Fig. 5
From: Explainable convolutional neural networks for assessing head and neck cancer histopathology
Exemplary tiles of the test dataset with corresponding Grad-CAMs and HR-CAMs, generated for the classification network. The Grad-CAMs (7\(\times\)7 pixels) were resized to 224\(\times\)224 pixels. Values close to one indicate a high importance of the respective image region. a Tiles containing both tumor and surrounding tissue with Grad-CAMs. b Tiles containing atypical cells with HR-CAMs. c Tiles containing squamous epithelium with HR-CAMs