ResUNet++:a comprehensive improved UNet++ framework for volumetric semantic segmentation of brain tumor MR images
Published in Evolving Systems, 2024
This study, utilizing data from BraTS challenges (2019-2021), aims to enhance automatic segmentation for high-grade (HGG) and low-grade gliomas (LGG). Instead of only modifying network architecture, the research emphasizes data preprocessing, augmentation, training, and testing strategies. The proposed ResUNet + + framework, an improved 3D encoder-decoder model based on ResNet50 and incorporating 3D dense convolutional blocks and convolutional transpose layers, outperforms existing models in segmenting WT, TC, and ET in HGG and LGG. Evaluated with five performance metrics, ResUNet + + shows superior effectiveness compared to state-of-the-art models.