Advancing neonatal respiratory distress syndrome diagnosis: A spectroscopic approach and disposable spectrometric silicon chip development

Abstract

Facilitating near-patient rapid diagnosis is vital for personalized precision medicine and improved patient outcomes. Unlike centralised analytical lab tests, which involve complex and laborious processes, rapid biological characterisation informing immediate clinical decisions is essential to improve care of critically ill patients. In cases like neonatal respiratory distress syndrome (nRDS) delays due to current diagnostic approaches, based on conventional lab analysis, exacerbate the disease and result in further harm to pre-term neonates. nRDS is caused by a deficiency in pulmonary surfactant, a mix of lipids and proteins that normally decrease the surface tension in lungs, making it easier to breathe. There is a requirement that nRDS can be detected early, at the point of care, so that effective and timely treatment can be provided. Despite knowing that nRDS can be diagnosed when the ratio of two biomarkers, lecithin(L)/sphingomyelin(S), is less than 2.2 there are no current point of care tests that can measure them within a clinically important two-hour time window. Optical methods such as Raman spectroscopy and attenuated total reflectance (ATR) coupled with Fourier transform infrared spectroscopy (FTIR) can identify and quantify such biomarkers, but current technology is expensive and not conducive to clinical use. This project sought to develop a low-cost, disposable silicon chip that requires a minimal sample volume to measure the concentrations of L and S in a manner consistent with bedside delivery of care. The silicon chip made use of surface enhancement features to increase the observed absorbance signal, permitting the measurement of smaller concentrations than possible with current methods. First, mid infrared spectra of millimolar concentrations of binary mixtures of synthetic pulmonary surfactant lipids were measured in dichloromethane. This measurement approach the baseline measurement and data processing pathway to develop partial least squares regression (PLSR) models with quantified uncertainty around the critical L/S 2.2 region of +0.26,-0.34 moles/mole. This informed a second study which generated a pulmonary surfactant lipids model and took a design of experiments (DoE) approach to generate lipid mixture samples that mapped the physiological space. The ATR-FTIR spectra collected were used to train PLSR models with a quantified uncertainty. Models for sphingomyelin and total phosphatidylcholine were used to predict the L/S ratio with an uncertainty of ±0.3 moles/mole. Shapely additive explanations (SHAP values) were reported and identified key spectral regions for which the enhanced optical platform should be targeted. Third, binary liposomes of L and S were measured using Raman and ATR-FTIR spectroscopies and PLSR models of their spectra were compared with models generated using high- and low-level fused data. It was demonstrated that the fused models were more accurate and precise than those from a single modality. The fourth study developed the design, fabrication and characterisation of the enhanced optical platform using sodium acetate. This demonstrated an absorption peak enhancement of 4.6 times in the target 1000 cm^(-1) to 1800 cm^(-1) region thus validating the design. The fifth study measured dried surfactant lipids extracted from adult humans using ATR-FTIR and verified the similarity of the spectra to synthetic samples previously tested. The sixth study developed a method to correct FTIR spectra that suffer from environmental interference by water and carbon dioxide. The correction was applied to spectra and used to generate a PLSR model which required 3 fewer latent variables when compared to uncorrected data, for models with similar performance, establishing that a more parsimonious model could be generated from applying the environmental correction and still maintain the required model performance. The contributions from this thesis achieve the goals of developing the measurement of analytes using vibrational spectroscopy and promises their application to point of care contexts for better patient healthcare.

More information

Identifiers

ORCID for Senthil Murugan Ganapathy: orcid.org/0000-0002-2733-3273

ORCID for James Wilkinson: orcid.org/0000-0003-4712-1697

Catalogue record

Export record