VABAM, a generative framework decoupling waveform shape and amplitude, outperformed existing methods across multiple benchmark datasets for controlled physiological signal generation.
The VABAM generative framework successfully decouples waveform shape and amplitude in physiological signals like ECGs, enabling targeted amplitude modulation and improved signal synthesis.
Pulsatile physiological signals, such as arterial blood pressure and electrocardiograms, encode cardiovascular dynamics through rhythmic variations in waveform shape and amplitude. Controlled synthesis of such signals is critical for advancing physiological understanding and clinical applications. However, most existing generative methods represent waveform shape and amplitude in a single, mixed form. This coupling constrains the ability to adjust one without affecting the other, thereby limiting controllability and interpretability in signal generation. We present VABAM, a generative framework that operates on a single physiological signal to decouple waveform shape and amplitude through cascaded filtering. This decoupling enables targeted amplitude modulation while preserving waveform shape. To assess the synthesis quality, we introduce four metrics that quantify waveform shape factorization, shape preservation, amplitude modulation controllability, and spectral similarity, alongside conventional reconstruction accuracy. Across multiple benchmark datasets, VABAM outperforms existing methods, demonstrating the significance of waveform shape-amplitude decoupling for controlled physiological signal generation. This may enable amplitude-targeted augmentation, uncertainty-quantified prediction, and enhanced real-time anomaly monitoring, thereby advancing clinical decision-making in physiological signal analysis.
Kim et al. (2026) studied Pulsatile physiological signals (arterial blood pressure, electrocardiograms). VABAM (generative framework) vs. Existing generative methods was evaluated on Waveform shape factorization, shape preservation, amplitude modulation controllability, spectral similarity, and reconstruction accuracy. VABAM, a generative framework decoupling waveform shape and amplitude, outperformed existing methods across multiple benchmark datasets for controlled physiological signal generation.