Elena Kalcheva-Yovkova, Technical University, Sofia
Abstract: This paper represents some results of a hybrid technique for kidney segmentation from axial computed tomographic images. The proposed technique combines edge based, region based and morphological watershed based algorithms with prior anatomical knowledge of the human abdomen. It includes kidney region extraction, coarse watershed based segmentation and adaptive refinement. The approach is tested on several CT datasets in DlCOM standard. The results of the segmentation of the left and right kidneys are quantitatively assessed using supervised evaluation technique including similarity, consistency and distance measures. The average values of specificity, sensitivity and accuracy are relatively high (above 0,96) and mean value of Dice coefficient is about 0, 95. The global consistency error values alter between 0,02 and 0,05 and the average Hausdorff distance is around 0,06.
Keywords: Computerized tomography images, contrast enhancement, morphological gradient, watershed segmentation, supervised evaluation.