Medical University of Innsbruck (Dlaska Lab)
ECG foundation model built on the xLSTM (extended LSTM) architecture: a bidirectional stack of nine alternating scalar- and matrix-memory LSTM blocks that scales linearly with sequence length, unlike the quadratic cost of transformer-based ECG models. Pretrained with SimDINOv2, a coding-rate-regularized self-distillation (DINO) objective adapted from computer vision to ECG time series, on roughly 8 million recordings from CODE, INCART, and Chapman-Shaoxing-Ningbo. Introduced alongside BenchECG, a standardized 8-dataset/10-task benchmark, on which xECG achieves the best average rank of any publicly available ECG foundation model, with particular strength on long-context tasks (30-minute ambulatory arrhythmia classification, multi-hour sleep-apnea segmentation) where transformer-based models are computationally limited.
Architecture
RNN / LSTM / GRU
Bidirectional xLSTM encoder (9 alternating sLSTM/mLSTM blocks, ~57M parameters) over non-overlapping temporal ECG patches; pretrained with SimDINOv2 teacher-student self-distillation
Framework
PyTorch
Added to catalog
2026-08-10
MIT
License for model weights only. Associated code may be licensed seperately, check code source for specific terms.
CODE (Clinical Outcomes in Digital Electrocardiology, full cohort)
Full restricted-access CODE cohort of Brazilian primary-care 12-lead ECGs, distinct from the public CODE-15% subsample already in this catalog; used for xECG pretraining.
12-lead ECG multi-label arrhythmia-classification dataset released for the 2018 China Physiological Signal Challenge.
45,152 12-lead, 10-second ECGs from Chapman University / Shaoxing People's Hospital / Ningbo First Hospital with arrhythmia diagnoses; ~56% male / 44% female.
48 half-hour excerpts of two-channel ambulatory ECG recordings from 47 subjects studied by the BIH Arrhythmia Laboratory; used as a cross-dataset, lead-missing/noise-robustness benchmark distinct from the pretraining data.
70 single-lead ECG recordings (approximately 7-10 hours each) with minute-by-minute sleep-apnea annotations; used in the BenchECG benchmark to test long-context segmentation performance.
75 annotated 30-minute 12-lead ambulatory ECG recordings from 32 subjects, collected at the St. Petersburg Institute of Cardiological Technics; one of three corpora (with CODE and Chapman-Shaoxing-Ningbo, ~8M ECGs total) used to pretrain xECG.
CPSC2018 multi-label diagnostic classification (9 classes)
MIT-BIH heartbeat-level 5-class arrhythmia classification (AAMI classes)
Sleep-apnea presence/absence segmentation in overnight single-lead ECG (1-minute resolution)
General-purpose ECG patch/signal-level representation embedding