Enhanced Gastrointestinal Disease Classification Using Hybrid Deep Learning on Multi-Class Endoscopic Images

Abstract

Gastrointestinal (GI) diseases are a major global health concern, with early and accurate diagnosis being crucial for effective treatment and improved patient outcomes. Automated computer-aided diagnosis (CAD) systems powered by deep learning have shown great potential in enhancing diagnostic accuracy and efficiency. In this paper, we propose a novel approach for automated gastrointestinal disease detection using hybrid deep learning architectures. Leveraging the multiclass Kvasir dataset, which includes eight distinct GI disease classes, we integrate convolutional neural networks (CNNs) with Swin Transformer models to improve classification accuracy, feature extraction, and diagnostic efficiency. These models combine the spatial feature extraction capabilities of CNNs with the global contextual learning ability of Swin Transformers to enhance disease classification performance. The results demonstrate significant performance improvements, with hybrid models achieving superior feature representation and reducing misclassification rates. This study not only underscores the potential of hybrid deep learning frameworks for computer-aided diagnosis but also contributes to the advancement of intelligent endoscopic image analysis.