TRANSFORMING GALLBLADDER CANCER SCREENING THROUGH DEEP LEARNING–ENABLED ULTRASOUND IMAGING: A PROSPECTIVE DIAGNOSTIC STUDY AT AYUB TEACHING HOSPITAL, ABBOTTABAD, PAKISTAN

Abstract

Gallbladder cancer (GBC) remains one of the most aggressive hepatobiliary malignancies, largely due to late-stage diagnosis and the absence of reliable, non-invasive screening strategies, particularly in low- and middle-income countries. Conventional ultrasound imaging is widely used as a first-line diagnostic modality; however, its effectiveness is highly operator-dependent and limited in detecting early-stage malignant changes. This study aims to evaluate the clinical utility of deep learning–enabled ultrasound imaging for the early screening and diagnosis of gallbladder cancer in a real-world tertiary care setting. A prospective diagnostic study was conducted at Ayub Teaching Hospital, Abbottabad, Pakistan, involving patients presenting with suspected gallbladder pathology. Ultrasound images were acquired using standardized imaging protocols and annotated by experienced radiologists. A deep learning framework based on convolutional neural networks was developed to automatically analyze ultrasound images and classify gallbladder lesions into malignant and non-malignant categories. The model was trained, validated, and tested using institution-specific datasets to ensure clinical relevance and robustness. Diagnostic performance was assessed using accuracy, sensitivity, specificity, precision, F1-score, and area under the receiver operating characteristic curve (AUC), with histopathology and expert consensus serving as reference standards. The proposed deep learning model demonstrated strong diagnostic performance, achieving high sensitivity and specificity in differentiating gallbladder cancer from benign conditions such as cholelithiasis and chronic cholecystitis. Notably, the AI-assisted system showed improved detection of subtle morphological features that are often overlooked in conventional ultrasound interpretation. Comparative analysis revealed that the deep learning–enabled approach outperformed routine ultrasound assessment, particularly in early-stage disease identification. The model also exhibited consistent performance across varying image qualities, highlighting its potential to reduce inter-observer variability and diagnostic subjectivity. This study provides prospective clinical evidence supporting the integration of deep learning–powered ultrasound imaging into gallbladder cancer screening workflows. The findings suggest that AI-assisted ultrasound can enhance diagnostic accuracy, facilitate early detection, and support clinical decision-making in resource-constrained healthcare environments. Adoption of such intelligent diagnostic systems may significantly improve patient outcomes through timely intervention and personalized management. Future work will focus on multi-center validation, explainable AI integration, and real-time deployment to further advance AI-driven gallbladder cancer screening.