FDP-TF: A fast two-pass trend filtering for ECG delineation

ECG delineation, which involves detecting key waveform features such as P, QRS, and T waves, is essential for accurate cardiac monitoring and diagnosis. In a recent study, the authors introduced the Adaptive Trend Filtering (ATF) algorithm, which effectively identifies local extrema in ECG signals for both denoising and delineation, demonstrating strong performance compared to state-of-the-art methods. However, its implementation using the Alternating Direction Method of Multipliers (ADMM) resulted in long execution times, limiting its practical application. This paper presents Fast Double-Pass Trend Filtering (FDP-TF), a complete reformulation of the original ATF algorithm that leads to a significantly faster and novel algorithm. The results show that the novel FDP-TF algorithm attains a speedup of several orders of magnitude with no loss in algorithm performance, thus making it a compelling alternative for fast off-line processing of very long ECG recordings, such as those encountered in the QT dataset, real-time processing of on-line signals, and even implementation in wearable devices. The results also show that this increase in speed is achieved without compromising the performance of the FDP-TF algorithm, which remains on par with current state-of-the-art leading approaches.