Abstract
In this proof-of-principle study, we systematically studied the potential of Raman spectroscopy for detecting pre-analytical delays in blood serum samples. Spectra from 330 samples from a liver cirrhosis cohort were acquired over the course of eight days, stored one day at room temperature, and stored subsequently at 4 °C. The spectra were then used to train Convolutional Neural Networks (CNN) to predict the delay to sample examination. We achieved 90% accuracy for binary classification of the serum samples in the groups "without delay" versus "delayed". Spectra recorded on the first day could be distinguished clearly from all subsequent measurements. Distinguishing between spectra taken in the range from the second to the last day seems to be possible as well, but currently, with an accuracy of approximately 70% only. Importantly, filtering out the fluorescent background significantly reduces the precision of detection.