Bridging the Feature Gaps for Retrieval of Multi-Dimensional Images

摘要

Content-based image retrieval (CBIR) refers to the use of visual features for images retrieval, and has become an attractive approach to managing biomedical image achieves. However, existing CBIR systems are typically designed for 2D single-modality imaging, and are restricted when multi-dimensional images are considered. With the advances in imaging scanners, image complexity and magnitude have rapidly expanded in both the temporal (time) and spatial (space) dimensions, i.e., dynamic PET provides physiological functions of the human body acquired in 3D volumes over multiple time sequences, and dual-modality imaging scanners that sequentially acquires co-aligned functional (PET) and anatomical (CT) images. This manuscript summarizes research in CBIR of multi-dimensional biomedical images with focuses on the feature extraction and retrieval techniques that utilize the information available in the image抯 multidimensional data spaces. Applications of multi-dimensional CBIR will be exemplified with our ongoing studies with 4D dynamic PET and dual-modal PET/CT images.