Radiofrequency ablation is being used increasingly in the treatment of liver tumors, and the detection of local recurrences on follow-up imaging is an important and occasionally challenging task. When the tumor is not associated with nodular enhancement, recurrence detection only relies on a precise identification of shape changes in the post-ablation area. In order to better characterize subtle shape changes, we present a computer-aided diagnosis tool based on a semi-automatic segmentation of CT data with a watershed algorithm. The 3D moment of inertia of the segmented area is computed and provides a quantitative criterion for the study of post-ablation changes over time. Preliminary results on two clinical cases demonstrate that our tool can effectively improve the radiologist’s ability to detect early tumor recurrence.