PhysioNet Research Code Competition: Digitization of ECG Images (www.kaggle.com)

PhysioNet has relaunched its ECG Digitization Challenge (start Oct 21, 2025) asking teams to build models that extract 12‑lead ECG time series from photos, scans and paper printouts. The goal is to convert billions of historical or image‑only ECGs into machine‑readable signals so modern diagnostic software and ML models can use legacy data to improve cardiovascular care, especially in settings that lack digital records. The task is technically challenging: imaging artifacts (rotation, scale, blur, reflections), variable lead placement across vendors, differing sampling rates and signal lengths, and noisy clinical traces complicate extraction. Submissions are scored with a modified signal‑to‑noise ratio (SNR) that first aligns predictions to ground truth by finding a horizontal shift up to 0.2 seconds (maximizing cross‑correlation) and removing any constant vertical offset; SNR is computed by summing signal and noise power across all 12 leads and converted to decibels. Entrants must submit notebook-based solutions (CPU/GPU runs ≤9 hours, internet disabled; public external data and pretrained models allowed) as submission.parquet or submission.csv. Key dates: entry/team deadline Jan 15, 2026, final submission Jan 22, 2026. Prize pool awards seven places ($10k down to $5k), and successful approaches could unlock vast historical ECG datasets for research and clinical deployment.