Robust Interactive Semantic Segmentation of Pathology Images with Minimal User Input

From the simple measurement of tissue attributes in pathology workflow to\ndesigning an explainable diagnostic/prognostic AI tool, access to accurate\nsemantic segmentation of tissue regions in histology images is a prerequisite.\nHowever, delineating different tissue regions manually is a laborious,\ntime-consuming and costly task that requires expert knowledge. On the other\nhand, the state-of-the-art automatic deep learning models for semantic\nsegmentation require lots of annotated training data and there are only a\nlimited number of tissue region annotated images publicly available. To obviate\nthis issue in computational pathology projects and collect large-scale region\nannotations efficiently, we propose an efficient interactive segmentation\nnetwork that requires minimum input from the user to accurately annotate\ndifferent tissue types in the histology image. The user is only required to\ndraw a simple squiggle inside each region of interest so it will be used as the\nguiding signal for the model. To deal with the complex appearance and amorph\ngeometry of different tissue regions we introduce several automatic and\nminimalistic guiding signal generation techniques that help the model to become\nrobust against the variation in the user input. By experimenting on a dataset\nof breast cancer images, we show that not only does our proposed method speed\nup the interactive annotation process, it can also outperform the existing\nautomatic and interactive region segmentation models.\n