Tolga D Dittrich, MD
@to_dittrich
European Stroke Organisation Conference
May 4-6, 2022
The session chaired by William Whiteley (UK) and Iris Grundwald (UK) focused on current applications and possible future areas of utilization of artificial intelligence (AI) in the field of stroke.
Helle Collatz Christensen from Denmark gave the opening lecture on early stroke detection in the emergency setting. Since experience has shown that emergency medical services do not always correctly recognize a relevant proportion of stroke symptoms at an early stage, the question for Professor Christensen was how to improve stroke recognition. “AI can help us,” Christensen said. Christensen and his team collected telephone contacts between emergency callers and the dispatcher. Based on this data, an AI algorithm was developed that analyzes conversations for stroke suspicious patterns to identify suspected stroke cases early and inform the dispatcher. Existing data on AI recognition of cardiac arrest is already promising. However, Christensen sees one of the challenges in that stroke is different from cardiac arrest due to various possible symptoms.
The second presentation was by Philip White (UK) on the strengths and weaknesses of current AI applications in stroke imaging. The main areas of application include automated ASPECTS scoring, penumbra evaluation on CT perfusion, large vessel occlusion, and hemorrhage detection. According to White, these approaches are promising as they could accelerate treatment pathways. However, caution is advised in clinical application: These tools should still be seen as decision support rather than primary diagnostic devices.
Signild Åsberg from Sweden presented insights into application areas of AI and big data in the AF context. “Big data is information, AI is treatment of information,” Åsberg explained. The application areas are broad and start with the prediction of AF, for example, by AI-enhanced ECGs. Here, AI-enhanced systems have been shown to identify individuals with AF from recordings of normal sinus rhythm. According to Åsberg, this example illustrates one of the constraints of AI technology: the lack of insight into the criteria by which AI-based categorizations are made. The “black box of unsupervised deep learning” is something to be aware of, Åsberg said. In summary, she sees AI technology as having the potential to improve stroke prevention in AF, but the role and the optimal methodology still need to be defined.
In his speech, Alfons Hoekstra (Netherlands) provided exciting insights into computational modeling possibilities for stroke acute therapy. The catchwords in this context were so-called “in silico clinical trials.” These are computer models using virtual patients to simulate treatment and brain tissue damage scenarios to estimate clinical outcomes. Preliminary research shows remarkable congruence in predictive accuracy with registry data and points in a promising direction.
The last speaker was Christian Gerloff from Germany, who introduced AI tools in stroke rehabilitation. In the future, AI could help optimize external assistance systems for mobility-impaired individuals. One of the current challenges is the fine adaptation of the devices. This is necessary for upper extremity assistive systems, for example, so that both the fast catching of a ball and the slow, fine grasping of an object become possible. Here, Gerloff sees a potential application of AI to simplify and improve the personalization of devices.
The session provided a fascinating overview of AI applications that have already been implemented in practice as decision support tools, especially in stroke imaging, and the wide range of possible future applications in the stroke domain.
