The accuracy of diagnosing an Anterior Cruciate Ligament (ACL) tear depends on the radiologist’s or surgeon’s expertise, experience, and skills. In this study, we contribute to the development of an automated diagnostic model for anterior cruciate ligament (ACL) tears using a lightweight deep learning model, specifically ResNet-14, combined with a Spatial Attention mechanism to enhance diagnostic performance while conserving computational resources. The model processes knee MRI scans using a ResNet architecture, comprising a series of residual blocks and a spatial attention mechanism, to focus on the essential features in the imaging data. The methodology, which includes the training and evaluation process, was conducted using the Stanford dataset, comprising 1,370 knee MRI scans. Data augmentation techniques were also implemented to mitigate biases. The model’s assessment uses performance metrics, ROC-AUC, sensitivity, and specificity. The results show that the proposed model achieved an ROC-AUC score of 0.8696, a sensitivity of 79.81%, and a specificity of 79.82%. At 6.67 MB in size, with 1,684,517 parameters, the model is significantly more compact than existing models, such as MRNet. The findings demonstrate that embedding spatial attention into a lightweight deep learning framework augments the diagnostic accuracy for ACL tears while maintaining computational efficiency. Therefore, lightweight models have the potential to enhance diagnostic capability in medical imaging, allowing them to be deployed in resource-constrained clinical settings.