Ecg Classification Eval

ecg-classification-eval

Convolutional Fourier Analysis Network (CFAN): A Unified Time-Frequency Approach for ECG Classification — Jeong et al. (2025) (arXiv:2502.00497, 2025)

What this evaluates

Evaluates the ability of deep learning architectures to accurately classify electrocardiogram (ECG) recordings into predefined physiological or pathological categories. The benchmark probes joint time-frequency feature extraction capabilities by comparing models that embed Fourier analysis directly into convolutional layers against traditional signal processing and baseline CNN approaches.

Datasets

• MIT-BIH — total ?; splits: train (-1), val (-1), test (-1)
• ECG-ID — total ?; splits: train (-1), val (-1), test (-1)
• Apnea-ECG — total ?; splits: train (-1), val (-1), test (-1)

Metrics

• accuracy (primary) — range: percent
  • The proportion of correctly classified ECG signals out of the total number of signals in the evaluation set. Calculated as (number of correct predictions) / (total number of predictions) and reported as a percentage.

Input / output format

Input: 1D ECG time-series signals (typically preprocessed, normalized, and segmented into fixed-length windows for the CNN architecture).

Output: Predicted class label for each ECG recording.

Scoring recipe

def calculate_accuracy(predictions, gold_labels):
    correct = sum(1 for p, g in zip(predictions, gold_labels) if p == g)
    return (correct / len(gold_labels)) * 100

Common pitfalls

• The provided text does not specify the exact train/val/test split ratios or cross-validation fold configuration, only mentioning 'per fold' timing in the results table.
• Statistical significance testing (p ≤ 0.02) is reported to validate improvements over baselines, but the specific statistical test (e.g., paired t-test, Wilcoxon signed-rank) is not detailed in the snippet.

Evidence (verbatim from paper)

CFAN outperforms state-of-the-art methods (SPECT, CNN1D, FFT1D, CNN1D-FAN) across ECG classification tasks—achieving 98.95% accuracy on MIT-BIH, 96.83% on ECG-ID, and 95.01% on Apnea-ECG—with statistically significant improvements (p ≤ 0.02) over the second-best method in two tasks.

Citation

@misc{jeong2025cfan,
  title={Convolutional Fourier Analysis Network (CFAN): A Unified Time-Frequency Approach for ECG Classification},
  author={Jeong et al. (2025)},
  year={2025},
  note={arXiv:2502.00497}
}

• arXiv: 2502.00497