Multimodal Deep Learning Based Brain Tumor Segmentation Using CT And MRI Scans

Abstract

Brain tumor segmentation is crucial in medical imaging for early diagnosis and treatment. This paper introduces a multimodal architecture for brain tumor segmentation that merges computed tomography (CT) and magnetic resonance imaging (MRI) data. Initially, models are trained independently for each image modality by completing preprocessing processes such as normalization and grayscale conversion. Subsequently, a Convolutional Neural Network (CNN)-based architecture is designed to effectively combine the processed CT and MRI data. Our training results show that the CT unimodal model achieved 92% accuracy, the MRI unimodal model achieved 73% accuracy, and the multi-modal model achieved an outstanding 95% accuracy. In the evaluation of model performance based on the confusion matrix, accuracy, recall, precision, and F1 score, the multimodal model exhibited greater efficacy compared to the unimodal models. These findings indicate that integrating CT and MRI scans into a multimodal model substantially boosts performance and yields significant benefits in important medical tasks like tumor segmentation, particularly in the field of medical imaging. © 2025 IEEE.