Automated Kidney Tumor Detection and Segmentation Using YOLO11 and Grad-CAM
5th International Conference on Informatics and Software Engineering, IISEC 2026, Ankara, Turkey, 5 - 06 February 2026, pp.388-393, (Full Text)
- Publication Type: Conference Paper / Full Text
- Doi Number: 10.1109/iisec69317.2026.11418526
- City: Ankara
- Country: Turkey
- Page Numbers: pp.388-393
- Keywords: deep learning, Grad-CAM, kidney tumor detection, segmentation, YOLO11
- Middle East Technical University Affiliated: Yes
Abstract
Early detection and clear delineation of the tumor in the kidney are of the greatest significance in the accurate diagnosis and treatment of the tumor. In this study, a method for identifying kidney tumors using the YOLO11 model series is proposed to enhance the accuracy of the model by utilizing the Grad-CAM technique for improved interpretability. In this work, the YOLO11 model is used for the accurate segmentation of the tumor in the kidney in real-time with high precision and sensitivity. All five variants of the YOLO11 model series are trained and tested for the segmentation capacity of each model, with specific emphasis on attaining the highest levels of accuracy and interpretability. The YOLO11m model achieved the best value of 99.18% precision, 99.19% mAP@0.5, and a Dice coefficient of 95.85%, while YOLO11l obtained the highest recall, which is 98.76%. In addition, using the Grad-CAM technique can improve the interpretability of the model, thus providing a clear visualization of the areas that can significantly affect the decision-making of the model. These results indicate significant improvements in model transparency and detection accuracy, providing a robust method for automatic kidney tumor diagnosis. This may lead to improvements in clinical processes through faster, more accurate and interpretable tumor identification. The result of this action would be improved patient outcomes.