Multimodal MRI Segmentation-Based Structural Feature Extraction for Glioma Survival Prediction
GBP Proceedings Series · 16.05.2026 · 28 · 89-98
DOI: 10.70088/z05vrw97 ↗Kurz gefasst
Fachartikel in der Zeitschrift GBP Proceedings Series, online erschienen am 16.05.2026. Verfasst von Shenda Qu.
Abstract (Originalsprache)
Glioma represents a highly heterogeneous primary brain tumor, and accurate prognosis assessment remains critically important for personalized treatment planning and effective clinical decision-making. Although multimodal magnetic resonance imaging (MRI)-based tumor segmentation has achieved substantial progress in recent years, existing studies predominantly focus on segmentation accuracy alone. Consequently, the intrinsic prognostic value of segmentation-derived structural information remains insufficiently explored in current literature. To address this critical gap, this study proposes a comprehensive glioma survival prediction framework based on multimodal MRI segmentation and advanced structural feature extraction. A robust 3D U-Net architecture was utilized to accurately segment the whole tumor, tumor core, and enhancing tumor regions from the BraTS 2020 multimodal MRI dataset. Based on these precise segmentation results, diverse structural descriptors—including volume ratios, shape features, spatial relationship features, and heterogeneity-related characteristics—were systematically extracted and employed for three-class survival prediction. Experimental results demonstrated that the segmentation model achieved impressive Dice scores of 0.901, 0.821, and 0.764 for the whole tumor, tumor core, and enhancing tumor, respectively. In the context of survival prediction, the extracted structural features significantly outperformed traditional clinical and radiomic features, achieving an accuracy of 0.653, an F1-score of 0.629, and an Area Under the Curve (AUC) of 0.734. Furthermore, the fused-feature setting improved the predictive performance to 0.684 accuracy, 0.661 F1-score, and 0.763 AUC. These compelling findings indicate that segmentation-derived structural features provide highly useful and interpretable prognostic information, serving as a practical foundation for imaging-based survival prediction in glioma patients.
Zitieren
APA
Qu, S. (2026). Multimodal MRI Segmentation-Based Structural Feature Extraction for Glioma Survival Prediction. GBP Proceedings Series, 28, 89-98. https://doi.org/10.70088/z05vrw97
Vancouver
Qu S. Multimodal MRI Segmentation-Based Structural Feature Extraction for Glioma Survival Prediction. GBP Proceedings Series. 2026;28:89-98. doi:10.70088/z05vrw97
BibTeX
@article{Qu2026,
author = {Qu, Shenda},
title = {Multimodal MRI Segmentation-Based Structural Feature Extraction for Glioma Survival Prediction},
journal = {GBP Proceedings Series},
year = {2026},
volume = {28},
pages = {89--98},
doi = {10.70088/z05vrw97}
}
EUVIMED