Abstract
<title>Abstract</title> <p>Background Differentiating distal ureteral stones from phleboliths and arterial calcifications on non-contrast CT is challenging due to overlapping imaging features. This study aims to develop and evaluate a deep learning model for automated classification of pelvic calcifications. Methods In this retrospective study, CT images were collected and labeled by expert radiologists into four classes: ureteral stone, vascular calcification, ureteral stone with vascular calcification, and normal cases. Images were preprocessed, resized, and augmented to improve generalization and address class imbalance. A convolutional neural network (CNN) model was trained using a 5-fold cross-validation approach. Model performance was assessed using accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (ROC-AUC). Results The model achieved an overall accuracy of 94.6%, with a mean sensitivity and specificity of 93.2% and 95.1%, respectively. The AUC values were 0.97 for stones, 0.95 for vascular calcification, 0.94 for ureteral stone with vascular calcification, and 0.96 for the normal class, indicating excellent discriminative performance. Conclusion The CNN-based approach can assist radiologists in accurately differentiating pelvic calcifications from distal ureteral stone, improving diagnostic confidence and clinical decision-making.</p>