Computer-Aided Diagnosis (CAD) systems have become common tools to assist medical professionals with accurate and fast diagnostic decisions. In particular, imaging methods are highly demanding on the quantity of generated data, where the output is volumetric data often together with multimodality, multiple contrast, or multiple time scans. In this contribution, we present our CAD pipeline for analysis of spine lesions in volumetric CT images, which helps physicians diagnose the disease, determine the appropriate treatment, and also observe its effectiveness during follow-up examinations. We showed that image segmentation steps are an integral part of CAD systems, and moreover, deep-learning methods outperform standard image processing ones and are able to work on the human precision level.