Table 2:

Image analysis challenges nominated by workshop participants.

Challenge	Problem Statement	Goal
Image registration	Deformations in tissue that arise in iterative staining approaches (e.g., CyCIF, 4i, IHC, MiFish) complicate alignment.	Align images in a “z” stack (e.g., across rounds of staining or serial sections) to maximize the correlation of nuclei stains between consecutive images in the stack.
Automated marker gating	Sample-sample variation in marker intensity owing to batch- and tissue-specific effects complicates efforts to compare markers in multiplex images across tissues and time points.	Automatically and consistently “gate” markers across tissues (i.e., to differentiate between expressed and unexpressed levels of marker expression). Predictions will be compared to manually-curated gating.
Quantification of epithelial polarity in organoids	Cells in organoids self organize and become polarized: specific proteins and organelles are oriented towards the center of the gland, whereas other cell components are expressed at the periphery. De-polarization may be associated with cancer.	Define a metric of organoid polarity and obtain values of this metric for each organoid in the provided data.
Cell segmentation	Developing a pipeline to segment nuclei and other cellular components (cell boundaries, subcellular structures, etc.) has traditionally been slow and incremental.	Develop general guidelines for successfully optimizing segmentation algorithms via a competition. Teams will attempt to develop optimized segmentation algorithms for four different segmentation tasks within the hackathon.
Quantification of PD1 asymmetry in immune cells	PD1 localization is asymmetric on a subset of cells during T-cell-tumor cell interactions and this asymmetry is thought to be related to T-cell activation.	Develop a suitable metric and analytical pipeline for measuring the amount of polarization of membrane-bound markers (e.g. PD1).
Rare cell type identification in noisy data	Known biology dictates the existence of certain rare cell subpopulations, but standard clustering methods fail to detect these robustly.	Given a cell type definition (specified as marker expression), accurately find all cells of that type.
Cell type inference from morphology and spatial distribution	Morphological characteristics may be relevant to discerning cell types but have not yet been used in cell type calling.	Predict cell types (defined by characteristic marker expression) using morphological features (area, perimeter, eccentricity, etc.) and cell-cell proximity information (encoded as an undirected graph).

Bolded challenges were selected for the hackathon.