Fig. 3
From: Explainable convolutional neural networks for assessing head and neck cancer histopathology
Performance of the classification and segmentation network on test data. a-c Receiver operating characteristic (ROC) curves and area under the curve (AUC). a Comparison of both networks, where the segmentation network’s pixel-level predictions were converted to tile-level predictions. b Model ensemble using voting. c Model ensemble using logistic regression. d-f Row-normalized confusion matrices. d Classification network. e Segmentation network with tile-level predictions. f Segmentation network with pixel-level predictions