Stroke has become an increasing problem. It has become the second most common cause of death and the most common cause of disability in adults in the European Union. Predictions show that stroke will continue to be a burden, as the number of people suffering stroke will increase over next 20 years.¹ Research on improving stroke care and prevention is therefore vital.

What is a stroke?

In general, strokes can occur in two different ways: through occlusion or through bleeding. The most common type of stroke is the ischemic stroke, where a blood clot (also known as a thrombus) blocks an artery leading to the brain. This occlusion cuts off the brain from the blood supply causing damage to brain cells due to a lack in oxygen and other nutrients. The second type of stroke is the haemorrhagic stroke happens when an artery inside the brain ruptures and begins to bleed. The build-up of blood disrupts the normal circulation of blood, preventing the brain from receiving oxygen and causing damage or kill the brain cells in that area.

Importance of stroke timing

Accurately identifying when a stroke occurred is crucial for selecting appropriate treatments. For ischemic stroke there are systemic thrombolysis which is most effective within 4.5 hours since symptom onset, and thrombectomy, which should be administered within 6 hours since symptom onset. The traditional method of determining the lesion age is known as the “Net Water Uptake Method” (NWU). It involves measuring the density of the tissue in the affected brain areas by conducting a CT scan and comparing the affected area to tissue in the healthy region on the opposite side of the brain. So, improving the way stroke age can be determined ensures that patients receive the most effective treatment within the optimal time window.

A study on AI for stroke timing

A recent study conducted by the Imperial College London, the University of Edinburgh, and Technische Universität München aimed to improve the estimation of ischemic stroke lesion age from unenhanced CT scans. The researchers developed an AI software that analyses CT-scans and estimates the time from symptom onset to imaging. The AI is trained to analyse a variety of markers, like density, texture, and shapes. The study shows that the program, also known as a convolutional neural network-radiomics (CNN-R) model, is roughly twice as accurate as the traditional method, NWU. The accuracy and reliability of this AI software can be a significant aid to stroke care.

The VALIDATE-project

Similar to the study on AI for stroke lesion age estimation, our project is also developing an AI model that aims to optimize treatment strategies for ischemic stroke by focussing on intravenous thrombolysis (IVT) and mechanical thrombectomy (MT). We are developing an AI-based clinical decision support system to assist healthcare professionals in selecting the most effective treatment method that is adapted to each individual patient, based on the patient’s, estimated, optimal health status three months after the stroke. Determining the timing of the stroke is a critical factor for the prediction of potential health outcomes, which is why the stroke lesion age study is exciting news to us. In the future, both AI softwares could work together to optimize stroke care for patients. The CNN-R model would accurately determine the lesion age and the VALIDATE AI could use that data to predict more precisely the patient’s health outcome three months later, in turn allowing healthcare professionals to choose the most effective treatment for the patient.

References

Luengo-Fernandez, R., Candio, P., Violato, M., & Leal, J. (2020). At What Cost? The Economic Impact of Stroke in Europe. Stroke Alliance for Europe. https://www.safestroke.eu/wp-content/uploads/2020/10/03.-At_What_Cost_EIOS_Full_Report.pdf Marcus, A., Mair, G., Chen, L. et al. Deep learning biomarker of chronometric and biological ischemic stroke lesion age from unenhanced CT. npj Digit. Med. 7, 338 (2024). https://doi.org/10.1038/s41746-024-01325-z