Skip to main content
Fig. 1 | BMC Cancer

Fig. 1

From: Leveraging artificial intelligence to predict ERG gene fusion status in prostate cancer

Fig. 1

Workflow schematic summarizing our algorithm development. a (Top panel) Whole slide images of H&E-stained prostate adenocarcinoma resections were spilt using QuPath into many 224 × 224 pixel patches for input into a convolutional neural network (CNN). Unknown yellow box indicates a separate subset of cases not included as part of the training subset. (Bottom panel) Patches labeled with ERG status were used for CNN training utilizing MobileNetV2. Final prediction of patches into ERG-negative or ERG-positive was based on highest probability. b MobileNetV2 convolutional block structure (adapted from Sandler et al.)

Back to article page