Accurate detection of gallbladder cancer (GBC) from ultrasound images is challenging due to subtle early signs and inter-observer variability. Inspired by the success of transformer architectures and attention mechanisms, we propose GallNet - a novel CNN-Transformer hybrid model with a Dual Attention Module for automated GBC detection. The Dual Attention Module consists of Global and Local Attention Branches to localize regions of interest and extract multi-scale spatial features. A pyramid transformer encoder captures global dependencies and contextual information at different resolutions. The attention-enhanced feature maps are fed into the transformer encoder to generate contextualized representations for classification. We train and evaluate GallNet on the GBCU dataset, achieving a level of performance that supports early-stage automated detection of gallbladder cancer. Our results demonstrate the potential of attention-based hybrid architectures for accurate GBC detection, with applicability to other medical image analysis tasks.