Correlator: an AI-driven tool for automated correlative registration of electron microscopy and NanoSIMS imaging
NanoSIMS has enabled chemical imaging of labeled therapeutics, greatly advancing our ability to study subcellular drug distribution. Nevertheless, interpreting these chemical signals critically depends on ultrastructural correlation with electron microscopy—a process that remains manual, labor‑intensive, and a major bottleneck in throughput and precision.
In this work, we introduce Correlator, an automated AI-driven pipeline that overcomes this limitation. Correlator integrates bidirectional optical flow, confidence-guided affine transformation, and automated template matching to robustly align multiscale electron microscopy and NanoSIMS images. It exploits morphology-rich ion channels (e.g., ³²S) to estimate transformations and propagates them to sparse therapeutic signals (e.g., ⁷⁹Br, ¹⁵N), effectively overcoming low signal-to-noise challenges and enabling organelle-precise molecular imaging. Correlator is provided as an easy-to-use ImageJ plugin and a Python package with GPU acceleration.
Please refer to the published paper for more details: https://www.biorxiv.org/content/10.64898/2026.04.30.721814v1.full
Demonstration of image registration with different secondary ion channels
Generalizability of the Correlator across diverse biological specimens
Large-field EM alignment via template matching using NanoSIMS images
Morphology channel-guided correlation for visualizing the nucleic acid therapeutics
Code and Plugin Avaliability
Source code of Python, FIJI plugin, and documents are available in:
Acknowledgement
This work is supported by the Innovation and Technology Fund (MHP/129/22), Hong Kong Research Grant Council General Research Fund (17102722, 17300523, 17302324), National Natural Science Foundation of China (32271445) to H.J; Australian Research Council (ARC) Early Career Industry Fellowship (IE230100042) and the National Health and Medical Research Council (NHMRC), Australia, Ideas Grants (2038782) to K.C.; and a grant from the Leducq Foundation (23CVD02) to S.G.Y. The work was conducted in the JC STEM Lab of Molecular Imaging, funded by The Hong Kong Jockey Club Charities Trust.