Evaluation and generalisability of deep learning systems
In this week’s NORA webinar, you will meet Andreas Kleppe from the Institute for Cancer Genetics and Informatics at the Oslo University Hospital. Andreas will introduce and discuss methods of evaluation and generalisability of deep learning systems, using examples from experiments where the goal is to predict whether a patient will eventually die from a cancer or survive following surgery.
Register by following the link below:
Many deep learning studies are not designed to provide unbiased estimation of the system's performance in the intended application. Reports of overoptimistic estimates and opportunities may inflate the expectation of what is currently possible, misguide resource allocation, and hamper the progression of the field. In this talk, we will look into how the performance of a deep learning system in an intended application could be estimated more reliably than what is currently common practice, even if restricted to using retrospective data. To exemplify how some choices of the learning setup may influence the generalisability of the system, results will be presented from experiments where the goal is to predict whether a patient will eventually die from a cancer or survive following surgery. The presentation builds upon recent publications in The Lancet and in Nature Reviews Cancer.
Andreas Kleppe is employed by the Institute for Cancer Genetics and Informatics at Oslo University Hospital, and a researcher for the Norwegian Research Council's ICT Lighthouse project called DoMore!, where the goal is to use medical computer vision to develop automatic systems that could improve the treatment of cancer patients. He has a secondary position as associate professor at the Department of Informatics at the University of Oslo. His research interests include machine learning and image analysis, in particular medical computer vision. In several of the projects he has been involved in, the goal has been to develop automatic systems that use various image data to predict whether a patient will recur and die from a cancer or survive following surgery.