In this paper, we propose a deep learning-based framework for detecting and delimiting kidney stones in CT images using YOLOv11. The model was trained on the publicly available Kaggle dataset [1], annotated with bounding boxes, and validated on 50 real CT scans collected under clinical conditions at Odorheiu Secuiesc Municipal Hospital. The method we present is a combination of CNN detection and threshold-based post-processing for segmentation. We experimented with various hyperparameter configurations and various YOLOv11 sizes to find the most suitable option. Our best-performing model (YOLOv11x) achieved a precision of 0.861 and a mAP@50 of 0.831 on the test set. Furthermore, its clinical diagnostic efficiency was 92.5%. The segmentation findings were validated by a radiologist and proved to be clinically applicable. The proposed approach demonstrates good performance and generalization across datasets, suggesting its applicability in real-world diagnostic scenarios.