GLI-Net: A global and local interaction network for accurate classification of gastrointestinal diseases in endoscopic images

The accurate classification of gastrointestinal diseases from endoscopic images is essential for early detection and treatment. However, current methods face challenges in effectively integrating both global and local features, which limits their ability to capture both broad semantic information and subtle lesion details, ultimately affecting classification performance. To address this issue, this study introduces a novel deep learning framework, the Global and Local Interaction Network (GLI-Net). The GLI-Net consists of four main components: a Global Branch Module (GB) designed to extract global image features, a Local Branch Module (LB) focused on capturing detailed lesion features, an Information Exchange Module (LEM) that facilitates bidirectional information exchange and fusion between the global and local features, and an Adaptive Feature Fusion and Enhancement Module (AFE) aimed at optimizing the fused features. By integrating these modules, GLI-Net effectively captures and combines multi-level feature information, which improves both the accuracy and robustness of endoscopic image classification. Experiments conducted using the Kvasir and Hyper-Kvasir public datasets demonstrate that GLI-Net outperforms existing state-of-the-art models across several metrics, including accuracy, F1 score, precision, and recall. Additionally, ablation studies confirm the contribution of each module to the overall system performance. In summary, GLI-Net’s advanced feature extraction and fusion techniques significantly enhance medical endoscopic image classification, highlighting its potential for use in complex medical image analysis tasks.