Super-resolution based image enhancement for magnetic resonance imaging

• Main Author: Chilla, Geetha Soujanya
• Other Authors: Manojit Pramanik
• Format: Theses and Dissertations
• Language: English
• Published: 2017
• Subjects: DRNTU::Engineering::Chemical engineering
• Online Access: http://hdl.handle.net/10356/69629
• Institution: Nanyang Technological University

Magnetic Resonance Imaging (MRI) is a versatile modality which is widely used for anatomical and physiological imaging in applications including oncology, neuroimaging, and cardiac angiography - to name a few. However, due to patient motion, hardware limitations and constraints on acquisition time, anisotropic images are acquired, sacrificing detail in through-plane. Multiplane super-resolution is a post-processing technique that uses data from multiplane anisotropic acquisition and reconstructs higher resolution MR images. However, certain gaps exist in the application of super-resolution in the context of MRI and research presented in this thesis attempts to address some of the unexplored areas in terms of clinical applications as well as applicability of current state-of-art super-resolution frameworks. The focus of this thesis is on the application of super-resolution for resolution improvement of MR images and development of a novel super-resolution framework that broadens its applicability to other MR applications. For this, three main studies were performed. In the first study, super-resolution was employed to enhance DWI resolution in the context of prostate cancer assessment. In this study, MRI data of 25 patients were acquired, isotropically reconstructed using super-resolution and analysed. Since it is unclear how through plane affects prostate cancer assessment, impact of through plane resolution improvement on prostate cancer diagnosis has been investigated. From the study, super-resolution reconstructions have been found to have increased SNR, sharpness and reduced volumetric error compared to anisotropic acquisitions. These isotropic super-resolution reconstructions have been also found to have increased accuracy for detection of clinically significant cancers and improved visualization in through-plane. In the second study, super-resolution reconstructions of DWI and super-resolution reconstructions of T2WI were fused to improve delineation of tumor and visualization. In this study, it was found out that fusion of super-resolution reconstructions of DWI and T2WI improved tumor coverage and visualization in through and oblique planes, which could potentially benefit targeted biopsy. In the third study, a novel SR framework, DIRSR, was designed and developed in which deformable registration is added to registration framework, along with rigid registration. Most super-resolution implementations use rigid registration, making the SR framework prone to artifacts in applications with non-rigid deformation/motion including lungs and heart. Using DIRSR, isotropic volumes were reconstructed from lung data and were evaluated. Results show significant improvement in alignment of images compared to Rigid-Registration SR frameworks. Reconstructions from DIRSR were also found to have increased sharpness and reduced error compared to Rigid-Registration SR framework and super-resolution using pre-registered inputs. In summary, the studies presented in the thesis indicate that SR techniques can be beneficial for clinical assessment and diagnosis, including MR guided biopsy.