ExplainabilityTransparencyAccountabilityeess.IV
GastroViT: A Vision Transformer Based Ensemble Learning Approach for Gastrointestinal Disease Classification with Grad CAM & SHAP Visualization
Sumaiya Tabassum, Md. Faysal Ahamed, Hafsa Binte Kibria, Md. Nahiduzzaman, Julfikar Haider, Muhammad E. H. Chowdhury, Mohammad Tariqul Islam
September 30, 2025
arXiv.org
4 citations
Key Takeaways
Key Takeaways from GastroViT Paper
-
AI can improve disease detection accuracy: The GastroViT model achieved 91.98% accuracy in classifying 23 different gastrointestinal diseases from endoscopic images, demonstrating that AI systems can potentially assist doctors in identifying GI conditions more reliably than individual models alone.
-
Ensemble approaches enhance reliability: The paper shows that combining predictions from multiple AI models (MobileViT_XS and MobileViT_V2_200) outperformed any single model, suggesting that responsible AI deployment may benefit from using multiple models together rather than relying on a single system.
-
Explainability tools are essential for clinical use: The researchers incorporated Grad CAM and SHAP visualization techniques to make the model's decisions interpretable, addressing a critical need in healthcare AI—doctors need to understand why the system made a particular diagnosis recommendation before trusting it in clinical practice.
-
Efficiency matters for real-world deployment: The final model uses only 20 million parameters despite handling 23 disease classes, indicating that effective AI systems don't require massive computational resources, making them more accessible and practical for deployment in diverse healthcare settings.
-
Data challenges reflect real-world complexity: The model was trained on a highly imbalanced dataset without data augmentation, mirroring actual clinical conditions where some diseases are rarer than others—this realistic testing approach better demonstrates how the system would perform in practice.