Evaluating the Generalizability of Predictive Classifiers Built from DESI Imaging Lipid Data across Mass Spectrometry Platforms.

Abstract

In this study, we evaluate the generalizability of predictive classifiers built from DESI lipid data for thyroid fine needle aspiration (FNA) biopsy analysis and classification using two high-performance mass spectrometers (time-of-flight and orbitrap) suited with different DESI imaging sources operated by different users. The molecular profiles obtained from thyroid samples with the different platforms presented similar trends, although specific differences in ion abundances were observed. When using a previously published statistical model built to discriminate thyroid cancer from benign thyroid tissues to predict on a new independent data set obtained, agreement for 24 of the 30 samples across the imaging platforms was achieved. We also tested the classifier on six clinical FNAs and obtained agreement between the predictive results and clinical diagnosis for the different conditions. Altogether, our results provide evidence that statistical classifiers generated from DESI lipid data are applicable across different high-resolution mass spectrometry platforms for thyroid FNA classification.