Image analysis frameworks for applications in bioinformatics

Images are widely used to represent experimental data in bioinformatics. Examples of these are microarray and 2-D electrophoresis, which are used in research of genomics and proteomics, respectively. Images provided by both methods have certain difficulties for image analysis. Therefore, it motivates computer scientists to study and find out robust computer vision methods to analyze these images.; We present a novel framework for solving the image segmentation problem in microarray images. The framework allows segmentation of multiple objects with similar shape, but different image qualities. In our method, contours are deformed to proper boundaries of spots based on a guiding scheme where contours of strong signal spots are used to control deformation of contours of weak signal spots. We then present an image analysis system for microarray images which utilizes our segmentation framework. The system offers important features such as handling of image alignment, array gridding, spot detection, and intensity extraction. An extension of the proposed framework, which is applicable to more varieties of image problems such as rotation of objects, partial occlusion, and weak edges, is also presented. In this dissertation, we then develop algorithms for spot detection and differential expression in 2-D electrophoresis images, which have different image problems compared to that of microarray images. We thus present a different spot detection method based on the analysis of pixels and regions. For differential expression, we exploit the advantages of a hierarchical method and image correlation to initialize pairs of corresponding spots. We also present an energy-optimization based algorithm to compute spot correspondence when some initial spot pairs are given. The energy incorporates similarity of spot local structure, spot image correlation, as well as overlapping of image constraints. We extensively tested all our methods with synthetic and real images. The results are quantitatively and qualitatively evaluated by comparing results with both the ground truth and existing techniques. The results show that our methods are robust and provide accurate information, useful for further analysis in the field of bioinformatics.