Computational Methods to Process and Analyze High-dimensional Imaging Mass Cytometry Datasets in Pathological Tissue Samples

Abstract

Highly multiplexed imaging methodologies generate high-dimensional images which allows for in situ spatial assessment of several protein markers in pathological tissue samples. Computational analyses of these images produce information of unprecedented depth and breadth that has been shown to correlate with disease progression and patient prognoses, thus proving valuable in basic science and clinical research. However, processing of such images presents many challenges, in particular the high dimensionality of the data, the presence of tightly clustered cells often difficult to segment, and the identification of cell populations which requires expert knowledge and specialized computational methods. In this thesis, state-of-the-art computational methods for every step of the processing of high-dimensional imaging mass cytometry (IMC) images were employed. Spillover compensation was implemented to improve data quality, cell segmentation delineated single cells in the images, cell phenotyping allowed the identification of cell types and states, and lastly, neighborhood and spatial analysis provided information about spatial organization and cellular interactions. This processing pipeline was applied to two multiplexed IMC datasets acquired from Hunner lesions and pancreatic ductal adenocarcinoma tissue samples. This study represents the first application of multiplexed images to interrogate Hunner lesions to date and provides a novel perspective into the cellular immune microenvironment of these lesions. Following the processing pipeline, the results were individually presented and the outputs from the two datasets were compared for assessing the generalizability of the computational methods. Altogether, our results show that the algorithms were able to effectively extract a depth of biological information from the images and generalize to both datasets.