To achieve accurate segmentation of glioma subregions from multiparametric MRI, which is essential for improved radiotherapy planning and assessment.
Key Findings:
DPEA-Net achieved mean Dice scores of 90.43% for whole tumor (WT) and 85.56% for tumor core (TC) on BraTS 2019 and 2020 validation sets, with scores of 81.89% for enhancing tumor (ET).
The model has a computational footprint of only 17.48 GFLOPs, enabling sub-2-second inference on standard clinical hardware.
A 1.5-fold TC weighting strategy enhances segmentation of the tumor core.
Interpretation:
Limitations:
Conclusion:
DPEA-Net is a tool for automated glioma subregion delineation.
