A Convolutional Neural Network and Transfer Learning Approach for Accelerated Quantitative Mass Spectrometry Imaging.

This research presents a novel approach to accelerate quantitative mass spectrometry imaging (qMSI) measurements using convolutional neural networks (CNNs) with a transfer learning approach. Current methods are time consuming, require numerous specific steps that allow for the introduction of error, and involve meticulous data analysis. The concept is to build a machine learning model that can be used for future qMSI experiments thereby saving time and decreasing variability. In essence, this model is a CNN trained by transfer learning on a smaller dataset of ion images with a known concentration of analyte, which is then applied to future qMSI studies when quantifying the same molecule in the same tissue type. This enhances the speed with which this analysis can be completed and, in theory, decreases the variability seen with previous methods. In this work, we demonstrate that, by using previously collected MSI data and a transfer learning approach, our CNN model allows for the accurate concentration to be determined on new tissues.