Skip to main content
Fig. 5 | Cancer Imaging

Fig. 5

From: Performance comparison between multi-center histopathology datasets of a weakly-supervised deep learning model for pancreatic ductal adenocarcinoma detection

Fig. 5

UMAP projection of features from TCGA + GTEx slides colored by label and dataset. Obtained features with the pretrained Resnet-50 on TCGA + GTEx are projected using the UMAP algorithm and visualized per label and dataset. The features are correctly separated between control and tumor. However, when the dataset plot is observed, control samples from TCGA are in the same cluster as the rest of the TCGA data, showing that the model is not accurately learning to differentiate between classes but between datasets

Back to article page

Cancer Imaging

ISSN: 1470-7330

Contact us