MRI Based Brain Tumor Segmentation with Ensemble Deep Learning

In order to analyze and track the development of the disease, automated segmentation of brain tumors from multimodal MR images is crucial. When compared to traditional, context-based computer vision techniques, deep learning algorithms perform better for semantic segmentation. Segmenting brain tumors with far more precision is now possible because to Convolutional Neural Networks, making them commonly used for biomedical image segmentation. This research introduces a notable yet simple combinatorial method that enhances the accuracy of predictions by merging two segmentation networks, specifically VGG16 and ResNet50. Before merging to generate a conclusive forecast, the two models underwent separate training on the BRATS dataset and were evaluated to generate segmentation maps. The segmentation of tumors differed significantly among the various sub-regions depicted in these maps. The suggested ensemble method (ES-BT) surpassed the most advanced architectures on the validation set, with an accuracy of 98.5%.