Fig. 1

Multi-instance learning (MIL) pipeline used in this work and experiments performed. Panel A Independent classifiers are trained with a MIL classifier, using the CPTAC and TCGA + GTEx datasets independently. The classifier is trained using a bag of tiles (100 tiles) and a Resnet-50 pretrained on Imagenet for feature extraction. Then, the features are averaged-pool and forwarded through a linear layer to perform the final prediction. Panel B Once the models have been pretrained on each dataset respectively, their performance is evaluated on the other dataset, testing the generalization capabilities of each model. Panel C The pretrained models on CPTAC and TCGA + GTEx are tested in the TMA dataset respectively, to compare their performance