Segmentation algorithms in spatial transcriptomics imaging play define the boundaries of cells or regions from which gene expression is quantified. Errors or variability in segmentation, such as over-segmentation, under-segmentation, or misassignment of transcripts, can substantially distort downstream analyses, including cell type identification, spatial pattern detection, and many others. Despite this impact, AFAIK, there is currently no widely accepted standard for assessing segmentation quality, making it difficult to compare methods or ensure reproducibility across studies. I see two key challenges: 1) developing robust quality metrics based on downstream data to evaluate segmentation performance, and 2) improving the segmentation algorithms themselves to better capture true biological structure. This gap suggests a valuable opportunity to explore these issues through systematic data exploration and benchmarking during a hackathon. Tackling this challenge can also inform what aspects of class definitions and interpretability with software does the segmentation (often in Python)
Segmentation algorithms in spatial transcriptomics imaging play define the boundaries of cells or regions from which gene expression is quantified. Errors or variability in segmentation, such as over-segmentation, under-segmentation, or misassignment of transcripts, can substantially distort downstream analyses, including cell type identification, spatial pattern detection, and many others. Despite this impact, AFAIK, there is currently no widely accepted standard for assessing segmentation quality, making it difficult to compare methods or ensure reproducibility across studies. I see two key challenges: 1) developing robust quality metrics based on downstream data to evaluate segmentation performance, and 2) improving the segmentation algorithms themselves to better capture true biological structure. This gap suggests a valuable opportunity to explore these issues through systematic data exploration and benchmarking during a hackathon. Tackling this challenge can also inform what aspects of class definitions and interpretability with software does the segmentation (often in Python)