Beyond image alignment: Challenges and emerging solutions for MSI-histology registration in multimodal spatial omics

Multimodal spatial omics integrates molecular distributions, cellular states, and tissue morphology within a shared spatial coordinate system, but reliable cross-modal registration remains a major barrier to reproducible quantitative analysis. Spatial omics modalities such as mass spectrometry imaging, spatial transcriptomics, spatial proteomics, and multiplexed imaging differ fundamentally from histology and medical imaging in signal domains, sampling strategies, and spatial resolution. These intrinsic gaps, together with section deformation, partial tissue loss, field-of-view mismatch, and batch effects, make visually plausible alignment insufficient for quantitative reliability.

This review advances the view that registration in multimodal spatial omics functions as an analytical calibration step rather than a purely visual preprocessing task. We organize the methodological discussion around four connected dimensions. First, we examine error sources under weak and partial correspondence and their consequences for region-of-interest quantification, tissue compartment assignment, and cross-section integration. Second, we discuss strategies for constructing cross-modal comparability through ion image representations, low-dimensional embeddings, tissue-domain and graph-based representations, semantic segmentation, and shared representation spaces. Third, we review similarity modeling choices, including mutual information, structural consistency, and semantic consistency, and discuss their practical scope across data types. Fourth, we summarize transformation and optimization frameworks, ranging from rigid or affine initialization to controlled nonrigid refinement, with explicit consideration of tearing, folding, missing regions, and incomplete overlap.

Emphasizing validation and quality control based on converging evidence from geometric accuracy, structural attribution, and downstream analytical consistency, this review provides a practical methodological map and evaluation framework for multimodal spatial omics. Aligning method selection with analytical objectives supports more reliable integration and interpretation in high-throughput spatial omics workflows.