Tubular structure extraction from (bio)medical images is a prerequisite for many applications in bio- engineering. The two main issues related to such extraction, namely denoising and segmentation, are generally handled independently and sequentially, by dedicated methods. In this article, we model the problem of tubular structure extraction in an optimization framework and we show how a Hessian- based vesselness measure can be embedded in the formulation, allowing in particular robust vessel extraction. Preliminary experiments of this method, on synthetic and retinal images, emphasise the potential usefulness of this approach.