Explainable AI-Based Graph Convolutional Network for Brain Tumor Diagnosis and Survival Risk Stratification with Cross-Dataset Validation

Brain tumor diagnosis requires accurate automated analysis of MRI data for reliable clinical decision-making. This paper suggests a Brain Tumor Classification and Survival Analysis framework based on a Multi-Attribute Graph Convolution Network (BTC-SA-MGCN). The model integrates GMMCKF denoising, DFCC segmentation, and RTQWT-based texture feature extraction for robust learning of tumor characteristics. Experiments were conducted using the BraTS 2018 dataset for training and validation, with additional cross-dataset evaluation on BraTS 2020 and BraTS 2021 to assess generalization. Under both 80:20 hold-out and 5-fold cross-validation strategies, the proposed model achieved up to 99.05% accuracy, 99.09% precision, 98.99% sensitivity, and 99.04% F1-score, outperforming recent CNN-, EfficientNet-, and ensemble-based approaches. Survival analysis further demonstrated strong prognostic capability with a C-index of 0.82 and significant risk stratification (p < 0.01). The results indicate that graph-based relational learning combined with multi-stage preprocessing provides a robust and clinically relevant framework for brain tumor classification and survival prediction.