In this work, a novel framework for automated, spatially-adaptive adjustment of active contour regularization and data fidelity parameters is proposed and applied for medical image segmentation. The proposed framework is tailored upon the isomorphism observed between these parameters and the eigenvalues of diffusion tensors. Since such eigenvalues reflect the diffusivity of edge regions, we embed this information in regularization and data fidelity parameters by means of entropy-based, spatially-adaptive `heatmaps'. The latter are able to repel an active contour from randomly directed edge regions and guide it towards structured ones. Experiments are conducted on endoscopic as well as mammographic images. The segmentation results demonstrate that the proposed framework bypasses iterations dedicated to false local minima associated with noise, artifacts and inhomogeneities, speeding up contour convergence, whereas it maintains a high segmentation quality.
