AI-ASSISTED ULTRASOUND IMAGING FOR EARLY DETECTION OF HEPATIC AND ABDOMINAL PATHOLOGIES: A DIAGNOSTIC ACCURACY STUDY IN RESOURCE-LIMITED RADIOLOGY SETTINGS OF PAKISTAN

Abstract

Artificial intelligence (AI)-assisted ultrasound may improve the detection of hepatic and abdominal pathologies, particularly in resource-limited settings with limited access to advanced imaging and specialist expertise. To evaluate the diagnostic accuracy of AI-assisted ultrasound for detecting hepatic and abdominal pathologies in resource-limited radiology settings in Pakistan. A prospective diagnostic accuracy study was conducted among 510 adults undergoing abdominal ultrasonography. AI-assisted interpretation was compared with conventional radiologist assessment against an appropriate reference standard. Sensitivity, specificity, predictive values, accuracy, AUC, and Cohen’s kappa were calculated. AI-assisted ultrasound demonstrated 91.5% sensitivity, 93.3% specificity, 85.4% PPV, 96.2% NPV, and 92.7% overall accuracy. Excellent diagnostic discrimination was observed (AUC = 0.94, 95% CI: 0.91–0.96), with high agreement with radiologists (κ = 0.82, 95% CI: 0.76–0.87). AI-assisted ultrasound showed promising diagnostic performance and potential as a second-reader and decision-support tool in resource-limited radiology settings. Prospective multicenter validation using locally representative datasets is warranted.