Renal calculi represent a widespread global health issue, impacting millions of people. Non-contrast computed tomography (CT) remains the definitive imaging technique for their identification. However, manual review of scans is labor-intensive and prone to inter-observer variability, underscoring the need for reliable automated systems. This investigation introduces a unified automated system that integrates YOLOv8-based object detection with ResNet-18 regression to enable accurate detection and quantitative severity evaluation of kidney stones in abdominal CT scans. We implemented a custom two-stage inference approach using the publicly available Kaggle kidney stone dataset, which contains 2603 annotated CT images. The dataset was partitioned into training (80%), validation (10%), and test (10%) subsets. In the initial stage, YOLOv8 localizes stones with a precision of 0.737, recall of 0.688, and mean Average Precision at an Intersection over Union threshold of 0.5 ([email protected]) of 0.707. The subsequent stage applies ResNet-18 regression to estimate continuous severity values, achieving a mean absolute error (MAE) of 0.1080. These scores are then converted into standard clinical categories (Mild, Moderate, Severe) using fixed thresholds. The complete pipeline delivers consistent results across anatomical regions, maintains a low false-positive rate, and generates outputs that are readily interpretable for clinical use. Overall, this method effectively addresses limitations in current standardized assessment practices and provides valuable support for diagnostic processes in radiology and urology departments.
