Leveraging unsupervised segmentation for semi-supervised renal calculi and carcinoma segmentation and classification

Abstract

Indeed it became crucial to develop an AI-driven system for detecting Renal illnesses spontaneously due to a healthcare issue of Renal failure. The global shortage of nephrologists is the core reason for this. This research concerns two crucial diseases: Renal Calculi and Carcinoma by using semi-supervised localization and unsupervised segmentation of a total 5477 CT images of axial point of view in order to use it in commonly used models. The gathered CT images are prepared by resizing and balancing via augmentation and the analysis shows that the mean color distribution of images was the same across all classes. We have used YoloV8 for kidney localization, quick shift and label merge for segmentation. Also, we manually annotated all the images with the supervision of a medical expert. Furthermore, we compare both datasets using VGG19, AlexNet, ResNet50. We got better output in Semi- Supervised localized and unsupervised segmented images rather than the annotated images via experts in terms of classification. We believe that reducing and focusing on the region of interest can easily achieve good output in commonly used models which will be very time efficient and cost effective as well.