GUnet3++: UNet-based architecture for brain MRI segmentation

In recent years despite progress in medical imaging and machine learning achieving precise and automated brain tumor segmentation remains challenging due to the diverse characteristics of tumors and inconsistencies in MRI scans. In this study, we employ Unet-based architecture to address these challenges and improve diagnostic outcome. GUnet3++ architecture integrates dense connectivity, The model leverages multi-scale feature extraction and attention mechanisms to improve its capacity for capturing intricate anatomical structures. Additionally, data augmentation methods are utilized to increase the variety within the limited training dataset, contributing to further enhancements in model accuracy and reliability and efficacy of GUNet3++ is demonstrated through extensive experiments, achieving a training accuracy of 99.52%, and Dice Coefficient is 99.18%. and the validation accuracy on (TCGA,TCIA) dataset it reached 98.26% and Dice Coefficient is 98.45%. These coming outcomes reflect substantial improvements in the segmentation precision and generalization capabilities over traditional methods, which underscores GUNet3++’s potential to revolutionize medical image analysis, paving the way for more precise diagnoses and improved treatment planning in neuroimaging.