Efficient Vertebrae Segmentation in MRI Using YOLOv8 and U-Net

Manual analysis of spinal MRI scans is time-consuming and prone to human errors due to subtle anatomical variations. Recent advancements in Artificial Intelligence (AI) provide tools for automating spinal segmentation, potentially improving diagnostic efficiency. In this study, we propose a hybrid framework that integrates YOLOv8 for vertebrae localization with a U-Net architecture and the watershed algorithm for segmentation. YOLOv8 is used for its high-speed detection and localization capabilities, while the combination of U-Net and watershed facilitates precise delineation of spinal structures. Experimental results indicate a mean Average Precision (mAP) of 99.1% for localization and a segmentation accuracy of 90% with minimal loss. The integrated method may contribute to improving diagnosis and treatment planning of spinal deformities. Moreover, the proposed approach is scalable and could be adapted to other medical imaging segmentation tasks.