Peking University (PKUDigitalHealth) (Zhang, Hong et al.)
Model weights not public. Contact creators for more information.
Patch-based masked-training framework for robust arrhythmia detection from digitized, multi-layout ECG images (e.g. 3x4, 2x6, 12x1 printed/scanned layouts), designed to handle the asynchronous lead timing and partial signal blackout that digitization introduces. An adaptive variable block-count masking strategy focuses model attention on key patches with cross-lead dependencies. Evaluated on PTB-XL digitized into multiple synthetic layouts and externally validated on 400 real digitized ECG images from Chaoyang Hospital, outperforming classical imputation baselines and the CNN foundation model ECGFounder.
Architecture
Vision Transformer
Patch-based masked-training framework with adaptive variable block-count missing-data representation learning, focusing attention on key cross-lead patches, layered on top of a digitized-ECG backbone (evaluated using ECGFounder and other baselines as the underlying encoder)
Framework
PyTorch
Added to catalog
2026-08-10
Chaoyang Hospital AF ECG Images
400 real digitized 12-lead ECG images used for external validation of atrial-fibrillation diagnosis on out-of-distribution printed/scanned layouts.
52% male / 48% female; age range 0-95 (median ~62). 21,837 10-second 12-lead ECG records.
PTB-XL 23-subclass diagnostic classification from digitized multi-layout ECG images
Atrial fibrillation diagnosis from real digitized ECG images (external validation)