Accurate registration and failure detection in tissue micro array images

Image-to-image registration is well-studied, but the challenge of automatically validating registration results remains. In many applications, registration performance is still visually inspected. In Tissue microarrays (TMAs), there are usually hundreds of tissues on a single glass slide, making visual inspection almost impossible. In cases where the TMA slide has to be removed from the microscope and replaced (e.g., in sequential staining, temporal studies, and change analysis), the images on the TMA need to be registered to the corresponding images in the previous steps. The registration has to be robust since undetected registration failure will lead to inaccurate results in subsequent analysis. In this work, we present an automated system to register images of corresponding tissue samples in a TMA, detect registration failures accurately, and automatically re-initialize registration in such cases. The system is image-independent, and reduces the number of cases where the images need to be visually validated. The system has been applied in a study of sequential fluorescent staining of tissue samples on a TMA slide, with 99.88% registration accuracy.