NeuroAssist: A Hybrid SEMoNet Framework for Brain Tumour Classification From MRI With Patient‐Centred Conversational Support

Brain tumour classification (BTC) remains a critical challenge in clinical neuro‐oncology due to the high variability of tumour morphology and the need for rapid, accurate diagnosis to support timely intervention. Magnetic resonance imaging (MRI) is the clinical standard for noninvasive brain tumour assessment; however, manual interpretation is time‐intensive and susceptible to interobserver variability. To address these challenges, this study presents Neuro‐Oncology Artificial Intelligence Assisted Support and Interpretation System (NeuroAssist), a hybrid artificial intelligence (AI) framework built upon Squeeze‐and‐Excitation Mobile Network (SEMoNet) for automated BTC, coupled with a patient support chatbot to enhance interpretability and engagement. SEMoNet integrates Mobile Network Version 2 (MobileNetV2) for computationally efficient feature extraction with Squeeze‐and‐Excitation ResNet50 (SE‐ResNet50) for attention‐driven deep feature representation. The fusion of lightweight and attention‐enhanced architecture enables robust learning of both local and global tumour characteristics. The proposed framework classifies MRI scans into four clinically relevant categories: no tumour, glioma, meningioma and pituitary tumour. Experiments conducted on a curated dataset of 7200 MRI images demonstrate that SEMoNet achieves a classification accuracy of 93.2%, outperforming several established convolutional architectures in terms of accuracy, precision, recall, F1 score and AUC‐ROC. To enhance clinical usability and patient‐centred care, NeuroAssist incorporates a ChatGPT‐powered patient support chatbot (PSC), which translates model predictions into clear, patient‐friendly explanations and provides nondiagnostic, supportive guidance for personalized health planning. This integration bridges the gap between AI‐driven diagnostics and human‐centred communication. The results confirm that NeuroAssist offers a balanced combination of diagnostic accuracy, computational efficiency, interpretability and scalability, highlighting its potential for sustainable deployment in real‐world neuro‐oncology workflows and resource‐constrained clinical environments.