The study compared four deep learning models for the automatic detection of kidney stones in CT images – YOLOv8, YOLOv5, Faster R-CNN and RetinaNet. The research included 4,000 annotated CT slices from 170 patients. Faster R-CNN achieved the highest localization accuracy with mAP@0.5 = 0.93, while YOLOv8 achieved mAP@0.91. YOLOv8 has proven to be the best for clinical practice due to the optimal balance between high detection accuracy and fast computational efficiency, enabling real-time processing. Automatic detection of kidney stones using these models can speed up diagnosis and reduce dependence on manual interpretation of CT images. Non-contrast computed tomography is the gold standard for detecting kidney stones, but its manual analysis is time-consuming and can be subjective.