Automated Single Cell Phenotyping of Time-of-Flight Secondary Ion Mass Spectrometry Tissue Images

Abstract

Existing analytical techniques are being improved or applied in new ways to profile the tissue microenvironment (TME) to better understand the role of cells in disease research. Fully understanding the complex interactions between cells of many different types and functions is often slowed by the intense data analysis required. Multiplexed Ion Beam Imaging (MIBI) has been developed to simultaneously characterize 50+ cell types and their functions within the TME with a subcellular spatial resolution, but this results in complex data sets that are challenging to qualitatively analyze. Deep Learning (DL) techniques were used to build the MIBIsight workflow, which can process images containing thousands of cells into easily digestible reports and plots to enable researchers to easily summarize data sets in a study and make informed conclusions. Here we present the three types of DL models that have been trained with annotated MIBI images that have been pathologist validated as well as the associated workflow for the evolution of raw mass spectral data into actionable reports and plots.