Image analysis and statistical modeling for applications in cytometry and bioprocess control

Abstract

Today, signal processing has a central role in many of the advancements in systems biology. Modern signal processing is required to provide efficient computational solutions to unravel complex problems that are either arduous or impossible to obtain using conventional approaches. For example, imaging-based high-throughput experiments enable cells to be examined at even subcellular level yielding huge amount of image data. Cytometry is an integral part of such experiments and involves measurement of different cell parameters which requires extraction of quantitative experimental values from cell microscopy images. In order to do that for such large number of images, fast and accurate automated image analysis methods are required. In another example, modeling of bioprocesses and their scale-up is a challenging task where different scales have different parameters and often there are more variables than the available number of observations thus requiring special methodology. In many biomedical cell microscopy studies, it is necessary to analyze the images at single cell or even subcellular level since owing to the heterogeneity of cell populations the population-averaged measurements are often inconclusive. Moreover, the emergence of imaging-based high-content screening experiments, especially for drug design, has put single cell analysis at the forefront since it is required to study the dynamics of single-cell gene expressions for tracking and quantification of cell phenotypic variations. The ability to perform single cell analysis depends on the accuracy of image segmentation in detecting individual cells from images. However, clumping of cells at both nuclei and cytoplasm level hinders accurate cell image segmentation. Part of this thesis work concentrates on developing accurate automated methods for segmentation of bright field as well as multichannel fluorescence microscopy images of cells with an emphasis on clump splitting so that cells are separated from each other as well as from background. The complexity in bioprocess development and control crave for the usage of computational modeling and data analysis approaches for process optimization and scale-up. This is also asserted by the fact that obtaining a priori knowledge needed for the development of traditional scale-up criteria may at times be difficult. Moreover, employment of efficient process modeling may provide the added advantage of automatic identification of influential control parameters. Determination of the values of the identified parameters and the ability to predict them at different scales help in process control and in achieving their scale-up. Bioprocess modeling and control can also benefit from single cell analysis where the latter could add a new dimension to the former once imaging-based in-line sensors allow for monitoring of key variables governing the processes. In this thesis we exploited signal processing techniques for statistical modeling of bioprocess and its scale-up as well as for development of fully automated methods for biomedical cell microscopy image segmentation beginning from image pre-processing and initial segmentation to clump splitting and image post-processing with the goal to facilitate the high-throughput analysis. In order to highlight the contribution of this work, we present three application case studies where we applied the developed methods to solve the problems of cell image segmentation and bioprocess modeling and scale-up.