Abtract

Thesis summarizes the developement of a Visual Analytics tool to understand the behaviour of Human Neuro-Stem cells, in terms of their motility (motion) and mitosis (reproduction) under different experimental environments during the course of their life cycle . The input images are microscopy images taken with Green Fluorescent Protein marker in the cells. A novel approach in terms of Visual Analysis and cell boundary extraction has been developed to accommodate various unique features that are displayed by stem cells in their nascent stage. Firstly, shape of the stem cells varies widely between cells at a given time and the same cell as it grows over time. The variation in fluorescent content causes the changes in brightness of images making it difficult to bracket the cell and non-cell region based on brightness values. The motility (spatial movement of cells) is not constant either, further complicating the tracking. In the system a biologist can mark the regions containing cells and the system automatically detects the cells in that region using a set of parameters (such as minimum brightness and size) using Watershed Algorithm. The parameters for the detection can be changed by the user very easily. Using these initial parameters, the systems predicts the region possibly containing the cell in the next image in sequence. It then uses a predictor corrector loop (by varying parameters to watershed) to delineate the cells from its background. Further, definite patterns occurring, leading to cell division, are exploited to predict and mark cell mitosis. In essence, we develop a two phase system, which initially segments and tracks cells, and then allows for visually analysing their behaviour using tools such as mark up frames and hyper-linked graphs to gather higher level biologically semantic data. The system implements an end-to-end pipeline from acquisition of images to building a statistical model of the cells’ life cycle from the microscopy images. The implementation is out-of-core in the sense that the volume of data processed (typically ranging in hundreds of gigabytes) does not affect

1

the performance as, only a part of the dataset is loaded in memory for segmentation, tracking and analysis.