Improved peaks automatic detection algorithm of noisy quasi-periodic physiological signals

The automatic multiscale-based peak detection (AMPD) method for quasi-periodic signals suffers from high computational complexity, substantial memory requirements, and occasional peak detection failures. To address these limitations, this paper proposes an Improved AMPD (IAMPD) algorithm that significantly enhances computational speed while maintaining strong noise robustness. The improvement in efficiency is attained through a frequency-informed scale constraint that narrows the search space using prior knowledge of the signal's frequency range, greatly reducing redundant computations. This is combined with an optimized computational process that integrates redundant operations and minimizes intermediate caching. Extensive evaluations on both simulated signals with varying signal-to-noise ratios (SNRs) and real physiological data demonstrate the superior performance of IAMPD. The algorithm achieves a speedup of over 160 times compared to the original AMPD while maintaining perfect detection performance across all evaluation metrics: Sensitivity (Se), Positive Predictivity (+P), and F1-score (F1), each consistently attaining 100% even at 0 dB SNR, thereby outperforming other benchmark methods.Importantly, IAMPD preserves the parameter-free advantage of AMPD, requiring only an estimated frequency range. Experimental results confirm its effectiveness in accurately detecting peaks in various physiological signals including heart sounds, electrocardiogram (ECG), pulse, and respiratory signals, making it well-suited for real-time applications.