Summary
Researchers report open-source MATLAB and Python software that recovered ECG waveforms with zero error from non-clipped GE MUSE vector PDFs in a test of 1,632 PDF–XML pairs. The method uses information retained in the PDFs’ vector encoding and is specific to MUSE-generated files.
Researchers have developed open-source MATLAB and Python software that reconstructs ECG waveforms from vector-encoded General Electric (GE) MUSE PDFs. In a medRxiv preprint, the team reports that the software recovered every non-clipped waveform in a test of 1,632 PDF–XML pairs with zero error against the corresponding raw XML data.
The approach is designed for ECG PDFs generated by MUSE, rather than for arbitrary scans or ECG images. It also identifies waveform clipping, a condition the researchers say prevents exact recovery.
Why a vector PDF can retain waveform data
ECG digitization often means estimating a signal from an image: software identifies the plotted trace and maps its position back to time and voltage. That process can introduce small differences from the original samples.
A vector PDF can contain waveform graphics derived directly from the underlying data, rather than only a grid of image pixels. The authors say values are rounded when ECG data are encoded into these PDFs, but that the relationship between the raw samples and the encoded values remains one-to-one. Their method uses this relationship, described as double quantization, to reverse the encoding and recover the original samples.
In practical terms, the software works from the vector representation and its encoding parameters, not by tracing pixels. The paper presents this as a way to avoid the approximation involved in rasterising a PDF and then digitising its image.
Tests and reconstruction checks
The researchers compared reconstructed waveforms with raw XML waveforms across 1,632 PDF–XML pairs, covering five layouts, each at three voltage gains. After cases identified by the software as clipped were excluded, all assessed waveforms were reconstructed with zero error.
The software also checks whether the parameters used for reconstruction are consistent with those used when the PDF was encoded. The team perturbed encoding parameters to test this certification step and reports that it was highly sensitive to even minor changes. When the check passes, the authors say it can certify bit-exact reconstruction without comparing the result against ground-truth XML.
The software supports multiple MUSE PDF layouts and voltage gains, and automatically flags clipping. Its source code is available at the project’s GitHub repository, BIVectors/pdf2ECG.
Scope of the result
The reported result concerns GE MUSE-generated vector PDFs. The authors state that their method does not replace general ECG image digitization, which remains relevant for ECGs available only as images. They consider their approach preferable to rasterization followed by image-based digitization for MUSE PDFs.
This work is a medRxiv preprint. The authors report that the datasets used in the study are not publicly available for ethical and legal reasons; underlying results may be provided by the authors upon request where possible.