This study addresses the critical need for early and accurate brain tumor diagnosis on MRI images by comparing five versions of the YOLO algorithm (YOLOv5, YOLOv7, YOLOv8, YOLOv9, and YOLOv12) with consistent parameters. Utilizing a pre-annotated Kaggle MRI brain dataset, the research meticulously verified annotations and employed data augmentation (flipping, rotation, blurring, noise) to expand the dataset from 801 to approximately 1362 images, enhancing model generalization and robustness. Models were trained and evaluated on metrics including precision, recall, mAP@0.5, mAP@0.5:0.95, and inference time. YOLOv12 demonstrated superior overall performance, achieving the highest recall (97.32%), mAP@0.5 (92.2%), and mAP@0.5:0.95 (76.57%), establishing its robustness for accurate detection and object localization. While YOLOv7 achieved the highest precision (96.89%) and excellent inference speed, its overall mAP and recall were surpassed by other iterations. YOLOv9 and YOLOv8 also showed strong competitive performance, indicating significant advancements in the newer YOLO generations. The findings confirm the efficacy of the YOLO algorithm for brain tumor detection and localization in MRI images, with YOLOv12 proving to be the most effective variant in this comparative analysis.
Satria et al. (Thu,) studied this question.