NeuroDiag AI: Automated Brain Tumor Segmentation and Multi-Class Classification

Abstract

Brain tumors represent one of the most life-threatening neurological disorders, with early and accurate diagnosis being critical for improved patient outcomes. Manual segmentation and classification of brain MRI scans is time-intensive, error-prone, and highly dependent on specialist expertise. This paper presents NeuroDiag AI, an automated deep learning framework for simultaneous brain tumor segmentation and multi-class classification from MRI scans. The proposed system integrates a U-Net-based segmentation backbone with a pre-trained ResNet-50 encoder and an attention-guided classification branch to jointly identify tumor regions and classify them into four categories: Glioma, Meningioma, Pituitary Tumor, and No Tumor. Trained on a curated dataset of 7,023 MRI images (T1, T2, FLAIR sequences) augmented to 24,000 instances, NeuroDiag AI achieves a classification accuracy of 97.8%, precision of 96.9%, recall of 96.5%, and F1-Score of 96.7%, outperforming existing baselines including VGG-16 (87.2%), ResNet-50 standalone (89.0%), U-Net (90.5%), and 3D CNN (91.2%). The proposed model demonstrates clinical-grade performance with a mean inference time of 2.1 seconds per scan, making it suitable for integration into real-time clinical diagnostic workflows.