Standardized Ultrafast Transcranial Ultrasound Acquisition in Acute Stroke: The HYPER-AI-SCAN Pilot Study With Exploratory Artificial Intelligence Analysis.
Authors
Affiliations (7)
Affiliations (7)
- School of Medicine and Health Sciences Universitat Autònoma de Barcelona (UAB) Barcelona Spain.
- Stroke Unit, Department of Neurology Vall d'Hebron University Hospital Barcelona Spain.
- Stroke Research Group Vall d'Hebron Research Institute (VHIR) Barcelona Spain.
- Department of Radiology Stanford University School of Medicine Stanford CA USA.
- Department of Neurology Centro Hospitalar e Universitário de Coimbra Coimbra Portugal.
- Department of Neurology IRCCS Humanitas Research Hospital Rozzano Milan Italy.
- Department of Biomedical Sciences Humanitas University Pieve Emanuele Milan Italy.
Abstract
Intracerebral hemorrhage requires rapid mechanism-specific treatment, yet early stroke triage relies on clinical severity scales that do not reliably distinguish hemorrhage from ischemic stroke. We assessed the feasibility of a standardized ultrafast transcranial ultrasound protocol designed to support subsequent artificial intelligence (AI)-based intracerebral hemorrhage detection. HYPER-AI-SCAN (Hyper-Acute AI-Assisted Sonographic Cerebral Hemorrhage Assessment Network) was a single-center prospective feasibility study with embedded exploratory AI analysis. Between March and December 2025, 102 adults with acute computed tomography (CT)-confirmed stroke were prospectively enrolled according to trained operator availability. Primary feasibility outcomes were protocol completion, operator-rated adequate transtemporal windows, and acquisition time. In patients with retrievable paired baseline CT, exploratory AI analyses compared CT-only, ultrasound-only, and cross-modal CT-guided ultrasound models using patient-level area under the receiver operating characteristic curve, sensitivity, and specificity. Two patients were excluded after acquisition because insonation depth was <19 cm, yielding 100 evaluable patients (33 intracerebral hemorrhage; 67 ischemic stroke). The protocol was completed in all evaluable patients; adequate transtemporal windows were identified in 87%, and median acquisition time was 22 seconds (interquartile range, 13-31). Paired baseline CT images were available in 91 patients. Mean area under the receiver operating characteristic curve was 0.86 (SD, 0.07) for the CT-only model, 0.85 (SD, 0.14) for the ultrasound-only model, and 0.92 (SD, 0.07) for the cross-modal model. Standardized ultrafast transcranial ultrasound acquisition was feasible at the bedside in acute stroke and generated brief cine-loops suitable for exploratory AI development. Multicenter external validation is required before diagnostic or triage application. Clinicaltrials.gov. Identifier: NCT07319013.