ECG Fiducial-Point Detection

Advancing Trustworthy Cardiac Biomarkers

This project and its associated publications are based on my doctoral dissertation, Advancing Trustworthy Cardiac Biomarkers: From Probabilistic to Deep Learning Methods for Robust ECG Peak Detection and HRV Recovery. The work focuses on trustworthy ECG biomarker development through robust R-, P-, and T-wave peak localization and improved heart rate variability analysis. Its main outcomes include browser-based annotation software, an annotated multi-rate ECG dataset, detection algorithms, and hardware deployment.

Overview

Pooling ECG recordings across devices means reconciling different sampling rates, and the long, expert-annotated P-wave and T-wave references needed to train detectors are scarce. This project addresses both: a dataset that pairs native and correctly resampled signals with verified peak annotations, the tool used to produce those annotations, and the models built on them. Each output below has its own permanent home and identifier.

Outputs

CardioAnnotate

Live

A self-contained, browser-based tool for viewing and editing ECG peak annotations. Runs entirely in the browser from a single HTML file.

RPT15 Database

Under review

A 42-subject single-lead ECG dataset with expert-verified R, P, and T peak annotations, released at native sampling rates and at five resampled rates produced with anti-alias and anti-imaging filtering. Derived from six open-access PhysioNet databases.

ECGformer

Patent pending

A deep learning model for ECG fiducial-point detection.

Details posted after the provisional patent application is filed. Preprint forthcoming.

Waveformer

Patent pending

A frequency-agnostic deep learning model for ECG fiducial-point detection across sampling rates.

Details posted after the provisional patent application is filed. Preprint forthcoming.

Hardware Deployment

In preparation

A study on deploying the detection models on embedded hardware.

Link forthcoming.

How to cite

Paper citation will be updated once the preprint or published version is available.

CardioAnnotate

Mojtahed, H., Rao, R., Paolini, C., Sarkar, M.. CardioAnnotate: An Interactive Toolkit for ECG Signal Processing, Annotation, Gaussian Process Correction, and Interoperable Data Exchange. Manuscript under review; [year].

Mojtahed, H. CardioAnnotate: Browser-Based ECG Annotation, RR Correction, HRV Analysis, and Export [software]. Zenodo; 2026. doi:10.5281/zenodo.20820831. Live application: https://cardioannotate.github.io/.

RPT15 Database

Mojtahed, H., Rao, R., Paolini, C., Sarkar, M.. RPT15: An ECG Database with Beat-Level R, P, and T Annotations at Native and Uniform Resampled Sampling Rates. Manuscript under review; [year].

Mojtahed, H.. RPT15: An ECG Database with Beat-Level R, P, and T Peak Annotations at Native and Uniform Resampled Sampling Rates [dataset]. PhysioNet; [year]. doi:[PhysioNet DOI — forthcoming].

Data sources & license

RPT15 is derived from six open-access PhysioNet databases (MIT-BIH Arrhythmia, Normal Sinus Rhythm, ST Change, Supraventricular Arrhythmia, and Sudden Cardiac Death Holter, and the European ST-T Database), each released under the Open Data Commons Attribution License v1.0. RPT15 is released under CCA 4.0. CardioAnnotate is released under CCA 4.0

Contact

Hamed Mojtahed, PhD
ORCID