially collapsed Gibbs sample [23]	3403	99.81	98.97	4.3 ± 20.8
k-nearest neighbor-based [30]	30 records	NA	NA	2.8 ± 18.6
TU complex analysis [28]	3528	92.60	NA	0.8 ± 30.3
Neural network and fixed-size least-squares SVM [19]	3542	NA	NA	−3.0 ± 16.9
L.EKF25 [42]	10 records	NA	NA	11 ± 39
N.L.EKF25 [42]				4 ± 23
L.EKF25 [42]	15 records	NA	NA	−17 ± 30
N.L.EKF25 [42]				−21 ± 19

NA: not available; L.EKF25: linear Kalman filter; N.L.EKF25: nonlinear Kalman filter.