Glioblastoma treatment planning commonly relies on the contrast-enhancing tumor volume identified on MRI. Infiltrating tumor cells, however, frequently extend beyond this volume, which impacts outcome. Accurately detecting and segmenting these infiltrative regions is therefore critical for optimizing surgical and radiotherapy planning. Histopathological studies indicate that the [18F]FET PET-positive volume may offer more complete coverage of the tumor and its infiltration, but is not always available. We combine [18F]FET PET labels with conventional MRI labels to enable a more comprehensive segmentation of the glioblastoma tumor volume and investigate whether such segmentation can be achieved using MRI alone. To achieve this, we introduced two novel hybrid tumor definitions: the hybrid tumor core and the hybrid whole tumor, derived from a combination of MRI and [18F]FET PET labels. We subsequently trained a nnU-Net-based segmentation model to predict these labels using combined MRI and PET data and implemented two strategies to handle missing PET data. The hybrid segmentation achieved DSC and NSD of 0.889 and 0.682 for HybTC, and 0.929 and 0.835 for HybWT, respectively, which is on par with separate segmentation. Results demonstrated that while the proposed strategies could partly compensate for the missing PET images, the performance remained significantly lower compared to the segmentation including PET. Since consistently predicting infiltration as defined by [18F]FET PET from MRI alone currently remains unfeasible, the findings support the continued inclusion of PET imaging in glioblastoma management.
De Sutter, S, Dirks, I, Raes, L, Geens, W, Everaert, H, Bourgeois, S, Duerinck, J & Vandemeulebroucke, J 2026, Segmentation of Glioblastoma Infiltration Using Hybrid Labels from MRI and [18F]FET PET. in Z Cui, I Rekik, H-IL Suk, X Ouyang, K Sun & S Wang (eds), Machine Learning in Medical Imaging - 16th International Workshop, MLMI 2025, Held in Conjunction with MICCAI 2025, Proceedings. Lecture Notes in Computer Science, vol. 16241 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 411-420, 16th International Workshop on Machine Learning in Medical Imaging, MLMI 2025 was held in conjunction with the 28th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2025, Daejeon, Korea, Republic of, 23/09/25. https://doi.org/10.1007/978-3-032-09513-8_40
De Sutter, S., Dirks, I., Raes, L., Geens, W., Everaert, H., Bourgeois, S., Duerinck, J., & Vandemeulebroucke, J. (2026). Segmentation of Glioblastoma Infiltration Using Hybrid Labels from MRI and [18F]FET PET. In Z. Cui, I. Rekik, H.-IL. Suk, X. Ouyang, K. Sun, & S. Wang (Eds.), Machine Learning in Medical Imaging - 16th International Workshop, MLMI 2025, Held in Conjunction with MICCAI 2025, Proceedings (pp. 411-420). (Lecture Notes in Computer Science; Vol. 16241 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-032-09513-8_40
@inproceedings{212feba4cc2e4e5eacbb5479542a6d2b,
title = "Segmentation of Glioblastoma Infiltration Using Hybrid Labels from MRI and [18F]FET PET",
abstract = "Glioblastoma treatment planning commonly relies on the contrast-enhancing tumor volume identified on MRI. Infiltrating tumor cells, however, frequently extend beyond this volume, which impacts outcome. Accurately detecting and segmenting these infiltrative regions is therefore critical for optimizing surgical and radiotherapy planning. Histopathological studies indicate that the [18F]FET PET-positive volume may offer more complete coverage of the tumor and its infiltration, but is not always available. We combine [18F]FET PET labels with conventional MRI labels to enable a more comprehensive segmentation of the glioblastoma tumor volume and investigate whether such segmentation can be achieved using MRI alone. To achieve this, we introduced two novel hybrid tumor definitions: the hybrid tumor core and the hybrid whole tumor, derived from a combination of MRI and [18F]FET PET labels. We subsequently trained a nnU-Net-based segmentation model to predict these labels using combined MRI and PET data and implemented two strategies to handle missing PET data. The hybrid segmentation achieved DSC and NSD of 0.889 and 0.682 for HybTC, and 0.929 and 0.835 for HybWT, respectively, which is on par with separate segmentation. Results demonstrated that while the proposed strategies could partly compensate for the missing PET images, the performance remained significantly lower compared to the segmentation including PET. Since consistently predicting infiltration as defined by [18F]FET PET from MRI alone currently remains unfeasible, the findings support the continued inclusion of PET imaging in glioblastoma management.",
keywords = "Glioblastoma, Magnetic resonance imaging, Positron emission tomography, Segmentation",
author = "\{De Sutter\}, Selene and Ine Dirks and Laurens Raes and Wietse Geens and Hendrik Everaert and Sophie Bourgeois and Johnny Duerinck and Jef Vandemeulebroucke",
note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.; 16th International Workshop on Machine Learning in Medical Imaging, MLMI 2025 was held in conjunction with the 28th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2025 ; Conference date: 23-09-2025 Through 23-09-2025",
year = "2026",
doi = "10.1007/978-3-032-09513-8\_40",
language = "English",
isbn = "9783032095121",
series = "Lecture Notes in Computer Science",
publisher = "Springer Science and Business Media Deutschland GmbH",
pages = "411--420",
editor = "Zhiming Cui and Islem Rekik and Heung-IL Suk and Xi Ouyang and Kaicong Sun and Sheng Wang",
booktitle = "Machine Learning in Medical Imaging - 16th International Workshop, MLMI 2025, Held in Conjunction with MICCAI 2025, Proceedings",
address = "Germany",
}