R-Net: A Reliable and Resource-Efficient CNN for Colorectal Cancer Detection with XAI Integration

Efficiency without sacrificing accuracy: R-Net achieved 99.37% accuracy for colorectal cancer detection while using significantly fewer computational resources than standard models like MobileNet (95.83%) and ResNet50 (96.94%), demonstrating that lightweight AI systems can maintain high performance—an important consideration for deploying AI in resource-limited healthcare settings.

Explainability is essential for clinical trust: The researchers integrated three explainability methods (SHAP, LIME, and Grad-CAM) to visualize which parts of medical images the AI model used to make decisions. This transparency is crucial for doctors to understand and trust AI recommendations before using them in patient care.

Responsible AI design reduces barriers to adoption: By creating a model that requires fewer computational resources and shorter training times, the researchers addressed practical obstacles to implementing AI in healthcare, making the technology more accessible to hospitals and clinics with limited IT infrastructure.

Comparative testing strengthens credibility: The paper tested R-Net against eight other established AI models on the same dataset, providing transparent benchmarking that helps stakeholders understand the model's relative strengths and limitations rather than relying on isolated claims.

Responsible AI requires multiple safeguards: The combination of efficiency, accuracy, and explainability demonstrates that responsible AI in healthcare isn't just about one factor—it requires balancing performance, interpretability, and practical feasibility to ensure the technology can be safely and equitably deployed.