Gastroenterological disease detection using transformer-based medical imaging for sustainable healthcare

Early detection of gastroenterological diseases significantly improves patient outcomes and reduces late-stage diagnostic burden, yet traditional CNN models show limitations in capturing complex patterns within medical imaging datasets, prompting investigation into transformer architectures like Vision Transformer (ViT). Application of the ViT technology in detecting gastroenterological diseases with the help of medical imaging has not been fully explored, despite the promising capabilities. In this paper, the effectiveness of the ViT-B16 structure for the identification of gastrointestinal abnormalities is considered using a combined dataset of Curated Colon Dataset and HyperKvasir Dataset (10,000 images across four classes), and compared with established methodologies. Our experimental results showed that ViT-B16 performed better when compared to alternative approaches; it achieved 99.5% classification accuracy compared to 99.1% by EfficientNetB5 and 97.1% by EfficientNetB2, with other supportive performance metrics including precision (99.4%), recall (99.4%), and F1-score (99.4%), AUC values ranged from 0.99 to 1.00 across all classes, reflecting very strong discriminatory power regarding disease classification tasks. These suggest that ViT-B16 has great potential for medical diagnosis applications, especially classification tasks in healthcare, where evidence-based decision-making and model interpretability are key considerations. The model also supports sustainable healthcare through computational efficiency and reduced diagnostic burden. However, there are several challenges that have not been addressed, including addressing ethical concerns about diagnostics, improving diagnostic accuracy for underrepresented disease classes, and validating the model across diverse clinical settings, which are essential directions for future research to continue developing gastroenterological disease-detecting techniques.