Explainable Unified CAD Framework for Brain Tumor Segmentation and Prediction Using Grad-CAM and SHAP
DOI:
https://doi.org/10.64149/J.Carcinog.24.7s.263-270Keywords:
Brain Tumor, Computer-Aided Diagnosis, 3D U-Net, CNN-LSTM, Explainable AI, Grad-CAM, SHAPAbstract
This study presents an explainable computer-aided diagnosis (CAD) framework for brain tumor detection and grading using multi-modal MRI. Building on prior advances in segmentation and classification, the proposed system integrates a 3D U-Net for voxel-wise tumor segmentation with a CNN-LSTM-based classifier for glioma grading. To improve clinical interpretability, we embed explainable AI techniques—Grad-CAM for spatial attention mapping and SHAP for feature attribution analysis. Evaluated on the BraTS 2020 dataset, the framework achieves expected performance of Dice scores of 0.88 (WT), 0.83 (TC), 0.80 (ET), and classification accuracy of 94% with an AUC of 0.96. Grad-CAM heatmaps align with tumor subregions (IoU ≈ 0.70), while SHAP ranks tumor volume, GLCM entropy, and mean intensity among the most influential predictors. The results demonstrate that integrating XAI with unified CAD enhances both trustworthiness and diagnostic accuracy, addressing a key barrier to clinical adoption.
