The following provides a detailed breakdown of the Whole Slide Image workflow. The ADP database was meticulously created and acquired from diverse Histological Tissue types originating from various organs. The justification and reasoning behind the inclusion and exculsion of certain data is also addressed below.
A total of 100 glass slides were selected from a larger size of 500 anonymized glass slides (each sized 1”x3” and 1.0 mm in thickness) by observing the slides under a Nikon H550L brightfield microscope using the 20x objective lens and 0.75 numerical aperture. We selected those slides with:
Acceptably few focus variations caused by tissue specimen thickness.
Diverse spectrum of color variations of tissue stains.
Acceptably few preparation imperfections such as air bubbles and tissue folding/crushing/cracks.
Different organs of origin, such as brain, kidney, breast, liver, and heart; with different diagnoses (i.e. disease or non-disease related).
The 100 selected glass slides were then digitized using a Huron TissueScope LE120 Whole Slide Image scanner at 40X magnification (0.25µm/pixel resolution, uncompressed TIFF file)
Each digitalized slide was then divided into a randomized subset of recognizable non-background patches of size 1088x1088 pixels with an overlap of 32 pixels.
In order for an extracted patch to be added to the database it had to pass the background detector and focus quality assessment.
Based on this condition, each glass slide produced 177 patches (minimum: 12 patches, maximum: 280 patches).
Given digital pathology slide patches containing visibly recognizable tissues, the team assumes that each patch originates from an unknown organ and can be classified with one or more histological types. Here, the team elaborates on the chosen taxonomy of histological tissue types and their organizing principles. In histology itself, there are two practical approaches:
Basic Histology: which studies tissue structure (or morphology).
Systematic (or Functional) Histology: which studies tissue functionality and organization into organs.
The Basic Histology approach is readily applicable to the slide patches because even a small visual field is sufficient to identify the tissue structure.
CONTINUE READING IN CVPR
This application simply determines if the image passed in is a background image. An image is classified as a background image if it contains more than 97.5% of pixels exceeding 85% intensity on all three RGB channels.
This application determines if the image passed in, is of a high enough quality whereby the information it represents is still recognizable from the image. Images without any recognizable information due to significant focus problems or non-tissue objects (e.g. dust specks) are disregarded
These individuals have been crucial to the success of the Atlas of Digital Pathology project
Konstantinos N. Plataniotis, Bell Canada Chair in Multimedia, is a Professor with the ECE Department at the University of Toronto.
Dr. Corwyn Rowsell is a Staff Pathologist at St. Michael’s Hospital and Associate Professor in the Department of Laboratory Medicine at University of Toronto.
Savvas become Chief Technology Officer for Huron in 2010. He joined the company (originally known as Biomedical Photometrics Inc.) in 1998 as project leader.
He is a post-doctoral fellow at the University of Toronto (UofT) and research scientist at Huron Digital Pathology in St. Jacobs, Ontario, Canada.