Component-trees can be used for the design of image processing methods, and in particular segmentation ones. However, despite their ability to consider various kinds of knowledge and their tractable computation, methodological deadlocks often forbid to efficiently involve them in real applications. In this article, we explore new solutions to some of these deadlocks, and more especially those related to (i) complexity of the structures of interest and (ii) multiple knowledge handling. The usefulness of the proposed strategies is illustrated by preliminary results related to vessel segmentation from 3-D angiographic data.