Predicting recovery after acute ischemic stroke using neurologists' judgments and AI

STOP-stroke: STroke Outcome Prediction in the Acute Treatment Setting - a Prospective, Single-center, Observational Study

This project tests whether neurologists or a deep-learning computer can better predict recovery after an acute ischemic stroke for patients who arrive within 24 hours using routine clinical and imaging data.

Quick facts

• Study type: Observational
• Enrollment: 250 (estimated)
• Sex: All
• Sponsor: University of Zurich (Academic / other)
• Locations: 1 site (Zurich)
• Trial ID: NCT06534645 on ClinicalTrials.gov

What this trial studies

This prospective observational project enrolls patients with suspected acute ischemic stroke presenting to a tertiary stroke unit and collects the clinical and imaging information used by treating neurologists. Treating neurologists will make real-time outcome predictions (NIHSS at 24 hours and 3 months, and mRS at 3 months), and deep-learning models will be given the same structured clinical and imaging inputs to generate predictions. The team will compare predictive performance, identify which baseline variables drive decisions for both humans and models, and perform error analyses on misclassifications. Findings will be used to refine the algorithm and improve handling of multimodal clinical and imaging data.

Who should consider this trial

Good fit: Adults (18 years and older) with suspected acute ischemic stroke who present to or are referred to the University Hospital Zurich within 24 hours of symptom onset (including wake-up or unclear onset) and who undergo clinical neuroimaging and can provide consent or have a proxy provide consent.

Not a fit: Patients who explicitly refuse use of their health data or cannot provide consent and have no available proxy, those with non-ischemic diagnoses (e.g., primary intracerebral hemorrhage), or those presenting outside the 24-hour window are unlikely to benefit from this project.

Why it matters

Potential benefit: If successful, the project could help doctors make faster, more accurate predictions of stroke recovery and support safe integration of trustworthy AI tools into clinical care.

How similar studies have performed: Previous machine-learning and deep-learning studies have shown promise for predicting stroke outcomes from imaging and clinical data, but multimodal, prospective comparisons against clinician predictions in routine care remain limited.

Eligibility criteria

Inclusion Criteria:

* Patients from up to 18 years years of age without any upper age limit.
* Patients with clinical suspicion of acute ischemic stroke (acute onset focal neurological deficit) at the discretion of the paramedic or treating physician within 24 hours of symptom onset including wake-up situation and unclear symptom onset planned for clinically indicated neuroimaging.
* Patients with externally performed neuroimaging before admission or referral to the USZ will be included from the time point N2 on if no refusal of use of data is documented.

Exclusion Criteria:

• Patients with documented objection of subsequent use of personal health data or patients who reject the use of personal health data during follow-up after initial informed consent by an independent physician in the acute setting. We will not include patients in the study if there is no written informed consent either from the patient her-/himself, the next of kin or the independent physician.

Where this trial is running

Zurich

• University Hospital Zurich, Department of Neurology — Zurich, Switzerland (Recruiting)

Study contacts

• Study coordinator: Laura Philine Westphal, MD

Conditions Stroke Outcome Prediction Supported by Deep Learning Algorithm; stroke outcome prediction; deep learning; AI in stroke