A Reliable Kidney Stone Detection Method Using Inductive Transfer-Based Ensemble Deep Neural Networks

Abstract

Chronic kidney disease is a significant global health concern, with kidney stones being a major contributing factor to kidney
dysfunction. Early and accurate detection is essential to prevent severe complications. This study proposes an efficient approach
using inductive transfer-based ensemble deep neural networks for kidney stone detection. A combination of classification models,
including DarkNet19, InceptionV3, ResNet101, and others, along with detection algorithms from the YOLO family, enhances
diagnostic accuracy. Feature extraction techniques such as ReliefF and validation methods like KNN and KFold improve model
performance. The integration of the Xception model further refines classification accuracy, while a user-friendly Flask-based front
end facilitates real-time testing with secure authentication. The proposed approach improves early diagnosis, reduces physician
workload, and enhances patient care.