Meniscal tears are common knee injuries requiring precise and timely diagnosis to allow efficient therapeutic interventions. The interpretation of MRI scans, however, is inherently subject-dependent, time-consuming, and vulnerable to variability. The current study tries to overcome these issues through the use of a dataset consisting of 3,059 labeled MRI scans belonging to the “Healthy,” “Partially Injured,” and “Completely Ruptured” classes. The data were preprocessed with the operations of resizing, augmentation, and early stopping to maximize the performance of the model. A series of YOLO-based approaches were tested. Between them, the best-performing models were YOLOv11s with a 0.781 mean Average Precision (mAP@0.5) and YOLOv8s with a 0.780. The worst-performing ones were YOLOv10s with 0.714. Comparing with the previous studies, the current models yield impressive results with better diagnostic accuracy since YOLOv11s incorporates the latest architecture combined with explainable AI elements to enhance the clinical applicability. The current study underlines the capability of deep learning for revolutionizing the meniscal injury diagnostic process while overcoming variability issues and imbalance issues within the dataset to promote extended clinical applicability.