Enhanced Brain Tumor Diagnosis through Multimodal Fusion and 3D Convolutional Neural Networks

Brain tumor detection has enhanced using a Multi-CNNs with Multimodal Information Fusion method has been proposed in this work. The proposed method leverages multiple modalities, like T1-weighted, T2-weighted, and FLAIR images, to create a fused representation of the input data. A deep neural network architecture using 3D Convolutional Neural Networks (3D-CNNs) is constructed to capture spatial and volumetric information present in medical images. The method also incorporates a real normalization layer and a weighted loss function to improve the learning process and prioritize accurate detection and localization of tumor lesions. Evaluating our method on a Kaggle dataset of 253 images (98 without tumor, 155 with tumor), we achieve superior results compared to existing approaches. Our proposed method attains an accuracy of 98.9%, sensitivity of 99.2%, specificity of 99.1%, recall of 98.7%, precision of 99.3%, and F1 score of 98.9%. Leveraging multimodal fusion, 3D-CNNs, and additional enhancements, our study contributes to more reliable brain tumor diagnosis, benefiting patients and healthcare providers.