Fig. 4
From: Deep learning for detection of radiographic sacroiliitis: achieving expert-level performance
Gradient-weighted Class Activation Maps (Grad-CAMs) of the model for selected images taken from the test dataset. Grad-CAMs provide visual explanations for the model decision, as they highlight the important image regions that led to the model’s decision. The examples presented show that the model almost exclusively and correctly focusses on the sacroiliac joints to predict whether definite radiographic sacroiliitis is present or not