Interactive learning based nodule detection in ct lung volumes

We present a novel method to automatically detect lung nodules in CT lung scans. Our method is generalized in the sense that it does not assume/depend a particular organ or a particular nodule type. hence it does not require an organ segmentation. We test our method in a challenging set (Anode09) that is comprised of low dose CT scans which include all types of nodules (solid, ground glass opacity, juxta-fissural, juxta-vascular) of less than 10mm in size. Our method produces 8 false positives per scan for true positive rate of 52%, which is comparable to the first 6 results from the contest.