An Optimized CNN–U-Net Framework for Kidney Segmentation in T2-Weighted MRI Scans

Segmentation of kidneys with T2-weighted magnetic resonance imaging (MRI) cannot be achieved easily with Chronic Kidney Disease (CKD) because kidney morphology is disordered, and tissue contrast is also low. The article presents a better CNN-U-Net framework of kidney segmentation in T2-weighted MRI. The proposed approach consists of a convolutional feature learning phase and a U-Net encoder-decoder and applies a progressive training to balance the segmentation. An experimental CKD MRI dataset was tested and the results assessed by using Dice coefficient and Intersection over Union (IoU). The skeleton gave a score of 0.72 in Dice score and 0.57 in IoU, and it quintessentially shows a consistent increment over that of the baseline U-Net model. The outcomes are not high but indicate the challenge of CKD MRI data processing and determine the efficacy of the provided optimization strategy. The study is at the applied evaluation of CNN -U-Net-based kidney segmentation of pathological cases using MRI.