Use of serial line-mapped infrared spectra across a heterogeneously dried urine sample enhances component-discriminability and accuracy in identifying pancreatic cancer

초록

A strategy of utilizing spatially-resolved line-mapped infrared (IR) spectra across a heterogeneously dried urine sample has been demonstrated as a tool for enhancing component-discriminability to identify pancreatic cancer (PCa). Since urine metabolites were heterogeneously distributed within a dried urine sample on a substrate by so called a coffee-ring effect, line-mapped spectra over the sample could contain more diverse and characteristic spectral features related to PCa. With this feasibility, 53 and 54 urine samples obtained from healthy subjects and patients with PCa, respectively, were prepared and, along with the individual line-mapped spectra of each sample, their average spectra (emulating a bulk spectroscopic sampling) were also employed for discrimination between the normal and cancer groups using machine learning models. Classes corresponding to each of the line-mapped spectra of a sample were determined and a final class of the sample was uniquely decided by vote of the determined classes. The discrimination accuracy based on Random Forest (RF) using the line-mapped spectra was 90.0%, superior to 85.8% using the average spectra. The 4th, 5th, and 6th principal component loadings of the line-mapped spectra differed from those of the average spectra owing to the recognition of detail component-discriminant features in the line-mapped spectra, thereby yielding the improved accuracy. The demonstrated scheme is further applicable to enhance the component-discriminability in other spectroscopic discriminant analysis of samples with complex composition.

키워드