MRI has been growing as a widely preferred modality for structural brain imaging.
Intensity inhomogeneity is one of the prime obstacles in the automatic MR image processing
algorithms. Retrospective correction strategies for the correction of this artifact
have been found to be more useful in clinical practices in comparison with prospective
methods. The retrospective correction strategies are applied directly on the image and
therefore can correct for intensity inhomogeneity arising from all the possible sources.
The retrospective correction strategies can be broadly classified into filtering, surface
fitting, histogram based and segmentation based methods.
This thesis contributes improved novel retrospective correction focusing on the adult
brain MR images. The methodologies are proposed with specific objectives of reduced
predefined parameters that critically affect the correction accuracy, enhancement of conventional
filtering methods and fully-automatic segmentation based method. Two novel
methodologies have been proposed with minimal user-defined parameters. The first
method has been implemented by formulating the inhomogeneity problem considering
sparsity constraints in suitable transform domains. While in the second method, a
compensation function and a modified diffusion strategy have been developed for intensity
inhomogeneity correction. The performance of conventional filtering methods
have been improved significantly by considering multiscale frameworks with the focus
on reducing edge artifacts. And finally, a segmentation based method has been introduced
which is fully-automatic that requires no manual interaction even for parameter
initialization.
