Cuff-less Arterial Blood Pressure Waveform Synthesis from Single-site PPG using Transformer & Frequency-domain Learning

We propose two novel purpose-built deep learning (DL) models for synthesis ofthe arterial blood pressure (ABP) waveform in a cuff-less manner, using asingle-site photoplethysmography (PPG) signal. We utilize the public UCIdataset on cuff-less blood pressure (CLBP) estimation to train and evaluate ourDL models. Firstly, we implement a transformer model that incorporatespositional encoding, multi-head attention, layer normalization, and dropouttechniques, and synthesizes the ABP waveform with a mean absolute error (MAE)of 14. Secondly, we implement a frequency-domain (FD) learning approach wherewe first obtain the discrete cosine transform (DCT) coefficients of the PPG andABP signals corresponding to two cardiac cycles, and then learn alinear/non-linear (L/NL) regression between them. We learn that the FD L/NLregression model outperforms the transformer model by achieving an MAE of 11.87and 8.01, for diastolic blood pressure (DBP) and systolic blood pressure (SBP),respectively. Our FD L/NL regression model also fulfills the AAMI criterion ofutilizing data from more than 85 subjects, and achieves grade B by the BHScriterion.