Optimized cross-scale feature matching for microscopic image registration of breast cancer immunohistochemical sections

Immunohistochemical staining is pivotal in pathology. For breast cancer biomarkers (ER/HER2/PR), spatial discrepancies in consecutive sections compromise quantitative analysis and clinical interpretation. To overcome this issue, we present an efficient cross-scale feature matching framework for registering IHC stained images of human breast cancer tissue sections. The method first performs rigid registration on low-magnification images using feature correspondence and localized affine transformations. These transformation parameters are then mapped onto high-magnification counterparts, enabling accurate alignment with minimal computational overhead. We further provide quantitative evaluations of the matching process and demonstrate a diverse set of successful registration cases. The results confirm that the proposed framework delivers robust and efficient performance across varying scales and evaluation metrics.