ColoXAI-RecomNet: Explainable Recommender Framework for Colorectal Cancer Classification Using Integrated CNN...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Pathology

Background:

• Accurate classification of colorectal diseases from colonoscopy images is crucial.
• Existing methods often lack interpretability, hindering clinical adoption.
• Explainable AI (XAI) offers a path towards trustworthy diagnostic tools.

Purpose of the Study:

• To propose and evaluate a five-stage explainable framework (ColoXAI-RecomNet) for multi-class colorectal image classification.
• To enhance predictive accuracy and provide clinically interpretable decision support.
• To leverage deep learning and machine learning for robust colorectal disease detection.

Main Methods:

• Investigated three parallel hybrid Convolutional Neural Network (CNN) ensemble models (RDV-2025, IEM-2025, DRE-2025) as feature extractors.
• Employed XGBoost for feature refinement (Stage 2) and a multi-class Support Vector Machine (SVM) with an RBF kernel for classification (Stage 3).
• Integrated LIME for visual explanations (Stage 4) and converted outputs to a clinically interpretable recommender format (Stage 5).

Main Results:

• The DRE-2025 CNN ensemble model showed strong performance in initial feature extraction.
• The combined DRE-2025 + XGBoost + SVM model achieved a high accuracy of 98.60%.
• The framework demonstrated excellent performance metrics: Precision (98.75%), Recall (98.50%), F1-score (98.62%), and AUC (99.30%).

Conclusions:

• The proposed ColoXAI-RecomNet framework effectively classifies colorectal diseases from colonoscopy images with high accuracy and interpretability.
• The integration of CNNs, XGBoost, SVM, and LIME provides a powerful and explainable approach for medical image analysis.
• This framework holds significant potential for improving diagnostic accuracy and clinical decision-making in gastroenterology.