3‐Dimensional Convolutional Neural Network for Detection of an Anterior Cruciate Ligament Injury in Knee Magnetic Resonance Imaging Scans

Problem Accurate diagnosis of anterior cruciate ligament (ACL) injuries from knee magnetic resonance imaging (MRI) is clinically essential but remains challenging due to subtle anatomical variations, limited labeled data, and severe class imbalance. Aim This study aims to develop a lightweight and computationally efficient deep learning model for automated detection of complete ACL tears from knee MRI scans. Methods A custom 3D convolutional neural network (3D‐CNN) with approximately 153 k trainable parameters was proposed. Two publicly available datasets included are Stanford ML Group and Clinical Hospital Centre Rijeka, which were combined to improve generalization. To address dataset heterogeneity and class imbalance, standardized preprocessing, conservative geometric data augmentation, and class weighting were employed. The model was trained using 5‐fold stratified cross‐validation. Results The proposed model achieved an average classification accuracy of 97.6% and an average AUC of 0.993, outperforming several state‐of‐the‐art deep learning architectures while maintaining a small model size (∼2 MB) and short training time (∼10 min per fold). Conclusion The results demonstrate that a lightweight 3D‐CNN can accurately and efficiently detect complete ACL tears from knee MRI scans, highlighting its potential for clinical decision support and deployment in resource‐constrained environments.