David is a full time NHS consultant pathologist and has been lead pulmonary and skin pathologist at the University Hospital Coventry since 1997. He has led the project to adopt digital pathology at UHCW for the last 5 years.
He now heads the UHCW NHS Trust Digital Pathology Centre of Excellence (CoE), an academic venture exploring the use of this technology in routine histopathology. The CoE collaborates closely with Professor Nasir Rajpoot and colleagues at the Computer Science School University of Warwick and is currently evaluating algorithms aimed at improving pathologist’s ability to accurately grade cancers and the automation of mundane time consuming quantitative tasks.
His other research interests are in improving early lung cancer diagnosis, and the pathology of neuroendocrine lung tumours.
After taking over as clinical director in 2009 he has established the Yvonne Carter Chair of Pathology at the University of Warwick and the centralisation of cellular pathology for Coventry and Warwickshire Pathology Services.
Title: Algorithm development and strategy for digital pathology in the future
Although it is now widely accepted that modern digital pathology solutions provide a viable alternative to conventional light microscopy for the purposes of diagnostic histopathology, conversion to the new modality is impeded by the lack of a return on the investment required.
The development of algorithms which successfully exploit the potential for computer assistance in the task of routine histopathology is feasible and once achieved will ensure digital pathology becomes the best practice modality.
In this talk we review the areas of diagnostic pathology which are likely to be exploited in algorithm development, the progress made to date, the strategy for implementation and likely effects this will have on the delivery of diagnostic histopathology services.
At the most basic level digital pathology algorithms are computational tools that function to interrogate the pixel level data in the digital pathology image in order to extract information which is of diagnostic or prognostic use. Although most of these tools have at their core the recognition of the same cellular vents and details with which pathologists are familiar, they broadly fall into two quite different groups. In the downstream group are pathologist supervised applications where the data from a region of interested detected by the pathologist can be quantified to produce a value range that has the potential to be more accurate and reproducible than the pathologist. This enables algorithms to be used in providing prognostic information and deciding treatment options for oncological patients and patients with progressive fibro-inflammatory conditions. The upstream group includes the algorithms based on pattern recognition and deployed to search through large volumes of pixel data for patterns of diagnostic interest. These types of algorithms are of use diagnostically the in automation of slide reading and the detection of rare events.
The successful adoption of these algorithms will have a number of implications for pathology services. Upstream algorithms are very likely to depend on the export of large numbers of whole slide images, each containing gigabytes of data, from the digital pathology database to separate servers where the algorithms will be run and then for the results re-imported back into the digital pathology archive. This enables centralisation of the upstream algorithm work which may be feasible for a region-wide approach and provide some economies of scale. These algorithms are going to be less stressed by speed requirements but have much greater scale of computational requirements. Conversely the downstream algorithms are going to be run by the pathologist as they identify regions of interest and will therefore need to be working on the diagnostic workstation and required to run at diagnostic speed in such a way as to minimise the disruption to the diagnostic workflow. The results need to be readily imported into the report and delivered to the laboratory information.