PI: Mark Bathe, Department of Biological Engineering, Anne Carpenter, Whitehead Institute for Biomedical Research, MIT
PI: Victor Lempitsky, Center for Computational and Data-Intensive Science and Engineering, Skoltech
This proposal is motivated by the call for new, powerful image analysis tools for biomedical applications, as well as by the deep learning‐driven revolution in image analysis and computer vision. Following these two trends, several groups specialized in deep learning have presented very impressive results obtained for challenging cell image analysis tasks. Yet, this class of methods remains inaccessible to biomedical specialists that lack expertise in deep learning. The main goal of the project will thus be to bridge this gap and to bring the power of deep learning based image analysis algorithms to biomedical researchers. Towards this end, we will develop and publish a toolbox that will allow users to build and to train deep learning models in an interactive way and without using programming/scripting languages. We anticipate that the flexibility of deep learning methods will allow us to combine the advantages of existing interactive cell image analysis systems popular among biomedical practitioners with powerful deep learning algorithms.
Additionally, the new approach will enable sharing and reuse of deep models trained for popular microscopy modalities in a similar vein to the way that deep models trained on photographs are shared and reused within the computer vision community. The initial stage of the project will focus on developing advanced image analysis software for immune microenvironment assays, which are actively used in developing new immunology‐based cancer treatments. Such assays possess a massive amount of valuable information that is hard to extract and quantify manually or with commonly used image processing techniques. The proposed project will bring together three groups with complementary expertise needed for its success.