Tolga D. Dittrich, MD
European Stroke Organisation-World Stroke Organization 2020 Virtual Conference
November 7-9, 2020
Scientific Session: “Artificial Intelligence in Stroke Imaging,” Sunday, November 8, 2020
Speakers: Susanne Wegener, Roland Wiest, Paul Bentley, Kim Mouridsen, Sook-Lei Liew
Chairs: Kim Mouridsen, Susanne Wegener
Machine learning (ML) methods as a component of artificial intelligence are a growing field in stroke imaging research. We are already familiar with such automated evaluation systems, such as ASPECT scoring or mismatch volume calculation. Nevertheless, clinicians are often confronted with a complex mixture of different clinical, laboratory, and radiological parameters that must be weighed against each other to make an individual therapeutic decision.
“Machine learning is a precise mathematical way in which we can do this in a reliable, objective manner,” said Paul Bentley, of Imperial College London. Unlike conventional image interpretation, an algorithm can evaluate radiological source data to derive applicable rules. ML approaches are particularly promising for objectifying imaging results and detecting subtle changes in the context of intricate radiological findings in acute ischemic stroke. However, to provide additional information, ML methods need a relatively large set of initial data. This limitation especially becomes relevant in the context of imaging-based prediction of stroke recovery and rehabilitation response, where imaging does not constitute a common clinical component, as Sook-Lei Liew from the University of Southern California emphasized in her lecture.
The potential of ML in both acute stroke and stroke rehabilitation imaging is broad. In the future, ML-based techniques, for example, in ischemic core imaging in the extended time window, could help us to identify better patients who could benefit from endovascular treatment.
