CORE -- A Cell-Level Coarse-to-Fine Image Registration Engine for Multi-stain Image Alignment

TL;DR

CORE presents a coarse-to-fine, multimodal WSI registration framework that first achieves fast global alignment using prompt-based tissue masks and deep feature matching, then refines alignment at the nuclei level via shape-aware point-set registration and CPD-based non-linear deformation. By coupling tissue-wide robust registration with cell-level precision, CORE generalises across stains (H&E, IHC, PAS, mIF) and datasets, providing accurate nuclei correspondence and efficient computation. The approach demonstrates superior accuracy (lower TRE/rTRE) and faster coarse registration, with effective visualisation and open-source availability, offering a practical foundation for integrated, high-resolution tissue analysis. This work advances computational pathology by enabling reliable cross-modality, cell-level mapping essential for spatial transcriptomics and multi-marker analyses.

Abstract

Accurate and efficient registration of whole slide images (WSIs) is essential for high-resolution, nuclei-level analysis in multi-stained tissue slides. We propose a novel coarse-to-fine framework CORE for accurate nuclei-level registration across diverse multimodal whole-slide image (WSI) datasets. The coarse registration stage leverages prompt-based tissue mask extraction to effectively filter out artefacts and non-tissue regions, followed by global alignment using tissue morphology and ac- celerated dense feature matching with a pre-trained feature extractor. From the coarsely aligned slides, nuclei centroids are detected and subjected to fine-grained rigid registration using a custom, shape-aware point-set registration model. Finally, non-rigid alignment at the cellular level is achieved by estimating a non-linear dis- placement field using Coherent Point Drift (CPD). Our approach benefits from automatically generated nuclei that enhance the accuracy of deformable registra- tion and ensure precise nuclei-level correspondence across modalities. The pro- posed model is evaluated on three publicly available WSI registration datasets, and two private datasets. We show that CORE outperforms current state-of-the-art methods in terms of generalisability, precision, and robustness in bright-field and immunofluorescence microscopy WSIs

Figures (11)

• Figure 1: A high-level overview of the proposed WSI registration method (CORE). (A) Block shows input source and target slides. (B) The coarse registration block illustrates the overlay before transformation, the estimated (rigid+non-rigid) coarse alignment using feature matching, and the resulting coarse-registered overlay. (C) Shows the coarse registered source output and target slides. (D) Using the target and coarsely registered source, nuclei point sets are extracted, and fine shape-aware point set registration is performed to estimate local nuclei-level deformations. (E) The TIAViz visualisation tool enables real-time application of the registration to full-resolution slides.
• Figure 2: Coarse Rigid Registration Block. (A).The preprocessing and tissue mask extraction workflow applies gamma correction, stain normalisation, and prompt-based mask extraction. (B) The TriMorph block computes translation via COM (centre-of-mass) estimation, scale factor and rotation angle from input images and tissue masks. (C) XFeat block estimates semi-dense feature from inputs and perform rigid transform (D) Coarse Non-Rigid registration processing block based on iterative optimisation and adaptive regularisation based deformation estimation.
• Figure 3: Proposed fine shape-aware nuclei point set registration. First coarse displacement field is applied on source WSI resulting in Coarse registered WSI. Nuclei point set are detected from Target and Coarse Registered WSI and then shape aware alignment using nuclei points is performed followed by local deform estimation.
• Figure 4: Multi-scale visualisation of CORE results on ACROBAT dataset. Left column: Low magnification views (0.625$\times$) of the target WSI (H&E), source WSI (IHC), and registered source WSI. Middle column: High magnification views of selected regions (indicated by green boxes) showing detailed tissue structures. Right column: Overlay comparison before and after registration, displayed at both low and high resolution (red boxes), demonstrating improved alignment of tissue structures following registration.
• Figure 5: Multi-scale visualisation of CORE results on ANHIR dataset. Left column: Low magnification views (0.625$\times$) of the target WSI (H&E), source WSI (ER), and registered source WSI. Middle column: High magnification views of selected regions (indicated by green boxes) showing detailed tissue structures. Right column: Overlay comparison before and after registration, displayed at both low and high resolution (red boxes), demonstrating improved alignment of tissue structures following registration.