Fig. 3.

Examples of correct predictions highlighted by Gradient-weighted Class Activation Mappings (Grad-CAMs) for radiographs with only one finding. Images A, C, E, G, I, and L are the original images showing a fracture (A), osteoarthritis (C), an endoprosthesis (E), dislocation (G), osteosynthesis material (I), and calcification (L). Images B, D, F, H, K, and M are the corresponding Grad-CAMs. Grad-CAMs highlight image areas that were important to the model’s decision. They can be used to verify that a pathology was correctly identified and that the decision was not based on image noise (e.g., due to overfitting) or a different finding (e.g., a model that should detect a fracture but instead detects only osteosynthesis)