Browsing by Subject "Attributed relational graphs"

Filter results by typing the first few letters
Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemOpen Access
    Perceptual watersheds for cell segmentation in fluorescence microscopy images
    (2012) Arslan, Salim
    High content screening aims to analyze complex biological systems and collect quantitative data via automated microscopy imaging to improve the quality of molecular cellular biology research in means of speed and accuracy. More rapid and accurate high-throughput screening becomes possible with advances in automated microscopy image analysis, for which cell segmentation commonly constitutes the core step. Since the performance of cell segmentation directly a ects the output of the system, it is of great importance to develop e ective segmentation algorithms. Although there exist several promising methods for segmenting monolayer isolated and less con uent cells, it still remains an open problem to segment more con uent cells that grow in aggregates on layers. In order to address this problem, we propose a new marker-controlled watershed algorithm that incorporates human perception into segmentation. This incorporation is in the form of how a human locates a cell by identifying its correct boundaries and piecing these boundaries together to form the cell. For this purpose, our proposed watershed algorithm de nes four di erent types of primitives to represent di erent types of boundaries (left, right, top, and bottom) and constructs an attributed relational graph on these primitives to represent their spatial relations. Then, it reduces the marker identi cation problem to the problem of nding prede ned structural patterns in the constructed graph. Moreover, it makes use of the boundary primitives to guide the ooding process in the watershed algorithm. Working with uorescence microscopy images, our experiments demonstrate that the proposed algorithm results in locating better markers and obtaining better cell boundaries for both less and more con uent cells, compared to previous cell segmentation algorithms.