AI-assisted Transcranial Duplex Sonography for Early Detection of Intracerebral Haemorrhage: HYPER-AI-SCAN

Overview

The goal of this observational study is to evaluate whether transcranial Doppler ultrasound, combined with artificial intelligence (AI), can help identify intracerebral haemorrhage (ICH) in people with acute stroke (both men and women, adults of all ages) within 48 hours of symptom onset. The main questions it aims to answer are: Is it feasible to perform standardized protocol transcranial ultrasound in acute stroke patients? Can AI models trained on ultrasound images accurately distinguish haemorrhagic stroke ("ICH suspected") from non-haemorrhagic stroke? There is no comparison group, because all participants will undergo both CT (as standard care) and ultrasound (research imaging), and the AI models will compare their ultrasound-based predictions against CT-confirmed diagnoses. Participants will: undergo a non-invasive transcranial ultrasound scan after CT confirms the type of stroke allow researchers to collect coded ultrasound images for AI model training provide clinical and imaging information (already collected as part of routine care) to help evaluate factors related to diagnostic accuracy No treatments or changes to clinical care will be introduced as part of the study.

Stroke is a medical emergency that can be caused either by a blocked blood vessel (ischaemic stroke) or by bleeding inside the brain (haemorrhagic stroke). These two types of stroke require very different treatments, and identifying which one is occurring as quickly as possible is essential. Currently, the only reliable way to distinguish between these two types of stroke is with a brain scan such as a CT scan. However, CT is not always available immediately, especially in prehospital settings or in hospitals without 24/7 imaging access. As a result, patients may experience delays before receiving the correct treatment. This study aims to explore whether a simple ultrasound scan of the brain, performed through the skull, can help identify haemorrhagic stroke more quickly. This technique is called transcranial Doppler ultrasound (TCD). It is fast, non-invasive, and uses no radiation. A total of 500 patients with suspected stroke within 48 hours of symptom onset will be included. After the standard CT scan confirms the diagnosis, each participant will undergo a brief ultrasound scan following a structured protocol. The ultrasound images will then be used to train and test artificial intelligence (AI) models, which will learn to recognize patterns associated with haemorrhagic stroke. These AI models will compare the ultrasound images with CT results and try to predict whether a bleed is present ("ICH suspected") or not. The main goals of the study are: To determine whether portable ultrasound can be performed reliably and consistently in real stroke patients. To evaluate whether AI can support clinicians by interpreting these ultrasound images and distinguishing between haemorrhagic and non-haemorrhagic strokes. All other clinical information-such as symptoms, timing of arrival, and medical history-will also be collected to understand which factors may influence the performance of ultrasound and AI. Importantly, the ultrasound does not replace standard medical care and will not influence the treatment that patients receive. It is performed only for research purposes. The CT scan remains the reference test for diagnosis. By combining ultrasound with AI, this project hopes to pave the way for future systems capable of assisting paramedics or physicians in identifying haemorrhagic stroke earlier, especially in settings where CT is not immediately available. Earlier recognition may help reduce delays in blood pressure management or treatment reversal for patients taking anticoagulants.