TY - GEN
T1 - Automated Gross Tumor Volume Segmentation in Meningioma Using Squeeze and Excitation Residual U-Net for Enhanced Radiotherapy Planning
AU - Matlala, Boitumelo
AU - van der Haar, Dustin
AU - Vadapalli, Hima
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.
PY - 2026
Y1 - 2026
N2 - Accurate segmentation of the gross tumor volume (GTV) in meningioma is essential for radiation therapy planning. This study evaluates several deep learning-based approaches, including U-Net, Attention U-Net, nnU-Net, and SE-ResUNet (Squeeze-and-Excitation Residual U-Net), for automated GTV segmentation on post-contrast MRI scans. Incorporating squeeze-and-excitation blocks and attention mechanisms improved performance, with SE-ResUNet achieving the highest average Dice score across datasets (80.5%), followed by Attention U-Net (79.3%), nnU-Net (77.6%), and U-Net (73.6%). These results demonstrate that SE-ResUNet provides the most accurate and robust segmentation, while all advanced models outperform the conventional U-Net. Automated segmentation with these methods can reduce clinical workload and variability, offering a reliable tool for planning radiation therapy in meningioma.
AB - Accurate segmentation of the gross tumor volume (GTV) in meningioma is essential for radiation therapy planning. This study evaluates several deep learning-based approaches, including U-Net, Attention U-Net, nnU-Net, and SE-ResUNet (Squeeze-and-Excitation Residual U-Net), for automated GTV segmentation on post-contrast MRI scans. Incorporating squeeze-and-excitation blocks and attention mechanisms improved performance, with SE-ResUNet achieving the highest average Dice score across datasets (80.5%), followed by Attention U-Net (79.3%), nnU-Net (77.6%), and U-Net (73.6%). These results demonstrate that SE-ResUNet provides the most accurate and robust segmentation, while all advanced models outperform the conventional U-Net. Automated segmentation with these methods can reduce clinical workload and variability, offering a reliable tool for planning radiation therapy in meningioma.
KW - Attention U-Net
KW - Gross Tumor Volume
KW - Meningioma
KW - Residual U-Net
KW - SE-ResUNet
KW - Segmentation
KW - nnU-Net
UR - https://www.scopus.com/pages/publications/105017225389
U2 - 10.1007/978-3-032-00652-3_5
DO - 10.1007/978-3-032-00652-3_5
M3 - Conference contribution
AN - SCOPUS:105017225389
SN - 9783032006516
T3 - Lecture Notes in Computer Science
SP - 57
EP - 67
BT - Artificial Intelligence in Healthcare - 2nd International Conference, AIiH 2025, Proceedings
A2 - Cafolla, Daniele
A2 - Rittman, Timothy
A2 - Ni, Hao
PB - Springer Science and Business Media Deutschland GmbH
T2 - 2nd International Conference on Artificial Intelligence on Healthcare, AIiH 2025
Y2 - 8 September 2025 through 10 September 2025
ER -