In this work, we explore the creation and evaluation of deep learning models for kidney stone formation prediction based on CT scan images. Moreover, the need for precision and rapid kidney stone determination is indeed pressing as more than a million of people in all over the globe has been diagnosed with this painful aliment. The method is built on a dataset of 3200 CT scan images collected from online repositories and live patient data. In the preprocessing of data, various techniques are implemented such as noise reduction, normalization, resizing the image. We train and evaluate three major deep learning architectures VGG16, VGG19, ResNet50 with the typical performance measures such as accuracy, precision, recall, and F1-score. From the results, it clearly shows that VGG19 provides highest accuracy of about 97.89% followed by VGG16 with an accuracy around (94.5%) and ResNet50 having lesser accuracy equal to (91.2%). Additional confusion matrices confirm the superior performance of VGG19 in discriminating kidney stone and non-stone images with an absence or minimal examples of both false positives and negatives. These comparative differences demonstrate the performance and efficiency in training time of VGG19, which make it applicable as a potential candidate for development automated diagnostic systems to early detect kidney stones. This concept is necessary for the future focus of medical imaging research in improving diagnostic accuracy with deep learning and patient care in urology.