Dr. Donal O’Shea is a digital pathology industry veteran having worked for 20 years in academia and commercial within the space.
Donal was co-founder of Slidepath in 2003, one of the pioneering companies in digital pathology application software. Donal managed the company through to its acquisition by Genetix PLC in 2009. As part of the management team, Genetix was sold to Danaher Corp in 2010 for $100m. Within Danaher’s Leica Microsystems and Leica Biosystems operating companies, Donal supported the integration of Genetix, successfully lead various business divisions, launched new products and was significantly involved in the acquisition of Aperio Technologies in 2012, helping to establish Leica Biosystems as the dominant market leader in the space.
Donal has recently founded Deciphex, a company primed to capitalise on the significant interest in application of artificial intelligence in diagnostic pathology. The company has grown rapidly from its foundation in 2017, attracting capital, filing intellectual property and building key strategic collaborations in healthcare and biopharma. The company plans to launch its first commercial product in 2019.
Donal has a Ph.D. in computer vision in lifescences and was a HRB Postdoctoral Fellow, assessing the potential of image analysis in the grading and staging of breast cancer. As an Academic, Donal has supervised and graduated 5 Ph.D. students (3 specifically in digital pathology) and published extensively in the space. Donal is on the board of a number of Irish start-ups in the lifescience and diagnostics space, including Oncomark, Microsynbiotix and GlowDx.
Title: “Finding Normals: Application and Implementation of AI driven Screening Processes in Pathology”
Abstract:
Developing deep learning models that account for all variants of abnormality in a given tissue slide can be a challenge requiring extensive and time consuming annotation of thousands of pathology slides by experienced pathologists.The process also needs to effectively account for tissue processing variance and digital pathology scanner selection, which is almost infinite due to the extensive range of interchangeable unit operations available for the pathology lab.
In scenarios where a high percentage of normal content presents, we propose a generalised model for abnormality detection to help identify abnormal content of interest and support the elimination of normal cases from the workflow. The generalised model as proposed is developed solely from normal cases, requires no annotation of slides as a result and can be rapidly tailored to account for inter-laboratory processing variance.
In his presentation Dr. O’Shea will discuss how such approaches can be leveraged, useful application segments for the technology and how he foresees this capability being embodied in software workflows.