Kidney and Tumor Segmentation using U-Net Deep Learning Model

Abstract

Medical Image Segmentation is a challenging field in the area of Computer Vision. In this paper U-Net deep learning model was used for semantic segmentation. The reason for shortlisting U-Net was its suitability on small data set and also it was originally designed for Biomedical Image segmentation process. Visual representations of the predicted results have shown promising results using U-Net. Experimental results were computed on two different cases. Case No 1, includes testing the method on images for which labelled information was available and considering only those slices where the presence of kidney was detected. Case No 2, involves testing the method on those images who’s labelled information was not available and applying the method on all the CT slices with respect to a patient. Experimental results was based on a metric called IOU (Intersection over Union) score which is one of the most commonly used metric in semantic segmentation.

Keywords: Kidney-tumor segmentation, U-Net model, IOU Dice score , Deep Learning Representation Learning