The goal is to empower users with computational technology that fuses with their knowledge and experience to analyze HCS images and achieve their scientific goals. The imCellPhen system will provide an integrated environment that enables users to visually explore large-scale HCS image databases in effective ways, retrieve images similar to those of their interests, detect novel phenotypes, and interactively construct phenotype recognizers.
The architecture of imCellPhen
Low-level image features are extracted from each image as its metadata and all computations are applied to metadata. The system will allow users to start with the novel phenotype detection step which initially relies only on a wildtype recognizer and clustering techniques. The discovered patterns are not expected to define phenotypes well enough and should be subject to visual examination and refinement. Users can select one pattern and then use the CBIRRF technology to train a mathematical model to recognize the phenotype. The trained phenotype recognizer will be added to the phenotype recognizer pool, which is one form of domain knowledge database, to improve the pattern discovery step. The more human experts interact with imCellPhen, the more knowledge it can accumulate and the more intelligent and accurate it will become.
