Modeling Single Cell Properties from Histological Images

Abstract

In modern pathology, digitized images of histological sections are routinely used in defining phenotypes and characteristics for different areas of the tissue. Digital images are created from stained histological slides by specialized scanners and are further analyzed with a computer. Traditionally this type of digital pathology analysis is limited to analyzing the tissue section in local patches or in the sub-sampled section level. We propose a novel approach derived from methodology in precision digital pathology and network analysis to study the single-cell level local neighborhoods of the tissue while preserving spatial information in the form of network connections. We show that our tool can successfully be used in advanced and precise assessment of local properties combining multiple stainings and further apply this method to a multi-stained histological mouse aortic root dataset.