Adam David Kapelner

Systems and Approaches
Machine learning, software engineering, image analysis, microscopy

Interests
image segmentation, spatial statistics, supervised machine learning

Summary
I am interested in using supervised machine learning to find objects of interest in images. This is also called "image segmentation" or "region of interest" identification. The Lee Group studies the immunological responses in lymphnodes of breast cancer patients. Tissue samples are stained for many markers via IHC and imaged globally using automated microscopy (some markers of interest are CD4, CD8, Granzyme B, and others). The big question posed is how do immune cells interact with cancer and each other? And, is this information useful in diagnosis or prognosis? In order to answer this question, each cell must be identified in the tissue, their locations resolved, and this data analyzed using spatial statistical or geometric methods. This is a specialized application of image segmentation because IHC-stained and imaged cells contain color information and are radially symmetric. I developed novel software that uses Mahalanobis distances of pixel values and Random Forests that not only counts, but pinpoints objects of interest. The procedure requires the user to train and classify iteratively until a desired result is reached. The software package includes training, classification and data analysis tools (see http://www.gemident.com for info, downloads, and source code).