CPP-UNet: Combined Pyramid Pooling Modules in the U-Net Network for Kidney, Tumor and Cyst Segmentation

Abstract

Renal carcinoma stands prominently as a significant contributor to global cancer-related mortality rates, highlighting the critical importance of early detection and diagnosis in the management of this ailment. Moreover, the rising incidence of kidney tumors poses a challenge in differentiating between malignant and benign lesions using radiographic methods. Therefore, we present CPP-UNet, an innovative convolutional neural network-based architecture designed for the segmentation of renal structures, including the kidneys themselves and renal masses (cysts and tumors), in a computed tomography (CT) scan. Particularly, we investigate the fusion of the Pyramid Pooling Module (PPM) and Atrous Spatial Pyramid Pooling (ASPP) for improving the UNet network by integrating contextual information across multiple scales. Our proposed method yielded promising outcomes in the Kidney and Kidney Tumor Segmentation challenge (KiTS21 and KiTS23) datasets, exhibiting Dice indices of 93.51% and 92.84% for Kidneys and Masses, 90.33% and 92.08% for Renal Masses, and 85.69% and 88.17% for Tumors, respectively.