Ahmed Sameh

Papers

• Beat-Synchronous Tokenization for ECG Transformers
  2026 · IEEE MLSP 2026

  University of Minnesota group asks a narrow, well-controlled question: does tokenizing an ECG waveform at cardiac-cycle (beat) boundaries instead of fixed time patches change Transformer performance, holding everything else (encoder architecture, SSL objective, evaluation protocol) constant? They compare fixed patches against three beat-synchronous tokenizer designs — resampled beats, adaptive-pooled beats, and resampled beats plus explicit R-R interval timing — on two settings: 12-lead diagnostic classification (PTB-XL after MIMIC-IV-ECG masked pretraining) and single-lead rhythm classification (Icentia11k after patient-level contrastive pretraining). No wrist signal, no PPG, no accelerometer — this is a pure ECG tokenization ablation.

  • Venue: IEEE MLSP 2026 (accepted)
  • Signals: ecg
  • License: arXiv perpetual non-exclusive license
• The Impact of Temporal Context Length and Encoding Strategies on Self-Supervised ECG Representation Learning
  2026 · arXiv preprint

  This study tests how input length and input encoding affect self-supervised ECG models, using single-lead Icentia11k with the same Transformer throughout. It compared 16-second, 1-minute, 5-minute and 10-minute inputs, and continuous patch embeddings against fixed quantized tokens. Longer context beat 16-second snapshots for rhythm detection and patient retrieval, with the 5- and 10-minute models best.

  • Published: arXiv preprint, 2026
  • Data: Icentia11k single-lead ECG
  • Variables: input length (16 s to 10 min), continuous vs VQ tokens

Co-authors