PERFORMANCE EVALUATION OF DIFFERENT YOLO MODELS FOR LUNG NODULE DETECTION

Abstract

Lung cancer is one of the leading causes of cancer-related deaths worldwide. The early diagnosis of this disease is critically important for the success of treatment. Computer-aided diagnosis systems and deep learning methods are widely used to ensure accuracy and speed in the automatic detection of lung nodules. In this study, the performance of medium models of four different YOLO architectures (YOLOv8, YOLOv9, YOLOv10, and YOLOv11) in lung nodule detection was comprehensively evaluated on the LUNA16 dataset. The models were compared using metrics such as precision, recall, F1-score, overall accuracy (mAP50, mAP5095), and processing speed. The obtained results have shown that YOLOv8 offers high speed and accuracy, YOLOv10 provides the best sensitivity, and YOLOv11 excels in overall accuracy. To our knowledge, this study presents one of the first comprehensive comparisons of the latest YOLO architectures under fair experimental conditions. By systematically analyzing the relationships between performance metrics, this study fills a gap in the literature. Furthermore, our study demonstrates that deep learning-based YOLO models can be reliable and effective tools for the early diagnosis of lung cancer. The findings obtained are of a nature that will contribute to accurate and rapid diagnostic processes in clinical applications.