Key result
CNNs maintain robust ECG classification across 100–500 Hz, while Patch Transformers improve at higher frequencies.
Why the study?
Electrocardiograms operate at varying sampling frequencies that influence bandwidth, storage, battery life, and potentially deep learning classification performance and computational cost.
Does ECG sampling frequency impact deep learning classification performance and computational cost across CNN and Patch Transformer architectures?
Does ECG sampling frequency impact deep learning classification performance and computational cost across CNN and Patch Transformer architectures?
CNNs demonstrate robustness to lower ECG sampling frequencies for diagnostic classification, whereas Patch Transformers benefit modestly from higher temporal resolution at increased computational cost.
Answers are grounded in this paper and cite their sources.
CNNs may suffice at lower ECG sampling frequencies to limit costs; leaves open whether Patch Transformers merit higher rates in practice.
BACKGROUND Electrocardiograms are deployed across a wide range of environments and frequently operate at varying sampling frequencies, directly influencing bandwidth consumption, storage demands, and battery life. OBJECTIVE This study aims to evaluate the impact of ECG sampling frequency on both classification performance and computational cost across convolutional neural network (CNN) and Patch Transformer architectures. METHODS Using the PTB-XL dataset, 15,000 eligible 12-lead ECGs were analyzed at 100, 300, and 500 Hz. A CNN and a Patch Transformer were trained for multilabel classification of five diagnostic superclasses: normal, myocardial infarction, ST/T-wave change, conduction disturbance, and hypertrophy. Performance was assessed along with training time, memory use, and inference latency. RESULTS CNN performance was relatively stable across sampling frequencies, with macro ROC-AUC values of 0.895, 0.886, and 0.881 at 100, 300, and 500 Hz, respectively. In contrast, the Patch Transformer showed modest improvement with increasing frequency, with macro ROC-AUC increasing from 0.856 at 100 Hz to 0.864 at 300 Hz and 0.877 at 500 Hz. Higher sampling frequencies increased computational cost for both architectures, with longer training times, greater memory use, and slower inference. Sensitivity and learning-curve analyses supported architecture-dependent trends across the tested configurations. CONCLUSION ECG sampling frequency had modest, architecture-dependent effects on deep learning classification performance. CNNs were robust across 100-500 Hz, whereas the Patch Transformer benefited incrementally from higher temporal resolution at greater computational cost. These findings support a task- and architecture-aware approach to ECG acquisition in resource-constrained settings.
No takes yet. Share an insight, caveat, or question.
Freedman et al. (2026) studied Electrocardiogram classification (n=15,000). Varying ECG sampling frequencies (100, 300, and 500 Hz) vs. Comparison across different frequencies and architectures (CNN vs Patch Transformer) was evaluated on Multilabel classification performance for five diagnostic superclasses measured by macro ROC-AUC. Convolutional neural network classification was robust across 100-500 Hz, whereas the Patch Transformer improved incrementally at higher frequencies.
Synapse has enriched 10 related papers. Consider them for comparative context: