Machine Learning -Assisted Detection of Respiratory Disease by Vibrational Spectroscopy.

Abstract

Respiratory infectious diseases have a high pathogenic potential due to their great infectious capacity. In the case of seasonal flu, it is responsible for between 3 and 5 million cases of severe illness and 290,000 to 650,000 deaths worldwide each year. Current in vitro diagnostic methods are reliant on the detection of specific molecules or biomarkers for clinical diagnosis. This project analyses the potential of using spectroscopy techniques for the diagnosis of infectious respiratory diseases, assuming that these diseases present a measurable physiological change in the whole body. Raman and absorbance spectra of plasma samples from subjects under different clinical conditions have been measured. Firstly, the different spectra have been analysed separately; secondly, the Raman and absorbance spectra were combined to increase the number of differential features. Using these data, a machine learning-assisted multivariate analysis was performed to generate diagnostic classifiers. The trained classifiers demonstrate a high diagnostic capacity both to discriminate healthy subjects from diseased patients and to differentiate between the infectious respiratory diseases studied. These results show the potential of using machine learning-assisted spectroscopy techniques to develop a rapid, low-cost and non-invasive diagnostic method.