Automated Multimodal Correlative Registration for Organelle-Specific Elemental and Isotopic Imaging

Abstract Mapping the subcellular partitioning of therapeutics is important for understanding their trafficking, mechanisms of action, and off-target effects. NanoSIMS provides chemical images of labeled therapeutics, but assigning these signals to cellular structures requires correlation with electron microscopy (EM). This correlation is commonly performed by manual landmark selection and is therefore time-consuming and operator-dependent. Here, we present an automated computational pipeline for registering chemical and ultrastructural images across multiple spatial scales. The method combines bidirectional RAFT optical-flow estimation, confidence-guided affine fitting, and template matching to locate a NanoSIMS field of view within a larger EM map. A morphology-rich ion channel (for example, 32S) is used to estimate the transformation, which is then applied to molecule-specific channels (for example, 79Br or 15N). Agreement with expert-guided manual registration was evaluated using PSNR and SSIM, and the workflow was applied to several cell and tissue specimens. Applications to labeled oligonucleotide and antibody-based therapeutics demonstrate how the registered images can support organelle-level interpretation of molecular distributions in vitro and in vivo.