|Case Western Reserve University
|Biomedical Engineering; multi-modal imaging, registration, spectral embedding,alpha-MI
|Full text PDF:
Mutual information (MI) based on image intensity is typically inadequate to accurately co-register multi-modal images that have substantial differences in appearance. Texture features, which are alternative image representations, at different length scales emphasize different tissue structures and also reveal similar underlying structural attributes between multi-modal images. These may not be discernible with intensity representation. However, using texture features as an alternative representation is challenging due to issues such as information redundancy and sub-optimal combination of feature representations. To address these issues, we introduce a novel approach called Multi-scale Spectral Embedding Registration (MSERg) that employs a new dimensionality reduced multi-scale image representations revealing similar attributes across different modalities with minimal redundancy. We compared MSERg with other representation-based registration methods including intensity-based registration, Combined Feature Ensemble Mutual Information and single-scale Spectral Embedding Registration for fusion of prostate MRI and ex vivo digitized histopathology images and establish that MSERg to be the most effective method. Advisors/Committee Members: Madabhushi, Anant (Committee Chair).