Kidney diseases affect individuals across all age groups and are a major global health concern. Pathological and other conditions, such as tumors, cysts, and stones, along with normal states of the kidneys, need to be detected as early as possible to improve treatment outcomes and quality patient care. This study looks into the use of computed tomography (CT) images for deep learning-based kidney disease classification. We evaluated four widely used convolutional neural networks (CNNs) such as VGG16, MobileNetV2, ResNet50, and InceptionV3 on a dataset of 12,456 CT images. Among the individual models, MobileNetV2 achieved the highest validation accuracy of 99.64%. As a novel contribution, we propose a hybrid deep learning model that combines MobileNetV2 and ResNet50 to enhance diagnostic performance. The hybrid architecture design led to superior results: 99.88% validation accuracy, 99.50% precision, 99.50% recall, 99.25% F1-score, and a reduced validation loss of 0.0090. Performance was further validated using confusion matrices, receiver operating characteristic (ROC) curves, classification reports, and 6-fold cross-validation to assess generalization. The proposed model demonstrates strong robustness and generalizability across kidney condition categories. As far as we are aware, not many research have looked into a hybrid combination of MobileNetV2 and ResNet50 for multi-class kidney CT classification.