Due to the many advantages of Artificial Neural Networks in medical image screening processes, this work conducts a comparative analysis of the DenseNet201, ResNet152V2, NASNetLarge, VGG19, Xception, InceptionV3, InceptionResnetV2 and MobileNetV2 architectures in the task of anterior talofibular ligament lesion classification in ankle magnetic resonance images, with the use of transfer learning. This work implements and comparatively analyzes the trained models with state of the art methods in the field of medical imaging, a little explored field in computer vision. Our work also uses CNNs as feature extractors, as this method is independent of the human choice of a specific feature extractor and the models can better learn the method of extraction. We obtained as high as 81% accuracy, 80% F1 score, 78% precision, 82% recall and 82% area under the receiver operating characteristic curve mean results with the VGG19 architecture, competitive results compared to previous works.