Towards Fast and High-quality Biomedical Image Reconstruction

Department of Computer Science and EngineeringReconstruction is an important module in the image analysis pipeline with purposes of isolating the majority of meaningful information that hidden inside the acquired data. The term ???reconstruction??? can be understood and subdivided in several specific tasks in different modalities. For example, in biomedical imaging, such as Computed Tomography (CT), Magnetic Resonance Image (MRI), that term stands for the transformation from the, possibly fully or under-sampled, spectral domains (sinogram for CT and k-space for MRI) to the visible image domains. Or, in connectomics, people usually refer it to segmentation (reconstructing the semantic contact between neuronal connections) or denoising (reconstructing the clean image). In this dissertation research, I will describe a set of my contributed algorithms from conventional to state-of-the-art deep learning methods, with a transition at the data-driven dictionary learning approaches that tackle the reconstruction problems in various image analysis tasks.clos