MV2I Swinnet: A Deep Learning Based Image Classification Network for Gastric Tissue Pathology Images

Timely and accurate diagnosis of gastric cancer is critical for effective treatment planning and improved patient outcomes. Histopathological examination remains the gold standard for diagnosis; however, existing image classification models often struggle to simultaneously capture fine-grained local tissue patterns and long-range contextual information, which may limit their diagnostic accuracy. To address this issue, we propose a hybrid image classification framework: MobileNetV2-Inception-Swin Transformer based Hybrid Network, termed MV2I SwinNet in short, where MV2I denotes the integration of MobileNetV2-based modules and an Inception-inspired multi-scale design. lightweight convolutional feature extraction with a Swin Transformer to jointly model local structural details and global contextual dependencies in gastric histopathological images. Specifically, redesigned MV2 Inception blocks are employed to enhance multi-scale feature representation with limited computational overhead, while a hierarchical Swin Transformer is used to capture global contextual information. Experimental results on the public GasHisSDB dataset demonstrate that MV2I SwinNet consistently outperforms classical convolutional neural networks, Transformer-based models, and existing hybrid approaches. These results indicate that the proposed method provides an effective and reliable solution for computer-aided gastric cancer diagnosis.