Multiple Instance Captioning: Learning Representations from Histopathology Textbooks and Articles

We present ARCH, a computational pathology (CP) multiple instance captioning\ndataset to facilitate dense supervision of CP tasks. Existing CP datasets focus\non narrow tasks; ARCH on the other hand contains dense diagnostic and\nmorphological descriptions for a range of stains, tissue types and pathologies.\nUsing intrinsic dimensionality estimation, we show that ARCH is the only CP\ndataset to (ARCH-)rival its computer vision analog MS-COCO Captions. We\nconjecture that an encoder pre-trained on dense image captions learns\ntransferable representations for most CP tasks. We support the conjecture with\nevidence that ARCH representation transfers to a variety of pathology sub-tasks\nbetter than ImageNet features or representations obtained via self-supervised\nor multi-task learning on pathology images alone. We release our best model and\ninvite other researchers to test it on their CP tasks.\n