Automated Analysis and Classification of Histological Tissue Features by Multi-Dimensional Microscopic Molecular Profiling

摘要: Characterization of the molecular attributes and spatial arrangements cells features within complex human tissues provides a critical basis for understanding processes involved in development disease. Moreover, ability to automate steps analysis interpretation histological images that currently require manual inspection by pathologists could revolutionize medical diagnostics. Toward this end, we developed new imaging approach called multidimensional microscopic profiling (MMMP) can measure several independent properties situ at subcellular resolution same tissue specimen. MMMP involves repeated cycles antibody or histochemical staining, imaging, signal removal, which ultimately generate information analogous flow cytometry on intact sections. We performed microarray containing diverse set 102 using panel 15 informative 5 stains plus DAPI. Large-scale unsupervised data, visualization resulting classifications, identified profiles were associated with functional features. then directly annotated H&E from series such canonical interest (e.g. blood vessels, epithelium, red cells) individually labeled. By integrating image annotation signatures specific annotations statistical models automatically classifying these The classification accuracy automated histology labeling was objectively evaluated cross-validation strategy, significant (with median per-pixel rate 77% per feature samples) de novo prediction obtained. These results suggest high-dimensional may advance computer-based systems parsing relevant cellular data arbitrary samples, provide framework resource spur optimization technologies.