Ultrasound Image-based Gallbladder Cancer Classification with Graph Neural Networks

Abstract

Gallbladder cancer (GBC) is a highly aggressive malignancy often diagnosed at advanced stages due to a lack of early detection methods. This study proposes a novel approach for GBC classification using ultrasound images, combining convolutional neural networks (CNNs) with graph neural networks (GNNs). The dataset, consisting of annotated ultrasound images, was preprocessed to extract regions of interest (ROIs) and organized into class-specific datasets. Initially, CNN models such as VGG-16 and ResNet-18 were employed for feature extraction and integrated with graph convolutional networks (GCNs) to explore spatial relationships. However, the performance remained suboptimal, with the best accuracy from ResNet-18 + GCN at 76%. To enhance accuracy, hybrid models combining CNNs with graph attention networks (GATs) were implemented, resulting in modest improvements, with ResNet- 18 + GAT achieving 83% accuracy. Significant advancements were achieved by integrating graph isomorphism networks (GINs) with CNNs, specifically VGG-16 and ResNet-18. The VGG-16 + GIN model achieved an accuracy of 93%, while the ResNet- 18 + GIN model demonstrated the best classification performance with an accuracy of 98.48%. This approach highlights the potential of graph-based learning in medical image analysis, providing an innovative solution for early and precise GBC detection. The findings pave the way for future research in applying hybrid GNN frameworks to medical diagnostics