A new QRS detector stress test combining temporal jitter and F-score (JF) reveals significant performance differences amongst popular detectors

Figures

• 

  Fig 1. MITDB heartbeat annotations (dots) not located on R-peaks (record 107) and jitter between annotations.

• 

  Fig 2. Examples of ECG detection errors using a fixed temporal window. d0…4 are sample numbers of detected QRS complexes. a0…4 are sample numbers of the annotated QRS complex (ground truth). w is the temporal window used for the classical performance measures sensitivity, positive predictivity, false positive rate and overall accuracy.

• 

  Fig 3. Experimental setup and example ECG traces. A: Wiring of the two wireless biosignal amplifiers (Attys). The first amplifier records V1 − V2 with a chest strap and the second amplifier Einthoven II & III with standard Ag/AgCl electrodes. B: Photo of the chest strap and the biosignal amplifier mounted on it. Signal comparison between C: chest strap and D: Einthoven II recording while the subject was jogging.

• 

  Fig 4. Graphical illustration of the JF algorithm. A: Time stamps al of the QRS annotations versus the time stamps of the actual QRS detections dk. pl links the index numbers of the annotations l = 0, 1, 2, 3 to the index numbers k of the detections dk. B: Mapping function Eq 11 which maps the average jitter Δ¯ to a score between zero and one which reaches 0.5 at a jitter of 12 ms.

• 

  Fig 5. Classical sensitivity values. A: Einthoven, B: Chest strap. Significantly above 90% indicated with an asterisk.

• 

  Fig 6. JF values. A: Einthoven, B: Chest strap. Significantly above 90% indicated with a “*”.

• 

  Fig 7. Comparing detector performance against increasing signal to noise ratios. The JF benchmark value was calculated for the different activities of sitting, doing a maths test, walking, operating a hand-bike and jogging. The 4 detectors EngZee, Elgendi, WQRS and Pan& Tompkins have been benchmarked. If a JF value is significantly above 90% it is indicated with an asterisk.