Comparative Study of Methods for Cycle Length Estimation in Fractionated Electrograms of Atrial Fibrillation

Figure 1. Comparative analysis of the preprocessing stage of each of the compared LAW detection methods applied on the same EGM: (a) Denoised EGM to be processed; (b) Result of AAT preprocessing; (c) Result of ACLI preprocessing; (d) Result of HT preprocessing. The dashed line in (d) corresponds to the application of the hyperbolic tangent.

Figure 2. Block diagrams of the LAW detection methods compared. Underlined values in the first blocks show modified thresholds with respect to traditional Botteron’s preprocessing method. BP: band-pass filtering; LP: low-pass filtering; HP: high-pass filtering; Pi: amplitude of the i-th peak; med{CL}: median cycle length.

Figure 3. Comparative example for calculating the CL estimation error. (A) EGM with experts’ (circle) and method’s annotations: (a) expert’s CL (b) method’s CL and (c) errors in CL estimation. Top: Situation 1 (squares) undersensing case (missing activation-red). The arrow shows the starting points in case of unequal CLs for the undersensing case; Bottom: Situation 2 (triangles) undersensing and oversensing case (false positive activation-red). (B) Error estimation for the two example situations. (left) Mean CL error; (right) Individual CL error. Notice how the simultaneous existence of under- and oversensing in Situation 2 may create erroneous CL calculation for the mean CL case.

Figure 4. One-second example of the three EGM types and the annotations made from the three algorithms as well as from the experts. Experts’ annotations are considered as ground truth. For AF type I, activations were clearly distinguished from the baseline and all the methods were able to identify them. For AF type II, although the baseline shows little perturbation, lower amplitude activations slightly complicate the procedure, leading to a mishit for the AAT algorithm. Finally, for AF type III EGMs, mishits and false positives are observed for all methods across the EGM.

Figure 5. Comparative analysis of errors estimating CL using the mean CL and the individual CL methods for the three LAW detectors compared: (a) Results for type I electrograms; (b) type II electrograms; and (c) type III electrograms. Values indicated with + stand for outliers.

Figure 6. Individual CL error depending on EGM length for the three LAW detection methods analyzed: (a) Results for type I electrograms; (b) type II electrograms; and (c) type III electrograms.