An end-to-end workflow for multiplexed image processing and analysis

Abstract

Simultaneous profiling of the spatial distributions of multiple biological molecules at single-cell resolution has recently been enabled by the development of highly multiplexed imaging technologies. Extracting and analyzing biologically relevant information contained in complex imaging data requires the use of a diverse set of computational tools and algorithms. Here, we report the development of a userfriendly, customizable, and interoperable workflow for processing and analyzing data generated by highly multiplexed imaging technologies. The steinbock framework supports image pre-processing, segmentation, feature extraction, and standardized data export. Each step is performed in a reproducible fashion. The imcRtools R/Bioconductor package forms the bridge between image processing and single-cell analysis by directly importing data generated by steinbock. The package further supports spatial data analysis and integrates with tools developed within the Bioconductor project. Together, the tools described in this workflow facilitate analyses of multiplexed imaging raw data at the single-cell and spatial level.1 .

Development of a high dimensional imaging mass cytometry panel to investigate spatial organization of tissue microenvironment in formalin-fixed archival clinical tissues

“…Segmentation of raw IMC data was done following the Steinbock framework with Docker container (Deepcell) [ 19 ]. Briefly, a panel.csv file was generated and channels for segmentation (nucleus and cytoplasm/membrane) were chosen.…”

Mapping the breast tumor microenvironment: proximity analysis reveals spatial relationships between macrophage subtypes and metastasis-initiating cancer cells

“…5 of 342 ROIs of poor quality were excluded from analysis. Image segmentation was carried out using a set of software tools within a standardized pipeline 36 based on the following softwares: Jupyter Notebook, CellProfiler 3.1.9 37 Ilastik 1.3.3post3 38 , and HistoCAT 1.76 39 . Stacks of images suitable for all downstream steps in the pipeline were first created using CellProfiler.…”

Segmentation aware probabilistic phenotyping of single-cell spatial protein expression data

“…As input, STARLING requires only the high-dimensional image along with imperfect cell segmentation that are summarized to a per-cell expression profile along with physical cell size in pixels (Fig. 1C) as is commonly output by a range of pipelines for highly multiplexed imaging data 26,27 . After optimization of the loss, STARLING outputs both the denoised cluster identifies representing the cellular phenotypes of the underlying cells in the input biological material, as well as a per-cell segmentation error probability that may be useful for follow-up visualization and analysis (Fig.…”