V1.2.0 Feature Release

V1.2.0 Feature Release

New Feature:
Pacemaker Spike Detection

If using raw ECG data e.g. as input for ML models, pacemaker ECGs often have to be evaluated separately. Our new feature adds a pacemaker ECG flag based on the detection of pacemaker spikes to the output metadata csv file.

Methods:
Pacing-spike detection is performed on the raw, unfiltered signal immediately after file import and before any filtering or resampling as conventional filters (e.g. notch, 0.5–40 Hz bandpass, wavelet/EMD denoising) might attenuate the 0.1–2 ms pacing artifact. Following the slope-steepness framework of Haq et al.[1], candidate spikes were enhanced by ~120 Hz high-pass filtering and projection onto the first principal component of the 12-lead signal, detected by adaptive peak picking, and confirmed by a 30%-envelope width criterion (2–30 ms) together with per-lead raw-slope confirmation (≥2 leads exceeding ~12 µV/ms), mirroring the automated detection pipeline described by Dhar et al. [2]. For recordings without an embedded sampling rate (e.g., CSV/ASC files), the user provided rate is used. Each ECG recording receives a single pacemaker flag (paced/non-paced/unknown) with a confidence score and spike count within the ecg_summary.csv output file.

Validation:
The pacemaker detection feature was validated end-to-end on the PTB-XL v1.0.3 dataset [3] using all 294 ECG records with the SCP PACE statement plus 400 randomly selected non-PACE controls (n = 694, 500 Hz). Our record-level pacemaker flag achieved a sensitivity of 0.962 (95% CI 0.933–0.978), a specificity of 0.872 (0.836–0.901), PPV of 0.841, a NPV of 0.970, and an accuracy of 0.909 (Wilson intervals), with an AUROC over the confidence score of 0.929 (bootstrap 95% CI 0.910–0.948). No record returned an indeterminate result (0/694). Error review showed 6 of 11 false negatives at exactly two detected spikes; just one below the set ≥3-spike record rule. False positives included artifact-driven cases. As limitation, the used enriched PTB-XL cohort (41% prevalence) might not be ideal for extrapolating to general screening populations.

References:
[1] Haq KT, Javadekar N, Tereshchenko LG. Detection and removal of pacing artifacts prior to automated analysis of 12-lead ECG. Comput Biol Med. 2021;133:104396. https://doi.org/10.1016/j.compbiomed.2021.104396.
[2] Dhar R, Kewalramani S, Tereshchenko L. Automated Detection and Removal of Pacing Artifacts from ECG Signal. PhysioNet. 2026. RRID:SCR_007345. https://doi.org/10.13026/hq2h-8p89.
[3] Wagner P, Strodthoff N, Bousseljot RD, Kreiseler D, Lunze FI, Samek W, Schaeffter T. PTB-XL, a large publicly available electrocardiography dataset. Sci Data. 2020;7:154. Data version 1.0.3: PhysioNet. 2022. https://doi.org/10.1038/s41597-020-0495-6.

V1.1.0 Bugfix Release

Bugfixes and improved robustness

V.1.0.0 Initial Release

First Release of the cross-platform GUI