Normal/Stone Renal CT Slices Detection Using Deep-Learning with Ensemble Features

Owing to its therapeutic importance, several Deep-Learning (DL) methodologies are frequently utilized in hospitals for disease detection through medical imaging. Renal CT (RCT) is an imaging modality utilized to assess abnormalities in the kidney. Kidney stones (KS) are a prevalent issue, and prompt detection and treatment are crucial. This research utilized a DL-method to classify the selected RCT slices into normal/KS category. The DL-tool comprises several stages: (i) image collection and resizing, (ii) deep-features extraction and classification using SoftMax, (iii) best three models selection and Ensemble Deep-Features (EDF) vector generation, and (iv) classification and 3-fold cross-validation. This study analyzed 1250 RCT images per class, revealing that conventional-features based identification achieves >88.5% accuracy, whereas EDF based detection attains an accuracy >98%. The results validate that this DL-tool functions effectively on the selected RCT database, and in the future, the performance of DL-tool can be assessed with clinically acquired RCT images.