Rapid Multi-stain Digital System for Histology based on Machine Learning
DescriptionHistology is usually the definitive step in disease diagnosis.Histology requires staining tissues for contrast prior tomicroscopic analysis, but staining is time intensive and consumesthe tissue. Many clinical scenarios would benefit greatly fromrapid staining that preserves the tissue (eg. for multiple stains).This project will develop a novel system that rapidly generatesstains of a tissue specimen on a digital computer. The system isbased on machine learning. Tissues will be fixed, embedded, andsectioned with a tissue processor for consistency. Note that thesystem can be adapted to frozen processing for increased speed.Unstained tissue sections will be imaged with a fluorescencemicroscope. The sections will then be stained and imaged with abrightfield microscope. The unstained autofluorescence andstained brightfield images will be used to train a machine learningalgorithm. After training, the algorithm will prospectivelygenerate “stained” images from unstained images WITHOUTchemical staining. This novel system, consisting of the processor,fluorescence microscope, algorithm, and computer, will initiallybe developed for research at the Hong Kong Hospital Authority.Licensing for clinical use and commercialization will follow. Thissystem is suited for time-sensitive scenarios that would benefitfrom multiple stains, not clinically possible at present.
|Effective start/end date||1/05/19 → …|