Evolution of Automatic Paper Electrocardiogram Digitization: Methods, Applications, and Future Perspectives

Cardiovascular disease remains a leading cause of global mortality, and the electrocardiogram (ECG) is among the most widely used diagnostic tools. However, a substantial number of ECG records remain available only as paper records or scanned images, which are susceptible to aging, fading, and other forms of physical degradation, leading to information loss. Automatic paper ECG digitization, which involves reconstructing machine-readable waveforms from visual records, offers an effective means of transforming historical ECG records into reusable digital data. This systematic review synthesizes the evolution of automatic paper ECG digitization following the PRISMA framework and analyzes 78 studies. We first summarize public ECG databases and emerging image-based datasets and then examine the digitization pipeline across three stages: image preprocessing, waveform localization and extraction, and signal calibration and reconstruction. Both traditional computer vision techniques and deep learning methods are systematically compared, highlighting a shift from handcrafted rules toward data-driven feature learning that improves robustness to noise, waveform overlap, and heterogeneous layouts. We further review major applications, including historical data recovery, automated diagnosis, monitor calibration, and multimodal digital health integration. Key challenges are identified, notably the scarcity of paired image–signal datasets, the absence of standardized evaluation protocols, limited interpretability, and gaps in clinical validation. Future progress will depend on standardized benchmarks, lightweight and interpretable models, and multimodal and foundation-model approaches to enable robust, generalizable, and clinically applicable ECG digitization. By transforming legacy paper and image-based ECG records into machine-readable signals, ECG digitization can also serve as an enabling technology for multimodal physiological signal processing and artificial intelligence-driven health monitoring, contributing to the broader digital health ecosystem.