Fig. 3From: A deep learning approach for projection and body-side classification in musculoskeletal radiographsInput radiographs resized to 256 × 256 pixels with corresponding Grad-CAM overlay of two correctly classified projections demonstrating the influential image regions (red overlay). 1a, 1b Clavicle anterior–posterior. 2a, 2b Clavicle obliqueBack to article page