Untargeted Swab Touch Spray-Mass Spectrometry Analysis with Machine Learning for On-Site Breast Surgical Margin Assessment

Direct sampling mass spectrometry (MS) has rapidly advanced with the development of ambient ionization MS techniques. Swab touch-spray (TS)-MS has shown promise for rapid clinical diagnostics. However, commercially available swabs are notorious for their high background signals, particularly in the positive ionization mode. Although changes to MS methods or precleaning of the swabs can serve as workarounds, this inherent issue still limits the clinical application of swab TS-MS. In this study, we report the use of the sterile-packaged OmniSwab as an alternative material for untargeted swab TS-MS analysis. As a proof of concept, breast surgical margins were swabbed in vivo during surgeries and analyzed using a compact mass spectrometer within the hospital. Subsequently, various machine learning algorithms were applied to the acquired MS spectra to determine the optimal model for classifying margins as normal or tumor. The Least Absolute Shrinkage and Selection Operator (LASSO) model yielded the highest prediction performance, with accuracies exceeding 90% in both testing and validation data sets. Notably, three out of four surgical margins involved with cancer cells were accurately identified. The entire workflow, from swab TS-MS analysis to margin prediction, can be completed within 5 min with high accuracy, demonstrating the feasibility of swab TS-MS to assist intraoperative decision-making.