Advances in computer technologies over the last decade have catalyzed the development of modern computerized schemes for lesion detections in radiological images. One biggest challenge is that computers generally lack sufficient perceptibility and intelligence in terms of discovering pathological patterns; the decision making process is hence hindered. As it is known that knowledge plays an indispensable role in computer vision and artificial intelligence, integrating anatomical knowledge into such computer system holds great promise for facilitating decision making and improving patient care in neuro-radiology.

Based on the assumption that the brain exhibits a high level of bi-fold symmetry and that this symmetry is violated in the presence of pathological conditions, a principle goal of this effort was motivated to construct a symmetry-based paradigm for automatic localization and segmentation of brain lesions. The framework of this methodology is grounded on the hypothesis that the systematic correlation between asymmetry and pathologies can be a key to the improvement of existing detection algorithms. Integrating symmetry/asymmetry information as the prior knowledge or heuristics into a computer aided diagnostic (CAD) system, ought to enhance the system performance in the analysis of brain pathologies.

The methodology of this study is two-fold: First a symmetry axis or, the symmetry plane needs to be spatially oriented because it is valuable for the correction of possible misalignment of radiological scans and for hemisphere-wise asymmetry evaluation. In a second step automatic detection and quantification of brain lesions such as stroke and tumor is required. In this dissertation, I explore and discuss the discriminating capacity of symmetry/asymmetry in the context of extracting pathological findings in various radiological applications with different modalities, such as MRI and CT. In other words, the first part of this research focuses on solving an image registration problem, and the second part relies upon the performance of pattern recognition and segmentation algorithms applied to asymmetry detection.

It should be noted, however, that the methods developed in this thesis for a set of particular neuro-applications may have a more general applicability since many other parts of human body, not only limited to the brain, are highly symmetrical in nature.