Fig. 6
From: Deep learning for histopathological segmentation of smooth muscle in the urinary bladder
Receiver Operating Characteristic (ROC) curve for VGG16 (A), ResNet18 (B), SqueezeNet (C), MobileNetV2 (D), U-Net (E), MA-Net (F), DeepLabv3 + (G) and FPN (H). Both patch-to-label models (A-D) and pixel-to-label models (E–H) were trained and tested using ninefold cross-validation. Seven TUR images were evaluated for each fold to calculate ROC curves and corresponding ROC-AUC are indicated in the caption. The mean ROC curve (blue) and standard deviation (grey shaded region) are provided for the models