Classification and Detection of Kidney Stone Using Hybrid CNN Model on CT Images

Kidney stones and tumors are among the most common and severe urological diseases across the globe. The success of clinical decision-making in these situations is heavily reliant on accurate and timely diagnosis via CT imaging. However, interpreting CT images manually is both time-consuming, and fraught with inaccuracies due to human error. This work proposed an automated technique utilizing a Hybrid Convolutional Neural Network (Hybrid CNN) to categorize kidney CT images into two categories: Normal and Stone.The images were obtained from a publicly available dataset from Kaggle, which contained labeled CT images organized by class. The model utilized categorical cross-entropy loss for training and employed the Adam optimizer, following the pre-processing of images through resizing and normalization. The proposed Hybrid CNN obtained an F1-score of 89.90 %, recall of 92.01 %, precision of 87.95 %, and total accuracy of 98.52 %. These results substantiate that deep learning can improve diagnostic accuracy and minimize mislabeling, while assisting radiologists with their clinical workflow. The illustrative results and evaluation metrics indicate the robustness and feasibility of applying the model to real-world medical image classification tasks.