Enhancing watershed segmentation of touching and weakly-connected features in biomedical images.

We report how to improve and control watershed segmentation of touching features in optical-microscope images of immunochemically stained cells and in three-dimensional (3D) reconstruction of weak-connected components in transmission electron microscopy images from ultra-thin slices of compact chromatin clumps, from rat lymphocytes. Our approach includes image processing of the distance-transform domain, and a discrete-boundary formulation of morphological operators, to speed up the 3D watershed segmentation. The adjustment for connectivity criteria, as well as other tuning parameters, come from the Nyquist sampling criterion, applied to spatial resolution, and are obtained from biological considerations, such as the average size of a normal cell. We also combined both enhancements in 3D and present the mathematical background as well as visual results.