We describe a new method for vascular image analysis that incorporates a generic physiological principle to estimate vessel connectivity, which is a key issue in reconstructing complete vascular trees from image data. We follow Murray's hypothesis of the minimum work principle to formulate the problem as an optimization problem. This principle reflects a global property of any vascular network, in contrast to various local geometric properties adopted as constraints previously. We demonstrate the effectiveness of our method using a set of microCT mouse coronary images. It is shown that the performance of our method has a statistically significant improvement over the widely adopted minimum spanning tree methods that rely on local geometric constraints.