Biological image analysis of model organisms


Model organisms are perhaps the most important experimental subjects in biology nowadays. These specimens have characteristics such as short life cycles and compact genome sequences that make them suitable for a number of experimental techniques at different biological levels. They are widely used to explore biology fundamentals and bioengineering products. Due to its importance, there is a need for methods to process the ever increasing amount of biological image data related to model organism research. In this PhD microscopic images of two model organisms, the C. elegans nematode and the A. thaliana plant, are studied. Biological images are inherently difficult to process due to noise, blur, clutter and optical effects. The main problem when sample measurements must be extracted either automatically or semi-automatically is specimen detection. The focus in this PhD has been the usability of several image processing tools that incorporate prior knowledge into the image processing chain. The first problem addressed is the detection of elongated specimens laying in isolation. For this, differential geometry and scale space principles are proposed to describe and detect linear objects. Differential geometry allows us to characterize the shape of an image surface in terms of image derivatives. Scale space provides a mathematical framework to describe image features according to the scale of the observation. As a result a set of features is proposed to detect individual specimens.Another part of this PhD research concerns with the extraction of statistics from adult C. elegans nematode populations imaged at low magnifications. The active contour framework is used to extract shape evidence and use it for detection.After convergence contour energies can be related to image and geometrical properties of the segmented object. We propose a detection technique that exploits this characteristic to extract a sample of individuals located in clusters. Population measurements are computed from these samples and compared to manual measurements.Finally, we concentrate on the problem of segmenting cells of A. thaliana epidermal tissue. These samples are imaged using Differential Interference Contrast technique. To generate sufficient image data, we utilize a revolving stage and different focus settings. A technique to enhance the cellular wall is proposed. It is demonstrated experimentally that the enhanced image can be effectively processed using a well-established segmentation algorithm.