Atrial Fibrillation from an Engineering Perspective by Leif Sörnmo

Author:Leif Sörnmo
Language: eng
Format: epub, pdf
Publisher: Springer International Publishing, Cham

(5.23)

where denotes the ensemble average. While the computation of is straightforward, it has been shown that is more sensitive to temporal misalignment of the beats in the ensemble than is , leading to an overestimation of [27]. In that study, a sampling rate of 3 kHz was recommended to ensure that sampling-related misalignment does not influence , i.e., a much higher sampling rate than what is used in most modern ECG devices (1 kHz). However, it is straightforward to perform digital interpolation to obtain the recommended sampling rate.

Since low-amplitude T waves are difficult to detect, and consequently to align, may not be reliably estimated. This problem can be handled by using the time-dependent weights in (5.22) when processing the QRS interval, whereas the fixed weights of ABS are used when processing the T wave interval [21]—a solution requiring two templates to accomplish f wave extraction. Separate processing of these two intervals has been considered before, but then in the context of spatiotemporal alignment, see Sect. 5.2.4.

As already noted, the performance of signal- and noise-dependent weighted averaging was evaluated using simulated f waves, generated by the harmonic model described in Sect. 3.​4, added to non-AF ECGs of the PTB database. The results showed that the normalized MSE between the extracted and the model f wave signals is about 25% lower than that of ABS. The main explanation to this improvement is related to the inclusion of the ensemble variance in the computation of , which should increase as the morphologic beat-to-beat variability increases. No information was provided on whether classification of beat morphology is required before forming the beat ensemble, and, therefore, it remains to be quantified to what extent beats with deviating morphology influence f wave extraction in terms of QRS-related residuals. However, when such beats are encountered, smaller QRS-related residuals are likely to result for than for the weights used in ABS or weighted ABS.

Figure 5.2 illustrates the performance of signal- and noise-dependent weighted averaging when an ECG signal with considerable variation in f wave amplitude is processed. It is interesting to note that the extracted f wave signal does not contain as much QRS-related residuals as does the ABS-produced f wave signal.

Fig. 5.2Signal- and noise-dependent weighted averaging. a Original ECG and extracted f wave signal obtained by b average beat subtraction, and c signal- and noise-dependent weighted averaging

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.