Applications of U-net to diffuse optical tomography data: Image reconstruction and superresolution

Di use optical tomography (DOT) is being investigated for effective functional brain imaging. It serves as a cheaper, less bulky and safer alternative to fMRI imaging which is the current gold standard. However, due to the complicated nature of the system, analytical reconstruction approaches begin to fail. Currently functional signal activations are reconstructed using an analytical approach but the quality of the image, especially in terms of resolution, is low. This is one of the primary barriers to making DOT the gold standard for functional brain imaging. This thesis presents the theory behind the inverse problem and discusses possible solutions to image reconstruction and superresolution problems in the DOT brain imaging space. With the growth in deep learning and its applications in medical imaging, a U-net based architecture is proposed to learn the mapping and estimate a higher resolution image. This work shows that the proposed deep learning model trained on simulated images from real-world fMRI images of the human brain can reconstruct higher resolution images while reducing the number of hallucinations.U of I OnlyAuthor requested U of Illinois access only (OA after 2yrs) in Vireo ETD syste