Adaptive information processing in hybrid imaging on the cloud

点击翻译按钮获取中文摘要

Background: Hybrid imaging is defined as the fusion of two or more imaging technologies into a single form of imaging. Ideally, this new form is synergistic-that is, more powerful than the sum of its parts. Hybrid imaging has grown from simply fusing two modalities into an independent research area. In the last decade, very wide ranges of hybrid imaging modalities have been becoming routinely available. Typical example would include PET (Positron Emission Tomography)-MRI (Magnetic Resonance Imaging), PET-CT (Computed Tomography), and multiple source radar-sonar images. To realize the full potential of hybrid imaging, diverse kinds of expertise must be brought together. Taking PET-MR-CT for an example: Traditionally PET is interpreted by human expert from nuclear medicine; CT and MRI are interpreted by different human experts from radiology. Hybrid PET-MRI brings significant challenges for human perceptions. This is especially true when dynamic hybrid imaging such as cardiac PET-MR and real time sonar-Radar. While the images provide complementary information, there is lack of computer based intelligence tools to handle this huge amount of data efficiently. This brings significant research opportunities for adaptive information processing platform. Motivation: Our research is motivated by: i) Within the last decade, there has been a growing increase of hybrid image acquisition, which is very challenging for human perception. The specialty of the hybrid images has required that the researcher apply particular methods to analysis rather than extend existing image processing methods. ii) Information extracted from the mutual modality contains valuable insight, which often requires further analysis. As an example, Cardiac CT and MRI images can output important information that can be used to medically diagnose disease, as well as track and monitor the progression of disease and therapy and PET can output muscle on molecular level, which can be used to track metabolism within muscle; and iii) the integrated registration and segmentation methods is motivated by the observation of the reciprocity between registration and segmentation and reciprocity between different modalities. OBJECTIVES: The main long term objective is to develop and evaluate an comprehensive adaptive information processing computer software platform for hybrid imaging. The system is able greatly reduce the current workload in the practice, further improve the accuracy; The secondary objective is to establish a publicly accessible hybrid imaging database for benchmarking to help the research community and to improve the current state-of-art of research in hybrid image analysis. Short term objectives: 1) Research and develop an intelligent PET-MRI-CT image processing software system, which covers image fusion, integrated segmentation and registration and information extraction system; 2) Research and develop multiple information based focus-of-attention mechanisms for generating region of interest (ROI) from the hybrid images segmentation and registration for further analysis; 3) Collect and share 50 cases of hybrid image database. SIGNIFICANCE: To the best of our knowledge, this is the first attempt to create a comprehensive information processing system for hybrid imaging. The proposed short term objective: PET-MR-CT information processing system leverage the existing leading facility in the institute, combined with the technical strengths of the group, fused with the user knowledge inherited in the group. The proposed methodology leverages the cutting edge techniques currently being developed in the group and fused with other leading technologies in the field, move towards a multidisciplinary, integrated research program.