In this work we present a new segmentation method named Smart Cage Active Contours (SCAC) that combines a parametrized active contour framework named Cage Active Contours (CAC), based on a ne trans- formations, with Active Shape Models (ASM). Our method e ectively restricts the shapes the evolving contours can take without the need of the training images to be manually landmarked. We apply our method to segment the caudate nuclei subcortical structure of a set of 40 subjects in magnetic resonance brain images, with promising results.