Pathology is the area of medical science that studies the causes, nature and effects of disease. Manual screening of microscopic slides by a pathologist is both labour-intensive and subjective. Several factors, including fatigue and inexperience can lead to false interpretations. These issues in manual screening have motivated the development of automated screening systems which make quantitative measurements of cellular objects directly on a glass slide. The downside to diagnosis of slide specimens is that physical access to the tissue on the glass slide is needed. A virtual slide, obtained by capturing and digitising the tissue specimen, can be viewed on an ordinary computer without the need of a microscope, with the added benefit that it does not fade or degrade over time. Although several systems that create virtual slides have been developed, none of them is currently approved for primary diagnosis.
The aim of this research is to show that a virtual image, which is assembled from hundreds of single field-of-view images, is a glass-faithful representation of the underlying microscopic specimen. Applying automated quantitative analysis techniques could subsequently lead to a widespread use of virtual microscopy for the primary diagnosis of pathology specimens. This work covers the most fundamental tasks which are, calibrating a microscope system by determining the microscope's lens parameters, and eliminating any illumination effects that can cause changes in the shape of objects, hence leading to incorrect measurements.
The acquisition and stitching of the entire microscopic specimen at high resolution is time consuming, especially in the case where the specimen only occupies a small area of the slide. This time can be significantly reduced if only the area of interest is acquired. The use of multi-resolution scanning algorithms can further reduce the acquisition time by adapting the scan layout. In this work, a reduction of one third to one half has been achieved by determining the minimal area to be scanned.
Autofocusing is another important step as manual focusing is subjective and requires human attention. However, it is important to make sure that objects of interest are focused rather than dust on the cover slip. Here, several autofocusing methods have been compared and the fastest and most accurate has been combined with a global maximum search to quickly find the focus position where most objects are in focus.
The outcomes of this work suggest that a virtual image is a glass-faithful representation of the underlying microscopic specimen. Further improvements to the background illumination correction method are necessary to prove the glass-faithfulness and to make sure that the automated quantitative analysis will yield correct measurements of microscopic objects.