Kidney tumor is one of the diseases threatening human health. Ultrasound is widely applied in kidney tumor diagnosis due to its high popularization, low price and no radiation. Accurate segmentation of kidney tumor is the basis of precise treatment. Kidney tumors often grow in the middle of cortex, so that segmentation is easy disturbed by nearby organs. Besides, ultrasound images own low contrast and large speckle, leading to difficult segmentation. This paper proposed a novel kidney tumor segmentation method in ultrasound images using adaptive sub-regional evolution level set models (ASLSM). Regions of interest are firstly divided into subareas. Secondly, object function is designed by integrating inside and outside energy and gradient, in which the ratio of these two parts are adjusted adaptively. Thirdly, ASLSM adapts convolution radius and curvature according to centroid principle and similarity inside and outside zero level set. Hausdorff distance (HD) of (8.75 ± 4.21) mm, mean absolute distance (MAD) of (3.26 ± 1.69) mm, dice-coefficient (DICE) of 0.93 ± 0.03 were obtained in the experiment. Compared with traditional ultrasound segmentation method, ASLSM is more accurate in kidney tumor segmentation. ASLSM may offer convenience for doctor to locate and diagnose kidney tumor in the future.
The background of abdominal computed tomography (CT) images is complex, and kidney tumors have different shapes, sizes and unclear edges. Consequently, the segmentation methods applying to the whole CT images are often unable to effectively segment the kidney tumors. To solve these problems, this paper proposes a multi-scale network based on cascaded 3D U-Net and DeepLabV3+ for kidney tumor segmentation, which uses atrous convolution feature pyramid to adaptively control receptive field. Through the fusion of high-level and low-level features, the segmented edges of large tumors and the segmentation accuracies of small tumors are effectively improved. A total of 210 CT data published by Kits2019 were used for five-fold cross validation, and 30 CT volume data collected from Suzhou Science and Technology Town Hospital were independently tested by trained segmentation models. The results of five-fold cross validation experiments showed that the Dice coefficient, sensitivity and precision were 0.796 2 ± 0.274 1, 0.824 5 ± 0.276 3, and 0.805 1 ± 0.284 0, respectively. On the external test set, the Dice coefficient, sensitivity and precision were 0.817 2 ± 0.110 0, 0.829 6 ± 0.150 7, and 0.831 8 ± 0.116 8, respectively. The results show a great improvement in the segmentation accuracy compared with other semantic segmentation methods.