Reproduce “Beat-ssl: Capturing Local ECG Morphology through Heartbeat-level Contrastive Learning with Soft Targets” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Muhammad Ilham Rizqyawan, Peter Macfarlane, Stathis Hadjidemetriou, Fani Deligianni
The BEAT-SSL framework leverages dual-context contrastive learning with novel soft targets to effectively learn representations from unlabelled ECG data, emphasizing both global rhythm and local heartbeat morphology. It achieves 93% of a large foundation model's performance in multilabel classification and surpasses other state-of-the-art methods by 4% in ECG segmentation tasks, demonstrating its capability in capturing fine-grained local features.
Key insights
- Optimal performance is achieved by combining rhythm-level `soft_1` contrasting (emphasizing strong similarities with an exponent of 50) and heartbeat-level `hard` contrasting, highlighting the synergistic benefits of dual-context learning.
- Soft targets effectively capture the continuous nature of feature similarity in ECGs, providing a more refined signal for contrastive learning compared to binary targets.
- Emphasizing heartbeat-level contrasting is crucial for learning fine-grained local morphological features, which are vital for precise ECG wave segmentation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.