An Efficient Gastrointestinal Cancer Classification Using Capsnet and Swin Transformer Model With Hybrid Classifiers

Gastrointestinal (GI) cancer remains a major challenge in global healthcare, necessitating the development of advanced diagnostic techniques that facilitate early detection and improve prognostic precision to enhance patient care. This study proposes a novelty framework to automatically detect GI cancer and classify it using Deep Learning (DL) models coupled with Machine Learning (ML) classifiers. The proposed framework integrates Capsule Networks for frame level extraction, Swin transformers for spatiotemporal analysis alongside hybrid classifiers including Support Vector Machine (SVM), Random Forest (RF), Extreme Gradient Boost (XG Boost) are employed to categorize GI disease stages such as dyed-lifted polyps, normal cecum, normal pylorus, polyps, and ulcerative colitis. By combining spatial and temporal information from video sequences, the integrated model enhances the classification accuracy. This approach addresses variability in video sequences by incorporating preprocessing techniques like data augmentation and frame extraction. The evaluation on GI cancer video dataset yielded a classification accuracy of 99.72 % accompanied by high F1-scores and AUC values significantly outperforming existing methods. This study demonstrates the potential of integrating DL architecture with robust ML algorithms for reliable, real-time application in endoscopic procedures.