A Novel Ensemble Learning Approach for Grouping the State-of-the-Art YOLOV10 and YOLOV11 Models for Kidney Stone Detection in CT and Ultrasound Images.

Despite its essential role in preserving healthy kidney tissue, kidney stone detection has received limited attention in academic literature. Physicians need to accurately and precisely detect the location of kidney stones in medical images, which is a challenging and time-consuming task. Deep learning techniques, which offer a powerful ability for object detection, can be utilized to address this problem. In this study, two different image modalities (CT and ultrasound imaging) of kidney stone images are utilized for performing a generalized overview. A novel ensemble framework combining the latest YOLOV10 and YOLOV11 models is proposed to minimize false negative and positive errors, thereby improving the performance of the individual models. Experiments show that the proposed deep learning ensemble model enhances the performance of individual models by 5.4%, 2.4%, and 1.3% of precision, recall, and F1-score, respectively, compared to the best individual model trained using the CT imaging modality. They also indicate that utilizing the ultrasound-based dataset improves the F1-score by 1% and the Map50 score by 1.34% compared to the individual models. Results show that the proposed approach exhibits enhanced performance and demonstrates that the ensemble framework outperforms state-of-the-art methodologies.