Analyzing images of biological specimen to extract relevant spatial information can be quite difficult. At the most basic level, an image is a grid of pixels, and each pixel holds a set of values either for display (e.g., RGB values in bright field images) or for individual biomarker intensities (e.g., fluorescent intensity values per channel in multiplex fluorescent images). Using this grid of values, we then need to determine where individual cells are, assigning a group of pixels and their associated values to the cell, in order to perform cell-based spatial analysis.
To do this, the Enable Medicine Insights Team has developed a standard workflow for analysis of multiplexed fluorescent images of tissue.
Overview of standard analysis workflow.
In brief, after acquisition of and uploading the images onto the platform, we perform cell segmentation, a process that allows the pixel-by-pixel biomarker signal to be abstracted to the cell level. We then perform a cellular level QC, filtering out segmentation artifacts (i.e., objects that are detected by the segmentation algorithm that are not real cells). Using the remaining detected cells, we then label the cells by phenotype. The phenotype-labeled cells can then be used in more detailed spatial analysis such as pairwise or interaction analysis or Cell Neighborhood analysis. Comparisons of these quantitative metrics can be made through the Explorer while qualitative observations can be highlighted in the Visualizer, to generate biological insights.
The following links will provide more detail about each portion of the workflow:
Phenotyping: Unsupervised Clustering
Spatial Analysis: Neighbor Distance
Spatial Analysis: Cellular Neighborhoods
Spatial Analysis: Cell-Cell Interactions
Written by Maha Rahim and Amy Lam
Edited February 28, 2023