Region and Learning Based Retrieval for Multi-modality Medical Images

We present a region-based image retrieval framework for multi-modality, positron emission tomography computed tomography (PET-CT), images. An image retrieval system can be used to assist the diagnostic process, by providing reference cases that contain similar scans to the interpreting clinicians. PET-CT scans are essential tools for the accurate staging of lung cancer and provide co-registered functional (PET) and anatomical (CT) information from a single scan; the complexity of these data, however, place new challenges in computerized image analysis and retrieval. The choice of a region-based method was inspired by the objective of retrieving images with similar patterns of disease involvement, where there is a parenchymal lung tumor and disease in regional lymph nodes. Our results on clinical data from lung cancer patients show a higher retrieval precision over the usual techniques and the other non-region based methods.