Related Experiment Videos
Direct Reconstruction of High-Fidelity Electrocardiogram Signals From Vector-Based PDF Files With Integrated Deep
Shian-Sen Shie1,2, Po-Yen Huang1,3, Ming-Shien Wen4
1Division of Infectious Diseases, Chang Gung Memorial Hospital, Linkou Medical Center, Taoyuan, Taiwan.
JMIR Formative Research
|July 24, 2026
Summary
This study introduces a new method to directly extract electrocardiogram (ECG) signals from PDF files, enabling accurate prediction of key ECG parameters using deep learning for efficient analysis.
Area of Science:
- Cardiology
- Biomedical Engineering
- Data Science
Background:
- Electrocardiograms (ECGs) are frequently stored as vector-based PDF files from ECG management systems.
- Previous research shows ECG signals can be reconstructed from vector graphics, enabling parameter derivation.
Purpose of the Study:
- To develop an integrated framework for direct, high-fidelity ECG signal reconstruction from vector-based PDFs.
- To simultaneously estimate multiple clinically relevant ECG parameters using deep learning.
Main Methods:
- Analyzed 50,000 twelve-lead ECG PDFs from a MUSE system (2015-2024).
- Developed a direct PDF parsing pipeline to extract vector path objects and reconstruct time-series signals.
- Utilized deep learning models (DualECGFormer, DualResNetECG) for parameter estimation, comparing performance against a rule-based approach (NeuroKit2).
Main Results:
- The direct parsing method achieved high reconstruction fidelity (MAE < 1x10-3 mV) and was ~4.6x faster than SVG conversion.
- Deep learning models outperformed the rule-based method for most parameters, with DualResNetECG showing strong performance (e.g., MAE of 1.11 bpm for heart rate, 7.27 ms for PR interval).
- Models demonstrated reliable detection of parameters like PR interval and P-wave axis (AUC up to 0.978).
Conclusions:
- An efficient, scalable framework for direct ECG signal extraction from MUSE PDFs and multiparameter estimation via deep learning was developed.
- The approach offers high reconstruction accuracy and competitive predictive performance for large-scale retrospective ECG analysis.