AI Research Examines ECG Sampling Frequency Impact on Atrial Fibrillation Detection Models
A comprehensive benchmark study investigated the influence of different electrocardiogram sampling frequencies on deep learning models designed for atrial fibrillation detection. Utilizing 12-lead, 10-second recordings from the PTB-XL dataset, researchers resampled the data to frequencies of 62, 100, 250, and 500 Hz. They examined two architectures: a traditional 1-D Convolutional Neural Network and a hybrid CNN-Long Short-Term Memory model. Performance was evaluated through a stringent patient-safe cross-validation framework. Results indicated that detection metrics are significantly affected by sampling frequency in an architecture-dependent manner. The hybrid CNN-LSTM model excelled at intermediate frequencies of 100-250 Hz, while the 1-D CNN exhibited varied performance across frequencies. This study, identified as arXiv:2604.16437v1, highlights the need to better understand how diverse ECG datasets impact model performance and reliability.
Key facts
- Deep learning models for atrial fibrillation detection are trained on heterogeneous ECG datasets with varying sampling frequencies
- The study conducted a systematic benchmark using 12-lead, 10-second recordings from the PTB-XL dataset
- Recordings were resampled to target frequencies of 62, 100, 250, and 500 Hz
- Two architectures were evaluated: a standard 1-D CNN and a hybrid CNN-LSTM model
- Evaluation used a rigorous patient-safe cross-validation framework
- Sampling frequency significantly impacts detection metrics in an architecture-dependent manner
- The hybrid CNN-LSTM model demonstrated optimal performance at intermediate frequencies (100-250 Hz)
- The 1-D CNN baseline exhibited different performance characteristics across frequencies
Entities
—