Stroke
Conditions
Keywords
Artificial Intelligence, G-FAST
Brief summary
This study aims to validate the clinical performance of an artificial intelligence (AI)-based automatic assessment system for the G-FAST score. The core comparison is the consistency and accuracy between AI-generated G-FAST results and standardized manual G-FAST assessments performed by trained professionals. The goal is to provide a convenient, efficient, and objective tool for acute stroke screening and early identification, reduce the subjective variability of manual scoring, and optimize the pre-hospital and in-hospital stroke assessment workflow.
Interventions
None listed
Sponsors
Xuanwu Hospital, Beijing
Capital Medical University
People's Hospital of Beijing Daxing District
Beijing Tiantan Hospital
The First Hospital of Fangshan District,Beijing
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
1. Aged ≥ 18 years, of either sex. 2. Clinically diagnosed with stroke, and confirmed by cranial CT/MRI to have ischemic or hemorrhagic stroke. 3. Onset within 7 days. 4. Alert and oriented, able to cooperate with standardized video and audio data collection. 5. The patient or their legally authorized representative understands the study and voluntarily provides written informed consent (including consent for audio-visual data collection).
Exclusion criteria
1. Neurological deficits caused by non-stroke etiologies (e.g., brain tumor, traumatic brain injury, encephalitis). 2. Patients with impaired consciousness, severe cognitive dysfunction, or psychiatric disorders that prevent cooperation with video collection and scale assessment. 3. Patients with severe visual or hearing impairment, or global aphasia, who are unable to follow instructions. 4. Critically ill patients requiring immediate cardiopulmonary resuscitation or endotracheal intubation, making video and audio data collection impossible. 5. Patients with severe facial or limb deformities, or large-area dressings that severely interfere with camera data collection. 6. Patients with unilateral or bilateral upper limb amputation, severe deformity, unhealed fracture, joint fixation, or severe contracture.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Agreement between AI-generated and physician-scored G-FAST scale assessments | within 7 days of acute stroke onset | The agreement between the scores generated by the artificial intelligence (AI) system and the scores assigned by neurologists on G-FAST scale will be evaluated using weighted Kappa coefficients. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Agreement of AI System vs. Neurologists in Binary G-FAST Classification (Score ≥3 vs. <3) | within 7 days of acute stroke onset | Kappa coefficient will be calculated to evaluate the agreement between the artificial intelligence (AI) system and neurologist experts in the binary classification of G-FAST scale scores, defined as high risk (total score ≥3) vs. low risk (total score \<3) for large vessel occlusion stroke. |
| Bland-Altman Agreement Limit Analysis | within 7 days of acute stroke onset | A Bland-Altman plot will be constructed, with the difference between manual scores and AI scores on the vertical axis and the mean of the two scores on the horizontal axis. The limits of agreement (mean difference ± 1.96 × standard deviation) will be calculated. |
| Diagnostic performance analysis | within 7 days of acute stroke onset | Taking the manual score as the gold standard, a 2×2 contingency table was constructed to calculate the sensitivity, specificity, positive predictive value, negative predictive value, accuracy, and Youden index of the AI scoring system for stratifying the G-FAST score (LVO ≥3 vs. non-LVO \<3). The ROC curve was plotted and the AUC was calculated. |
Contacts
CONTACTQingfeng Ma, MD
CONTACTZixin Wang, MD Candidate
STUDY_CHAIRQingfeng Ma, MD
Xuanwu Hospital, Beijing
Outcome results
None listed