Kidney Segmentation with LinkNetB7

Abstract

Cancer is a deadly disease for which early diagnosis is very important. Cancer can occur in many organs and tissues. Renal cell carcinoma (RCC) is the most common and deadly form of kidney cancer. When diagnosing the disease, segmentation of the corresponding organ on the image can help experts make decisions. With artificial intelligence supported decision support systems, experts will be able to achieve faster and more successful results in the diagnosis of kidney cancer. In this sense, segmentation of kidneys on computed tomography images (CT) will contribute to the diagnosis process. Segmentation can be done manually by experts or by methods such as artificial intelligence and image processing. The main advantages of these methods are that they do not involve human error in the diagnostic process and have almost no cost. In studies of kidney segmentation with artificial intelligence, 3d deep learning models are used in the literature. These methods require more training time than 2d models. There are also studies where 2d models are more successful than 3d models in organs that are easier to segment on the image. In this study, the LinkNetB7 model, which has not been previously used in renal segmentation studies, was modified and used. The study achieved a dice coefficient of 97.20%, precision of 97.30%, sensitivity of 97%, and recall of 97%. As a result of the study, LinknetB7 was found to be applicable in kidney segmentation. Although it is a 2d model, it is more successful than UNet3d and some other 2d models.