A Bayesian Multi-Scale Framework for Photoplethysmogram Imaging Waveform Processing

Abstract

Photoplethysmography imaging (PPGI) is an increasingly popular
technique for remotely creating signals with a plethora of medical
information, referred to as PPGI waveforms. However, PPGI waveforms
are often heavily affected by illumination variation and motion
artefacts. Current PPGI waveform processing methods are useful
for estimating heart rate, however, structural detail is not preserved,
rendering the signal incapable of providing additional medical information.
For this reason, we propose a multi-scale framework based
on the Bayesian residual transform which aims to suppress noise
and preserve structural details necessary for extracting cardiovascular
information beyond the scope of heart rate. Experiments conducted
on a dataset consisting of 24 different PPGI waveforms and
corresponding PPG waveforms captured via a finger pulse oximeter
suggests a high level of noise and ambient illumination variation
suppression is achieved while signal fidelity is largely retained.