3D U-TFA: A Deep Convolutional Neural Network for Automatic Segmentation of Glioblastoma

Shang Wu,Zhencheng Chen,Peng Sun
DOI: https://doi.org/10.1016/j.bspc.2024.106829
IF: 5.1
2025-01-01
Biomedical Signal Processing and Control
Abstract:Glioblastoma segmentation faces numerous challenges, including varying information provided by multiple MRI sequences and the blurred tumor boundaries, especially in infiltrative areas. This study proposes a groundbreaking deep convolutional neural network, 3D U-TFA (3D U-Transformer with FEA Attention), aimed at enhancing segmentation accuracy and efficiency to provide more reliable support for glioblastoma diagnosis and treatment. The decoder of this method can access multi-level feature maps, thereby improving segmentation precision. At the bottom of the encoder, we introduce a 3D Transformer module to efficiently capture and enhance global features. Additionally, a 3D attention mechanism incorporating both channel and spatial attention is added before each skip connection, significantly improving feature selection accuracy. The encoder employs an innovative block design, further optimizing the depth and breadth of feature extraction. This method demonstrates excellent performance in handling complex tumor boundaries and multimodal MRI data, providing strong technical support for the precise diagnosis and treatment of glioblastoma. The 3D U-TFA was evaluated on the Brain Tumor Segmentation (BraTS) Challenge datasets for the years 2019, 2020, and 2021. In BraTS2019, the Dice coefficients for ET (enhancing tumor), TC (tumor core), and WT (whole tumor) were 0.853, 0.911, and 0.928, respectively. In BraTS2020, the Dice coefficients for ET, TC, and WT were 0.832, 0.926, and 0.926, respectively. In BraTS2021, the Dice coefficients for ET, TC, and WT were 0.863, 0.901, and 0.930, respectively. Experimental results indicate that 3D U-TFA achieves significant performance improvements in automatic glioblastoma segmentation, demonstrating its potential and effectiveness in practical applications.
What problem does this paper attempt to address?