A Novel Lightweight Explainable Deep Learning Framework for Automated Brain Tumor Segmentation and Classification

Abstract

This research work presents an explainable lightweight deep learning approach to automate brain tumor extraction and classification using MR images. This method incorporates lightweight feature extractor sharing, joint segmentation and classification branches, and explainability using Grad-CAM. The proposed approach will aim to identify the relevant features of the tumor in an accurate way with minimal computational cost. Explainable artificial intelligence (XAI) techniques will be incorporated to identify and highlight critical regions within the tumor, thereby enhancing the interpretability of the model’s predictions. The performance of the method would be assessed in terms of the evaluation metrics used for segmentation and classification problems. The proposed framework achieved 98.34% classification accuracy, 98.02% precision, 97.72% recall,98.55% specificity and 97.32 F1-score indicating improved classification performance on the BRISC 2025 dataset. This framework also achieves high precision and recall with 98.6% specificity, 95.3% DSC, and 91.0% IoU, demonstrating accurate tumor segmentation with effective identification of non-tumor regions. The contributions of each module to the research have been further verified via ablation studies. So, this framework is anticipated to yield an accurate, effective, and interpretable method for computer-based detection of brain tumors.