Atlas-based semi-automatic kidney tumor detection and segmentation in CT images

摘要

Manual segmentation of tumors in a medical image is difficult and time-consuming, while automatic segmentation which does not need interactions has many challenges due to poor contrast between different regions. We therefore present a semi-automatic kidney tumor detection and segmentation method. Our method firstly segments the kidney from the whole image using single atlas based segmentation. Then we apply a supervoxel segmentation to generate an over segmentation, and estimate the abnormal probability of each voxel respectively, using the segmentation result above. We visualize the probability by using the volume rendering technique, highlighting the abnormal regions, which is easy to observe and select for users. This method is more reliable and flexible comparing to automatic methods.