• Madsen, Jens
• Postle, Anthony
• Vincent Veluthandath, Aneesh
• Clark, Howard W.
• Senthil Murugan, Ganapathy
• Ahmed, Waseem
• Wilkinson, James

Abstract

Point of care devices present an attractive proposition for a rapid, evidenced based, diagnosis to be provided at the patient bedside, and give clinicians access to almost real-time information about a patient's condition.Devices based on attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIR) can provide rapid, label free measurements consistent with delivery of bedside care.Neonatal respiratory distress syndrome (nRDS) affects some pre-term neonates from their first breath and delays in treatment are associated with poor clinical outcomes.nRDS can be diagnosed by analysis of the lecithin/sphingomyelin ratio (L/S ratio) of the lung surfactant obtained from bronchoalveolar lavage.Following on from our work on mid-infrared spectroscopy for the diagnosis of nRDS, where we established a data processing methodology to evaluate machine learning algorithms used for determining L/S ratios of simple mixtures, this work develops the process, by increasing the number of constituents and using smaller calibration steps to more closely match the patient sample. We will show the performance of machine/deep learning algorithms to predict the concentrations of the constituents present and their L/S ratio along with prediction intervals indicating the uncertainty in the measurement.The results will further inform calibration procedures for a proof-of-principal ATR-FTIR based point-of-care device that can be used in a clinical setting to provide a rapid indication of the L/S ratio of patient samples.

Topics

• Fourier transform infrared spectroscopy
• surfactant
• machine learning