Akademik Veri Yönetim Sistemi
SIGNAL IMAGE AND VIDEO PROCESSING, cilt.19, sa.18, 2025 (SCI-Expanded, Scopus)
Detection of kidney stones is an essential aspect of medical diagnosis, considering the growing incidence of nephrolithiasis worldwide and its potential for developing serious complications in case of lack of treatment. This work presents a two-stage deep learning pipeline for automatic kidney stone detection on a large-scale, augmented, and annotated dataset of 35,457 axial CT images, divided into stone and non-stone cases. In the first stage, a U-Net-based convolutional neural network was trained for segmenting the kidney region from the CT images. The dataset was hand-annotated by expert radiologists by delineating the kidney regions, and their binary masks were employed as the ground truth for supervised segmentation. The segmentation model that was trained showed a high Mask Intersection over Union (Mask IoU) of 0.97, reflecting high precision in kidney region detection. The trained segmentation model was employed for automatically extracting the kidney regions of interest from all images for downstream analysis. In the second stage, cropped regions from the segmented kidney areas were used as input for a tailored Convolutional Neural Network (CNN) to classify the presence or absence of kidney stones. This classification model resulted in high precision of 97.5% with localised kidney region consideration, reflecting its efficacy in improved diagnosis with precision. Through the use of high-quality expert annotations, large-scale data augmentation, and modular deep learning, this work presents an efficient, high-throughput, and clinically pertinent solution for automated screening of kidney stones in CT scans. The suggested system can be helpful for reducing the interpretation time and diagnostic variability in real-world clinical settings.