A Synthetic-Data-Driven LSTM Framework for Tracing Cardiac Pulsation in Optical Signals

This dataset accompanies the paper "A Synthetic-Data-Driven LSTM Framework for Tracing Cardiac Pulsation in Optical Signals" (https://doi.org/10.1364/BOE.574286). It contains physiologically realistic synthetic time-series data designed to train and evaluate deep learning models for cardiac pulsation tracing and motion artifact removal.Signals were generated using parameterized models of optical pulsatile waveforms (NIRS, PPG, DCS) and augmented with diverse artifacts including amplitude modulations, spikes, oscillations, baseline drifts, and noise sources. The data are organized into three pulse morphology types (single peak, closer double peaks, and further double peaks) to reflect physiological variability. All signals are sampled at 50 Hz, with segment lengths of 60 seconds.