Another thing is I noticed you have analyzed frozen tissue images (ones that have TCGA--BS or TCGA--TS in file names; e.g., TCGA-10-0928-01A-02-BS2.facd92de-6893-4650-a327-1f15554a01fc.svs). Frozen tissue images are different from diagnostic images (one with TCGA-*-DX in file names; e.g., TCGA-3C-AALI-01Z-00-DX1.F6E9A5DF-D8FB-45CF-B4BD-C6B76294C291.svs). There are differences between those two types of images in terms of staining protocols, how tissue is fixated on glass slides, etc. Our models are trained with data from diagnostic slides only. They may not perform well on frozen tissue images — we don't analyze frozen tissue images in our scientific projects. We can use them in our joint project to scale up input data size for measuring I/O performance, etc, but we shouldn’t use them to compare the accuracy of the deep learning models. To compare the menndl model with our model, you can use diagnostic images from luad, paad and prad cancers.
anagainaru
commented
2 years ago 
Another thing is I noticed you have analyzed frozen tissue images (ones that have TCGA--BS or TCGA--TS in file names; e.g., TCGA-10-0928-01A-02-BS2.facd92de-6893-4650-a327-1f15554a01fc.svs). Frozen tissue images are different from diagnostic images (one with TCGA-*-DX in file names; e.g., TCGA-3C-AALI-01Z-00-DX1.F6E9A5DF-D8FB-45CF-B4BD-C6B76294C291.svs). There are differences between those two types of images in terms of staining protocols, how tissue is fixated on glass slides, etc. Our models are trained with data from diagnostic slides only. They may not perform well on frozen tissue images — we don't analyze frozen tissue images in our scientific projects. We can use them in our joint project to scale up input data size for measuring I/O performance, etc, but we shouldn’t use them to compare the accuracy of the deep learning models. To compare the menndl model with our model, you can use diagnostic images from luad, paad and prad cancers.