Multimodal clinical integration in deep learning-enhanced CT imaging for automated detection of gallbladder stones

Gallbladder stones (cholelithiasis) represent one of the most common hepatobiliary conditions, affecting millions worldwide and contributing to substantial morbidity. Ultrasound is the primary imaging modality for detecting gallstones; however, its diagnostic accuracy decreases in obese patients and in cases of small or obscured stones. Computed tomography (CT) imaging provides higher anatomical resolution and is frequently employed when ultrasound findings are inconclusive or complications are suspected. Yet, CT interpretation is labor-intensive, operator-dependent, and subject to inter-observer variability. Recent advances in artificial intelligence (AI), particularly deep learning, have opened new avenues for automated and reproducible gallstone detection. This study proposes a novel framework that leverages convolutional neural networks (CNNs), attention mechanisms, and multimodal clinical integration for CT-based gallbladder stone diagnosis. The architecture incorporates EfficientNet-B3 as a backbone, a channel-spatial attention module to refine feature maps, and clinical data fusion to enhance diagnostic performance. On a curated dataset of annotated gallbladder CT images, the model achieved an area under the curve (AUC) of 0.96, sensitivity of 93.5%, and specificity of 91.2%, outperforming baseline models. Grad-CAM heatmaps confirmed that predictions were driven by regions corresponding to gallstones, enhancing clinical interpretability. These findings underscore the potential of AI-driven CT analysis to augment radiological workflows, improve diagnostic accuracy, and support clinical decision-making in gallbladder stone disease.