By Pietro Caliandro
Chairs: Prof Iris Grunwald, United Kingdom and Prof William Whiteley, United Kingdom
Artificial intelligence applications in acute stroke recognition
Presented by prof Helle Collatz Christensen, Denmark
Prof Christensen emphasizes how difficult it is to correctly identify stroke patients during phone calls at the emercengy number and how it is important that operators properly recognize patients and activate the stroke management chain. Only 35% of strokes are recognized at the time of the phone call and this aspect needs to be improved to facilitate access to treatment. Artificial intelligence algorithms can be instructed to recognize a patient as a stroke patient by interpreting the phone calls that are recorded during the distress call. This procedure requires the creation of linguistic models that the algorithm should be able to recognize during phone calls. An important issue is linked to the management of sensitive patient data.
Strengths and weaknesses of current AI tools for stroke imaging diagnosis
Presented by prof Philip White, United Kingdom
Prof Philip White presented the different AI tools currently able to diagnose stroke by interpreting neuroimaging. He underlined how the introduction of some AI models such as RAPID or VIiz in LVO has made it possible to significantly improve stroke treatment by making revascularization times faster and therefore more effective. Prof White underlines that AI tools are a great promise as decision support tools but caution is required in its use because we need regulatory rules and more robust evidence.
Artificial intelligence and big data in stroke prevention due to AF
Presented by prof Signild Åsberg, Sweden
Prof Åsberg began his interesting presentation by explaining the concepts of AI and big data. When we talk about big data we refer to a considerable amount of information that must be managed appropriately, while AI represents the set of calculation tools that allow information to be processed. She then demonstrated how these concepts can be applied to improve the identification and treatment of stroke patients in whom atrial fibrillation is suspected. AI algorithms can create predictive models of the presence of atrial fibrillation, identify atrial fibrillation in patients with stroke, and monitor NOAC therapy. She underlined that AI has the potential to improve stroke prevention in AF but its role, circumstances of its application, and the optimal methods need to be defined.
Computational modelling of acute stroke therapy
Professor Alfons Hoekstra, Netherlands
Professor Hoekstra illustrated the concept of in silico trial and highlighted how AI algorithms can be used to create virtual patient cohorts in which to define the localization and the extent of the ischemic lesion, the type of revascularization treatment to which they are subjected and evaluate the outcome based on the results of real clinical trials such as MR-CLEAN. A similar approach makes it possible to simulate the clinical conditions of the individual patient and predict the evolution in advance based on the information obtained from the trials. Furthermore, Prof Hoekstra highlighted how modeling tools can be useful for designing new real clinical trials.
Looking to the future – AI tools in rehabilitation and re-integration
Professor Christian Gerloff, Germany
Professor Christian Gerloff begins his stimulating report by recalling the results of NETS trial which highlighted the lack of efficacy of tDCS in improving the motor function of the upper limb in stroke patients. He invites us to reflect on the possible causes that led to this result and underlined how the location of the lesion, skull and skin thicknesses, different strutural anatomy, different co-morbiditie and so on may have prevented the effectiveness of tDCS since the stimulus parameters rwere not personalised according to subjective characteristics of the patient. AI can be a powerful mean of calculation which, based on typical elements of the individual subject, can allow us to customize the stimulation parameters. He then illustrated how this personalization can be applied to rehabilitation approaches such as upper limb support tools that can be customized by integrating multiple data recorded by motion sensors positioned on the patient’s upper limb. The last example presented is the application of AI as a tool for predicting the outcome after the recanalization procedure in a “real world” population and not selected as that typical of clinical trials.
Artificial intelligence increasingly enters our daily life and its use in the treatment of stroke is a frontier to be explored. It is not a question of delegating the physician’s function to an algorithm, but implementing the tools available to the physician in order to make increasingly personalized and effective therapeutic choices. Artificial intelligence applied to stroke must be a patient-centric and privacy-preserving tool whose development requires the involvement of physicians, patients and caregivers in order to meet the needs of the end user.
