Venous Component Separation from a Pulse Oximeter Signal: using Blind Source Separation with Ultrasound Imaging as Reference

Abstract

Intradialytic hypotension (IDH) is a major drawback of haemodialysis and occurs in more than every fifth dialysis treatment. There is a need for real-time measurement, with an ability to detect IDH before it occurs. Photoplethysmography (PPG) may provide sufficient information about hemodynamic status to prevent IDH. In particular, the venous component may be useful.

This thesis aims to investigate suitable signal separation methods for venous component extraction, and the wavelengths for optimal venous signal extraction. Furthermore, the effect of different body positions and measurement location-combinations are examined. Moreover, the thesis aims to answer if a better signal can be provided using additional wavelengths as a complement to the two conventional wavelengths used in standard pulse oximeters (I.e., red and infrared). Lastly, potential applications of the venous component are searched for in literature, apart from blood volume dynamics.

For signal separation, principal component analysis (PCA) was, with regards to morphology, found to give similar results as the second-order blind identification (SOBI) for venous extraction. Optimal wavelengths for venous component were established to be somewhere between 500-1000 nm, based on the limitations. The lower limit is due to bad signal to noise ratio and shallow penetration depth, whereas the higher limit was due to water absorption. Regarding penetration depth and absorption, another subject to consider is the effect of melanin and other chromophores absorbing differently for various wavelengths. Based on literature findings and LED-availability this thesis continued to investigate venous component extraction using a customized sensor, provided by Baxter. The sensor used the wavelengths 515 nm (blue green), 577 nm (yellow green), 660 nm (red) and 940 nm (infrared). For reference, ultrasound measurements of nearby blood vessels were conducted simultaneous to PPG-measurements using ultrasound equipment provided by the Department of Biomedical Engineering at LTH. Combinations of PPG and ultrasound reference were done in three different setups: PPG measurements of the thumb, forehead, and neck. Respectively, ultrasound reference was acquired from wrist, external jugularis and internal jugularis. The components from PPG were compared with the ultrasound reference using correlation and visual inspection. Likewise, comparison of separation using only the two conventional versus all four of the available sensor channels was done utilizing correlation. Additionally, a comparison of the ratios of the principal component eigenvalues were done and summarized.

No concluding results could be shown from comparing correlation with ultrasound reference using two conventional wavelengths versus four. Ultrasound images proved to be unexpectedly difficult to obtain and no satisfactory setup of PPG- and ultrasound measurement could be found. At times, signal separation on the PPG-signals did indeed reveal components with venous-like pulsations which correlated well with the ultrasound reference signal, but these results were never consistently acquired. Extraction of the venous component using PPG could be useful as a non-invasive option for Peripheral Intravenous Analysis. Moreover, it might also be used as an estimator of Central Venous Pressure if the measurement location is on the neck.