Efficient Kidney Tumor Classification and Segmentation with U-Net

Abstract

A novel approach to kidney tumor identification is introduced, which integrates kidney tumor classification and segmentation algorithms. The multi-faceted approach commences with the classification of kidney images, adaptly discerning between normal and tumor instances. For individuals identified as tumor-positive, a sophisticated UNet-based architecture is intricately employed to achieve precise segmentation, capturing nuanced details of both kidney and tumor regions. Many models, including VGG16, MobileNetV3, DenseNet50, and others, were tested in order to achieve this. Among these, MobilenetV3 performs better than the others in terms of accuracy, with a 99.1% accuracy rate and a 99% precision rate for classification. In this research, we applied a novel U-Net model to accurately segregate kidney and kidney tumor from CT scan data. With this, an average dice coefficient score of 0.9445 is obtained. In advancing the landscape of kidney tumor analysis, this proposed strategy not only bridges classification and segmentation but also showcases a significant leap toward refined clinical applications.