Improving Kidney Stone Detection with YOLOV10 and Channel Attention Mechanisms in Medical Imaging

Accurate and timely detection of kidney stones is crucial for effective medical intervention and treatment planning. However, existing detection methods often struggle with challenges related to sensitivity, precision, and the ability to process complex and variable medical images. In this study, an advanced kidney stone detection system is developed using the latest object detection algorithm, You Only Look Once version 10 (YOLOv10), integrated with channel attention mechanisms to enhance model performance. This combination aims to improve detection accuracy by enabling the network to focus more precisely on critical regions in medical images, particularly in Computed Tomography (CT) scans, where kidney stones may appear in varying shapes, sizes, and intensities. The proposed system begins with data augmentation techniques, such as rotation, scaling, and contrast adjustments, to enhance the model’s generalization ability across different image conditions and patient profiles. YOLOv10 was selected due to its lightweight architecture, high detection speed, and enhanced performance in small object detection tasks. To further improve feature extraction, channel attention mechanisms such as Squeeze-and-Excitation (SE) blocks or Efficient Channel Attention (ECA) modules are incorporated. These modules enable the network to selectively focus on the most informative feature channels associated with kidney stone regions, while suppressing irrelevant background information, thereby improving the distinction between stones and surrounding tissues. The model is trained and fine-tuned using a diverse CT scan dataset containing various types and sizes of kidney stones. Evaluation results demonstrate that the proposed model achieves a high detection accuracy of 93.7% with a very low loss of 0.18. It exhibits stability without issues like overfitting, underfitting, or local minima entrapment, making it a highly reliable tool for clinical applications.