Automated Classification of Gastrointestinal Conditions from Capsule Endoscopy Images Using EfficientNetB2

Early diagnosis of gastrointestinal diseases is crucial for effective treatment and improved patient outcomes, as delays can significantly impact therapy efficacy. This study utilizes the Wireless Capsule Endoscopy (WCE) Curated Colon Disease Dataset from Kaggle, comprising 6000 high-resolution endoscopic images evenly distributed across four classes: Normal, Ulcerative Colitis, Polyp, and Esophagitis. The dataset was split into training, validation, and test sets with a 70:20:10 ratio. Utilizing the EfficientNetB2 model, enhanced with preprocessing and data augmentation techniques, the proposed approach achieves an impressive accuracy of 96.67% and a loss of 0.1078 on the test set. The Normal class serves as the baseline, while the other classes highlight critical pathological traits, underscoring the importance of early detection. This work demonstrates the potential of AI-driven diagnostic tools to automate and enhance medical imaging analysis, reducing the burden on healthcare professionals. The results advocate integrating such systems into clinical workflows to optimize diagnosis and treatment planning.