Sampling Frequency Has Architecture-Dependent Effects on ECG Deep Learning Classification: CNNs Are Robust, Transformers Benefit Modestly

This study investigates how varying ECG sampling frequencies (100, 300, and 500 Hz) affect the classification performance and computational costs of convolutional neural networks (CNNs) and Patch Transformers. The results indicate that CNNs maintain stable performance across different frequencies, while Patch Transformers show modest improvements with higher frequencies but at increased computational costs. The findings suggest that ECG acquisition strategies should consider both the task requirements and the architecture used, especially in resource-limited environments.