Vision-Language-Based Large-Scale Model for Emergency Screening

This is a single-center, prospective, stratified 1:1 parallel-group randomized superiority diagnostic trial at The First Hospital of Jilin University. Eligible emergency patients aged ≥18 who received routine thoraco-abdominal non-contrast CT were randomly assigned to two groups. The experimental group adopted a self-developed vision-language large model (powER) for auxiliary CT interpretation, while the control group received standard radiological reading without AI assistance. All participants completed standardized 60-day clinical follow-up to obtain clinical gold-standard outcomes. The primary objective is to compare the diagnostic performance of the AI model versus conventional radiology interpretation for multiple acute thoraco-abdominal emergency lesions.

Inclusion Criteria: * Adult patients aged 18 years or older presenting to emergency department requiring thoraco-abdominal CT examination for suspected acute intra-thoracic or intra-abdominal emergency conditions. * Clinically indicated emergency thoraco-abdominal CT scan ordered by emergency physicians. * Patient or legal representative provides written informed consent for participation in this prospective study. * Availability for clinical follow-up to establish reference-standard diagnosis. Exclusion Criteria: * Age under 18 years old. * Contraindications for contrast-enhanced CT examination (severe contrast allergy, uncorrectable severe renal dysfunction). * Inability to complete CT acquisition according to protocol due to critical clinical instability. * Refusal to participate or inability to obtain valid informed consent. * Prisoners or vulnerable populations who are inappropriate for enrollment.