Anterior Cruciate Ligament (ACL) tear is a common knee injury that can be diagnosed using Magnetic Resonance Imaging (MRI). In this study, a novel approach has been proposed for improving the accuracy of ACL tear detection using an optimized Gated Recurrent Unit (GRU) model. The GRU model is a type of Recurrent Neural Network (RNN) that can capture the temporal features of MRI images. The optimization of the GRU model is based on a modified metaheuristic technique, called the Flexible Fitness Dependent Optimizer (FFDO). The FFDO is a flexible and problem-dependent algorithm that can find the best solutions to optimize problems by generating and evaluating a set of candidate solutions based on a fitness function. The proposed approach is applied to the MRNet dataset, and the results are compared with 17 state-of-the-art methods. The results show that the proposed approach can achieve a high accuracy of97.95%for ACL tear detection, which is higher than the existing deep learning and machine learning methods. The proposed approach can also achieve high values for other performance metrics, such as specificity, sensitivity, precision, MCC, F1-score, and AUC. The proposed approach can effectively extract the essential features of knee MRI images that are indicative of ACL tears with high diagnostic accuracy.