A Unified Deep Learning Approach for Brain Tumor Detection Using Multimodal Radiological Imaging

Brain tumour classification is an essential radiological task requiring an accurate and efficient diagnostic system. Traditional single-modality imaging, such as MRI, CT, and PET, provides valuable but limited insights, often leading to inadequate diagnoses. This research proposes a unified deep learning approach integrating multimodal imaging and deep ensemble to enhance brain tumour detection. A standardized dataset was created to harmonize MRI, CT, and PET scans, addressing class imbalance and feature discrepancies. Three deep learning models, EfficientNetB3 (MRI Model), ResNet50 (CT Model), and DenseNet121 (PET Model), were trained individually, achieving high classification accuracies of 99.06%, 91.40%, and 98.89%, respectively. However, single models exhibited biases, limiting generalization. To overcome this, a deep ensemble approach was implemented, a fusion of the individual models’ predictions achieving a state-of-the-art performance with 99.66% accuracy, 99% precision, and 99% recall, outperforming the individual model. This study demonstrates that our proposed unified approach significantly enhances brain tumour classification and provides radiologists with AI-powered diagnostic systems.