MCMICRO: a scalable, modular image-processing pipeline for multiplexed tissue imaging

Abstract

Highly multiplexed tissue imaging makes detailed molecular analysis of single cells possible in a preserved spatial context. However, reproducible analysis of large multichannel images poses a substantial computational challenge. Here, we describe a modular and open-source computational pipeline, MCMICRO, for performing the sequential steps needed to transform whole-slide images into single-cell data. We demonstrate the use of MCMICRO on tissue and tumor images acquired using multiple imaging platforms, thereby providing a solid foundation for the continued development of tissue imaging software.

UnMICST: Deep learning with real augmentation for robust segmentation of highly multiplexed images of human tissues

“…Representative cores for lung adenocarcinoma, non-neoplastic small intestine, normal prostate, colon adenocarcinoma, glioblastoma, non-neoplastic ovary, and tonsil were extracted from image mosaics and down-sampled by a factor of 2 to match the pixel size of images routinely acquired and analyzed in MCMICRO 33 . Images were then cropped to 256 x 256-pixel tiles, and in-focus DNA and NES were imported into Adobe Photoshop to facilitate human annotation of nuclear boundaries.…”

“…Nuclear pleomorphism (variation in nuclear size and shape) is even used in histopathology to grade cancers 32 . To account for variation in nuclear morphology we generated training, validation, and test datasets from seven different tissue and tumor types (lung adenocarcinoma, non-neoplastic small intestine, normal prostate, colon adenocarcinoma, glioblastoma, non-neoplastic ovary, and tonsil) found in 12 cores from EMIT (Exemplar Microscopy Images of Tissue 33 , RRID : SCR_021052), a tissue microarray assembled from clinical discards. The tissues had cells with nuclear morphologies ranging from mixtures of cells that were large vs. small, round cells vs. narrow, and densely and irregularly packed vs. organized in clusters.…”

“…Images were segmented using the UnMICST-U model trained on DNA with NES data and real augmentations. Mean fluorescence intensities across 45 markers for 27,847 segmented nuclei were quantified as described in 33 . E-cadherin positive and CD45 positive cells were identified using Gaussian-mixture models on log-transformed data.…”

Stitching and registering highly multiplexed whole slide images of tissues and tumors using ASHLAR

“…High-plex mosaic images represent the key “Level 2 or 3” data type for all subsequent visualization and quantitative data analysis. The data level concept was introduced by dbGAP for genomics (Tryka et al, 2014) and its implementation to tissue imaging is described in detail in the MITI guidelines (Schapiro et al, 2022a). In this context, “data levels” denote different degrees of data processing, with Level 1 corresponding to single, raw image tiles, Levels 2 data to stitched, illumination corrected mosaics and Level 3 to mosaic images that have also been subjected to manual or automated quality control to improve interpretability and accuracy.…”

“…We describe ASHLAR’s design and implementation and compare its performance to existing tools using high-plex CyCIF images. ASHLAR is available as a Docker or Singularity container and has been incorporated into MCMICRO (Schapiro et al, 2022a), the Nextflow-based image processing pipeline developed by HTAN; as part of MCMICRO, ASHLAR has been tested with several hundred CyCIF, CODEX and mxIF images acquired from 12 types of mouse and human tissues at seven different institutions on five different microscopes and slide scanner platforms ( Supplementary Table 1 ). ASHLAR is therefore a robust and practical tool for use with diverse spatial profiling methods.…”