A Novel DenseSIFT U-NET Based Approach to Perform Kidney Tumor Semantic Segmentation

Kidney tumors pose a significant health concern, particularly in senior age groups, and are a leading cause of morbidity and mortality. Current clinical practices rely on noninvasive methods, such as Computed Tomography (CT), Ultrasound (US), Magnetic Resonance Imaging (MRI), for the identification and diagnosis of kidney tumors. In this research paper, we introduce a novel DenseSIFT integrated U-Net based network for end-to-end kidney and kidney tumor segmentation. The proposed architecture employs DenseSIFT image as an input to the encoder-decoder structure, where the encoder and decoder are interconnected by skip connections and learns robust features to enable accurate segmentation. To validate the effectiveness of our approach, we also conduct performance comparisons with the traditional Vanilla U-Net. The experimental results on the 2019 Kidney and Kidney Tumor Segmentation (KiTS19) challenge dataset demonstrate that our method achieves an impressive mean Intersection over Union (meanIoU) of 91.98%, showcasing its potential for enhanced kidney tumor segmentation.