Enhancing ECG signal classification through pre-trained stacked-CNN embeddings: a transfer learning approach

Area of Science:

• Biomedical Engineering
• Computational Biology
• Artificial Intelligence in Healthcare

Background:

• Accurate electrocardiogram (ECG) classification is vital for clinical decision-making.
• Existing computational models face challenges in balancing performance and efficiency for ECG analysis.

Purpose of the Study:

• To develop an innovative computational framework for optimized ECG classification.
• To combine transfer learning with traditional machine learning for improved efficiency and accuracy.

Main Methods:

• Utilized a pre-trained Stacked Convolutional Neural Network (SCNN) for high-dimensional feature embedding.
• Evaluated embeddings using various machine learning classifiers, including Multilayer Perceptrons (MLPs).
• Employed transfer learning from diverse datasets to enhance classifier discrimination.

Main Results:

• MLPs demonstrated a strong balance of computational efficiency and performance.
• Achieved high F1-scores: 0.94 (multi-class) and 1.00 (binary) on CinC2017; 0.85 (multi-class) and 0.99 (binary) on CPSC2018.
• The proposed framework outperformed existing methods in ECG classification benchmarks.

Conclusions:

• The synergy of deep learning feature extraction and transfer learning offers a robust ECG classification strategy.
• This approach addresses critical research gaps, enhancing efficiency and adaptability in healthcare.
• The findings pave the way for future advancements in automated ECG analysis.