Kidney tumor segmentation using an ensembling multi-stage deep learning\n approach. A contribution to the KiTS19 challenge

Abstract

Precise characterization of the kidney and kidney tumor characteristics is of\noutmost importance in the context of kidney cancer treatment, especially for\nnephron sparing surgery which requires a precise localization of the tissues to\nbe removed. The need for accurate and automatic delineation tools is at the\norigin of the KiTS19 challenge. It aims at accelerating the research and\ndevelopment in this field to aid prognosis and treatment planning by providing\na characterized dataset of 300 CT scans to be segmented. To address the\nchallenge, we proposed an automatic, multi-stage, 2.5D deep learning-based\nsegmentation approach based on Residual UNet framework. An ensembling operation\nis added at the end to combine prediction results from previous stages reducing\nthe variance between single models. Our neural network segmentation algorithm\nreaches a mean Dice score of 0.96 and 0.74 for kidney and kidney tumors,\nrespectively on 90 unseen test cases. The results obtained are promising and\ncould be improved by incorporating prior knowledge about the benign cysts that\nregularly lower the tumor segmentation results.\n