Machine Learning-Based Cell Segmentation Enables Spatial Single-Cell Analysis of the Tumor Immune Microenvironment, Jasper Lee, UG '21 (2308192)
The tumor microenvironment contains a complex collection of immune cells which can infiltrate cancer tissues to varying degrees. Studies into tumor immune cell populations are highly clinically relevant, given that the presence of immune cells has been shown to play a significant role in disease progression and resistance to therapy. Experimental techniques, including the CODEX multiplexed immunofluorescence platform, have been developed in order to determine their location of immune cells at single-cell resolution. However, these efforts are limited by inferior cell segmentation results which negatively affect the quality of downstream analysis. To address this issue, we developed a cell segmentation pipeline capable of highly accurate segmentation across a wide variety of normal and tumor samples. We utilized the NucleAIzer machine learning-based algorithm in order to segment nuclei. Following post-processing in ImageJ, the segmented nuclei were then used as seed points for a secondary watershed segmentation algorithm capable of extracting the whole cell area. Our algorithm effectively segments cells with minimal mis-segmentation events, making this approach useful when a high degree of segmentation accuracy is required. We also developed a method to acquire cell marker intensity information within segmented cell regions, allowing us to assemble a spatial map of marker intensities. This information allows us to accurately probe tumor samples in order to determine the immune cell populations present in the microenvironment. Our computational pipeline will help enable further research into the tumor immune microenvironment, by providing a method for highly accurate cell segmentation and detection of immune cell populations.