Batch-Agnostic Self-Supervised Discovery of Tissue Niches in Multiplexed Imaging

Abstract

Multiplexed imaging enables rich, high-resolution characterization of the tumor microenvironment in intact tissues. However, analysis of such data traditionally requires cell segmentation and phenotyping pipelines, which can be labor-intensive and error prone. In this thesis, we present BASSL-MI, a batch-agnostic, self-supervised framework for discovering tissue niches directly from multiplexed imaging data. BASSL-MI operates on image patches, eliminating the need for explicit cell segmentation and phenotyping. Furthermore, built on a modified contrastive block disentanglement architecture, BASSL-MI learns dual latent representations that separate biologically informative features from batch-dependent factors through spatially guided augmentations and batch-invariance objectives, mitigating image-, sample-, or batch-specific artifacts. We validate BASSL-MI on a 56-marker colorectal cancer CODEX dataset and demonstrate that BASSL-MI trained embeddings markedly reduce image-specific variability and recover biologically interpretable spatial niches. Notably, it uncovers CD20- rich follicular regions that are associated with improved patient survival, outperforming published findings from cell segmentation-driven analysis. Our results demonstrate that self- supervised, patch-based learning captures clinically relevant spatial organization within the tumor microenvironment, providing a scalable, automated alternative to traditional cell-centric analysis.