Lung cancer detection in a high-risk screening cohort using machine learning-assisted ATR-FTIR spectroscopy of serum

We report lung cancer detection using liquid biopsy. Machine learning-assisted ATR-FTIR spectroscopy of serum achieved 84.6% specificity and 78.6% sensitivity. This method requires no labelling or processing beyond isolating serum from blood. Samples from 52 participants with confirmed lung cancer and 52 samples from participants matched by age, sex and current smoking status were analysed. ATR-FTIR spectroscopy was used to measure 9 μL samples in liquid form. Specificity and sensitivity were investigated using different machine learning classifiers and figures of merit. Combining measured spectra with medical history and a random forest classifier gave optimum performance. Our strategy for improving specificity and sensitivity and for point-of-care integration will be discussed which may lead in the future to use as a triage for patients to determine which patients require further investigation for lung cancer.