Automatic detection of abnormalities or lesions that are attached to other anatomies in medical image analysis is always a very challenge problem, especially when the lesions are small. In this paper a novel method for the automatic detection of lung cancers or nodules attached to vessels in high-resolution multi-slice CT images is presented. We propose to use volume projection analysis to mimic physicians’ practices in making diagnosis. The volume projection analysis is performed on 1-dimensional curves obtained from the 3-dimensional volume. A multi-scale detection framework is proposed to detect nodules of various sizes. A set of features for characterizing nodules is defined. Results of experimental evaluation of the method are presented.