Acute Ischemic Stroke, Branch Atheromatous Disease, Cerebral Infarction
Conditions
Keywords
Branch atheromatous disease, Acute ischemic stroke, Artificial intelligence, Machine learning, Deep learning, Diagnostic model, Multimodal imaging, Magnetic resonance imaging, Lenticulostriate artery, Paramedian pontine artery
Brief summary
Branch atheromatous disease (BAD)-related stroke is an important subtype of acute ischemic stroke involving penetrating arteries and is associated with early neurological deterioration. Early recognition and standardized diagnosis remain challenging in routine clinical practice because clinical symptoms are often non-specific and the diagnosis requires integrated clinical and imaging assessment. This multicenter prospective observational study will collect demographic, clinical, laboratory, electrocardiographic, ultrasound, and multimodal neuroimaging data from adults with acute ischemic stroke within 1 week of symptom onset. Participants will receive routine clinical care determined by their treating physicians; no treatment or management strategy will be assigned by the study protocol. An independent central clinical-imaging adjudication committee will classify participants as BAD-related stroke or non-BAD acute ischemic stroke according to predefined diagnostic criteria. The study aims to develop and externally validate artificial intelligence-assisted screening and diagnostic models for BAD-related stroke and to evaluate their discrimination, calibration, and potential clinical utility.
Detailed description
Branch atheromatous disease (BAD)-related stroke has been increasingly recognized as a clinically meaningful subtype of acute ischemic stroke. It typically presents as a single subcortical infarction in the territory of penetrating arteries, especially the lenticulostriate arteries and paramedian pontine arteries. Because BAD-related stroke is not well captured by conventional etiologic classification systems and because its early diagnosis requires standardized interpretation of clinical and neuroimaging features, delayed or inconsistent recognition may limit subsequent precision-management research. This study is designed as a multicenter, prospective, observational cohort study. Eligible adults with acute ischemic stroke will be enrolled within 1 week after symptom onset or last known well time. Multisource data will be collected, including demographics, vascular risk factors, baseline neurological assessments, laboratory tests, electrocardiography, carotid/cardiac ultrasound, routine brain MRI, intracranial vascular imaging by MRA/CTA/DSA when available, high-resolution vessel wall MRI when available, ASL perfusion imaging when available, acute-phase treatment information, early neurological deterioration, and 90-day functional outcomes. The study will include two predefined diagnostic cohorts: participants with BAD-related stroke and participants with non-BAD acute ischemic stroke. BAD-related stroke will be adjudicated by an independent central clinical-imaging committee according to predefined imaging and etiologic criteria. The reference diagnosis will be based on baseline and follow-up clinical information, neuroimaging, vascular imaging, cardiac evaluation, and 90-day follow-up information when applicable. Artificial intelligence-assisted models will be developed and validated to support early screening and diagnostic classification of BAD-related stroke. The early screening model will use non-imaging or routinely available acute-phase clinical information, whereas the diagnostic model will integrate multisource clinical and imaging information. Model performance will be evaluated in an external validation cohort using discrimination, sensitivity, specificity, accuracy, calibration, and decision curve analysis. The study protocol does not assign any therapeutic intervention, diagnostic procedure beyond routine or protocol-specified observational assessments, or clinical management strategy. All treatments will be determined by the treating physicians according to local practice and applicable guidelines.
Interventions
DIAGNOSTIC_TESTMultisource clinical-imaging artificial intelligence diagnostic assessment
The diagnostic assessment consists of artificial intelligence-assisted analysis of routinely collected clinical, laboratory, cardiovascular, and multimodal neuroimaging data to estimate the probability of BAD-related stroke. The model output will be compared with an independent central clinical-imaging reference diagnosis. The model will not determine treatment assignment in this observational study.
Sponsors
Peking Union Medical College Hospital
Institute of Automation, Chinese Academy of Sciences
Beijing Zhongke Ruiyi Information Technology Co., Ltd.
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to 80 Years
Healthy volunteers
No
Inclusion criteria
* Age 18 to 80 years. * Diagnosis of acute ischemic stroke. * Time from symptom onset to enrollment ≤ 1 week; if the onset time is unknown, time from last known well to enrollment ≤ 1 week. * Availability of required baseline clinical and neuroimaging assessments according to the study protocol. * Written informed consent provided by the participant or legally authorized representative. Participants will be classified into the BAD-related stroke cohort if they meet all predefined BAD-related stroke diagnostic criteria, including: * A single isolated deep subcortical infarct on diffusion-weighted imaging. * The presumed culprit perforating artery is the lenticulostriate artery or the paramedian pontine artery. * For lenticulostriate artery territory infarction: a comma-shaped lesion extending from inferior to superior direction on coronal DWI or involvement of ≥3 axial DWI slices with 5-7 mm slice thickness. * For paramedian pontine artery territory infarction: a lesion extending from the deep pons to the ventral surface of the pons on axial DWI. * No ≥50% stenosis of the corresponding parent artery, confirmed by MRA, CTA, or DSA. Participants with acute ischemic stroke who do not meet BAD-related stroke criteria will be classified into the non-BAD acute ischemic stroke cohort.
Exclusion criteria
General
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Overall diagnostic accuracy of the AI-assisted model for identifying BAD-related stroke | Baseline acute phase, after completion of required clinical and neuroimaging assessments, within 7 days after symptom onset or last known well | Overall diagnostic accuracy will be calculated as the proportion of participants correctly classified as BAD-related stroke or non-BAD acute ischemic stroke by the AI-assisted diagnostic model, using the independent central clinical-imaging adjudication as the reference standard. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Overall accuracy of the early screening model for identifying possible BAD-related stroke | At enrollment, using non-imaging clinical data available within 24 hours after admission | Overall accuracy will be calculated as the proportion of participants correctly classified as possible BAD-related stroke or non-BAD acute ischemic stroke by the early screening model at a prespecified decision threshold. The model will use only prespecified non-imaging clinical data available at enrollment or within 24 hours after admission. The reference standard will be independent central clinical-imaging adjudication according to predefined diagnostic criteria. |
Countries
China
Contacts
CONTACTShengde Li, MD
PRINCIPAL_INVESTIGATORJun Ni, MD
Peking Union Medical College Hospital
Outcome results
None listed