An Interactive Technique of Fast Vertebral Segmentation for Computed Tomography Images with Bone Metastasis

Abstract

Computer-aided diagnosis (CAD) system can assistant radiologists to diagnose bone metastasis, which not only reduces burden on workload but also improves diagnostic accuracy. As key step in CAD system, vertebral segmentation can directly affect diagnostic results. In order to obtain high accurate segmentation results, we propose a connected component Labeled Graph Cuts (LGC) algorithm. The proposed method is tested on 100 computed tomography (CT) slices. The assessed quantitatively of experimental results is compared with those by radiologist. The proposed method has a 96.72[%] of True Positive Rate (TPR), and 1.84[%] of False Positive Rate (FPR), which have better performance than conventional Graph Cuts algorithm, 90.07[%] of TPR and 2.32[%] of FPR.