An automatically initialized level-set approach for the segmentation of proteomics images

This work introduces an automatically initialized level-set approach for the segmentation of proteomics images. Level-set methods are natural candidates for this segmentation task, since their topological adaptability is a prerequisite to identify the boundaries of thousands of protein spots present. However, they require manual initialization, whereas their convergence is highly sensitive to the position of the initial contour. The proposed initialization process is based on the formation of a level-set surface of multiple cones, which are centered at regional intensity maxima positions, associated with protein spots. This initialization process, as is the case with the subsequent contour evolution, is aided by means of histogram equalization and morphological processing. The experimental results indicate that the proposed level-set approach facilitates quicker convergence than the one obtained by the straightforward application of the Chan-Vese model. In addition, the identified spot boundaries are more plausible than the ones obtained by the application of state-of-the-art proteomics image analysis software.