The paper introduces a method for automated diagnosis of knee injuries using magnetic resonance imaging (MRI) sequences. MRI examinations serve as a crucial function in detecting knee ailments, yet the procedure is lengthy and susceptible to inaccuracies. The proposed approach utilizes sophisticated deep learning methodologies, particularly convolutional neural networks (CNNs) like AlexNet and SqueezeNet, to scrutinize MRI sequences. Different techniques, including attention mechanisms, are employed to reduce sequence information and predict injuries such as ACL and meniscus tears. The study demonstrates that a multi-model ensemble, combining the strengths of various networks, surpasses solitary frameworks concerning the Area Beneath the Receiver Operating Characteristic (ROC) Curve for varied categories of harm. Future work could explore diverse feature extraction methods and more advanced ensemble techniques for better performance.