Machine Learning Assisted Real-Time Label-Free SERS Diagnoses of Malignant Pleural Effusion due to Lung Cancer

Abstract

More than half of all pleural effusions are due to malignancy of which lung cancer is the main cause. Pleural effusions can complicate the course of pneumonia, pulmonary tuberculosis, or underlying systemic disease. We explore the application of label-free surface-enhanced Raman spectroscopy (SERS) as a point of care (POC) diagnostic tool to identify if pleural effusions are due to lung cancer or to other causes (controls). Lung cancer samples showed specific SERS spectral signatures such as the position and intensity of the Raman band in different wave number region using a novel silver coated silicon nanopillar (SCSNP) as a SERS substrate. We report a classification accuracy of 85% along with a sensitivity and specificity of 87% and 83%, respectively, for the detection of lung cancer over control pleural fluid samples with a receiver operating characteristics (ROC) area under curve value of 0.93 using a PLS-DA binary classifier to distinguish between lung cancer over control subjects. We have also evaluated discriminative wavenumber bands responsible for the distinction between the two classes with the help of a variable importance in projection (VIP) score. We found that our label-free SERS platform was able to distinguish lung cancer from pleural effusions due to other causes (controls) with higher diagnostic accuracy.

Keywords

• lung cancer
• pleural effusion
• surface-enhanced Raman spectroscopy
• clinical study
• chemometrics
• diagnosis